Undergraduate Final Year Project Proposals: 2014 - 2015 PID Supervisor 1400914 Aldhaher,S. (with Mitcheson,P.D.) Quadcopters with infinite flight time Room Description Quadcopters can be both a fun toy and a serious piece of kit for surveillance. Their main limitation is payload, and this in turn limits flight time due to poor battery performance. Wireless power transfer should be capable of overcoming these problems and we wish to create a large volume of EM field to fly and charge the copter in. The project will require knowledge of electromagnetics, devices and circuit design skills, an interest in high frequency power electronics design and ability to do analysis both on paper and in Matlab. Circuit simulation skills are also a must. 4T Due to commercial activity surrounding this work (in which the student will be able to engage though the Imperial spin out, Drayson Wireless), the student will be required to sign an NDA and IP agreement. Please talk to me for more details. 1400870 Angeli,D. 1111 Maximization of Power Extraction in Wind-farms with Wake Interactions 3E4D4T3I4J 1400871 Angeli,D. Robust Economic Model Predictive Control 1400872 Angeli,D. Integration of multiple appliances in demand-side frequency regulation 1400878 Angeli,D. Predictive control of multiple storage devices and energy sources 11 October 2014 1111 3E4D4T3I4J 1111 3E4D4T3I4J 1111 3E4D4T3I4J Different wind speeds entail the selection of appropriate operating points for wind turbines in order to maximize the efficiency of power extraction. In a wind-farm, however, due to proximity of wind turbines, the power extracted at one turbine may affect the wind experienced by nearby turbines, especially along the direction of the wind. Therefore, overall power maximization is not necessarily achieved simply by making sure that each turbine works at the point of maximum efficiency. The goal of the project is to build a model of a wind-farm to include wake interactions and exploit the model in simulation to test non-model based optimization algorithms for global maximization of power extraction. Economic Model Predictive Control is a model-based technique for profit maximization in the operation of plants or systems in general. It requires exact knowledge of a plant's governing equations and, under such assumptions, may provide near-optimal performance. The goal of this project is to investigate the impact of model uncertainty in economic revenues, and design possible solutions to cope with it. Simulations will be performed in matlab. Operating the power grid entails regulation of frequency within a small range centered around the nominal value of 50Hz. This is normally achieved from the generation side, by closely matching demand and supply of power. In recent years, an alternative solution to this problem has been to consider the possibility of using smart appliances to contribute automatically to the balance by deferring their operation in time. Appliances suitable for this task include refrigerators, air conditioning devices, battery chargers and others. The goal of this project is to test an algorithm based on stochastic hybrid automata for the integration of multiple appliances in the task of frequency regulation. Simulations will be carried out in matlab. Renewable energy sources (whose availability cannot always be relied on) or fluctuating energy prices make the possibility of using storage devices economically appealing. In this project we will investigate the optimal management (based on Economic Model Predictive Control) of multiple storage devices (such as batteries and ultracapacitors) in conjunction with renewable generation and uncertain power demand. The project will entail understanding the basics of Model Predictive Control, developing a quantitative mathematical model of the system to be managed and implementing in MATLAB simulations of the Model Predictive Control Algorithm. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 1 of 63 PID Supervisor Room Description 1400855 Angeli,D. (with Astolfi,A.) 1111 We consider a novel method for clustering points in the plane. The proposed algorithm is based on the notions of clustering function and level lines; the clusters are identified as the level sets corresponding to a reference value of the clustering function. The core idea is to regard the clustering function as a Hamiltonian function and to determine the level lines as the trajectories of the associated Hamiltonian system. The method is useful in pattern identification problems and weather forecast problems. Hamiltonian Data clustering 1400854 Astolfi,A. Optimal Model Reduction 1400879 1400997 1110A The problem of sharing the control of a system between a human operator and an automatic controller is considered. The goal of the project is to design a shared control algorithm for a simple mobile robot, or similar autonomous vehicle, and test it via simulations. 1110A 3E4D4T3I4J Clemow,P.R. Lab Scale HVDC Breaker Design and Build The model reduction problem consists in determining a "simple" model of a "complex" system which however describes accurately its behaviour. Aim of this project is to compute optimal reduced order models from a class of models which interpolate the given system. Optimality in terms of frequncy response indicators or time-domain indicators will be studied. 3E4D4T3I4J Astolfi,A. (with Angeli,D.) Hamiltonian Data clustering 1110A 4T Astolfi,A. Shared control 1400855 3E4D4T3I4J Mauric 4T We consider a novel method for clustering points in the plane. The proposed algorithm is based on the notions of clustering function and level lines; the clusters are identified as the level sets corresponding to a reference value of the clustering function. The core idea is to regard the clustering function as a Hamiltonian function and to determine the level lines as the trajectories of the associated Hamiltonian system. The method is useful in pattern identification problems and weather forecast problems. High Voltage DC (HVDC) is a technology that enables practical and economical transfer of bulk energy over large distances, especially undersea. As such, HVDC links are often proposed for interconnecting grids (such as the UK and France) and for connecting large offshore wind farms to their host grids (the Borwin project in Germany is a good example of this). Presently the links are most commonly point to point transfers, however there is a lot of interest in multi-terminal HVDC (MTDC) setups as this can enable extra functionality with fewer converters. An example of a proposed setup is the creation of a North Sea Supergrid connecting the wind farms of the various North Sea countries to each other. This could allow the aggregation of wind power across a large area (reducing much of the uncertainty involved) and the use of other technologies, such as hydro power in Norway, to counter the remaining wind variations). A key stumbling block for MTDC setups is the lack of availability of a DC breaker for disconnecting lines that have a fault. There are a number of solutions proposed to solve this problem but none have been installed in a real link as yet. This project will be to design a lab-scale HVDC breaker for testing on the MTDC setup in the power lab. This project would suit a student with good practical design and build experience. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 2 of 63 PID Supervisor Room Description 1400862 Evangelou,S. 1108B The objective of this project is to develop a simulation model that will describe accurately the dynamics of a hybrid vehicle. The task involves the modelling of the individual subsystems of the vehicle and its powertrain. A key issue is the accurate description of the interdependencies between the subsystems which will allow proper integration to the full model. The modelling problem is naturally multidisciplinary. Parametric modelling, control and optimisation of hybrid vehicles - battery degradation control 4D4T4J Key characteristics of the model are descriptions for the hybrid-electric drive powertrain, regenerative braking, vehicle dynamics, aerodynamics, driver and component- and vehicle-level controls. It is believed that existing platforms do not yet include sufficiently detailed component descriptions to allow the model to be used for the purposes to which it is built. These include to help identify which aspects of the design need to be modified for optimal overall performance and also to improve performance by designing good control schemes. This project will aim to develop flexible subsystem models in the form of equations derived from first principles, rather than models that rely on static look-up tables and performance maps, or simple power-request dynamics that do not allow for accurate scaling or for the adjustment of critical design parameters. A number of projects ran successfully in previous years within this framework but further work is required especially on the development of control schemes that take into account battery degradation. 1400863 Evangelou,S. Parametric modelling, control and optimisation of hybrid vehicles - converter control 1108B 4D4T4J The objective of this project is to develop a simulation model that will describe accurately the dynamics of a hybrid vehicle. The task involves the modelling of the individual subsystems of the vehicle and its powertrain. A key issue is the accurate description of the interdependencies between the subsystems which will allow proper integration to the full model. The modelling problem is naturally multidisciplinary. Key characteristics of the model are descriptions for the hybrid-electric drive powertrain, regenerative braking, vehicle dynamics, aerodynamics, driver and component- and vehicle-level controls. It is believed that existing platforms do not yet include sufficiently detailed component descriptions to allow the model to be used for the purposes to which it is built. These include to help identify which aspects of the design need to be modified for optimal overall performance and also to improve performance by designing good control schemes. This project will aim to develop flexible subsystem models in the form of equations derived from first principles, rather than models that rely on static look-up tables and performance maps, or simple power-request dynamics that do not allow for accurate scaling or for the adjustment of critical design parameters. A number of projects ran successfully in previous years within this framework but further work is required especially on the control of the DC link that connects the various electrical components of the powertrain of a series hybrid vehicle - generator (rectifier), battery (dc-dc converter) and motor (inverter). 1400864 Evangelou,S. Active variable geometry suspension for road vehicles 1108B 4T Variable geometry suspension systems for vehicles operate on the principle of varying by an actuator the geometry of passive force-producing elements, conventionally springs and dampers. They are of interest in terms of their potential to achieve performances close to ideal active suspensions but without giving away many of the benefits enjoyed by passive systems. Thus variable geometry implies low actuation forces, low power requirements, low energy consumption, fail-safe operation and potential to use much of the existing passive technology. This project will look into the application of fuzzy logic and neural network control for a variable geometry active suspension of a high performance sports car, to improve vehicle comfort and handling. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 3 of 63 PID Supervisor Room Description 1400865 Evangelou,S. 1108B The prediction of rider intention is very much essential for the supervisory energy control system in a modern small twowheeled vehicle. As these type of vehicles are predominantly used for personal commuting in urban scenarios, it is essential to recognise the effect of uncertainties caused by traffic and road conditions. Such conditions can influence the intentions of the rider, therefore these need to be identified from at least partial knowledge of the vehicle dynamic state, to manage optimally the powertrain energy flow. Prediction of rider intention for a two-wheeled vehicle based on limited on-board information 4D4T4J Research available in the literature has considered various approaches for predicting the driving load: a) Combined cascade neural networks with node decoupled extended Kalman filtering b) Discrete cosine transform (DCT) together with support vector machines (SVM) c) Combining road information from a static map with historical driving data d) Model based approach using discrete time Markov process In the first part of project, the methods mentioned above for driving load predictions will be implemented and compared for accuracy and real time implementation. In the second part of the project, data mining techniques will be explored to propose driving load prediction functions that are simpler to implement. The knowledge gained in the initial part of the project will be applied in developing this predictive function. 1400868 Evangelou,S. (with Silversides,R.) Torque vectoring and regeneration control for the Racing Green electric vehicle (EV3) 1400869 Evangelou,S. (with Silversides,R.) Active flap control for the Racing Green electric vehicle (EV3) 1400880 Jaimoukha,I.M. Online fault detection 1108B 4D4T4J 1108B 4D4T4J 1113 4D4T4J This project will look into the torque vectoring and regeneration control for the new Racing Green electric vehicle (EV3). The vehicle is planned to be a rear wheel drive electric vehicle with independent drive to each rear wheel. The project will look at the control strategies (and develop software/simulations) to demonstrate the potential benefits of using torque vectoring and regeneration control. This project will look into the active flap control for the new Racing Green electric vehicle (EV3). The vehicle will have an extensive aero package, therefore it is hoped to introduce some active flap control to improve handling on corners and reduce drag on straights. It is expect that this will be mainly a control problem, possibly using feedback from the suspension travel/downforce sensors in the chassis. This project aims to investigate fault detection schemes for linear dynamic systems subject to disturbances. The aim of the project is to develop and compare both online and offline optimization algorithms for the design of a condition monitoring system that attenuates the disturbances and is able to reliably detect faults. The student is expected to have a good background in optimization. All programs will be developed in Matlab and Simulink. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 4 of 63 PID Supervisor Room Description 1400881 Jaimoukha,I.M. 1113 Observers are used to estimate system states from system input and output measurements. Such observers are, however, vulnerable to potential faults in the measured signals. Design of fault tolerant observers 4D4T The aim of this project is to modify current observer design techniques to ensure a minimum level of performance under all probable fault scenarios and to optimize performance for the fault-free situation. The main challenge to overcome is the exponential increase in the number of fault scenarios as the number potentially faulty sensors increases. The student is expected to have a good background in control theory. All programs will be developed in Matlab and Simulink. 1400882 Jaimoukha,I.M. Invariant sets for model predictive control 1113 4D4T4J Model predictive control is an advanced control scheme that is model-based, applied in real time, incorporates optimal control and can handel hard inout and state constraints. A drawback of the method is the need to solve an online optimization problem to compute the control law at every sampling instant. The aim of the project is to compute an invariant set such that, if the state of the system is inside this set (so that on the one hand, the system is near the desired equilibrium, and on the other, it is guaranteed to remain in the set) then a constant precomputed state-feedback control law is applied and there is no need to carry out any extra online optimizations. The project uses robust control theory techniques, linear matrix inequalities and game theoretic approaches. All algorithms will be developed using Matlab and Simulink. 1400883 Jaimoukha,I.M. Robust global optimization problems in finance 1113 4D4T4J Many problems in finance, such as portfolio optimization, can be posed as optimization problems involving games between several players. These problems can be formulated as minimax optimization problems with mixed constraints and special classes of global optimization techniques are used for their solution. A main drawback of these techniques is their inability to handle uncertainties in the problem data due to lack of information available to the different players. This project aims to investigate a class of approximation techniques for problems involving uncertainties where the constraints of the problem are relaxed resulting in linear matrix inequality optimization problems which give upper or lower bounds on the exact solution. These techniques will be tested for accuracy and ease of solution on case studies from the field of finance. The student is expected to have a good background in linear algebra. All programs will be developed and tested in MATLAB. 1400844 Junyent-Ferre,A. Single-phase load symmetrising active filter design 1400845 Junyent-Ferre,A. Analysis of an energy-efficient computation system with renewable-supply-following capabilities 11 October 2014 1107 3E4D4T3I4J 1107 3E4D4T3I4J Most large electrical loads are designed to be connected to three-phase AC; however, some specific applications (AC railway, trolley bus, etc) involve large single-phase loads. The classic solution to make a single-phase load draw symmetrical current from a three-phase system is the Steinmetz circuit. This circuit is purely passive; therefore, it can't adapt to varying loads and voltage disturbances. Active thyristor-controlled variations of this circuit exist but their application is still limited to certain types of loads and they generate low-order harmonics. This project will study IGBT-based alternatives for microgrid applications. The project is recommended for students interested in FACTS devices, power quality and power electronics. The growth in demand for online computation services has raised concerns about the environmental impact of data centres and super-computers. This project will explore the concept of generation-following computation, where the power demand of a computer will be adjusted in real-time following the availability of renewable energy resources in order to make the system autonomous while minimising its energy storage requirements. The project will combine analytical and practical work, including the experimental identification of the energy requirements of a computing platform and the implementation of the proposed system. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 5 of 63 PID Supervisor Room Description 1400847 Junyent-Ferre,A. 1107 Offshore wind power is one of the most promising technologies for large scale renewable energy generation. Current state-ofthe-art wind turbines in the multi-megawatt range enable building wind farms with power ratings comparable to those of large conventional power plants. However, fundamental differences exist between wind farms and power plants when it comes to their dynamics. The future power system will require greater coordination between wind turbines to deliver high performance support services to the power system. This project will tackle the problem of performing transient simulations of full wind farms by decoupling different elements of their dynamics to simulate them in parallel. The project is recommended for students interested in power system simulation and wind power. Distributed simulation of large offshore wind farms 1400921 3E4D4T3I4J Junyent-Ferre,A. Watt-scale multi-purpose inverter design 1107 4T Power electronics are a technology enabler for renewable power generation and long distance power transmission. The use of power electronics can lead to a more flexible, more reliable and more efficient electric system. However their large scale integration in the network still poses a number of technological challenges. The aim of this project is to design a multi-purpose three-phase inverter to be used as a small emulator for power converter grid integration prototyping purposes. The target design will have the following features: it will operate at low voltages (safety), it will have a moderate cost, it will be programmable from Matlab Simulink and it will have communication capabilities via Ethernet bus. The candidate student must have a high motivation to learn practical design of power electronic converters. 1400925 Junyent-Ferre,A. Three-phase inverter control with imbalanced DC operation capabilities 1107 3E4T Three-phase AC to DC voltage-source converters (VSCs) are one of the basic building blocks found in a range of high-power applications, e.g. photovoltaic inverters, industrial motor drives, etc. In current practice, either the AC side or the DC side of the converter are grounded. This enables to balance the voltages around earth potential by passive means with ease. However, new applications of DC transmission at high, medium and low voltage have recently raised interest in converter topologies which enable to actively control the DC voltage balance to enable degraded operation when one conductor is faulty and asymmetric connection of loads. This project will focus on the design of the controller for a three-phase inverter for a DC transmission application with DC imbalance operation capabilities. This project is suitable for students with high motivation to learn about power converter modelling, simulation and control design. 1400889 Kuenzel,S.T.G.I. (with Pal,B.C.) Validation of linearization method used by Matlab vs. Algebraic solution technique 11 October 2014 1107 3E4D4T3I4J Motivation The project investigates the accuracy of linearization methods implemented in Matlab/Simulink. In the past perturbation methods were used by programs such as Matlab/Simulink to linearize the algebraic differential equations of a dynamic system. Since these methods suffer from errors due to signal scaling, recent Matlab/Simulink versions used pre-programmed Jacobians to circumvent the problem. Objective The aim is to compare and validate the linearized solution of a system with synchronous and wind generation and AC and DC transmission using Matlab/Simulink with the algebraic solution of the same algebraic differential equations. Methodology This project will be able to build on previous work, which should enable a good progress. The system will be modelled in Matlab/Simulink and linearized using Matlab’s linmod. The same system is represented by differential and algebraic equations in Maple, which can be linearized analytically. Learning outcome The project will use Matlab/Simulink and Maple, which are very useful tools for any student to get familiar with. The student will gain a good understanding of different techniques used for linearization. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 6 of 63 PID Supervisor Room Description 1400890 Kuenzel,S.T.G.I. (with Pal,B.C.) 1107 Motivation The layout of a wind farm is a non-trivial task, which can be affected by terrain, wind conditions and reliability considerations. Optimal wind farm cabling in particular is a challenging task, since longer cable connections increase the cost and losses, while wind turbines require to be spaced out to provide a good power output. Objective This project will optimize the wind farm cabling taking different constraint factors into account. Methodology The optimization problem should be formalized with care. The optimal cable layout is a topological problem, which will provide quite different solutions, depending on which optimization objectives are selected and which assumptions are made. The optimization algorithm itself is another design choice, which should be made carefully. Learning outcome The student will expand their Matlab knowledge, in particular regarding available optimization tools. Further, knowledge will be gained regarding the wake effect and wind farm layout, in particular optimal cabling. Optimal wind farm layout, with the main emphasis on optimal cabling 1400915 3E4D4T Mitcheson,P.D. Semiconductor device health monitoring 1112 3E4D4T3I4J Power electronics is generally a reliable technology, but in some circumstances, reliability needs to improve because the cost to fix things or schedule maintenance is very high - eg wind turbines, HVDC converters. This project will aim to implement an intelligent gate drive that monitors device status for degradation such as increased onstate resistance, changes in capacitance etc, which may be indicative of device failure. Good circuit design skills and an understanding of devices and instrumentation are required. 1400914 Mitcheson,P.D. (with Aldhaher,S.) Quadcopters with infinite flight time 1112 4T Quadcopters can be both a fun toy and a serious piece of kit for surveillance. Their main limitation is payload, and this in turn limits flight time due to poor battery performance. Wireless power transfer should be capable of overcoming these problems and we wish to create a large volume of EM field to fly and charge the copter in. The project will require knowledge of electromagnetics, devices and circuit design skills, an interest in high frequency power electronics design and ability to do analysis both on paper and in Matlab. Circuit simulation skills are also a must. Due to commercial activity surrounding this work (in which the student will be able to engage though the Imperial spin out, Drayson Wireless), the student will be required to sign an NDA and IP agreement. Please talk to me for more details. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 7 of 63 PID Supervisor Room Description 1400932 Mitcheson,P.D. (with Clerckx,B.) 1112 Wireless sensor network (WSN) has attracted a lot of attention in various machine-to-machine (M2M) and Internet of Things (IoT) applications such as smart home networking, smart metering, and healthcare due to its low deployment costs coupled with a high potential to sense and collect object-oriented data through a large number of cheap wireless sensor nodes. One of the main challenges in WSN is that nodes are resource-constrained, mainly due to battery limitation. To enable autonomous wireless devices, the project will investigate the feasibility of wireless power transfer. While receiver design for wireless power transfer has been relatively well researched, the design of suitable waveforms is much less known. In the project, the student will investigate how the waveform design influences the performance of wireless power transfer and identify suitable waveforms. The student will get the opportunity to progressively get familiar with the latest research in wireless communications, RF design and power transfer, discuss and imagine the basic concepts of next generation wireless power networks. Waveform Design for Wireless Power Transfer 3E4D4T3I4J Skills required: strong interests in communication/signal processing and RF design, algorithm development, analysis, simulations, RF measurement, matlab and PSpice programming The following references are helpful: • A. S. Boaventura and N. B. Carvalho, “Maximizing DC Power in Energy Harvesting Circuits Using Multisine Excitation,” 2011 IEEE MTT-S International Microwave Symposium Digest (MTT). • M. Pinuela, P. Mitcheson and S. Lucyszyn, “Ambient RF energy harvesting in urban and semi-urban environments” IEEE Trans. on microwave theory and techniques, vol 61, no 7, july 2013. • J. Park and B. Clerckx, “Joint Wireless Information and Energy Transfer in a Two-User MIMO Interference Channel,” IEEE Trans. Wireless Commun., vol. 12, no. 8, pp. 4210–4221, Aug. 2013. 1400913 Mitcheson,P.D. (with Yates,D.C.) Charge your phone in your pocket 1112 4T Wireless power transfer is now available via the Qi standard for charging phones, but only over very short distances on charging pads. The project will look at the possibility of charging a phone in your pocket as you sit as your office desk. The project will require knowledge of electromagnetics, devices and circuit design skills, an interest in high frequency power electronics design and ability to do analysis both on paper and in Matlab. Circuit simulation skills are also a must. Due to commercial activity surrounding this work (in which the student will be able to engage though the Imperial spin out, Drayson Wireless), the student will be required to sign an NDA and IP agreement. Please talk to me for more details. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 8 of 63 PID Supervisor Room 1400908 Pal,B.C. 1104 Description Motivation A Stochastic Method for the Operation of Active Distribution Networks 3E4D4T The penetration of distributed generators (DGs) in distribution networks could improve the system efficiency, reliability and security. Nevertheless, they could impact the system voltage, power quality, fault level and interact with the operation of capacitors and voltage regulators. The intermittency and variability of renewable DGs (e.g., wind and PV) impose challenges when operating distribution systems. Objective A stochastic method for the operation of active distribution network under active management schemes such as coordinated voltage control and adaptive power factor control is proposed in order to evaluate the active and reactive power of renewable and non-renewable DGs considering 1) uncertainties related to solar irradiance and load demand, 2) different operational status of DGs (multi-configurations), and 3) capability curve of PV inverters. Total costs are composed of active and reactive cost of renewable and non-renewable DGs and active power losses cost and the cost of imported/exported power from/to the grid. Methodology The simulation software can be MATLAB or GAMS. The first step is gathering load data, renewable and non-renewable DGs and a real distribution network data. The second step is formulating the problem and implementing it in one of the above mentioned softwares. Learning outcome Programming in MATLAB or GAMS can be a great learning experience and understanding the challenges of distribution systems operation with integration of renewable DGs. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 9 of 63 PID Supervisor Room Description 1400909 Pal,B.C. 1104 Motivation A Deterministic Method for the Planning of Active Distribution Networks 3E4D4T Under a passive network scheme, distributed generators (DGs) are usually operated with fixed power factors and on-load-tapchangers (OLTCs) are limited to only regulate the secondary voltages while the emerging active network management schemes have proved to be beneficial for distribution network operators (DNOs) compared to passive network management. Owing to the increasing trend toward integrating more DGs in transmission and distribution networks, the planning scope of maximizing DG penetration is taking the priority of DNOs in most countries. Objective A deterministic multi-objective multi-period multi-configuration method in active distribution networks under active management schemes such as coordinated voltage control and adaptive power factor control is proposed. The multi-objective method simultaneously minimizes the total cost and the total active power losses of the transmission lines from the point of view of DNOs in order to assess the active and reactive power of renewable and non-renewable DGs and active power losses of the lines considering 1) variability of load demand and wind generation profiles (multi-period scenarios), 2) different operational status of DGs (multi-configurations), 3) demand response (demand control) and 4) capability curve of doubly fed induction generator (DFIG) of wind turbines. Total costs are composed of active and reactive cost of renewable and non-renewable DGs, cost of load demand control and active power losses cost. Methodology The method can be implemented in MATLAB or GAMS. The first stage is gathering the data of load, renewable and nonrenewable DGs and a real distribution network. Second stage is formulating the problem and implementing it in the above mentioned software. Learning outcome Programming in MATLAB or GAMS can be a great learning experience and understanding the challenges of distribution systems planning with integration of renewable DGs. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 10 of 63 PID Supervisor Room Description 1400911 Pal,B.C. 1104 Motivation Voltage instability occurs in power systems when the system is unable to keep an acceptable voltage profile under increasing load demand and/or configuration changes. In today’s distribution systems (DSs), characterised by an increased number of distributed generation connections and a sharply rising load demand due to economic and environmental pressures, the operating conditions of the system more frequently can get closer to the voltage stability boundaries. Since DSs typically experience frequent changes in the load pattern, which are beyond the control of the system operator (SO), the problem of voltage stability is quite significant and challenging. Objective From a theoretical point of view, the project aims at systemising the problem of voltage stability in the DS, through a review of the different aspects and issues at both primary circuit level (sub-station) and secondary service (delivery point) levels. From an applicative (practical) point of view, the aim of this project is investigating the effect of realistic load variations on distribution voltage stability by means of a parametric static analysis. The study can be conducted by means of a load flow (LF) model developed in MATLAB platform. Methodology The first step will consist in a review of the voltage stability problem at the DS level, highlighting and differentiating aspects and concerns at both primary circuit level (sub-station) and secondary service (delivery point) levels. With respect to the LF application: a) load variations will be simulated by use of load patterns or “load directions”, defined as the rate of load increase from the base load at each bus; b) based on the network characteristics, voltage limits at each bus will be defined; c) repeated LF tests will be conducted in MATLAB on the selected DS at different load levels, to assess the trend of the voltage at each bus compared to the stability conditions. Results of the study will be presented in a 3D form. Learning outcome The student will gain a complete understanding of the voltage stability issue at the DS level. The student will practice with MATLAB modelling and analysis platform. At the end of the project, the student is expected to master the LF tool. Static analysis of distribution system voltage sensitivity to load variations 11 October 2014 3E4D4T Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 11 of 63 PID Supervisor Room Description 1400912 Pal,B.C. 1104 Motivation In an isolated (thus, off-grid) microgrid (MG), load demand variations and limited power capacity of generators can cause power balancing problems and alter the regime of circulation of reactive power in the network. Such a system can benefit of the application of power electronics (PE) interfaced battery energy storage (BES), which can support power balancing, as well as provide reactive power compensation through the PE interface, for power flow management and voltage control. Objective The scope of this project is investigating how the regime of circulation of reactive power changes within an isolated MG due to the more or less significant contribution of BES. The study will rely on a systemisation of the BES technology and of its application in a MG. A load flow (LF) model will be developed in MATLAB platform, in order to carry out a parametric analysis of the BES-controlled network flows and voltages within the off-grid MG under different load conditions. Methodology The first step will consist in a background study of the BES technology and of the basic paradigms of MG systems, as well as on a review of the PE part. The voltage/reactive power analysis will be conducted using a complete LF model of the MG developed in MATLAB. The voltage and reactive power flow in the network will be assessed under different load conditions, with the PE interface connected to the BES able to adjust the reactive power in the MG, either injecting reactive power when the voltage drops from the nominal value or absorbing reactive power when the voltage rises. Learning outcome At the end of the project, the student is expected to have a complete understanding of the LF tool and to master its modelling in MATLAB. The student will acquire familiarity with the BES technology and its application, with different MG paradigms, and with the representation of the applied technologies and components (generation, load, BES, PE interface) in a LF model. Static analysis of battery energy storage-based Volt/VAr control in an isolated micro grid under different load conditions 1400889 Pal,B.C. (with Kuenzel,S.T.G.I.) Validation of linearization method used by Matlab vs. Algebraic solution technique 11 October 2014 3E4D4T 1104 3E4D4T3I4J Motivation The project investigates the accuracy of linearization methods implemented in Matlab/Simulink. In the past perturbation methods were used by programs such as Matlab/Simulink to linearize the algebraic differential equations of a dynamic system. Since these methods suffer from errors due to signal scaling, recent Matlab/Simulink versions used pre-programmed Jacobians to circumvent the problem. Objective The aim is to compare and validate the linearized solution of a system with synchronous and wind generation and AC and DC transmission using Matlab/Simulink with the algebraic solution of the same algebraic differential equations. Methodology This project will be able to build on previous work, which should enable a good progress. The system will be modelled in Matlab/Simulink and linearized using Matlab’s linmod. The same system is represented by differential and algebraic equations in Maple, which can be linearized analytically. Learning outcome The project will use Matlab/Simulink and Maple, which are very useful tools for any student to get familiar with. The student will gain a good understanding of different techniques used for linearization. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 12 of 63 PID Supervisor Room Description 1400890 Pal,B.C. (with Kuenzel,S.T.G.I.) 1104 Motivation The layout of a wind farm is a non-trivial task, which can be affected by terrain, wind conditions and reliability considerations. Optimal wind farm cabling in particular is a challenging task, since longer cable connections increase the cost and losses, while wind turbines require to be spaced out to provide a good power output. Objective This project will optimize the wind farm cabling taking different constraint factors into account. Methodology The optimization problem should be formalized with care. The optimal cable layout is a topological problem, which will provide quite different solutions, depending on which optimization objectives are selected and which assumptions are made. The optimization algorithm itself is another design choice, which should be made carefully. Learning outcome The student will expand their Matlab knowledge, in particular regarding available optimization tools. Further, knowledge will be gained regarding the wake effect and wind farm layout, in particular optimal cabling. Optimal wind farm layout, with the main emphasis on optimal cabling 1400891 Pal,B.C. (with Singh,A.) Dynamic parameter estimation of a wind farm connected to a large scale power system 3E4D4T 1104 3E4D4T Motivation: An increased penetration of new renewable sources of energy into power systems requires that the reliability and stability of the system is not compromised by this penetration. Solar and wind energy form a chief component of these sources; but as these sources are highly intermittent, non-deterministic and stochastic in nature, it is very difficult to either analyze or control the negative effects that they can have on system reliability and stability. A fundamental step towards addressing this issue is dynamically identifying the equivalent parameters of a wind farm or a solar park, which can then be utilized for further estimation and control purposes. This project studies this identification problem, focusing specifically on a wind farm. Objective: The aim of this project is to study and develop algorithms and methods for dynamic parameter estimation of a wind farm (comprising hundreds of wind turbines) which is connected to a large scale power system. These dynamic parameters can be the equivalent inertia, damping coefficient or reactance of the wind farm. The student can also suggest and estimate some other relevant dynamic parameters. Methodology: The student would develop large scale models for both a generic power system and a wind farm, integrate the two models and perform simulations to get voltage and power measurements from the wind farm for various scenarios of wind input. These measurements would then be utilized for parameter estimation utilizing old/novel stochastic filtering and stochastic signal processing techniques. The estimated parameters will be compared with actual parameters for the all the simulated scenarios and conclusions will be reported. Matlab and DigSilent are suggested software for model development and simulation. Learning outcome: Successful completion of the project would give the student an in-depth understanding of the working of a power system. The student will also acquire a thorough knowledge of how a wind farm works. Model development and simulation will provide him a good understanding of differential and algebraic equations and their implementation through coding in the selected software. In all, this project will not only be an enriching research experience for the student, it will also benefit the power systems community by providing a possible solution to a real and challenging problem. 1400897 Parisini,T. Diagnosis of drift faults in sensors of industrial processes 11 October 2014 1114 4D4T The project deals with the problem of diagnosing faults and malfunctions that act as "drifts" in the measured physical quantities provided by sensors in process control systems encountered in industry. The specific goal of this project concerns the formulation of the diagnosis problem in a linear and a nonlinear context, the development of suitable algorithms and the simulation evaluation on an industrial benchmark. Extension to distributed faults is possible Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 13 of 63 PID Supervisor Room Description 1400898 Parisini,T. 1114 The project deals with estimating the characteristic of multiple sinusoidal signals in measurements affected by noise, bias and drifts. The application context is the one of health monitoring in mechanical systems. In the project, a suitable problem formulation has to be addressed, specific algorithms have to be devised and simulation evaluation has to be carried out. Multiple sinusoidal estimation in sensor noisy measurements 1400899 Parisini,T. Multi-sinusoidal adaptive repetitive control 1400900 Parisini,T. Fault diagnosis in electrical networks 1400868 Silversides,R. (with Evangelou,S.) Torque vectoring and regeneration control for the Racing Green electric vehicle (EV3) 1400869 Silversides,R. (with Evangelou,S.) Active flap control for the Racing Green electric vehicle (EV3) 11 October 2014 4D4T 1114 4D4T 1114 4D4T Mauric 4D4T4J Mauric 4D4T4J The project deals with designing an adaptive control scheme based on a repetitive control approach. In particular, a control system affected by multiple-sinusoidal disturbances has to be considered. An estimator of the characteristics of these disturbances has to be developed and the analysis of the behaviour when connected with the repetitive controller has to be carried out. Simulations showing the performance of the overall adaptive system have to be developed as well. The project deals with devising a fault diagnosis scheme for electrical networks. In particular, a problem formulation has to be developed in which failures on electrical components (e.g., capacitors, etc.) are taken into account, detected and identified in real-time. A simulation model has to be developed and tested on some specific scenarios This project will look into the torque vectoring and regeneration control for the new Racing Green electric vehicle (EV3). The vehicle is planned to be a rear wheel drive electric vehicle with independent drive to each rear wheel. The project will look at the control strategies (and develop software/simulations) to demonstrate the potential benefits of using torque vectoring and regeneration control. This project will look into the active flap control for the new Racing Green electric vehicle (EV3). The vehicle will have an extensive aero package, therefore it is hoped to introduce some active flap control to improve handling on corners and reduce drag on straights. It is expect that this will be mainly a control problem, possibly using feedback from the suspension travel/downforce sensors in the chassis. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 14 of 63 PID Supervisor Room Description 1400891 Singh,A. (with Pal,B.C.) 1105 Motivation: An increased penetration of new renewable sources of energy into power systems requires that the reliability and stability of the system is not compromised by this penetration. Solar and wind energy form a chief component of these sources; but as these sources are highly intermittent, non-deterministic and stochastic in nature, it is very difficult to either analyze or control the negative effects that they can have on system reliability and stability. A fundamental step towards addressing this issue is dynamically identifying the equivalent parameters of a wind farm or a solar park, which can then be utilized for further estimation and control purposes. This project studies this identification problem, focusing specifically on a wind farm. Dynamic parameter estimation of a wind farm connected to a large scale power system 3E4D4T Objective: The aim of this project is to study and develop algorithms and methods for dynamic parameter estimation of a wind farm (comprising hundreds of wind turbines) which is connected to a large scale power system. These dynamic parameters can be the equivalent inertia, damping coefficient or reactance of the wind farm. The student can also suggest and estimate some other relevant dynamic parameters. Methodology: The student would develop large scale models for both a generic power system and a wind farm, integrate the two models and perform simulations to get voltage and power measurements from the wind farm for various scenarios of wind input. These measurements would then be utilized for parameter estimation utilizing old/novel stochastic filtering and stochastic signal processing techniques. The estimated parameters will be compared with actual parameters for the all the simulated scenarios and conclusions will be reported. Matlab and DigSilent are suggested software for model development and simulation. Learning outcome: Successful completion of the project would give the student an in-depth understanding of the working of a power system. The student will also acquire a thorough knowledge of how a wind farm works. Model development and simulation will provide him a good understanding of differential and algebraic equations and their implementation through coding in the selected software. In all, this project will not only be an enriching research experience for the student, it will also benefit the power systems community by providing a possible solution to a real and challenging problem. 1400913 Wireless power transfer is now available via the Qi standard for charging phones, but only over very short distances on charging pads. The project will look at the possibility of charging a phone in your pocket as you sit as your office desk. The project will require knowledge of electromagnetics, devices and circuit design skills, an interest in high frequency power electronics design and ability to do analysis both on paper and in Matlab. Circuit simulation skills are also a must. Yates,D.C. (with Mitcheson,P.D.) Charge your phone in your pocket 4T Due to commercial activity surrounding this work (in which the student will be able to engage though the Imperial spin out, Drayson Wireless), the student will be required to sign an NDA and IP agreement. Please talk to me for more details. 1400926 Bouganis,C. Real-time Face Super-Resolution 904 3E4D4T3I4J Super-resolution is an image processing technique that aims to improve the resolution of an image. In this project, the aim is to achieve real-time super-resolution when the target image is a human face. Using a single camera, a set of low resolution images of a human face will be captured by employing an appropriate face detection algorithm (given) and the aim is to estimate a high resolution image of the human face in real-time. The project will investigate the trade-off between computational complexity and quality of the final super resolved image, as well as efficient mappings to the embedded system. The target embedded system can be a SoC system (ARM processor with FPGA fabric) or a system composed by an ARM and a GPU device (i.e Tegra K1). 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 15 of 63 PID Supervisor Room Description 1400927 Bouganis,C. 904 The project aims to the development of a tool that will automate the generation of a cascade support vector machine classifier targeting FPGAs. The novelty of the work is based on the custom precision arithmetic supported by an FPGA. The first part of the cascade is a low precision classifier that aims to classify the "easy" data, where the second part is a full precision classifier that classifies the "difficult" data points. A tool for generation of Support Vector Machine classifier IP core targeting FPGAs 1400928 Bouganis,C. Detection of underwater activity 3E4D4T3I4J The development of the tool can be performed in VHDL or HLS. After the successful development of the tool, the project will focus on the integration of a cascade SVM classifier with a bigger system in order to demonstrate the efficiency of the IP core. 904 3E4D4T3I4J The motivation of this project is based on the need to capture the eating habits of a specific bird that eat fish. The birds can fly for hours before they find a fish to capture. Thus, the envisioned system needs to be able to detect the moment that the bird dives in order to capture the fish. Moreover, low-power and low-weight requirements impose further restrictions to the design of such system. The project focuses on the design of a low-power embedded system that is able to detect when it goes underwater in order to start recording. The project involves algorithmic exploration in Matlab or C in order to evaluate the detection accuracy of the methods and the required computational time, as well as the implementation of the system in an embedded system which can be a micro controller, FPGA or other computational platform. Knowledge of image processing, embedded systems, and Matlab or C would be advantageous for the project. 1400929 Bouganis,C. SLAM on an embedded system 904 3E4D4T3I4J Modern robots need to be able to create an internal representation of the environment, as well as be able to localise themselves in that environment. The project aims to build on an existing approach for localisation an mapping that is called SLAM++ and has been developed by the Department of Computing. The current version of the software requires high computational power and the aim of the project is to investigate and propose an efficient mapping of the system in a SoC system (i.e. embedded ARM processor with FPGA fabric). On the successful mapping of the SLAM++ to the embedded processor, the work will focus on the integration of an existing FPGA design responsible for the acceleration of part of the algorithm with the rest of the system. Further optimisations and mappings of SW routines to the FPGA fabric will be investigated. The aim is to have an embedded system in place that runs in realtime and is able to localise a camera in a known environment. 1400930 Bouganis,C. Autonomous UAV with collision detection 11 October 2014 904 3E4D4T3I4J The project focuses on the development of a UAV system that is able to detect possible collisions in real-time. By extracting the depth of objects using Structure from Motion techniques, the UAV should be able to detect and avoid possible collision with other objects. The project will involve the development and implementation of a Structure of Motion submodule in C, its integration with a real drone (i.e. AR Drone) and its possible acceleration and mapping in a embedded SoC system (ARM core with an FPGA or GPU). Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 16 of 63 PID Supervisor Room Description 1400954 Bouganis,C. 904 Solving an overdetermined system of linear equations or computing the residuals of the system is a very common problem in many scientific fields, including difficult problems in computational statistics (e.g. statistical genetics). An overdetermined system of linear equations is represented by the following equation: AB = Y where A (m x n) and Y (m x k) are known matrices, B (n x k) is unknown and m > n (overdetermined). We either want to solve the system (i.e. find the B that minimizes Y - XB) or find the residuals of the system (i.e. the value of Y - XB when B is the solution). Both problems can be solved using a linear algebra method called QR decomposition. When the dimensions of the problem (m, n and k) are large and there are a lot of systems to solve (such as in big-data applications in genetics), we typically have to use parallel hardware to accelerate computations. GPU libraries like Nvidia's CULA provide routines for QR decomposition and other necessary operations. Nevertheless, CULA routines achieve peak performance only for square matrices and process each system separately. This leads to under-utilization of the GPU device when the matrix A is tall-skinny (m >> n) or when there are thousands of medium-sized systems to solve (which have to be processed sequentially). This project aims at developing a new versatile GPU library which overcomes these limitations and will allow us to efficiently tackle systems of any size as well as multiple problems in parallel. The student will base his or her work on existing CULA routines and routines from the CAQR library (specialized for tall-skinny problems). The main directions of the project will be: High performance, adaptive GPU library for computation of residuals of linear systems with varying matrix dimensions 3E4D4T3I4J 1)Develop code for solving systems and finding residuals which will be able to pick the best routine (between CPU and multiple GPU implementations) depending on the system's dimensions and the shape of A (square or tall-skinny). 2)Explore ways in which many small or medium-sized systems (either with different A matrices or different Y matrices or both) can be solved in parallel by designing new GPU kernels or by streaming many kernels into the GPU using the related CUDA functionality. 3)Optimize auxiliary computations and procedures (e.g. matrix multiplications, data transfers) to improve speed. Skills required: 1)Basic familiarity with parallel programming concepts and techniques. GPU programming and CUDA knowledge are a plus but not required. Expected acquired skills: 1)Experience with GPU programming, CUDA programming and its toolflow used on a real-world problem 2)Familiarity with widely used linear algebra techniques, their pros and cons and issues regarding their implementation on GPUs and CPUs 3)Research experience and contribution to the group's output. The results of the project will be used in real genetic problems research and will likely lead to a publication. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 17 of 63 PID Supervisor Room Description 1400967 Chen,G. (with RodriguezVillegas,E.) 906 Monitoring of human behaviours in free individuals is arguably one of most popular application in wearable technology. The aim of this project is to explore 1) a potential set of human behaviours could be recognised from tracheal sound by using our awardwinning miniaturized wearable tracheal sound monitor and 2) lightweight/real-time pattern recognition techniques to differentiate those behaviours on an microcontroller. The student is expected to have good programming skills in C/C++ or Matlab, and background on the embedded systems. For more information on the sensor used in this project please see http://www.acupebble.com 609 System-on-chip FPGAs combine dual ARM processors with flexible, uncommitted programmable logic. By integrating these components designers hope to achieve better power efficiency and performance. Human Behaviour Monitoring Using a Single Wearable Sensor 1400957 Cheung,P.Y.K. (with Stott,E.) Reconfigurable system-on-chip 4T 4D4T4J This project will develop a computing platform, based on an SoC FPGA, that reduces power consumption by dispatching tasks to custom kernels implemented in soft logic. The project will feature algorithm development in software and digital logic. Familiarity with Linux, FPGAs and hardware description languages will be an advantage. Once the basic hardware-software interaction is developed there are many interesting questions that could be addressed to achieve higher marks. For example, when are the benefits of a hardware kernel outweighed by the penalties of data transfer? How can hardware components be standardised and built into libraries for easy reuse? 1400938 Clarke,T.J.W. An Educational ARM Emulator with Data Visualisation 615 4D4T4J This project aims to write a tool that will make learning ARM assembly much simpler for 1st and 2nd year undergraduate students. The tool will implement a limited but highly useful subset of ARM code and provide a GUI which illustrates what happens when instructions, loops, etc are executed. Data flow and control flow will be illustrated diagrammatically as well as conventionally, and simulating with symbolic data ('x' instead of 0x100, 'x'+1 instead of 0x101) will be supported The project does NOT require a competent ARM programmer, but needs a student who is capable of programming in at least one modern programming language (Java, ruby, python, or one of the many good post-Java languages). Part of the challenge in this project is how best to visualise the changes in data that occur as a fragment of ARM code is executed. The project would be suitable, also, for a student who was a very competent programmer and wished to learn a new language Java or any of the many post-Java languages - during the project. Prerequisites: Some familiarity with OO programming. Proven ability to write medium-size programs Interest in learning to use new programming libraries, languages, etc 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 18 of 63 PID Supervisor Room Description 1400939 Clarke,T.J.W. 615 This search for and display a mashup of contextual information about property for sale/rent in the UK scraping information from the web. Specifically the Zoopla, Nestoria, and OS OpenSpace APIs will be used to obtain information and other information added. An Advanced UK House Search Agent 4D4T4J There is a very wide variety of possible mashups in this area (see the many mashups based on Zillow in the US for example). My preference is for information relevant to properties in the country where outdoor pursuits etc are important but I'm willing to adapt to some other niche given a good case. This project would suit any student with very good programming skills. 1400849 Constandinou,T. (with Liu,Y.) 2-wire Power/Data Interface for Implantable Medical Devices 901 4T Implantable medical devices typically employ transcutaneous (i.e. through the skin) telemetries to transmit power and data (inductively), thus avoiding the risk of infection due to breaching the skin barrier. The implanted device(s) however use wired connections between multi-module implants. Examples include: deep brain stimulators (DBS) for Parkinson's/dystonia/essential tremor, cochlear implants for hearing, etc. A key challenge in ALL such systems is the mechanical reliability of interconnects in general fewer wires are preferable. This project aims to develop a new 2-wire interface for communicating between a neural (brain) implant (inside the skull) and a processing unit (situated in the chest cavity). The interface should implement full-duplex (simultaneous bidirectional) communication of up to 1Mbps, with error detection/correction, and recover a reference clock, power supply, and bias references within the neural implant module. An additional requirement is that several modules can share this communication interface, i.e. can be connected and are addressible in parallel. Key challenges are: (1) there must be no static electric field (i.e. no DC), (2) electronics must occupy a minimal volume (minimal components), (3) minimal power dissipation. This project will firstly develop new concepts for this 2-wire protocol and implement these in a prototype (hardware) demonstrator. It is expected that any novel contributions will be published in relevant peer-reviewed journals/conferences. Ideal student background: good practical ability required - familiarity and experience with microcontrollers, embedded programming, analogue and digital electronics, PCB design, communication methods and protocols, power management/regulation. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 19 of 63 PID Supervisor Room Description 1400850 Constandinou,T. (with Liu,Y.) 901 Brain Machine Interfaces (BMIs) are a direct communication pathway between the brain and an external device. BMIs are often directed at assisting, augmenting, or repairing human cognitive or sensory-motor functions. Calibration-free Real-time Neural Spike Streaming 4T Implanted BMIs use tiny electrodes inserted into the tissue to monitor the spiking activity of individual neurons (action potentials) or local populations of neurons (local field potentials). However, the observed recordings vary from electrode to electrode (both in signal amplitude and background noise level), and may change over time. In order to effectively utilise these recordings, amplifier gains and analogue to digital converter resolutions are currently individually tuned (manually) such as to maximise the useful dynamic range. Future BMIs will have 1000s of channels to decode our thoughts and provide useful information for controlling external devices, for example, prosthetic limbs, computers, etc. It is however unpractical to individually manually calibrate each channel to the observed signal dynamics, particularly as these change over time. Future BMIs will therefore have to adapt and self calibrate to the signal dynamics. This project will develop a computationally-efficient (i.e. minimal computation and memory) adaptive algorithm to effectively detect these spike signals without needing any calibration. The algorithm will then be implemented in an ultra low power embedded platform to demonstrate real-time adaptive spike streaming. This will then be used by a research team based at Imperial that is implementing a multi-channel BMI implant chip. It is expected that any novel contributions will additionally be published in relevant peer-reviewed journals/conferences. Ideal student background: strong in Matlab and C, good experience required with microcontrollers and embedded programming. 1400848 Constandinou,T. (with Nicolaou,N.) "Real-time" awareness monitoring during anesthesia 901 4D4T4J Modern anesthesia is a "cocktail" of chemical agents. The anesthetist administers this "cocktail" of agents following general dosage guidelines, which are then adjusted based on patient characteristics. However, it is possible that sometimes the wrong amounts are administered, resulting to either over-administration (very deep anesthesia) or under-administration (possibility of regaining awareness during surgery). Based on a number of studies, it is estimated that at least 1,873,600 people are likely to have experienced intra-operative awareness in one year, who are deeply traumatized by this experience and can suffer severe psychological consequences. Therefore, monitoring the patient level of hypnosis during surgery provides vital information to prevent such cases from happening. Such monitoring is best achieved through the patient's electrical brain activity, whose patterns differ during wakefulness and anesthesia. The project involves the implementation of a system for "real-time" monitoring of awareness during anesthesia. The system will consist of the following parts: (1) Input: electrical brain activity (EEG) from a database of signals collected from patients during surgery; (2) Processing: FPGA implementation of a particular methodology that has been shown to distinguish between awareness and anesthesia using the EEG activity; and (3) Output: a visual representation of the underlying patient state. Some experience with Matlab and FPGA programming is required. The project is co-supervised by Dr. Timothy Constandinou and Dr. Nicoletta Nicolaou. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 20 of 63 PID Supervisor Room Description 1400856 Constantinides,G.A. 912 There have been a number of tiny universal Turing machines proposed, e.g. http://alvyray.com/CreativeCommons/TuringToysdotcom.htm. These are so small that the "machine" part could fit in one or two LUTs on an FPGA. The idea of this project is to fill an FPGA device with such Turing machines (how many cores can we fit?) and to produce a compiler for such machines. An extension is to propose (and implement) a method for passing information between neighbouring machines on the FPGA, and the appropriate software support. NIL Tiny Turing Machines This project should be great fun for someone who: - Has a real interest in computation - Is at least familiar with the idea of universal Turing machines - Already has FPGA implementation experience - Is an accomplished software engineer 1400958 Constantinides,G.A. (with Stott,E.) Power measurement breakdown in digital logic 912 4D4T4J The power consumed by a digital circuit is dependent upon a number of parameters, including supply voltage, switching activity and leakage current. It is easy to measure the overall power but how can we get a breakdown of the proportion of power used by different parts of the circuit? Knowing this would allow us to optimise operation of complex systems-on-chips to improve efficiency. This project will find out if it is possible to estimate power consumption in digital circuits by probing a small proportion of the internal and external signals and analysing the statistical properties of the trace that is returned. The main skills required are digital circuit design with hardware description languages and data analysis. Generating some indicative results should be straightforward, but to achieve high marks the candidate could consider how the probe locations could be selected while the circuit is being designed and produce tools that would automate the process. 1400951 Drane,T. (with Thomas,D.B.) Using Polynomials in Formal Verification 4D4T4J This is an industrially led research project that focuses on the crucial activity of formal verification when designing hardware; proving functional correctness of large hardware designs is a challenge and verification takes up a significant proportion of design cycles. However, recent advances in the area of algebraic geometry offer a new angle on such problems, with the potential to verify the seemingly unverifiable: - http://www.ece.utah.edu/~kalla/papers/86-XA763.pdf - http://cas.ee.ic.ac.uk/people/gac1/pubs/TheoDACKC11.pdf This project would suit students strong in mathematics, and will involve the use of bespoke algebraic geometry tools. This project will involve working with an industrial researcher (as well as the supervisor), so please come for discussion in room 903 on Monday 13th 12:30-13:30 when they will also be available. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 21 of 63 PID Supervisor 1400952 Drane,T. (with Thomas,D.B.) Constrained Input Domain Synthesis Room 4D4T4J Description This is an industrially led research project that addresses the observation that within deep arithmetic pipelines there may be redundant hardware; for example there may be branches which are actually never entered, variables which are constant, or stuck at bit, all regardless of which inputs enter the pipeline. It is also common that the set of possible inputs to the pipeline is restricted i.e. has a constrained input domain. Synthesis tools don’t optimise these redundancies, but exploiting these opportunities can provide significant hardware quality improvements. The idea is that one could hook up a verification and synthesis tools to discover these situations and thus automatically 'scrub' code of dead branches and logic. This project would suit students with either experience or an interest in RTL (VHDL/Verilog), synthesis and/or verification. This project will involve working with an industrial researcher (as well as the supervisor), so please come for discussion in room 903 on Monday 13th 12:30-13:30 when they will also be available. 1400953 Drane,T. (with Thomas,D.B.) Power Estimation of RTL 3E4D4T3I4J Power measurement of RTL typically requires long simulations, using tools with very expensive licenses usage, so it would be preferable if analytic probabilities could be calculated for switching information. Techniques do exist which attempt to perform this task but require further research in order to allow tools which deliver usable results on industrial strength benchmarks in a reasonable time. This project will look into the use of BDDs and interval arithmetic, in order to try to develop such tools. Some background papers are: http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=1542949&url=http%3A%2F%2Fieeexplore.ieee.org%2Fiel5%2F10358%2 F32955%2F01542949 - http://www.cs.virginia.edu/~robins/papers/DAC_2012_final_Gabe.pdf This project would suit students interested in the fundamentals of digital logic synthesis, with good mathematical skills, decent programming skills (in any language, matlab is fine), and some knowledge of RTL simulation. This project will involve working with an industrial researcher (as well as the supervisor), so please come for discussion in room 903 on Monday 13th 12:30-13:30 when they will also be available. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 22 of 63 PID Supervisor 1400919 Evans,B. (with Nikolic,K.) Machine Learning tasks on the SpiNNaker platform Room 1A1C1S Description SpiNNaker is a massively parallel computing platform inspired by biological neural tissue. SpiNNaker was primarily developed to model and efficiently simulate networks of spiking neurons. The aim of this project will be to implement a biologically-inspired machine learning algorithm on a SpiNNaker board. The ultimate aim will be to eventually make a robot which can learn to catch balls, but in this project you will initially work on the camera-processors communication and the learning algorithm. To this end you will use a special type of low-power camera, an “event-based Dynamic Vision Sensor” (DVS128), which responds to relative changes in illumination at each pixel by creating “spikes” in the form of events. The DVS128 is a 128x128 pixel temporal contrast CMOS sensor that generates asynchronous data representing quantised changes in illumination. The output is a continuous stream of “ON” and “OFF” type “events” containing the time-stamp and pixel address at which the illumination change occurred. This Address-Event Representation (AER) data will form the input signal for the SpiNNaker neuromorphic hardware. The methodology developed here will be applicable to a wide variety of practical real-time power-constrained machine vision applications, which have to process scenes spanning temporal scales from microseconds to days and illumination conditions from starlight to sunlight. If interested, please email me: [email protected] 1400920 Evans,B. (with Nikolic,K.) Modeling Neural Networks on the SpiNNaker platform 4D4T SpiNNaker is a massively parallel computing platform inspired by biological neural tissue. The SpiNNaker hardware can be used to model and efficiently simulate spiking neural networks. To aid in its use with a variety of existing neural network simulators, a special simulator-independent model description language has been developed, named PyNN. It would be particularly useful to endow such simulators with the ability to model optogenetic ion channels and explore the efficiency of their simulation on different hardware platforms. Specific tasks will include implementing new types of optogenetic ion-channels in different neural simulators (such as Brian and NEST), incorporating them into specific brain circuit topologies and simulating them (through PyNN) on multiple platforms including SpiNNaker. The aim of this project is to develop skills to work with neuromorphic hardware that have applications in neuroscience, robotics and computer science. If interested, please email me: [email protected] 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 23 of 63 PID Supervisor 1400961 Evans,B. (with Nikolic,K.) Machine Learning tasks on the SpiNNaker platform Room 4D4T Description SpiNNaker is a massively parallel computing platform inspired by biological neural tissue. SpiNNaker was primarily developed to model and efficiently simulate networks of spiking neurons. The aim of this project will be to implement a biologically-inspired machine learning algorithm on a SpiNNaker board. The ultimate aim will be to eventually make a robot which can learn to catch balls, but in this project you will initially work on the camera-processors communication and the learning algorithm. To this end you will use a special type of low-power camera, an “event-based Dynamic Vision Sensor” (DVS128), which responds to relative changes in illumination at each pixel by creating “spikes” in the form of events. The DVS128 is a 128x128 pixel temporal contrast CMOS sensor that generates asynchronous data representing quantised changes in illumination. The output is a continuous stream of “ON” and “OFF” type “events” containing the time-stamp and pixel address at which the illumination change occurred. This Address-Event Representation (AER) data will form the input signal for the SpiNNaker neuromorphic hardware. The methodology developed here will be applicable to a wide variety of practical real-time power-constrained machine vision applications, which have to process scenes spanning temporal scales from microseconds to days and illumination conditions from starlight to sunlight. If interested, please email me: [email protected] 1400922 Georgiou,P. A wearable device for Muscle Fatigue Detection in rehabilitation of athletes 902 4D4T Electromyography (EMG) is a technique used to evaluate the electrical activity of muscles. One interesting aspect of the EMG signal is the ability to extract muscular fatigue information by applying specific signal processing techniques. Evaluating muscle fatigue has numerous applications including medical research (orthopaedic, gait analysis), rehabilitation (post-surgery, physical therapy), ergonomics (risk prevention, ergonomic design), biomechanics (motor control) and sports science (strength training, movement analysis) applications. We have developed a novel ASIC which implements a revolutionary algorithm for muscle fatigue detection. The aim of this chip is to monitor and evaluate fatigue and lactic acid levels in the muscles of professional athletes. At the moment, the chip is a standalone device. The aim of this project is to integrate the Muscle Fatigue Chip in a complete wearable device that can be easily attached on athletes for human trials. Furthermore, the developed monitoring platform needs to be interfaced and controlled by a user through a smartphone app. The project is composed of three primary goals. First goal involves small scale PCB design for attaching the IC, power management, front-end calibration and Bluetooth communication with a handheld device. The second goal involves software development for a customized smartphone application to read data from and control the ASIC. The last goal of the project is the design and manufacturing of a customized casing for attaching the complete monitoring platform on patients. Required Expertise: Previous experience on smartphone-app and microcontroller programming and PCB design are desirable. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 24 of 63 PID Supervisor Room Description 1400923 Georgiou,P. 902 We have developed a handheld medical device called an artificial pancreas that is responsible for controlling blood glucose in Type 1 diabetic subjects. The system communicates with a continuous glucose monitor and insulin infusion pump to regulate glucose in a healthy range. The current system has finished clinical trials and over the next year we will be developing the system for use outside of the hospital and into the home. Engineering the Artificial Pancreas for the home: Design of a low power and robust system for safe home use 4D4T The aim of this project is design a handheld unit for the artificial pancreas which allows it’s safe and reliable operation in a home environment. The system needs to be robust and self aware of any faults which may occur and guarantee the safe and reliable delivery of insulin. The system must also be optimised for low power consumption as we are aiming for 7 days continuous operation on a single charge. This is a design and build project that involves microcontroller programming and PCB design. Students will gain significant experience in medical device design and be part of a multidisciplinary team in this effort. Required Expertise: Previous experience on microcontroller programming and PCB design are desirable. 1400924 Georgiou,P. Engineering the Artificial Pancreas for the home: A reliable meal detection system for meal announcement 902 4D4T We have developed a handheld medical device called an artificial pancreas that is responsible for controlling blood glucose in Type 1 diabetic subjects. The system communicates with a continuous glucose monitor and insulin infusion pump to regulate glucose in a healthy range. The current system has finished clinical trials and over the next year we will be developing the system for use outside of the hospital and into the home. When we eat a meal our pancreas releases some anticipatory insulin known as the Cephalic phase which helps reduce blood glucose from reaching high levels. In a person with diabetes this is absent. With our current system when a person with diabetes eats a meal we need to announce this on the system through a process of meal announcement. This requires the patient to push a button. This Project involves designing a wearable device that is capable of automatically detecting and classifying when a patient has eaten a meal. The objective is to automate meal announcement. The device needs to be unobtrusive and run an algorithm to be able to classify that someone has eaten a meal with greater than 90% confidence. This is a fairly open project and the design and sensor choice is up to the student. Required Expertise: Previous experience in MATLAB and microcontroller programming and PCB design are desirable. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 25 of 63 PID Supervisor Room Description 1400999 Georgiou,P. (with Herrero Vinas,P.) 902 An automatic closed-loop system, also referred to as an artificial pancreas (AP), provides the potential to improve glycemic control in diabetes. An AP consists of a continuous glucose sensor, an insulin pump and a control algorithm that links the two previous devices. In the last decade, there has been unprecedented technological progress in the development of closed-loop control systems for the AP and their clinical efficacy, in a supervised environment (i.e. in hospital), has been demonstrated. The next logical step is to move the AP to a home environment, but for this purpose, a robust fault-tolerant control architecture is required to guarantee patient’s safety. The aim of this project is to develop a robust fault-tolerant control architecture for the Imperial College Bio-inspired Artificial Pancreas (BiAP) in order to validate it in a home environment. A Robust Fault-Tolerant Control Architecture for the Bio-inspired Artificial Pancreas 4D4T4J The proposed architecture incorporates the following systems: • A system to detect faults in continuous glucose sensor and insulin pump • A system to detect unannounced perturbations, such as meal intakes and exercise • A system to predict hypo- and hyperglycaemia (i.e. low and high blood glucose level) • An alarm management system • A supervision system that allows switching between closed-loop and open-loop control modes when required The supervision architecture will be implemented on a microcontroller embedded in the BiAP handheld unit. Requirements: fundamentals of control theory; basic programming skills in Matlab and C. 1400999 Herrero Vinas,P. (with Georgiou,P.) A Robust Fault-Tolerant Control Architecture for the Bio-inspired Artificial Pancreas 4D4T4J An automatic closed-loop system, also referred to as an artificial pancreas (AP), provides the potential to improve glycemic control in diabetes. An AP consists of a continuous glucose sensor, an insulin pump and a control algorithm that links the two previous devices. In the last decade, there has been unprecedented technological progress in the development of closed-loop control systems for the AP and their clinical efficacy, in a supervised environment (i.e. in hospital), has been demonstrated. The next logical step is to move the AP to a home environment, but for this purpose, a robust fault-tolerant control architecture is required to guarantee patient’s safety. The aim of this project is to develop a robust fault-tolerant control architecture for the Imperial College Bio-inspired Artificial Pancreas (BiAP) in order to validate it in a home environment. The proposed architecture incorporates the following systems: • A system to detect faults in continuous glucose sensor and insulin pump • A system to detect unannounced perturbations, such as meal intakes and exercise • A system to predict hypo- and hyperglycaemia (i.e. low and high blood glucose level) • An alarm management system • A supervision system that allows switching between closed-loop and open-loop control modes when required The supervision architecture will be implemented on a microcontroller embedded in the BiAP handheld unit. Requirements: fundamentals of control theory; basic programming skills in Matlab and C. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 26 of 63 PID Supervisor 1401000 Kulasekeram,N. (with Toumazou,C.) A Wireless Transceiver for a nerve, chemical sensor and GUI 1401001 4D4T4J Kulasekeram,N. (with Toumazou,C.) A wireless transceiver for a nerve, chemical sensor and battery power management 1400960 Room 4D4T4J Levine,J.M. (with Constantinides,A.G.) Brewing the perfect espresso 4D4T4J Description This project is part of exciting work that's aim is to develop technology that taps into the nervous system. By doing so it aims to monitor and regulate appetite in overweight people by tapping into the nerve responsible for it. This project is an exciting opportunity to work with neuroscientists and engineers of many disciplines. The aim of the project is to design a wireless interface to a chemical sensor which is expected to provide a maximum output voltage of 10 millivolts at a frequency of 10Hz. This signal needs to be transmitted to a computer terminal or mobile handheld device . The intention is to use commercially available wireless modules for the purpose of transmitting and receiving the required data as well as develop custom PCB’s for interfacing with the sensor. The next stage is to write a software GUI which would interface with data received from a wireless receiver, The GUI would be used to display the real time chemical sensor data which has been transmitted from a maximum of 100 metres Skills to be gained include, Interfacing, Wireless protocol, Wireless modules, Software GUI and API programming, This project is part of exciting work that's aim is to develop technology that taps into the nervous system. By doing so it aims to monitor and regulate appetite in overweight people by tapping into the nerve responsible for it. This project is an exciting opportunity to work with neuroscientists and engineers of many disciplines. The aim of this project is to design and develop a chemical sensor (PCB) and an energy efficient method to transmit chemically sensed data which is only made available every 100 milliseconds. So a solution which is based on storing large amounts of data and transmitting it, to a receiver, which could be either a computer terminal or a mobile hand held device. The wireless solution needs to be battery powered, as the end solution needs to be portable. The battery should be used in the most efficient way possible. Skills to be gained include, Interfacing , Wireless modules, Wireless protocol, Power management The taste of espresso depends upon a variety of brewing parameters, including how long the dry ground coffee is pre-infused, the temperature of the water and the profile with which pressure is applied. Selecting these variables in somewhat of an art form but there is surely scope to create an electronic, expert barista. You will design and build hardware and a controller capable of accurately adjusting brewing parameters. It should be able to produce sample brews for taste experiments; allowing the best-tasting formula to be researched. The skills that could be employed include electronic and electromechanical system design, embedded software development, control and scientific experiment design. At a minimum, the student will need to create hardware that can brew an espresso to a user specification. There are many opportunities for achieving higher marks, particularly if you can discover and publish the formula for the perfect espresso. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 27 of 63 PID Supervisor 1400996 Levine,J.M. (with Constantinides,A.G.) Temperature and voltage characterisation of devices Room 4D4T4J Description The performance and lifetime of an integrated circuit is dependent upon operating and environmental conditions, which include temperature and supply voltage. To produce reliable devices it is important to characterise behaviour under all scenarios that are likely to be encountered in the field. The information can be used to develop techniques for making circuits more efficient, better performing and more reliable. This project will design an experimental rig for precisely and safely controlling the voltage and temperature of a circuit-undertest. An API will enable the hardware to be easily controlled by a top-level experiment procedure . The rig will then be used to carry out an investigation into performance and reliability in FPGA devices. The project will feature both hardware and software development and encompasses aspects from mechanical components all the way up to a potential web interface . Thus it can be adapted to fit a student's skills, though for high marks I would like to see a fully-integrated, vertical solution that could be released under an open-source license for use by the research community. To demonstrate its usefulness, the system could be used to carry out (and possibly publish the results of) accelerated life testing or statistical characterisation of devices. 1400849 Liu,Y. (with Constandinou,T.) 2-wire Power/Data Interface for Implantable Medical Devices 4T Implantable medical devices typically employ transcutaneous (i.e. through the skin) telemetries to transmit power and data (inductively), thus avoiding the risk of infection due to breaching the skin barrier. The implanted device(s) however use wired connections between multi-module implants. Examples include: deep brain stimulators (DBS) for Parkinson's/dystonia/essential tremor, cochlear implants for hearing, etc. A key challenge in ALL such systems is the mechanical reliability of interconnects in general fewer wires are preferable. This project aims to develop a new 2-wire interface for communicating between a neural (brain) implant (inside the skull) and a processing unit (situated in the chest cavity). The interface should implement full-duplex (simultaneous bidirectional) communication of up to 1Mbps, with error detection/correction, and recover a reference clock, power supply, and bias references within the neural implant module. An additional requirement is that several modules can share this communication interface, i.e. can be connected and are addressible in parallel. Key challenges are: (1) there must be no static electric field (i.e. no DC), (2) electronics must occupy a minimal volume (minimal components), (3) minimal power dissipation. This project will firstly develop new concepts for this 2-wire protocol and implement these in a prototype (hardware) demonstrator. It is expected that any novel contributions will be published in relevant peer-reviewed journals/conferences. Ideal student background: good practical ability required - familiarity and experience with microcontrollers, embedded programming, analogue and digital electronics, PCB design, communication methods and protocols, power management/regulation. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 28 of 63 PID Supervisor 1400850 Liu,Y. (with Constandinou,T.) Calibration-free Real-time Neural Spike Streaming Room Description Brain Machine Interfaces (BMIs) are a direct communication pathway between the brain and an external device. BMIs are often directed at assisting, augmenting, or repairing human cognitive or sensory-motor functions. 4T Implanted BMIs use tiny electrodes inserted into the tissue to monitor the spiking activity of individual neurons (action potentials) or local populations of neurons (local field potentials). However, the observed recordings vary from electrode to electrode (both in signal amplitude and background noise level), and may change over time. In order to effectively utilise these recordings, amplifier gains and analogue to digital converter resolutions are currently individually tuned (manually) such as to maximise the useful dynamic range. Future BMIs will have 1000s of channels to decode our thoughts and provide useful information for controlling external devices, for example, prosthetic limbs, computers, etc. It is however unpractical to individually manually calibrate each channel to the observed signal dynamics, particularly as these change over time. Future BMIs will therefore have to adapt and self calibrate to the signal dynamics. This project will develop a computationally-efficient (i.e. minimal computation and memory) adaptive algorithm to effectively detect these spike signals without needing any calibration. The algorithm will then be implemented in an ultra low power embedded platform to demonstrate real-time adaptive spike streaming. This will then be used by a research team based at Imperial that is implementing a multi-channel BMI implant chip. It is expected that any novel contributions will additionally be published in relevant peer-reviewed journals/conferences. Ideal student background: strong in Matlab and C, good experience required with microcontrollers and embedded programming. 1400848 Nicolaou,N. (with Constandinou,T.) "Real-time" awareness monitoring during anesthesia 4D4T4J Modern anesthesia is a "cocktail" of chemical agents. The anesthetist administers this "cocktail" of agents following general dosage guidelines, which are then adjusted based on patient characteristics. However, it is possible that sometimes the wrong amounts are administered, resulting to either over-administration (very deep anesthesia) or under-administration (possibility of regaining awareness during surgery). Based on a number of studies, it is estimated that at least 1,873,600 people are likely to have experienced intra-operative awareness in one year, who are deeply traumatized by this experience and can suffer severe psychological consequences. Therefore, monitoring the patient level of hypnosis during surgery provides vital information to prevent such cases from happening. Such monitoring is best achieved through the patient's electrical brain activity, whose patterns differ during wakefulness and anesthesia. The project involves the implementation of a system for "real-time" monitoring of awareness during anesthesia. The system will consist of the following parts: (1) Input: electrical brain activity (EEG) from a database of signals collected from patients during surgery; (2) Processing: FPGA implementation of a particular methodology that has been shown to distinguish between awareness and anesthesia using the EEG activity; and (3) Output: a visual representation of the underlying patient state. Some experience with Matlab and FPGA programming is required. The project is co-supervised by Dr. Timothy Constandinou and Dr. Nicoletta Nicolaou. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 29 of 63 PID Supervisor Room Description 1400919 Nikolic,K. (with Evans,B.) IBE B4 SpiNNaker is a massively parallel computing platform inspired by biological neural tissue. SpiNNaker was primarily developed to model and efficiently simulate networks of spiking neurons. The aim of this project will be to implement a biologically-inspired machine learning algorithm on a SpiNNaker board. The ultimate aim will be to eventually make a robot which can learn to catch balls, but in this project you will initially work on the camera-processors communication and the learning algorithm. Machine Learning tasks on the SpiNNaker platform 1A1C1S To this end you will use a special type of low-power camera, an “event-based Dynamic Vision Sensor” (DVS128), which responds to relative changes in illumination at each pixel by creating “spikes” in the form of events. The DVS128 is a 128x128 pixel temporal contrast CMOS sensor that generates asynchronous data representing quantised changes in illumination. The output is a continuous stream of “ON” and “OFF” type “events” containing the time-stamp and pixel address at which the illumination change occurred. This Address-Event Representation (AER) data will form the input signal for the SpiNNaker neuromorphic hardware. The methodology developed here will be applicable to a wide variety of practical real-time power-constrained machine vision applications, which have to process scenes spanning temporal scales from microseconds to days and illumination conditions from starlight to sunlight. If interested, please email me: [email protected] 1400920 Nikolic,K. (with Evans,B.) Modeling Neural Networks on the SpiNNaker platform IBE B4 4D4T SpiNNaker is a massively parallel computing platform inspired by biological neural tissue. The SpiNNaker hardware can be used to model and efficiently simulate spiking neural networks. To aid in its use with a variety of existing neural network simulators, a special simulator-independent model description language has been developed, named PyNN. It would be particularly useful to endow such simulators with the ability to model optogenetic ion channels and explore the efficiency of their simulation on different hardware platforms. Specific tasks will include implementing new types of optogenetic ion-channels in different neural simulators (such as Brian and NEST), incorporating them into specific brain circuit topologies and simulating them (through PyNN) on multiple platforms including SpiNNaker. The aim of this project is to develop skills to work with neuromorphic hardware that have applications in neuroscience, robotics and computer science. If interested, please email me: [email protected] 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 30 of 63 PID Supervisor Room Description 1400961 Nikolic,K. (with Evans,B.) IBE B4 SpiNNaker is a massively parallel computing platform inspired by biological neural tissue. SpiNNaker was primarily developed to model and efficiently simulate networks of spiking neurons. The aim of this project will be to implement a biologically-inspired machine learning algorithm on a SpiNNaker board. The ultimate aim will be to eventually make a robot which can learn to catch balls, but in this project you will initially work on the camera-processors communication and the learning algorithm. Machine Learning tasks on the SpiNNaker platform 4D4T To this end you will use a special type of low-power camera, an “event-based Dynamic Vision Sensor” (DVS128), which responds to relative changes in illumination at each pixel by creating “spikes” in the form of events. The DVS128 is a 128x128 pixel temporal contrast CMOS sensor that generates asynchronous data representing quantised changes in illumination. The output is a continuous stream of “ON” and “OFF” type “events” containing the time-stamp and pixel address at which the illumination change occurred. This Address-Event Representation (AER) data will form the input signal for the SpiNNaker neuromorphic hardware. The methodology developed here will be applicable to a wide variety of practical real-time power-constrained machine vision applications, which have to process scenes spanning temporal scales from microseconds to days and illumination conditions from starlight to sunlight. If interested, please email me: [email protected] 1400969 Papavassiliou,C. Student self proposed Design and build 1400970 Papavassiliou,C. Steering the beam of an antenna array 1400971 Papavassiliou,C. Circuits utilising Memristor devices 1400972 Papavassiliou,C. Interfaces for multilevel Resistive RAM 11 October 2014 915 A design and build project, preferably one which includes a radio. 3E4D4T3I4J 915 The sensitivity pattern of an antenna array can be steered by controlling the impedance matching to receivers connected to its elements. In this project we will use this method of beam steering to increase the angular resolution of an antenna array. 3E4D4T3I4J 915 3E4D4T3I4J 915 3E4D4T3I4J The memristor is a resistor whose value of resistance is determined by the recent history of signals applied to it. Memristors can be used to make variable gain amplifiers, power meters, pulsed transmitters (neuron emulators) and more. Resistive Random Access memories have been proposed as a future direction of memory technology. They are easy to manufacture, have a very small unit cell and can support several bits per cell. The main challenge is to devise a simple enough and accurate enough interface which will make this technology practical. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 31 of 63 PID Supervisor Room Description 1400973 Papavassiliou,C. 915 Groups of coupled oscillators develop collective modes, which are spectrally purer than each individual oscillator. In fact, the output properties of a group of coupled oscillator depend more on the interconnection details (strength of interactions and delays) than on the properties of the individual oscillators. To see this consider the ring oscillator widely used in integrated electronics, which IS a rather good oscillator, despite consisting of non-oscillating inverters! Arrays of oscillators 3E4D4T3I4J In this project we will investigate, through SPICE or MATLAB simulations, the noise properties of lattices (two-dimensional arrays) of oscillators. 1400988 Papavassiliou,C. Monolithic parametric amplifiers 1400989 3E4D4T3I4J 914 1A1S Rodriguez-Villegas,E. System for remote identification of whooping cough 1400976 915 Rodriguez-Villegas,E. Heart rate estimation from tracheal sounds II 1400974 3E4D4T3I4J Papavassiliou,C. Monolithic reflection amplifiers 1400968 915 UHF and microwave RFID systems for position and location tracking on wearable devices 11 October 2014 4T In this project we will study monolithic implementations of parametric amplifiers. A voltage divider with a negative arm can amplify; they are often called reflection amplifiers. Reflection amplifiers can often operate at extremely high frequencies. An example is the laser amplifier, operating at the frequency of vIR or visible light. Some devices, and circuits, exhibit locally, at some operating point a negative differential resistance and can therefore used to build a reflection amplifier. In this project we will study and design monolithic reflection amplifiers. Heart sounds have been a source of information for diagnosis on patients’ conditions. This project aims to implement an automatic real time heart rate estimation algorithm by using heart sounds obtained from the suprasternal notch. The student is expected to have a good background in embedded signal processing with C/C++ or Matlab. All programs will be developed in Matlab and tested in a microcontroller. For more information on the sensor used in this project please see http://www.acupebble.com 914 Whooping cough is a disease that affects 48.5 million people in the world annually. It is characterized by severe coughing ending in a whooping sound when the person breathes. The disease lasts for a very long time, and because it is highly contagious, it is important to detect it early to treat with antibiotics and hence prevent spreading. The aim of this project is to create new algorithms to automatically identify whooping cough from an acoustic signal sensed with a novel technology we have developed. This would potentially allow for whooping cough to be remotely diagnose, so that adecuate treatment can be provided with a minimum delay. The student taking this project should have excellent Matlab skills (or equivalent) and be willing to develop algorithms from scratch. 914 Comprehensive literature review on the state of the art position and location tracking systems using RFID technology will be undertaken. An implementation of RFID location tracking shall be developed as part of this project for wearable devices to identify the location and possibly the position of various devices on the body using off the shelf components. MATLAB, C 4T Rodriguez-Villegas,E. A modulated nonlinear capacitor can act as an amplifier. Parametric amplifiers do not involve transistors, are power efficient are narrowband and operate at extremely high frequencies. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 32 of 63 PID Supervisor Room Description 1400967 Rodriguez-Villegas,E. (with Chen,G.) 914 Monitoring of human behaviours in free individuals is arguably one of most popular application in wearable technology. The aim of this project is to explore 1) a potential set of human behaviours could be recognised from tracheal sound by using our awardwinning miniaturized wearable tracheal sound monitor and 2) lightweight/real-time pattern recognition techniques to differentiate those behaviours on an microcontroller. The student is expected to have good programming skills in C/C++ or Matlab, and background on the embedded systems. For more information on the sensor used in this project please see http://www.acupebble.com 914 This project looks at the implementation of an RFID system in microwave and/or UHF frequency range for high bit rate data transmission in wearable devices. Applications such as monitoring heart rate and body temperature will be investigated. The student will undertake a comprehensive review of current state of the art for wearable devices along with the state of the art UHF/Microwave RFID systems. Implementation of a programmable UHF/MW RFID system for ultra-low power high bit rate data transmission will be investigated using off the shelf components. MATLAB, C 4T Human Behaviour Monitoring Using a Single Wearable Sensor 1400975 Rodriguez-Villegas,E. (with Saremi-Yarahmadi,S.) 3E4T UHF and microwave RFID systems for wearable devices 1400977 Rodriguez-Villegas,E. (with Saremi-Yarahmadi,S.) Rodriguez-Villegas,E. (with Saremi-Yarahmadi,S.) 3E4T Saremi-Yarahmadi,S. (with Rodriguez-Villegas,E.) 1A Flexible coil antenna structures I 1400978 the student will look at the design, simulation, and experimental measurements on a range of flexible antenna and inductors coils. This will be complemented by simulation and measurements on the loading effects of human body on such structures. The aim is to construct an experimentally validated simulation model for such antenna and inductor structure. MATLAB, ANSYS HFSS, AWR Microwave Office Saremi-Yarahmadi,S. (with Rodriguez-Villegas,E.) UHF and microwave RFID systems for wearable devices 1400977 914 4T Flexible coil antenna structures 1400975 the student will look at the design, simulation, and experimental measurements on a range of flexible antenna and inductors coils. This will be complemented by simulation and measurements on the loading effects of human body on such structures. The aim is to construct an experimentally validated simulation model for such antenna and inductor structure. MATLAB, ANSYS HFSS, AWR Microwave Office 1A Flexible coil antenna structures I 1400978 914 Saremi-Yarahmadi,S. (with Rodriguez-Villegas,E.) Flexible coil antenna structures 11 October 2014 4T This project looks at the implementation of an RFID system in microwave and/or UHF frequency range for high bit rate data transmission in wearable devices. Applications such as monitoring heart rate and body temperature will be investigated. The student will undertake a comprehensive review of current state of the art for wearable devices along with the state of the art UHF/Microwave RFID systems. Implementation of a programmable UHF/MW RFID system for ultra-low power high bit rate data transmission will be investigated using off the shelf components. MATLAB, C the student will look at the design, simulation, and experimental measurements on a range of flexible antenna and inductors coils. This will be complemented by simulation and measurements on the loading effects of human body on such structures. The aim is to construct an experimentally validated simulation model for such antenna and inductor structure. MATLAB, ANSYS HFSS, AWR Microwave Office the student will look at the design, simulation, and experimental measurements on a range of flexible antenna and inductors coils. This will be complemented by simulation and measurements on the loading effects of human body on such structures. The aim is to construct an experimentally validated simulation model for such antenna and inductor structure. MATLAB, ANSYS HFSS, AWR Microwave Office Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 33 of 63 PID Supervisor Room Description 1401015 Spence,R. (with Witkowski,C.M.) 913A The instrument could, for example, be worn by the performer. The development and evaluation of a novel and unusual musical instrument to be constructed using a 3D printer 1401016 Spence,R. (with Witkowski,C.M.) Product Space Explorer for Smart Purchasing 1401017 Spence,R. (with Witkowski,C.M.) Computed Aided Design Space Exploration 1401018 Spence,R. (with Witkowski,C.M.) Looking-Glass 4D4T4J 913A The invention, design, implementation and evaluation of an interface (e.g., web page) to permit the exploration of many products (e.g., 1000 cameras) and the eventual selection of one to purchase. 4D4T4J 913A 4D4T4J 913A 4D4T4J The development of an interface to support the design of an artifact (e.g., electronic circuit) or scheme (e.g., a plan to reduce crime in a selected geographical area) in which the numbers of designable parameters and performances of interest are large (at least 10). There will be tolerances on parameters; there may be flexibility in performance requirements; and there will be huge benefits to be gained by allowing a designer to explore (e.g., the effect of parameter variation) interactively in order to gain some insight into, for example, inherent trade-offs in performances. Person A is seated in front of an LCD in transparent mode. The view captured by video camera A is transmitted - perhaps hundreds of miles - to the LCD (in display mode) that person B is viewing. This condition lasts (say) 1/30th of a second. In the next such interval the reverse occurs, and the sequence is repeated continuously. The effect is that each person sees the other as if they were on the other side of a glass wall. It would be very useful if both persons could draw on the wall and interact with the drawn diagram or text. For a simulation of the system see http://www.youtube.com/watch?v=yncgaaKvpQY and start at 7minutes. The simulation ends at 8.04 minutes 1400957 System-on-chip FPGAs combine dual ARM processors with flexible, uncommitted programmable logic. By integrating these components designers hope to achieve better power efficiency and performance. Stott,E. (with Cheung,P.Y.K.) Reconfigurable system-on-chip 4D4T4J This project will develop a computing platform, based on an SoC FPGA, that reduces power consumption by dispatching tasks to custom kernels implemented in soft logic. The project will feature algorithm development in software and digital logic. Familiarity with Linux, FPGAs and hardware description languages will be an advantage. Once the basic hardware-software interaction is developed there are many interesting questions that could be addressed to achieve higher marks. For example, when are the benefits of a hardware kernel outweighed by the penalties of data transfer? How can hardware components be standardised and built into libraries for easy reuse? 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 34 of 63 PID Supervisor 1400958 Stott,E. (with Constantinides,G.A.) Power measurement breakdown in digital logic Room 4D4T4J Description The power consumed by a digital circuit is dependent upon a number of parameters, including supply voltage, switching activity and leakage current. It is easy to measure the overall power but how can we get a breakdown of the proportion of power used by different parts of the circuit? Knowing this would allow us to optimise operation of complex systems-on-chips to improve efficiency. This project will find out if it is possible to estimate power consumption in digital circuits by probing a small proportion of the internal and external signals and analysing the statistical properties of the trace that is returned. The main skills required are digital circuit design with hardware description languages and data analysis. Generating some indicative results should be straightforward, but to achieve high marks the candidate could consider how the probe locations could be selected while the circuit is being designed and produce tools that would automate the process. 1400866 Thomas,D.B. A Garbage Collector for Heterogenous Embedded Systems 903 3E4D4T3I4J Contemporary FPGAs used in embedded systems contain both programmable logic and tightly coupled ARM CPUs, allowing software to talk directly to hardware in the same chip. This is encouraging the development of increasingly complex System-on-Chips (SoC), involving multiple CPUs and accelerators developed using High Level Synthesis (HLS) tools. As HLS tools become more sophisticated it is becoming necessary to dynamically allocate memory blocks in hardware, then share them between the hardware and software logic in the SoC, but there is little system support for this. The goal of this project is to create a garbage collector (dynamic memory manager) which spans both software and hardware compute devices, can track the movement of references between the two, and reclaim memory when no more references exist. Ideally this garbage collector would run in hardware, and proceed in parallel with the users of dynamic memory allocation. The target platform would be a Xilinx Zynq device, and the main languages used would be C/C++, VHDL, and one or more HLS tools (e.g. Vivado HLS or LegUp). Basic competence in C is necessary, and some experience in VHDL preferred. Experience with embedded systems and/or parallel programming would also be useful. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 35 of 63 PID Supervisor Room Description 1400867 Thomas,D.B. 903 Work-stealing is a technique for scheduling parallel programs in multi-core hardware, which is very easy to program in software (using languages such as Cilk or Intel Threaded Building Blocks), and can be efficiently implemented in modern shared memory CPUs. A particular advantage is that work-stealing allows parallelism to be described and exploited at run-time: programmers merely describe what could run in parallel, then it is up to the work-stealing library to determine how best to take advantage of the CPUs available. Work-stealing for Zynq FPGAs 3E4D4T3I4J The goal of this project is to take these ideas and apply them to the field of hardware-software co-design. The main target is the Xilinx Zynq chip, which combines two ARM cores and a large amount of FPGA logic resources in one chip. Usually programmers would explicitly decide which parts run in software, which parts in the FPGA logic, when exactly each part should run, and whether they run in parallel. However, here we will develop a run-time scheduling approach based on work-stealing, whereby software indicates tasks which are ready to run, and hardware will steal those tasks when idle. An existing pure software work-stealing run-time is available, so the goal is to extend it to enable scheduling of hardware tasks at run-time. Stretch goals would be to enable hardware tasks to generate tasks which can be stolen by software, and to optimise the hardware responsible for managing stealing of tasks. This is essentially an embedded systems project, so knowledge of C is required, and a good working knowledge of digital logic is needed. Experience with VHDL or Verilog is reccomended, but not a requirement. Courses which would be useful (but not required) for this project would be VHDL and Logic Synthesis, and High Performance Computing for Engineers. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 36 of 63 PID Supervisor Room Description 1400888 Thomas,D.B. 903 It is common to build soft processors out of FPGA logic, such as Altera's NIOS and Xilinx's Microblaze. Soft processors are usually relegated to running an OS and co-ordinating other accelerators in the FPGA, as they have limited single-threaded performance. This project will explore the creation of new soft processors which are designed to achieve multi-threaded performance, with ISA level support for lightweight threads, and the ability to exploit fine-grain task-level parallelism. Multi-threaded soft processors 3E4D4T3I4J This project requires the design and creation of a CPU from scratch, with the expected outcome to have a multi-threaded CPU working in hardware. Stretch goal would be to have it optimised and running some real applications. Some knowledge of CPU architecture is necessary, and decent digital analysis skills will be needed in order to design the micro-architecture. Experience with advanced CPU architecture and VHDL would be useful, along with a basic grounding in parallel programming. By far the most important thing is to just be interested in designing and building your own CPU. 1400951 Thomas,D.B. (with Drane,T.) Using Polynomials in Formal Verification 903 4D4T4J This is an industrially led research project that focuses on the crucial activity of formal verification when designing hardware; proving functional correctness of large hardware designs is a challenge and verification takes up a significant proportion of design cycles. However, recent advances in the area of algebraic geometry offer a new angle on such problems, with the potential to verify the seemingly unverifiable: - http://www.ece.utah.edu/~kalla/papers/86-XA763.pdf - http://cas.ee.ic.ac.uk/people/gac1/pubs/TheoDACKC11.pdf This project would suit students strong in mathematics, and will involve the use of bespoke algebraic geometry tools. This project will involve working with an industrial researcher (as well as the supervisor), so please come for discussion in room 903 on Monday 13th 12:30-13:30 when they will also be available. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 37 of 63 PID Supervisor Room Description 1400952 Thomas,D.B. (with Drane,T.) 903 This is an industrially led research project that addresses the observation that within deep arithmetic pipelines there may be redundant hardware; for example there may be branches which are actually never entered, variables which are constant, or stuck at bit, all regardless of which inputs enter the pipeline. It is also common that the set of possible inputs to the pipeline is restricted i.e. has a constrained input domain. Synthesis tools don’t optimise these redundancies, but exploiting these opportunities can provide significant hardware quality improvements. Constrained Input Domain Synthesis 4D4T4J The idea is that one could hook up a verification and synthesis tools to discover these situations and thus automatically 'scrub' code of dead branches and logic. This project would suit students with either experience or an interest in RTL (VHDL/Verilog), synthesis and/or verification. This project will involve working with an industrial researcher (as well as the supervisor), so please come for discussion in room 903 on Monday 13th 12:30-13:30 when they will also be available. 1400953 Thomas,D.B. (with Drane,T.) Power Estimation of RTL 903 3E4D4T3I4J Power measurement of RTL typically requires long simulations, using tools with very expensive licenses usage, so it would be preferable if analytic probabilities could be calculated for switching information. Techniques do exist which attempt to perform this task but require further research in order to allow tools which deliver usable results on industrial strength benchmarks in a reasonable time. This project will look into the use of BDDs and interval arithmetic, in order to try to develop such tools. Some background papers are: http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=1542949&url=http%3A%2F%2Fieeexplore.ieee.org%2Fiel5%2F10358%2 F32955%2F01542949 - http://www.cs.virginia.edu/~robins/papers/DAC_2012_final_Gabe.pdf This project would suit students interested in the fundamentals of digital logic synthesis, with good mathematical skills, decent programming skills (in any language, matlab is fine), and some knowledge of RTL simulation. This project will involve working with an industrial researcher (as well as the supervisor), so please come for discussion in room 903 on Monday 13th 12:30-13:30 when they will also be available. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 38 of 63 PID Supervisor Room Description 1401002 Toumazou,C. 912A Antimicrobial resistance has become an epidemic. This project will investigate for the first time development of a non-invasive platform to measure electromagnetic signals released from bacterial DNA. This will be used for detection of specific bugs which are thought to be resistant to antibiotics. Wireless detection of bacterial DNA 1401003 Toumazou,C. An Integrative Biological and Quantitative Approach to Analysing Wrinkles 1401004 4D4T 912A 4D4T Toumazou,C. 912A The See-Hear device for Dementia 4D4T4J 1401000 Toumazou,C. (with Kulasekeram,N.) A Wireless Transceiver for a nerve, chemical sensor and GUI 1401001 Toumazou,C. (with Kulasekeram,N.) A wireless transceiver for a nerve, chemical sensor and battery power management 11 October 2014 912A 4D4T4J 912A 4D4T4J The project involves the development of a software algorithm to complement a commercialised GENEU database which works out the difference between ageing skin based upon skin DNA. The database comprises a number SKIN DNA Genes that enabled selection of personalised active ingredients for skin care. The algorithm will assist in the selection of the correct antiageing serum for that individual. This project will involve the design and development of an optical based hearing device which prompts people with dementia to recognise individual friends and family members. The device can be worn on either wireless headsets, spectacles or around the ear. This project is part of exciting work that's aim is to develop technology that taps into the nervous system. By doing so it aims to monitor and regulate appetite in overweight people by tapping into the nerve responsible for it. This project is an exciting opportunity to work with neuroscientists and engineers of many disciplines. The aim of the project is to design a wireless interface to a chemical sensor which is expected to provide a maximum output voltage of 10 millivolts at a frequency of 10Hz. This signal needs to be transmitted to a computer terminal or mobile handheld device . The intention is to use commercially available wireless modules for the purpose of transmitting and receiving the required data as well as develop custom PCB’s for interfacing with the sensor. The next stage is to write a software GUI which would interface with data received from a wireless receiver, The GUI would be used to display the real time chemical sensor data which has been transmitted from a maximum of 100 metres Skills to be gained include, Interfacing, Wireless protocol, Wireless modules, Software GUI and API programming, This project is part of exciting work that's aim is to develop technology that taps into the nervous system. By doing so it aims to monitor and regulate appetite in overweight people by tapping into the nerve responsible for it. This project is an exciting opportunity to work with neuroscientists and engineers of many disciplines. The aim of this project is to design and develop a chemical sensor (PCB) and an energy efficient method to transmit chemically sensed data which is only made available every 100 milliseconds. So a solution which is based on storing large amounts of data and transmitting it, to a receiver, which could be either a computer terminal or a mobile hand held device. The wireless solution needs to be battery powered, as the end solution needs to be portable. The battery should be used in the most efficient way possible. Skills to be gained include, Interfacing , Wireless modules, Wireless protocol, Power management Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 39 of 63 PID Supervisor 1401013 Ye,R. Effort-Efficient Approximate Instructions Design for Microprocessors Room 3E4D4T3I4J Description Approximate computing is an emerging design paradigm that is able to trade off computation quality (e.g., accuracy) and computational effort (e.g., energy) by exploiting the inherent error resilience of a wide spectrum of applications, such as multimedia, digital signal processing, and the broad class of applications referred to as Recognition, Mining, and Synthesis (RMS). By relaxing the numerical equivalence between the specification and implementation of such applications, designers could use more energy/performance-efficient imprecise hardware systems and thus achieve significant energy saving and/or performance improvement. As a necessary part of microprocessors, Instruction Set Architecture (ISA) is visible to compilers, serving as the interface between software and hardware. To support approximate computing in microprocessors, it is essential to design a sophisticated approximate ISA that is able to conduct effort-efficient computations with very limited degradation of computation quality, such that compilers can accordingly generate “approximate” programs. Dr. Rong Ye, Room 905EE This project will investigate how to design a set of approximate arithmetic instructions for microprocessors. To achieve high marks, the candidate should consider how to reuse the data paths of existing instructions and hence reduce hardware overhead of approximate instructions design. Skills expected: The skills expected from students include digital circuit design with hardware description languages and background knowledge on computer architecture. 1400934 Brookes,D.M. Speech level estimation in acoustic noise 1400935 Brookes,D.M. Adaptive power spectrum estimation of non-stationary acoustic noise 1400936 Brookes,D.M. Optimal IIR filter design 1400937 Brookes,D.M. Robust Plosive Detection in Noisy Speech 11 October 2014 814 4D4T4J 814 4D4T4J 814 4D4T4J 814 3E4D4T3I4J Many speech enhancement algorithms need to know the active level of the wanted speech signal, i.e. the average power during periods of speech. This is easy to determine for noise-free speech but much harder when noise is present. The aim of the project is to develop and evaluate a reliable algorithm for estimating the active level of a speech signal that is corrupted by significant background noise. The project will be done in MATLAB. A crucial part of speech enhancement algorithms is the estimation of the power spectrum of the background acoustic noise. Conventional enhancement systems assume that the noise has a quasi-stationary power spectrum that changes slowly with time. In fact it is often the case that the background includes transient high amplitude noise sources as well. The goal of this project is to extend a conventional noise estimation algorithm to account for such transient noise components. Many of the filters that are needed when processing speech and other acoustic signals are specified as continuous time filters in the s-domain. Before they can be used, they need to be converted into discrete-time digital filters operating at a prescribed sampling frequency. This project will develop and evaluate methods of performing this translation that are optimal in a least squares sense. The project will use MATLAB. Plosives are sounds in human speech such as "p", "b", "t" in which the vocal tract is briefly blocked so that pressure can build up behind the blockage. The pressure is then abruptly released resulting in an impulsive burst of acoustic energy. Plosives are characterised by their short duration and their broadband spectrum; they are quite hard to detect because they are very short and realtively low energy. The goal of this project is to detect plosives reliably even in high levels of acoustic noise. The project will use MATLAB. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 40 of 63 PID Supervisor Room Description 1400931 Clerckx,B. 816 Wireless communication and power networks have enabled a plethora of novel applications in the last years. Both make use of the same and unique RF medium, but have been so far designed independently from each other. In this project, we will investigate the potential of performing a Joint Wireless Information and Power Transfer that wirelessly transfers energy jointly with information in wireless networks. The area, if found feasible, will create a new paradigm shift in future capacity and energy efficient wireless communication and energy networks (and in particular in machine-to-machine and Internet of Things applications), by viewing them as a single network designed under a unified framework and by overcoming the energy constraint of wireless devices through the transfer and harvesting of RF energy. Contrary to current wireless communication networks, interference is viewed as a source of energy to be harvested rather than mitigated. In the project, the student will investigate the role of interference in joint wireless information and power networks and identify the fundamental limits of such networks as well as suitable interference management techniques. The student will get the opportunity to progressively get familiar with the latest research in wireless communications, optimization and energy harvesting, discuss and imagine the basic concepts of next generation wireless communication and power networks. Simultaneous Wireless Information and Power Transfer 3E4D4T3I4J Skills required: strong interests in communication and signal processing, algorithm development, analysis, simulations, matlab programming The following references are helpful: • J. Park and B. Clerckx, “Joint Wireless Information and Energy Transfer in a Two-User MIMO Interference Channel,” IEEE Trans. Wireless Commun., vol. 12, no. 8, pp. 4210–4221, Aug. 2013. • J. Park and B. Clerckx, “Joint Wireless Information and Energy Transfer in a K-User MIMO Interference Channel,” accepted to IEEE Trans. On Wireless Comm. arXiv:1310.6870 • H. Son and B. Clerckx, “Joint Beamforming Design for Multi-User Wireless Information and Power Transfer,” accepted to IEEE Trans. On Wireless Comm. 1400933 Clerckx,B. MIMO for 5G Wireless Communication System 816 3E4D4T3I4J Wireless communication systems rely more and more on the use of multiple antennas at both the transmitter and the receiver (MIMO). MIMO is the core technology of 4G networks and is nowadays massively expanded to become a core element of 5G. In multi-user MIMO, multiple users share the same time/frequency resources and are separated in the spatial domain through the use of multiple antennas. Multi-user MIMO heavily relies on accurate channel state information (CSI) feedback, but having perfect transmitter side CSI (CSIT) is a challenging issue. Recently, a new transmission strategy has been shown to be more robust to imperfect CSIT than conventional approaches used so far in 4G systems. The objective of the project is to investigate the performance gain and potential benefits of that transmission strategy. Skills required: strong interests in communication and signal processing, algorithm development, analysis, simulations, matlab programming The following references are helpful: • C. Hao and B. Clerckx, “MISO Broadcast Channel with Imperfect and (Un)matched CSIT in the Frequency Domain: DoF Region and Transmission Strategies,” PIMRC 2013. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 41 of 63 PID Supervisor Room Description 1400932 Clerckx,B. (with Mitcheson,P.D.) 816 Wireless sensor network (WSN) has attracted a lot of attention in various machine-to-machine (M2M) and Internet of Things (IoT) applications such as smart home networking, smart metering, and healthcare due to its low deployment costs coupled with a high potential to sense and collect object-oriented data through a large number of cheap wireless sensor nodes. One of the main challenges in WSN is that nodes are resource-constrained, mainly due to battery limitation. To enable autonomous wireless devices, the project will investigate the feasibility of wireless power transfer. While receiver design for wireless power transfer has been relatively well researched, the design of suitable waveforms is much less known. In the project, the student will investigate how the waveform design influences the performance of wireless power transfer and identify suitable waveforms. The student will get the opportunity to progressively get familiar with the latest research in wireless communications, RF design and power transfer, discuss and imagine the basic concepts of next generation wireless power networks. 3E4D4T3I4J Waveform Design for Wireless Power Transfer Skills required: strong interests in communication/signal processing and RF design, algorithm development, analysis, simulations, RF measurement, matlab and PSpice programming The following references are helpful: • A. S. Boaventura and N. B. Carvalho, “Maximizing DC Power in Energy Harvesting Circuits Using Multisine Excitation,” 2011 IEEE MTT-S International Microwave Symposium Digest (MTT). • M. Pinuela, P. Mitcheson and S. Lucyszyn, “Ambient RF energy harvesting in urban and semi-urban environments” IEEE Trans. on microwave theory and techniques, vol 61, no 7, july 2013. • J. Park and B. Clerckx, “Joint Wireless Information and Energy Transfer in a Two-User MIMO Interference Channel,” IEEE Trans. Wireless Commun., vol. 12, no. 8, pp. 4210–4221, Aug. 2013. 1400960 Constantinides,A.G. (with Levine,J.M.) Brewing the perfect espresso 618 4D4T4J The taste of espresso depends upon a variety of brewing parameters, including how long the dry ground coffee is pre-infused, the temperature of the water and the profile with which pressure is applied. Selecting these variables in somewhat of an art form but there is surely scope to create an electronic, expert barista. You will design and build hardware and a controller capable of accurately adjusting brewing parameters. It should be able to produce sample brews for taste experiments; allowing the best-tasting formula to be researched. The skills that could be employed include electronic and electromechanical system design, embedded software development, control and scientific experiment design. At a minimum, the student will need to create hardware that can brew an espresso to a user specification. There are many opportunities for achieving higher marks, particularly if you can discover and publish the formula for the perfect espresso. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 42 of 63 PID Supervisor Room Description 1400996 Constantinides,A.G. (with Levine,J.M.) 618 The performance and lifetime of an integrated circuit is dependent upon operating and environmental conditions, which include temperature and supply voltage. To produce reliable devices it is important to characterise behaviour under all scenarios that are likely to be encountered in the field. The information can be used to develop techniques for making circuits more efficient, better performing and more reliable. Temperature and voltage characterisation of devices 4D4T4J This project will design an experimental rig for precisely and safely controlling the voltage and temperature of a circuit-undertest. An API will enable the hardware to be easily controlled by a top-level experiment procedure . The rig will then be used to carry out an investigation into performance and reliability in FPGA devices. The project will feature both hardware and software development and encompasses aspects from mechanical components all the way up to a potential web interface . Thus it can be adapted to fit a student's skills, though for high marks I would like to see a fully-integrated, vertical solution that could be released under an open-source license for use by the research community. To demonstrate its usefulness, the system could be used to carry out (and possibly publish the results of) accelerated life testing or statistical characterisation of devices. 1400892 Dai,W. Faster GPS 1400893 Dai,W. Radar Imaging via Compressed Sensing 811 3E4D4T3I4J 811 3E4D4T3I4J The Global Positioning System (GPS) is widely used in everyday life. In the process of calculating the distance from satellites, a GPS receiver computes hundreds of millions of multiplications, which can be slow and power consuming. The key ingredient in reducing the computational complexity is the FFT which translates correlations in time domain into multiplications in the frequency domain. Despite its huge success, recent advances in sparse reconstruction suggest that the computational cost can be further reduced if one exploits the sparse nature of the synchronization problem. This project aims at getting familiar with the relevant concept and theory, implementing the fast synchronization algorithm in Matlab (or C), and demonstrating the idea using synthetic data. As an extension, synchronization with the presence of multipath signals may also be considered. The goal of this project includes 1) getting familiar with the basic principles behind radar imaging, in particular, Synthetic Aperture Radar (SAR) imaging; 2) learning the background of compressed sensing; 3) understanding how to apply compressed sensing to radar imaging; 4) if time allows, possible improvements when applying recent progresses in compressed sensing techniques. References to start: * Baraniuk, R. and P. Steeghs, "Compressive radar imaging," Proc. IEEE Radar Conf., 128-133, Boston, MA, Apr. 2007. * Herman, M. A. and T. Strohmer, "High-resolution radar via compressed sensing," IEEE Transactions on Signal Processing, Vol. 57, No. 6, 2275-2284, Jun. 2009. * W. U. Bajwa, K. Gedalyahu and Y. C. Eldar, "Identification of Parametric Underspread Linear Systems and Super-Resolution Radar", IEEE Transactions on Signal Processing, vol. 59, no. 6, pp. 2548-2561, June 2011. * O. Bar-Ilan and Y. C. Eldar, "Sub-Nyquist Radar via Doppler Focusing",to appear in IEEE Transactions Signal Processing. 1400894 Dai,W. Learning from Data 11 October 2014 811 3E4D4T3I4J This will be a self-designed project by the student in the sense that the student will need to select a dataset that she/he is most interested in and identify technical tools to learn useful information from the data. The data could be the publicly available data from data.gov.uk or financial data or any other things that are legal. The student will be required to study the literature, identify the technical tools that can be used to analyse the data, and justify them. The programming platform (Matlab or C etc.) for numerical tests is also the student's choice. The course "Topics in Large Dimensional Signal Processing" is not required but highly recommended. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 43 of 63 PID Supervisor Room Description 1400895 Dai,W. 811 This will be a self-designed project by the student in the sense that the student will need to select a dataset that she/he is most interested in and identify technical tools to learn useful information from the data. The data could be the publicly available data from data.gov.uk or financial data or any other things that are legal. The student will be required to study the literature, identify the technical tools that can be used to analyse the data, and justify them. The programming platform (Matlab or C etc.) for numerical tests is also the student's choice. The course "Topics in Large Dimensional Signal Processing" is not required but highly recommended. Learning from Data 1400896 Dai,W. Sparse Channel Estimation 3E4D4T3I4J 811 3E4D4T3I4J Channel estimation is one of the keys for reliable communications. There have been many techniques proposed for sparse channel estimation in the literature. This project is designed to get familiar with the benchmark methods, compare and test them, and explore new techniques for modern communication systems, e.g. massive MIMO systems. Matlab programming will be required. References to start: * Waheed U. Bajwa, Jarvis Haupt, Akbar M. Sayeed, and Robert Nowak, "Compressed Channel Sensing: A New Approach to Estimating Sparse Multipath Channels". * Christian R. Berger, Shengli Zhou, James C. Preisig, and Peter Willett, "Sparse Channel Estimation for Multicarrier Underwater Acoustic Communication: From Subspace Methods to Compressed Sensing". 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 44 of 63 PID Supervisor Room Description 1400842 Dragotti,P.L. 802 Image Based Rendering (IBR) is a promising technique to render novel views from a set of available multi-view images. Instead of rendering views of 3-D scenes by projecting objects and their textures, new views are rendered by interpolating available nearby images. The advantage of such a method is that it produces convincing photorealistic results since the interpolated viewpoints are obtained through combinations of real images. The main drawback is the fact that a huge amount of data needs to be captured. Clearly, knowledge of the scene geometry reduces the number of images required. The estimation of depth and geometry has been normally achieved using passive stereo with multiple cameras. Unfortunately, 3-D reconstruction techniques from passive cameras, are still not reliable and do not work well in many cases. This fact has profoundly limited the use of IBR ideas. Recent advances in sensing technologies may soon allow large-scale deployment of 3-D cameras using active depth sensing systems. These cameras are able to estimate depth and geometry with good accuracy and reliability, and for this reason can be very useful in IBR. Image Based Rendering with depth sensors 3E4T3I4J In future mobile phones will be equipped with both a camera and an active depth sensing device. This will allow IBR to be performed easily and effectively on a mobile device. The aim of the project is to build a simple 3-D active sensor device which would mimic devices that will be put on mobile phones and then try to combine the data provided by the active device with the images obtained in order to achieve good IBR results. 1400843 Dragotti,P.L. Sparse Representations of Signals in the Union of the Discrete Cosine and Haar Bases 802 4D4T4J The problem of reconstructing signals or images from a small number of measurements is an old and important one, and finds application in many areas of signal processing and communication. Standard approaches are, to some extent, based on classical Fourier theory and, for this reason, yield poor reconstruction when the signal is not bandlimited or is under-sampled. Recently, it has been shown that it is possible to reconstruct many signals from a limited number of measurements. This is possible because the signals have a sparse representation in a certain basis or frame, or in a certain parametric space. The reconstruction process is based on reconstruction algorithms with linear or polynomial complexity. Because of the lowcomplexity of the reconstruction process, these new schemes can potentially have a profound impact in many signal/image processing applications such as medical imaging, image super-resolution and spread-spectrum communication. The aim of this project is to understand this very advanced techniques and to use these new schemes to enhance the resolution of low quality signals or images. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 45 of 63 PID Supervisor Room Description 1400861 Dragotti,P.L. 802 The neurophysiological image usually comes in the form of three-dimensional images (x-y-time). The analysis of neurophysiological images often faces two problems: (i) identifying the region of interest (ROI) (ii) Inferring the timing of action potentials (APs) (spike trains) from ROI. Traditionally, these two problems have been treated independently. The project considers a unified approach based on the observation that each neuron occupies a spatially localized area, and the temporal signals from that area correspond to one spike train. Moreover, since each neuron is spatially localized and has branch-like dendrites, the ROI can be inferred by forcing this spatial structure on the pixels related to a certain neuron. This project will exploit the latest single-pixel spike detection technique (finite rate of innovation (FRI) methods) together with classical machine learning /image processing techniques for identifying the ROI. The aim of the project is to introduce the student to the FRI theory and develop a simple algorithm that combines the ROI detection and AP detection together. 3E4T3I4J Spike detection with adaptively identified ROI for biological image analysis 1400945 Evers,C. (with Naylor,P.A.) 3D Rendering of Acoustic Maps for Robot Auditiion 4D4T Robotic applications, including human-robot interaction, are heavily dependent on the robot’s ability to explore and learn the surrounding environment. Mapping is used to integrate the acquired information into a holistic representation of the robot's world view. Acoustic maps reflect the robot’s belief of surrounding speakers and sound sources. Furthermore, sound reflections off walls and obstacles can be used to infer knowledge about the surrounding room geometry. In this project, the student develops a tool set for 3D rendering of acoustic maps for data representation in robot audition. In order to evaluate the approach, the student needs to consider practical aspects such as spatial resolution of the resulting maps as well as computation / rendering time. 1400998 Leung,K.K. Communications Among Vehicles on the Roads 612 3E4D4T3I4J Inter-vehicle communication networks or Vehicular Ad-hoc Networks (VANETs), a sub-class of Mobile Ad-hoc Networks (MANETs), is an area attracting a lot of interests in the research community recently, as they will perform crucial functions in road safety, detection of traffic accidents and reduction of traffic congestions. However, automotive ad-hoc networks will behave in fundamentally different ways from the predominated models in MANET research. Driver behaviour, mobility constraints and high speeds create unique characteristics in VANETs. These characteristics have important implications on the designs of efficient communication protocols and network control functions for VANETs, and it is therefore of vital importance that they are designed and studied carefully. As parts of the research projects www.commsp.ee.ic.ac.uk/~wiser/message/ and www.commsp.ee.ic.ac.uk/~wiser/phdscheme, our overall goal is to design network protocols and control functions to ensure efficient communications in VANETs. In particular, the medium-access-control (MAC) protocol, which controls vehicles accessing the communication channels, in the IEEE 802.11p standard is practically identical to that in other types of WiFi networks. That is, the standard does not consider or exploit the unique characteristics of VANETs such as vehicular traffic density and velocity in the protocol designs, which may provide possible performance gain. The main objective of this project is to examine the network protocols in the IEEE 802.11p standard, identify possible ways to enhance them and quantify their performance gain by network simulation or mathematical analysis. There is no specific pre-requisition for this project, but basic knowledge in wireless networks and network protocols would be helpful. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 46 of 63 PID Supervisor Room Description 1400916 Ling,C. 815 OFDM provides greater immunity to multipath fading and impulse noise, and eliminates the need for equalizers. However, a major drawback of OFDM transmission is its high peak-to-average power ratio (PAPR) of the transmit signal. This project will investigate some of the important PAPR reduction techniques for OFDM such as amplitude clipping and filtering, coding, partial transmit sequence, selected mapping, interleaving, tone reservation, tone injection, and active constellation extension. Students who wish to get Distinction would have to address the problem of PAPR reduction in OFDMA and MIMO-OFDM. Required backgrounds are wireless communications and coding. Peak-to-Average Power Reduction for orthogonal frequency-division multiplexing (OFDM) 1400917 Ling,C. MIMO Wireless Communication 1400918 Ling,C. Semi-Deterministic compressed sensing 1400991 3E4D4T3I4J 815 3E4D4T3I4J 815 3E4D4T3I4J Looney,D. (with Mandic,D.P.) Complexity science meets visual perception: Selective attention in the perceptions of images and videos 3E4D4T3I4J Multi-input multi-output (MIMO) communication significantly increases the bandwidth of wireless communications. In this project, you will study a few techniques of MIMO detection where the signals are jointly detected. These include zero-forcing, minimum mean-square error, interference cancellation, and maximum-likelihood detection. The aim is to deal with the issue of implementation complexity. It’s MATLAB-based and good background on wireless communication is required. Compressed sensing is a revolutionized technique for simultaneous sampling and compression of signals. It goes beyond the classical Nyquist sampling and predicts that a sparse vector in high dimensions can be recovered from what was previously believed to be incomplete information. Sensing matrices play a crucial role in compressed sensing. Although random matrices (i.i.d. Gaussian or Bernoulli) have been proved to good for compressed sensing, they are heavy in computation and storage; worse, they offer no guarantees in practice. In this project, you will study deterministic sensing matrices by using coding theory, sequence design, or de-randomizing structured random matrices, and explore the applications to MRI, Terahertz imaging etc. Programming skills are required for successful completion of this project. The trajectories of eye gaze can be related to different cognitive tasks, as illustrated in the seminal "Yarbus experiment" in 1960s. For instance, causal scanning of an image produces relatively random eye-tracks, whereas upon executing a cognitive instructions (look for ages of people, how affluent the family is) the eye gaze is modulated according to the instruction. The student will use our gaze tracker to perform dynamical complexity analysis of eye-gaze, according to different cognitive instructions. This will be performed for various images (from natural to abstract) and over a range of subjects and measured variables (x,y eye gaze, pupil diameter, eye dwelling). In this way, the 'complexity loss theory', which states that the complexity of physiological responses in living organisms reduces under constraints (ageing, illness) will be rigorously tested. Applications in 'guilty knowledge' scenarios in interview situations, together with those in fatigue monitoring, and the quality of web-design will be investigated. References: [1] A. Yarbus "Congitive Visual Attention", 1971 [2] M. U. Ahmed and D. P. Mandic, "Multivariate multiscale entropy: A tool for complexity analysis of multichannel data'', Physical Review E, vol. 84, no. 6, pp. 061918-1 -- 061918-10, 2011. [3] D. Looney, M. U. Ahmed and D. P. Mandic, "Human Centred Multivariate Complexity Analysis," Natural Intelligence Magazine, vol 1, no. 3, pp. 40-42, 2012. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 47 of 63 PID Supervisor Room Description 1400990 Mandic,D.P. 813 Standard complex-valued adaptive filters combine the amplitude and phase information to produce their estimates. However, in many applications the amplitude information is either not important or can be misleading (e.g. due to sensor different signal levels in Electroencephalography), and it is the phase information that needs to be estimated accurately. This project will investigate a least-mean-phase approach to adaptive filtering, which uses phase to perform adaptive estimation of signal and system parameters. This is natural in both human-centred studies (e.g. human hearing system is tuned for phase information) and also in e.g. communication systems. The project will introduce widely linear adaptive filters in the phase domain, in order to make such estimation optimal for both complex circular and complex noncircular data. The project will involve a case study on frequency and phase estimation in smart grid, and a preliminary study of a direct phase estimation without the necessity for the Clarke transform. The phase-only vs. amplitude-only estimation will be investigated via the least-mean-magnitude-phase (LMMP) approach. Phase-only adaptive filtering in smart grid applications 3E4D4T3I4J Familiarity with adaptive filtering is a prerequisite. References: [1] A. Tarighat and A. H. Sayed, "Least mean-phase adaptive filters with application to communications systems", IEEE Signal Processing Letters, vool. 11, no. 2, pp. 220-223, 2004. [2] D. P. Mandic and V. S. L. Goh, "Complex valued nonlinear adaptive filters: Noncircularity, widely linear, and neural models", Wiley 2009. [3] Y. Xia, S. C. Douglas and D. P. Mandic, "Adaptive frequency estimation in smart grid applications: Exploring noncircularity and widely linear estimators'', IEEE Signal Processing Magazine, vol. 29, no. 5, pp. 44-54, 2012. [4] S. C. Douglas and D. P. Mandic, "The Least Mean Magnitude Phase Adaptive Filter", Proc. ICASSP, 2011. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 48 of 63 PID Supervisor Room Description 1400992 Mandic,D.P. 813 This project will investigate adaptive prediction schemes suitable for the operation on vector sensors, such as the threedimensional wind signal. The forecasting will be conducted in the quaternion domain, benefiting from the desirable properties of quaternion algebra (division algebra), and for a class of stochastic gradient adaptive prediction algorithms. Short term forecasting of threedimensional wind profile for renewable energy applications 4D4T4J Recent advances in the statistics of quaternion variable (pseudocovariance, involutions, noncircularity) will be used to design enhanced learning algorithms. In addition, the use of an additional atmospheric variable (air density, temperature) within the four-dimensional quaternionic model will be investigated. The algorithms will be tested on real world 3D wind recordings. The student is also expected to conduct their own recordings based on our own 3D anemometers. Simulations will include several case studies to investigate the usefulness of the approach in the control and vibration suppression of wind turbines. The student should be familiar with spectrum estimation and adaptive signal processing. Literature: [1] C. Cheong-Took and D. P. Mandic, 'The Quaternion LMS Algorithm for Adaptive Filtering of Hypercomplex Processes', IEEE Transactions on Signal Processing, vol. 57, no. 4, pp. 1316–1327, 2009. [2] D. P. Mandic et al. 'Complex Valued Prediction of Wind Profile Using Augmented Complex Statistics', Renewable Energy, vol. 34, no. 1, pp. 196–201, 2009. 1400993 Mandic,D.P. Short term forecasting of threedimensional wind profile for renewable energy applications 813 4D4T This project will investigate adaptive prediction schemes suitable for the operation on vector sensors, such as the threedimensional wind signal. The forecasting will be conducted in the quaternion domain, benefiting from the desirable properties of quaternion algebra (division algebra), and for a class of stochastic gradient adaptive prediction algorithms. Recent advances in the statistics of quaternion variable (pseudocovariance, involutions, noncircularity) will be used to design enhanced learning algorithms. In addition, the use of an additional atmospheric variable (air density, temperature) within the four-dimensional quaternionic model will be investigated. The algorithms will be tested on real world 3D wind recordings. The student is also expected to conduct their own recordings based on our own 3D anemometers. Simulations will include several case studies to investigate the usefulness of the approach in the control and vibration suppression of wind turbines. The student should be familiar with spectrum estimation and adaptive signal processing. Literature: [1] C. Cheong-Took and D. P. Mandic, 'The Quaternion LMS Algorithm for Adaptive Filtering of Hypercomplex Processes', IEEE Transactions on Signal Processing, vol. 57, no. 4, pp. 1316–1327, 2009. [2] D. P. Mandic et al. 'Complex Valued Prediction of Wind Profile Using Augmented Complex Statistics', Renewable Energy, vol. 34, no. 1, pp. 196–201, 2009. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 49 of 63 PID Supervisor Room Description 1400994 Mandic,D.P. 813 Physiological sensing involves algorithms with multiple and redundant inputs, for instance the heart rate and respiration are coupled quantities. The form in which the useful information is encoded within the available data is unknown, and therefore data-mining algorithms are required. Multivariate Partial Least Squares for the modelling of stress based on human physiological data 1400995 Mandic,D.P. Harnessing the power of renewable energy sources: Complexity science answers 3E4D4T3I4J This project will analyse the data using partial least squares (PLS), an algorithm that maps both the output and input variables to a new "shared" latent component space, thus identifying common features between the independent and dependent variables that aid the regression. A starting point would be to implement multivariate regression using PLS, and then to derive a PLS version of well-known algorithms such as the minimum variance portfolio. The project will also involved recording from humans using the electroencephalogram and respiration sensors. 813 4D4T The dynamics and latent coupling between the atmospheric variables (wind, temperature, solar irradiation, air density, humidity) have a direct influence on the power output from the renewables (wind, solar). While the numerical weather forecasts are precise, they are obtained from satellite images and may not be accurate for every particular site. This project will look into joint dynamic complexity of atmospheric variables, in order to assess the relationship between the amount of 'structure' in data and the extent to which renewable energy can be generated from the wind and solar plants. [1] Costa M, Goldberger AL, Peng C-K. "Multiscale entropy analysis of complex physiologic time series". Physical Review Letters 2002;89:068102. [2] M. U. Ahmed and D. P. Mandic, "Multivariate multiscale entropy: A tool for complexity analysis of multichannel data'', Physical Review E, vol. 84, no. 6, pp. 061918-1--061918-10, 2011. [3] N. Rehman, D. Looney, T. M. Rutkowski, and D. P. Mandic, "Data-adaptive multivariate complexity analysis", Bulletin of the Polish Academy of Sciences, vol. 50, no. 3, pp. 433-466, 2012 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 50 of 63 PID Supervisor Room Description 1400991 Mandic,D.P. (with Looney,D.) 813 The trajectories of eye gaze can be related to different cognitive tasks, as illustrated in the seminal "Yarbus experiment" in 1960s. For instance, causal scanning of an image produces relatively random eye-tracks, whereas upon executing a cognitive instructions (look for ages of people, how affluent the family is) the eye gaze is modulated according to the instruction. Complexity science meets visual perception: Selective attention in the perceptions of images and videos 3E4D4T3I4J The student will use our gaze tracker to perform dynamical complexity analysis of eye-gaze, according to different cognitive instructions. This will be performed for various images (from natural to abstract) and over a range of subjects and measured variables (x,y eye gaze, pupil diameter, eye dwelling). In this way, the 'complexity loss theory', which states that the complexity of physiological responses in living organisms reduces under constraints (ageing, illness) will be rigorously tested. Applications in 'guilty knowledge' scenarios in interview situations, together with those in fatigue monitoring, and the quality of web-design will be investigated. References: [1] A. Yarbus "Congitive Visual Attention", 1971 [2] M. U. Ahmed and D. P. Mandic, "Multivariate multiscale entropy: A tool for complexity analysis of multichannel data'', Physical Review E, vol. 84, no. 6, pp. 061918-1 -- 061918-10, 2011. [3] D. Looney, M. U. Ahmed and D. P. Mandic, "Human Centred Multivariate Complexity Analysis," Natural Intelligence Magazine, vol 1, no. 3, pp. 40-42, 2012. 1400979 Manikas,A. Flexible Array Signal Processing: For groups of drones, satellites and underwater towed arrays 1400980 Manikas,A. Massive MIMO Communication for 5G Mobile Communications 1400981 Manikas,A. MIMO Communication Systems for Spacecraft and Space Battle 1400983 Manikas,A. Green Communication Systems 11 October 2014 801 3E4D4T3I4J 801 3E4D4T3I4J 801 3E4D4T3I4J 801 Flexible arrays refer to array of "sensor" with time varying geometry such as (i) groups of drones used in weather prediction and rapid disaster damage assessment and (ii) microsatellite formations for terrestrial imaging and navigation, planet finding and stellar imaging. This project concerns the study of detection, estimation and reception of signals received on such flexible arrays. This implies looking at objectives such as, for instance, obtaining an improved resolution vegetation study image from a group of drones and localising a planet using a group of satellites. Keywords: Massiva MIMO, Big Data, 5G. The next generation of mobile communication system is planning to extensively increase the number of antenna per handset and per base-station. The gain in performance would be huge if all the extra degrees of freedom could be exploited. There is a fundamental trade-off between the amount of data to be generated by the large number of antenna and the exploitation of these degrees of freedom. Keywords: MIMO communication, Outer Space communication, Outer Space Communication. MIMO Communication Systems are still underused in the context of Spacecraft-to-spacecraft communication. These could greatly improve the communication capabilities and energy efficiency. Keywords: MIMO Communication System, Spectral Efficiency, Energy Efficiency, Bio Safety, Electromagnetic Pollution. 3E4D4T3I4J Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 51 of 63 PID Supervisor Room Description 1400984 Manikas,A. 801 Keywords: MIMO Communication System, Differential Geometry, System Capacity. All the parameters of the MIMO systems are not included in the common expression of their capacity, which is under/over evaluated. There is an opportunity to characterise the capacity more efficiently by employing the tool of differential Geometry. 3E4D4T3I4J MIMO Systems Capacity and Differential Geometry 1400985 Manikas,A. 801 3E4D4T3I4J Jamming Capabilities of MIMO System 1400986 Manikas,A. 801 3E4D4T3I4J Highly Secret MIMO System 1400987 Keywords: Jamming of MIMO System, Military Application. Manikas,A. 801 3E4D4T3I4J Localisation of Wi-Fi and Mobile Signals Keywords: MIMO Communication, Very High Secrecy. Knowing the geometrical characteristics of a MIMO communcaition system, there is an opportunity of integrating the spatial parameters in the coding to make the signals readable at a particular position only. Question: Could it be possible to include the position of the multiple antennas in the coding scheme? This project will use National Instrument's Universal Software defined Radio Boards and will involve 3 main parts: 1)Localisation using strip line (PCB) antennas and 8 channel synchronous ADC 2)Localisation using an 8-element uniform linear antenna array on a single PCB with active (powered) LNA, down-converter and local oscillator 3)Localisation using 8 or 16-element uniform circular array with a DOA testing platform 1400943 Required course: DSP 3.07 Recommended course: Speech Processing Moore,A. (with Naylor,P.A.) Realtime binaural spatialisation for hearing aids 4T Many public places have an induction loop which transmits a clean speech signal captured by a microphone positioned close to a talker. A hearing aid user can select a special mode on their devices to listen to this signal. However the same signal is played equally to both ears. As such it contains no localisation cues and is not perceived to originate from the true source of the signal (I.e. the talker's mouth). The aim of the project is to process the clean speech signal to make the sound consistent with the visual cues. It is proposed that the binaural room impulse response can be obtained by adaptively estimating a pair of filters which transform the clean signal into the signals which reach the microphones in each of the hearing aids. Filtering the clean signal with the first part of the BRIR will then restore the localisation cues without the room reverberation. Key challenges in this approach will be the speed of convergence of the adaptive filters such that they can track rapid changes in the BRIR caused by head motion. Performance will also be effected by the level and type of background noise and the amount of reverberation. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 52 of 63 PID Supervisor 1400944 Moore,A. (with Naylor,P.A.) Room 1400946 Required Course: DSP 3.07 Recommended course: Speech Processing 4T Image source localisation using intensity vectors Naylor,P.A. When sound travels from a speaker to a microphone it follows many reflected paths as well as the direct path. Determining the directions of these reflections is very useful for estimating the room geometry and for dereverberation. Using a spherical microphone array, instantaneous intensity vectors can be calculated which give the direction of energy flow. This is the vector sum of energy flowing from each reflection. The aim of the project is to investigate different methods of decomposing measured intensity vectors to determine the directions of arrival of the constituent reflections. In the first instance Principal Components Analysis could prove useful but there is scope to apply a variety of techniques. 803 4D4T Wearable Microphone Arrays Description Pre-requisite course: DSP 3.07 Recommended course: Speech Processing This project studies the way in which microphone arrays (an organised collection of 2 or more sensors) could be designed in a wearable configuration. There are two aims: to capture sound from the environment surrounding the wearer for communications applications; to capture sounds from within the wearer's body for medical applications. The study will investigate: microphone technology array signal processing for ad hoc microphone arrays (for which the location of the sensors is not known and may vary) mechanical noise due to movement of the microphone against clothing or skin. The work will involve some detailed technical study as well as simulating and testing at least one array signal processing algorithm in Matlab. It is also desirable to construct a prototype wearable array. 1400945 Naylor,P.A. (with Evers,C.) 3D Rendering of Acoustic Maps for Robot Auditiion 803 4D4T Robotic applications, including human-robot interaction, are heavily dependent on the robot’s ability to explore and learn the surrounding environment. Mapping is used to integrate the acquired information into a holistic representation of the robot's world view. Acoustic maps reflect the robot’s belief of surrounding speakers and sound sources. Furthermore, sound reflections off walls and obstacles can be used to infer knowledge about the surrounding room geometry. In this project, the student develops a tool set for 3D rendering of acoustic maps for data representation in robot audition. In order to evaluate the approach, the student needs to consider practical aspects such as spatial resolution of the resulting maps as well as computation / rendering time. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 53 of 63 PID Supervisor Room Description 1400943 Naylor,P.A. (with Moore,A.) 803 Required course: DSP 3.07 Recommended course: Speech Processing 4T Realtime binaural spatialisation for hearing aids Many public places have an induction loop which transmits a clean speech signal captured by a microphone positioned close to a talker. A hearing aid user can select a special mode on their devices to listen to this signal. However the same signal is played equally to both ears. As such it contains no localisation cues and is not perceived to originate from the true source of the signal (I.e. the talker's mouth). The aim of the project is to process the clean speech signal to make the sound consistent with the visual cues. It is proposed that the binaural room impulse response can be obtained by adaptively estimating a pair of filters which transform the clean signal into the signals which reach the microphones in each of the hearing aids. Filtering the clean signal with the first part of the BRIR will then restore the localisation cues without the room reverberation. Key challenges in this approach will be the speed of convergence of the adaptive filters such that they can track rapid changes in the BRIR caused by head motion. Performance will also be effected by the level and type of background noise and the amount of reverberation. 1400944 Naylor,P.A. (with Moore,A.) Image source localisation using intensity vectors 1401009 Stathaki,P.T. Object detection in cluttered environments 1401010 Stathaki,P.T. Object tracking in cluttered environments 11 October 2014 803 4T Required Course: DSP 3.07 Recommended course: Speech Processing When sound travels from a speaker to a microphone it follows many reflected paths as well as the direct path. Determining the directions of these reflections is very useful for estimating the room geometry and for dereverberation. Using a spherical microphone array, instantaneous intensity vectors can be calculated which give the direction of energy flow. This is the vector sum of energy flowing from each reflection. The aim of the project is to investigate different methods of decomposing measured intensity vectors to determine the directions of arrival of the constituent reflections. In the first instance Principal Components Analysis could prove useful but there is scope to apply a variety of techniques. 812 4D4T 812 4D4T This project lies within the Image Processing and Computer Vision research areas. The problem is how to detect a prespecified type of object, as for example a vehicle or a human in complex environments. These could be indoor or outdoor crowded areas like underground stations, streets and others. We will use various Image Processing and Computer Vision algorithms. Good programming skills are required. Preferred programming language is C++ but MATLAB is also acceptable. This project lies within the Image Processing and Computer Vision research areas. The problem is how to track a pre-specified type of object, as for example a human in cluttered environments and most importantly without manual initialization of the object's location. Cluttered environments refer to indoor or outdoor crowded areas like underground stations, streets and others. We will use various Image Processing and Computer Vision algorithms. Good programming skills are required. Preferred programming language is C++ but MATLAB is also acceptable. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 54 of 63 PID Supervisor Room Description 1400873 Barria,J.A. 1012 Pervasive wireless sensor networks offer many mission-oriented opportunities. In this project we will evaluate the potential benefits of combining fixed and mobile sensor nodes as part of a mission-oriented, participatory sensor network deployed on an urban environment that has the objective of detecting the extent of the emission of a substance of interest, for example, pollutants. Distributed coverage and field estimation using mobileassisted WSN 4D4T4J This project will investigate sweep coverage strategies for different network design features like e.g. number and position of fixed sensors, and number and velocity of mobile sensors. In a sweep coverage problem, for a given set of point of interest (POI), the available sensors will need to monitor these POIs under certain operational constraints. We will evaluate the system by assuming simple mobility models for sensors with and without enhanced local intelligence, and more realistic models that seek to reduce operating costs. 1400874 Barria,J.A. Microscopic traffic models and anomaly detection 1012 4D4T4J This project will investigate a novel anomaly detection and classification algorithm that combines the spatiotemporal changes in the variability of microscopic vehicular traffic variables, namely, relative speed, inter-vehicle time gap, and lane changing. First a micro-simulation tool will be used and implemented in MATLAB environment. Then the performance of the classification algorithm on several anomaly scenarios will be assessed (in rural as well as urban environments). Different early warning mechanisms will be investigated. Decentralised anomaly detection mechanisms will be developed and assessed. Once the algorithms have been developed and tested their performance will be validated using real world data from, e.g., microscopic trajectory data extracted from CCTV cameras (NGSIM: http://ops.fhwa.dot.gov/trafficanalysistools/ngsim.htm) 1400875 Barria,J.A. Location-based routing algorithms in mobile ad hoc networks 1400876 Barria,J.A. Probe-based inference of network delay distributions 1012 3E4T 1012 4D4T4J In a MANET environment the dynamic topology is a key feature and may change rapidly and unpredictably due to relative node movements which would affect the transfer of information. The aim is to assess the impact of random events on the real-time availability of connections which will degrade the performance of routing protocols that make use of the position of nodes to disseminate information. In this project we will investigate nodes’ mobility estimations using, for example, range-based relative speed estimation and novel on-line estimation, prediction and classification techniques. The algorithms will be tested via simulation in terms of accuracy of estimation and computational complexity. Inference of network internal characteristics has become an increasing important issue for communication network operation. This project will investigate probe-based technology to establish a mechanism to infer and estimate internal link level delay distribution from end-to-end path measurements. We will investigate discrete as well as continuous approximations. The algorithms will be tested via simulation in terms of accuracy of estimation and computational complexity. Cat.: Mathematical/Analytical, Software. 1400877 Barria,J.A. Smart grid resilient communication infrastructure and protocols 11 October 2014 1012 4D4T4J Review of industrial protocols for streaming data and communication protocols. Review and analyse protocols like e.g. IEC 61850 communication protocol and alternative/complementary ones. Design and implement a client server communication link to investigate the performance of these protocols for high reliable service requirements. Implement distance testing capabilities. Explicit agreement from supervisor required Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 55 of 63 PID Supervisor Room Description 1401005 Gunduz,D. 1016 The past decade has witnessed a rapid growth in the mobile network data traffic, driven by the tremendous increase in the variety of mobile devices and data-rich applications, such as live video streaming and online gaming. Optimal Content Placement and Cache Capacity Allocation 4D4T4J It has been observed that a significant portion of the mobile data traffic is caused by a relatively small number of highly popular content, such as viral video files, news videos, etc., downloaded by millions of users. In today's systems, each time such a request is received, the network fetches the content from the central server, and delivers it to the user through the access point (i.e., base station, WiFi router) it is connected from. A smarter solution is to “prefetch” and “cache” some of the most popular files at the sBSs or WiFi APs, and serve the users directly from these nodes on the network edge. This will reduce not only the traffic in the core network, but also the latency for users. This project will explore the optimum way of assigning files to the caches, in order to minimize the expected downloading time. To this end we will explore simple yet informative models of such systems, formulate the above as an optimisation problem (e.g. a linear programme) and propose efficient algorithms to solve them. Reference: N. Golrezaei, K. Shanmugam, A. G. Dimakis, A. F. Molisch, and G. Caire, “Femtocaching: Wireless video content delivery through distributed caching helpers,” in INFOCOM, 2012 Proceedings IEEE. IEEE, 2012, pp. 1107–1115 (available at http://arxiv.org/pdf/1109.4179v4.pdf 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 56 of 63 PID Supervisor Room Description 1401007 Gunduz,D. 1016 Sensor networks are formed by small battery-powered devices that monitor the environment (e.g., weather monitoring, structures monitoring and health monitoring) and transmit the important data wirelessly to the network. For many sensor network applications it is either costly or impossible to replace drained batteries, such as structural sensors embedded in bridges, or sensors distributed over a rainforest. Energy harvesting technology resolves this limitation by harvesting the ambient energy (e.g., solar, wind or electromagnetic energy), promising seemingly perpetual operation. However, classical communication protocols, which are designed to improve the network lifetime based on limited power or energy resources, are not compatible with energy harvesting sensor networks, whose energy resources are sporadic and random. Energy harvesting communication system optimisation 4D4T4J In this project we want to investigate the following problem: there is an energy harvesting transmitter, which wants to transmit data to its destination. The transmitter can transmit through several orthogonal channels, but does not know the states of the channels (e.g., if the channels are good or bad for transmission). The transmitter can spend energy to probe some channels and estimate their states, but this will leave less energy for data transmission. Hence, the transmitter faces a trade-off between exploration (i.e., probing new channels and spending energy) and exploitation (i.e., transmitting on the best channel so far and use all the remaining energy for transmission). This trade-off has to be balanced with the random arrival of the energy as well. References: 1) P. Chaporkar and A. Proutiere. Optimal joint probing and transmission strategy for maximizing throughput in wireless systems. IEEE Journal on Selected Areas in Communications, 26(18):1546–1556, Oct 2008. 2) P Blasco, D Gunduz, M Dohler. A Learning Theoretic Approach to Energy Harvesting Communication System Optimization. IEEE Transactions on Wireless Communications 12 (4), 1872-1882. The perfect candidate must have background on statistics and stochastic processes as well as wireless communications. The project will involve reading and understanding some technical papers, building up a mathematical model for the problem at hand, and Phyton or Matlab coding of the problem. 1401008 Gunduz,D. Proactive Multimedia Caching to Smartphones 1016 4D4T4J A significant portion of today's mobile data traffic is caused by multimedia files downloaded from servers such as Netflix or YouTube. Existing systems are "reactive" in downloading these files; that is, each file is downloaded at the time of request. This might cause significant energy consumption, especially when the channel conditions are poor (user far from access point, heavy traffic, etc.). The main idea in this project is to develop a "proactive" caching algorithm, which learns user's demand profile, and proactively downloads those files that the user is likely to request, exploiting better channel conditions. For example, the news video that is watched by the user every morning through a 4G connection at a tube station can be downloaded through WiFi before the user leaves home. This project has several aspects and the student may focus on one or more of them depending on her/his interests and background: i) A smart learning algorithm can be developed to predict user's demands; ii) Proactive caching algorithm can be simulated using real traffic patterns and video content; iii) A smartphone application that implements proactive caching can be developed. 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 57 of 63 PID Supervisor Room Description 1401011 Gunduz,D. 1016 Recent progress in white LED (WLED) technology has boosted research interest in optical wireless (OW) communications. WLEDs can be modulated at high speed; and hence, can be simultaneously used for illumination and data communication purposes. In visible light communications (VLC), communication is performed in unregulated and unlicensed bandwidth by modulating the visible light rather than the RF signals. It is especially useful to create indoor high-speed links, has many advantages regarding licencing, security, and interference with other equipment. Visible Light Communication System Implementation 4D4T This project involves implementing a VLC system using off-the-shelf LEDs and Universal Software Radio Peripheral (USRP) devices. While some knowledge and desire to work with hardware is required, most of the implementation will be on software using LabView. 1400884 Kim,T-K. Multi-Pedestrian Tracking 1400885 Kim,T-K. Comparative study of 3D pose estimation methods in a new multi-instance RGB-D dataset containing clutter and occlusion 1400886 Kim,T-K. Object recognition in videos 1400887 Kim,T-K. Unified framework for face image analysis 11 October 2014 1017 4D4T4J 1017 4D4T4J 1017 4D4T4J 1017 4D4T4J Object tracking is the process of locating a moving object (or multiple objects) over time in a video. It has a variety of applications such as video editing, video communication and compression, augmented reality, security and surveillance, and human-computer interaction. In object tracking, a major challenge is handling appearance changes of the target objects in highly cluttered backgrounds due to factors such as changing pose, illumination and deformation. Recently a class of techniques treating tracking as a classification framework has been successful. A classifier needs to be highly time-efficient and iteratively updated to reflect object appearance and environmental changes in each frame. The participant is expected to have reasonable program skills in C/C++ and/or Matlab, and backgrounds to study and implement relevant cutting-edge technologies, to pursue novel contributions. Semantic segmentation is an important topic in computer vision with great application across many domains. Some examples include content based retrieval and augmented reality. Semantic segmentation possesses many challenges due to object intra-class variation, view-point changes and occlusion. In this project we will explore 1) efficient methods for depth image segmentation and 2) 3D object detection and pose estimation techniques. The student is expected to have good programming skills in C/C++ and/or Matlab, and background to study computer vision and machine learning tools. Object categorisation is one of the most important studies in computer vision. While considerable advances have been made in single-shot based object recognition (using a single ’query’ image), successful recognition in natural scenes still remains a challenging problem. Rather than a single query image that captures only a limited amount of information about a target object, a video (of moving objects and/or a camera) yields more evidence that can be accumulated for better recognition. With increasing use of mobile devices that have an embedded camera, there is a great motivation to develop visual recognition methods that take a video as input. In this project, we explore an extension of image based object cateogorisation techniques, into those for videos. Relevant works include Kim et al’s BMVC 2010 paper and TPAMI 2007 paper, available from Dr. T-K Kim’s website. The student is expected to have good program skills in C/C++, and backgrounds to study and implement relevant cutting-edge technologies, and to explore novel contributions. Face is a key object in image and video analysis. People have tackled various sub-problems of face recognition: face identity recognition, face expression recognition, facial pose estimation and facial feature point detection. In this project, we explore the combined effect by simultaneously tackling multiple sub-problems. By coupling one problem to the other, we expect to see improved accuracy. E.g. knowing present facial expression would provide a good prior for face feature point locations. The student is expected to have good program skills in C/C++ and/or Matlab, and backgrounds to study and implement relevant cutting-edge technologies, to explore novel contributions. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 58 of 63 PID Supervisor Room Description 1400964 Pitt,J.V. 1013 The of the project is to design and implement a multi-agent simulation of a decentralised Community Energy System (CES). A CES is an energy generation, distribution and storage system involving local community ownership and participation. The first part of the project is to write a simulator that models the co-dependent generation and storage systems in an 'islanded' system. The second part will involve scaling this simulator to inlcude nestec CES and composite CES, i.e. systems with a combination private and public partnership including CHP and large(r) scale generation. Finally, if time allows, there will be an opportunity to investigate an 'ecosystem' of interacting agencies in CES. Simulator for Decentralised Community Energy Systems 4D4T4J The multi-agent simulation platform to be used will be the Imperial developed PreSage-2. This is a very powerful tool but not without a learning curve. Therefore, software engineering and programming skills are necessary. 1400965 Pitt,J.V. 1013 Shared Space App 4D4T4J This project is concerend with the design and implementation of an App for participating in the management of a shared space, like a student flat, an open plan office or a public park, as a common-pool resource. The project will entail user-centred design, interaction, interface and affordance design, data and datflow design, protocol design, and architectural design, with implementation of corresponding functionality. The project will suit a student with an interest in software engineering, human-computer interaction, and app development. 1400966 Pitt,J.V. Algorithmic Self-Governance, Social Capital and Cryptocurrencies 1400947 Holmes,A.S. Control of a MEMS scanning mirror for display applications 1013 4D4T4J The project will investigate the relationship between electronic forms of social capital, created in self-organising electronic institutions, and the cryptocurrencies like Bitcoin, Ven, Ripple etc. On that basis, it will then examine the relationship between self-governing electronic institutions and some notion of "sovereignty". The project will involve programming and will suit someone interested in declarative programming, although other programming paradigms are possible. 701 3E4D4T In OSD Group we have recently developed a large area silicon MEMS scanning mirror for head-up display applications. This is based on a torsional mirror excited by electrostatic actuators which exhibit highly non-linear behaviour. The project will involve analysis and simulation of the device dynamics, followed by design and manufacture of an optimised driver. This will be a closed loop controller that can monitor the mirror position by capacitive sensing and generate appropriate excitation signals. Good circuit design skills will be required. 1400948 Holmes,A.S. Miniature turbine for energy harvesting in wireless sensor networks 701 3E4D4T This project is linked to ongoing research in the EEE Department on miniature turbines for energy harvesting from air flows. A cm-scale turbine has been developed and already forms the basis of a prototype wireless flow sensor. However, the turbine performance is currently sub-optimal. Building on earlier work, this project will use an establised modelling approach to produce alternative turbine rotor designs aimed at higher performance. These designs will be fabricated using prototyping facilities in the OSD Group and then tested in a wind-tunnel. This is a challenging, and potentially highly rewarding, project requiring strong practical engineering skills and a willingness to learn something about aerodynamics. 1400949 Holmes,A.S. Power conditioning for micropower energy harvesters 11 October 2014 701 3E4D4T Small-scale energy harvesters that extract power from their surroundings have many potential applications in wireless sensor networks. However, processing the raw output from such devices, in order to convert it into a more useable format, is a challenge. Typically the total power available is measured in microWatts, and the voltage may be too low to drive conventional circuitry. This project will investigate possible solutions to this problem, through design, simulation and prototyping. The project is suitable only for students with a strong interest in analogue electronics. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 59 of 63 PID Supervisor Room Description 1400950 Holmes,A.S. 701 This project will explore the viability of energy harvesting power generators based on vibrating/flapping structures. It is well known that an obstruction placed in an airstream will, under the correct conditions, generate a wake with instabilities that will cause an element placed downstream to flap or vibrate. Energy harvesting based on this principle has been demonstrated, but to date the devices produced have not shown good performance. The project will investigate the reasons for this, with a view to improving on existing designs. The aim will be to design and build a prototype harvester at cm scale. Good practical engineering skills, together with an interest in mechanics, will be required. Energy harvesting from air flow using vibrating structures 1400901 Lucyszyn,S. Moonbounce Project: Antenna Tracking System 1400902 Lucyszyn,S. Development of a thermal infrared spectrometer 1400903 Lucyszyn,S. Development of a secure thermal infrared door entry system 1400904 Lucyszyn,S. Electromagnetic Resonator Modelling for Mine Detection 1400905 Lucyszyn,S. Millimetre-wave Photonic Crystals using Novel Metamaterials 1400906 Lucyszyn,S. Infrared tracking system for covert surveillance 3E4D4T 602 3E4D4T3I4J 602 3E4D4T3I4J 602 3E4D4T3I4J 602 3E4D4T3I4J 602 3E4D4T3I4J A harware+software system designed to track the sun and moon was developed last year by an excellent project student. This used predictive methods for tracking. This year I am looking for a similar hard working student that can improve the hardware and software to accurately lock-onto the celestial target. This is a demanding project for a student that is able to write simple software and use this to control the digital hardware and mechanical actuators. This project is ideal for an all-rounder. This project is in collaboration with an existing PhD student. The student must develop the sensor and data acquisition system and write the code to analyze the data. The student must have an interest in analogue and digital circuit design, as well as implement A/D conversion and associate software development. This project is best suited for an all rounder. The student must develop a novel secure door entry system, based on thermal infrared technologies being developed by PhD students. The student must used A/D and D/C chip sets with be-spoke analogue and digital circuit design to design the whole secure door entry system. This project is best suited for an all rounder. The student should have an interest in microwave engineering and/or electromagnetics. The student will first develop numerical models of simple resonator structures, using commercial software, which is then extended to more realistic structures. Then experimental validation of the modelling will be performed, with the development of a simple sensor. This project is idea for a pro-active student interested in CAD and experimentation. In association with the Imperial College Centre for Terahertz Science and Engineering (in collaboration with the Department of Physics and Materials), this project investigates ultra-high quality (Q)-factor resonators for next generation RF electronics applications. The student must design and test transitions from the 2D metamaterial crystal structures to conventional coplanar waveguide probes and microfluidic sensors. The student should ideally have studied, or be currently studying, the E3.18 Microwave Technology course. 602 3E4D4T3I4J This project aims at developing a system for tracking moving targets that emit radiation in the infrared part of the frequency spectrum. The system must first detect a moving target and then follow its movements while taking images in both the visual and infrared parts of the frequency spectrum. The student must have an interest in image processing, control theory and have an active interested in developing computed-controlled actuators, under the influence of input data from various sources. This project was a success for an MSc student last year and so many of the technical challenges have been solved. However, what is needed is to improve the every aspect (resolution, speed and functionality). 11 October 2014 Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 60 of 63 PID Supervisor Room Description 1400907 Lucyszyn,S. 602 This is a practical project for a student that wants to design, build and test an antenna array. The objective of this project is to be able to transmit and receive signals to the moon and back. This project is ideal for someone that does not mind getting their hands dirty! Moonbounce Project Antenna 1400857 Sydoruk,O. Monitoring breathing activity 3E4D4T3I4J 603 4D4T Breathing, like heart pulse, is a vital function. But unlike heart pulse, breathing is difficult to monitor without disturbing a patient. The student working on this project will design theoretically a system that could be used as breathing monitor. The system will consist of resonant metallic loops magnetically coupled to each other. The loops could be placed on a flexible belt worn around the chest. The student will study the electric properties of the flexible system. The student will also find out whether these properties can be used to measure the lung volume. The project is theoretical and will involve numerical simulations in Matlab and analysis of RF circuits. It will suit someone with interest in biomedical and electronic devices. 1400858 Sydoruk,O. Random magneto-inductive communication 603 3E4D4T3I4J Optical and Semiconductor Devices Group leads a strong research programme in magneto-inductive waveguides. These waveguides are based on discrete magnetically coupled resonant LC-circuits. Their potential applications lie in the fields of magnetic resonance imaging and signal transmission in challenged environments, in particular, underground. The LC-circuits in a waveguide are usually arranged in a regular periodic manner. However, the arrangement can be broken, for example, in an industrial accident. In contrast to conventional cables, magneto-inductive waveguides would not be destroyed and could still be used for communication. This project is theoretical and will investigate how well magneto-inductive waveguides can work when some of their elements are displaced or destroyed. It will involve analytical calculations and numerical simulations in Matlab. It will suit a student with interest in waveguides. 1400859 Sydoruk,O. Distributed magneto-inductive power delivery 11 October 2014 603 4D4T Batteries can be charged (or power can be delivered) without cables by means of mutual induction. Existing solutions (such as artificial heart or phone charging) are stationary: the receiving system does not move. Proposals also exist to power large moving vehicles. This theoretical project will investigate the possibility of delivering power to a small moving object. In contrast to a large vehicle, the receiver should be small, suggesting the use of higher frequencies. As a power delivery system, the project will study a transmission line loaded periodically with magnetic loops. The modified properties of the transmission line will need be calculated analytically and simulated numerically. The project is computation-heavy and will suit someone with good knowledge of the transmission-line theory and strong analytical skills. Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 61 of 63 PID Supervisor Room Description 1400860 Sydoruk,O. 603 Optical and Semiconductor Devices Group leads a strong research programme in magneto-inductive waveguides. These waveguides are based on discrete magnetically coupled resonant LC-circuits; their potential applications lie in the fields of magnetic resonance imaging and signal transmission in challenged environments. Further development of the magnetoinductive technology requires a number of waveguide devices. Coupler for magneto-inductive waveguides 4D4T This project will aim at designing a coupler for magneto-inductive waveguides. The student will design lumped-element circuits that will target both standard coupler characteristics (loss, coupling, directivity, isolation) and the characteristics of the magnetoinductive technology (band width). The project is theoretical and involves analytical calculations and simulations in Matlab and/or a circuit simulator. It will suit a student with good analytical skills and knowledge of and interest in wave propagation and the transmission-line theory. 1400940 Syms,R.R.A. Control systems with intermittent feedback 1400941 702 3E4D4T3I4J Syms,R.R.A. RF-safe cables for implanted electronic systems 1400962 3E4D4T3I4J Syms,R.R.A. Modelling of nonlinear combat scenarios 1400942 702 702 3E4D4T3I4J Syms,R.R.A. Array-type detection of MRI signals 11 October 2014 702 1A Linear control systems conventionally have continuous feedback (for example, of position or velocity in a dynamical system). Under some circumstances, however, it may not be possible to observe the position continually, for example if a feedback signal is only available near the extremities of a travel range. The aim of this project is to investigate the circumstances under which a working control system may still be constructed, and evaluate the performance of the resulting controller. The project would suit someone with an interest in control theory and an ability to program in MATLAB and SIMULINK. Military scenarios such as the recruitment and deployment of forces and the attrition of forces resulting from combat can be modelled using systems of coupled differential equations. Because the equations are nonlinear, many surprising solutions are possible. The aim of this project is to investigate the circumstances under which stable, cyclic and chaotic solutions are obtained to combat scenarios involving more than two sets of forces. The project would suit someone with an interest in modern world events and control theory, and an ability to program in MATLAB or SIMULINK. Implanted electronic systems are increasingly being used to control medical conditions that are beyond the use of drugs, including heart arrhythmia, Parkinson's disease, epilepsy, chronic pain and incontinence. Unfortunately, the electrical cables that are generally used to link the controller with a stimulating electrode can be heated very rapidly by strong electric fields in a magnetic resonance imaging scanner. The aim of this project is to use simple electrical models to investigate tradeoffs in the design of RF-safe cables. The project would suit someone with an interest in medical physics and the ability to program in MATLAB or SPICE. The detection of RF signals in a magnetic resonance imaging (MRI) scanner is carried out using resonant L-C circuits. Because the signals are weak, is important to minimise the pickup of any noise, which arises mainly from the human body itself. Rather than using one large coil (which picks up noise from the whole body), an array of smaller coils is used. Each element detects local signals, together with a much smaller amount of noise. The outputs are then combined together, so that the signals add coherently while the noise adds incoherently. For small numbers of coils, the result is increased signal-to-noise ratio. Unfortunately, it is difficult to prevent each coil being detuned by mutual inductance with its neighbours. The aim of this project is to investigate the effect of feedback in suppressing the coil currents, and hence in reducing mutual inductance. The project would suit someone interested in medical physics and analog circuit design Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 62 of 63 PID Supervisor Room Description 1400963 Syms,R.R.A. 702 The detection of RF signals in a magnetic resonance imaging (MRI) scanner is carried out using resonant L-C circuits. Because the signals are weak, is important to minimise the pickup of any noise, which arises mainly from the human body itself. Rather than using one large coil (which picks up noise from the whole body), an array of smaller coils is used. Each element detects local signals, together with a much smaller amount of noise. The outputs are then combined together, so that the signals add coherently while the noise adds incoherently. For small numbers of coils, the result is increased signal-to-noise ratio. Unfortunately, it is difficult to prevent each coil being detuned by mutual inductance with its neighbours. The aim of this project is to investigate the effect of feedback in suppressing the coil currents, and hence in reducing mutual inductance. The project would suit someone interested in medical physics and analog circuit design. Array-type detection of MRI signals 11 October 2014 3E4D4T3I4J Key: 3E= EE BEng, 4D= EE MEng EM, 4T=EE MEng T, 3I = ISE BEng, 4J = ISE MEng Page 63 of 63
© Copyright 2024