PhD Projects 2020
- Charles Khoury: Application of deep learning power amplifier architectures for mmWave communications
- Konstantinos Lekkas: Advanced all-optical interconnect technologies for secure data Centre networks
- Alex Pitt: Efficient and linear wideband radio frequency power amplifiers
- Emma Valla: Robust consensus in swarm robotics
Charles Khoury
Application of deep learning to radio frequency cyber security
The recent deployment of 5G infrastructure has marks a new area of communication allowing for more connectivity and new applications including critical ones such self-driving vehicles. This implies the need for secure and reliable mobile networking at the physical layer as the network needs to be available to users, avoiding denial of service, and data secrecy and integrity needs to be maintained. Current approach to study the physical channel is based on improving one aspect of the communication pipeline via statistical understanding of the channel or hardware compensation. This approach has it's limitation as the system is not modelled holistically making it hard to adapt to changes in the wireless medium.
The field of Deep Learning and Machine Learning has allowed researcher to solve more complex problems due to the improvements in algorithm sand in the case of supervised learning applications an abundance of data. Moreover, the hardware needed to perform training and/or inference tasks has seen a significant increase in power and efficiency with graphics processors and dedicated accelerator scaling from data centres to mobile terminals.
Konstantinos Lekkas
Advanced all-optical interconnect technologies for secure data Centre networks
Transparent optical networking is essential for traditional transport optical networks telecommunication applications but has also evolved as the key enabling technology for future dynamic quantum secure optical communications and hyperscale data centres and supercomputers. Specifically, all - optical switches are very good candidates for high speed communications as the light signals are not required to be converted to electrical signals as it happens in traditional electronic switches, while they are also inherently transparent to the signals bit rate allowing them to support the ultra high bandwidth links of current and future data centre networks. On the other hand, low loss optical switches are the only means to switch quantum signals towards the realization of dynamic quantum key distribution topologies that allow their deployment beyond point-to-point links in a real mesh network. Therefore, the objectives of this PhD research are:
1. Develop the concept of dynamic quantum key distribution using low loss optical switches
2. Develop new optical data centre networking architectures that take advantage of high radix ultra-fast switches
3. Design and develop novel high-radix, high speed, low loss optical switching technologies that support future optical data centers and quantum networks.
Alex Pitt
Efficient and linear wideband radio frequency power amplifiers
The aim of the PhD will be to improve the bandwidth of radio frequency power amplifiers whilst maintaining a sufficient level of linearity and efficiency at input power level back off. Simulation tools will be used to investigate state of the art methods to understand their limitations and areas for improvement. This investigative work will then be used to formulate a new method in constructing a radio frequency power amplifier with good ultra wideband linearity and efficiency performance. This method will then be validated through the design of power amplifier hardware and then through laboratory test and measurement. This research will see an application in future communication systems beyond 5G where it will allow users to achieve higher data speeds and system operators to gain monetary savings through better system efficiency.
Emma Valla
Robust consensus in swarm robotics
Robot swarms are collections of a large number of relatively simple and inexpensive robots, which cooperate in order to carry out a task. A fundamental challenge in swarm robotics is to coordinate such swarms. It is typically not desirable or feasible to have a central controller, and the robots in the swarm have to achieve coordination on their own. There has been a large body of research developing decentralised algorithms for coordination in robot swarms. It is however often taken for granted that such algorithms will be robust to the failures or faulty behaviours of a small number of robots, but this has not been rigorously tested.
The goal of this project is to explore how robustness can be increased in consensus-based behaviours with robot swarms. The lens through which robustness will be primarily explored is that of the swarm network topology, with the aim to create mechanisms for self-correcting swarm networks. This includes determining the relationship between a network's topology and structure and its robustness to partial agent failure, total agent failure, noisy environments and communication loss. Implementation will include optimising for the most appropriate metric of network structure and developing a paired decentralised algorithm to correct for issues in the network. The Bristol Robotics Laboratory Kilobot Swarm will be used in physical simulations of the developed algorithm. The full scope of the project also includes benchmarking leading consensus-based algorithms for similar robustness to failure.