Imaging, Data and Communications
Advanced electronic/optoelectronic technologies designed to allow stable, intimate integration with living organisms will accelerate progress in biomedical research; they will also serve as the foundations for new approaches in monitoring and treating diseases.
Research Theme
Multi-agent Systems and Data Intelligence
Aim
Develop algorithms to infer when an agent becomes aware of a sensor from multi-sensor data and optimise the decisions of either the monitor or the agent being monitored.
Objectives
- Model how awareness of a sensor changes an agent's behaviour and the data seen by other sensors.
- Develop change detection methods to estimate when awareness occurs and quantify uncertainty when data are noisy or missing.
- Optimise sensor use and observation schedules to reduce the risk of discovery while retaining useful information.
- Optimise an agent's behaviour to infer sensor presence and limit the information available to the monitor.
Description
An agent may change its behaviour when it notices a sensor, altering the data available to other sensors. We will investigate whether these changes reveal when the agent became aware of the sensor, and how either side should respond.
We will model awareness as a latent state that influences behaviour and measurements from multiple sensors. Change detection methods will estimate when awareness occurs and quantify uncertainty when data are noisy or missing. We will use these estimates to choose sensors and observation times while balancing information gained against the risk of discovery. We will also optimise the monitored agent's actions to infer sensor presence and limit what it reveals.
Controlled simulations will test the methods when sensor coverage is incomplete and agents learn over time. We will compare performance when optimising for the monitor and when optimising for the agent, measuring detection delay, false alarms, uncertainty calibration and the information obtained by the monitor.
Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree.
Further information on English language requirements for EU/Overseas applicants.
Fully funded position.
Research Theme
Aim
To develop fast randomised adaptive processing algorithms for next-generation digital-array airborne radar, with performance guarantees stated in operational detection terms rather than abstract mathematical error bounds.
Objectives
- Determine how far the adaptive processing can be sketched or compressed without missing targets, with certified detection performance.
- Automatically detect and remove misleading or contaminated training data.
- Assess and enhance the algorithm's robustness to jamming.
- Validate the algorithms on open real radar data against standard simulated baselines.
Description
Next-generation airborne radars digitise every antenna element. As a result, the adaptive filters used to detect slow-moving ground vehicles (Space-Time Adaptive Processing, STAP) must now handle covariance matrices of dimension close to a million, recomputed every few milliseconds. Classical theory demands roughly twice that many statistically identical training samples, which real, heterogeneous terrain cannot supply. Fortunately, ground clutter is strongly low-rank, which is exactly the regime where randomised numerical linear algebra excels.
This project develops randomised Nyström STAP algorithms whose key novelty is certification. Instead of matrix-norm error bounds (tail concentration), guarantees will be expressed in radar context terms that are intuitive to radar engineers, e.g. probability of detection at a fixed false-alarm rate, and minimum detectable velocity. These will be delivered in conjunction to randomised sketch size, training budget and latency.
The student will also use statistical leverage scores and diversity-promoting sampling to screen contaminated training data without the circular need for a covariance inverse. Depending on progress, a further strand could investigate how re-randomising the sketch every processing interval makes the adaptive response unpredictable to a jammer. Methods will be validated on benchmark datasets such as KASSPER/DARPA against established baselines. The outcome is an assurable alternative to learned methods for ISR aircraft operating in cluttered, contested electromagnetic environments.
A strong background in linear algebra, probability and statistics.
To reduce risk the desired PhD candidate should have a first-class degree in Mathematics with a solid foundation in numerical methods, and programming. An MSc in a numerate discipline will be considered an advantage.
Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree.
Further information on English language requirements for EU/Overseas applicants.
This position is fully funded.
Research Theme
Aim
To use AI/ML methods to adapt and optimise the PNT solution in the absence of reliable GNSS for varying sensor suites in varying environments, whilst remaining agnostic regarding inputs and platform type.
Objectives
- Obtain sufficient labelled data, real and synthetic, for PNT from sensor systems in varying platforms and environments to apply Machine Learning (ML) techniques to.
- Identify suitable AI/ML methods to be applied to the data with realistic metrics as to what constitutes a “good” solution regardless of environmental and sensor suit constraints.
- Develop the selected methods on the sensor and digital world-model data available to provide system level PNT solutions for varying GNSS conditions and platform types.
- Validate the methods against independent data with GNSS “truth” available.
Description
Positioning, Navigation and Timing (PNT) without reliable GNSS is a complex problem for which a wide variety of solutions are being developed, from signals of opportunity to mutli-sensor fusion. What methods can be applied vary significantly depending on the platform type and the sensor suite available on that platform and with the environment the sensors and platform operate in. Moreover, the sensor information available may vary even within a mission depending on what other tasks the sensors have to undertake and the environment (physical and electromagnetic) may change over the course of the mission. Ensuring some level of agnosticism enables a solution to maintain a highly accurate output level, giving the AI/ML solution a wider use case. Varying input types and numbers of is to be considered, as to not cause any sensor suit constraints for the wider platform. The information the sensors provide, must also be referenced against the “worldview” that the AI/ML has been trained on e.g. digital terrain models, locations of features, emitters, etc, to provide and alternative to GNSS in a global reference frame. These models can also be used to determine which alternative PNT methods are optimum in specific environments. Building on work begun on other SPADS CDT projects and making use of data available, the project will look at how ML can be used to train a PNT solution to use different data sources over the mission to meet the required accuracy of PNT for the mission to be carried out successfully and to learn from previous missions to improve PNT performance over time. The system will not be concerned with GNSS signal validity and will aim to be ‘activated’ as a primary navigation system upon GNSS interferences. This means the system should be able to operate using all levels of GNSS available, providing the most notable PNT accuracy increase.
Research Theme
Aim
This project will develop theory and efficient computational methods to enable radar detection in complex sensing scenarios by long-time coherent integration.
Objectives
- Develop coherent track-before-detect algorithms for different radar modes, such as range-ambiguous and Doppler-ambiguous modes.
- Investigate the errors that arise when using machine learning to predict background clutter.
- Investigate statistical detection tests and performance bounds accounting for the uncertainty in the background clutter statistics identified in objective 2.
- Investigate computationally efficient approximate schemes for low-SWAP implementations.
Description
This project will address algorithmic challenges modern multi-mode radars face in the case of hard-to-detect objects in complex background environments. It will propose coherent long-time integration (CLTI) algorithms inspired by Bayesian filtering (e.g. coherent track-before-detect) and machine learning approaches to leverage data for learning complex backgrounds.
CLTI combines evidence of object existence across measurements collected over time, thereby improving the signal-to-noise-plus-background clutter ratio (SNCR) and consequently the probability of detection. Recent works have demonstrated the benefits of coherent track-before-detect in performing CLTI and neural network models in self-supervised learning of background clutter statistics in complex environments to characterise the “null hypothesis” of the detection problem. This project aims to build upon these two strands of work and develop CLTI algorithms for different radar modes (e.g. range-ambiguous and Doppler-ambiguous modes). A key focus will be investigating statistical detection tests to use with the CLTI outputs and related performance bounds that account for errors in the background clutter statistics. These results will be used to develop computationally efficient approximations of the proposed algorithms for implementation on low size-weight-and-power (SWAP) radar systems.
Minimum entry qualification - an Honours degree at 2:1 or above (or International equivalent) in a relevant science or engineering discipline, possibly supported by an MSc Degree.
Further information on English language requirements for EU/Overseas applicants.
This position is fully funded.