Machine Learning-aided Integrated Sensing and Communications

A new wave of transformative wireless applications is emerging, including augmented and virtual reality, smart factories, vehicle to everything (V2X) connectivity, high speed rail, and low altitude economy. These applications place unprecedented demands on future wireless networks, requiring not only enhanced communication capabilities but also accurate radar like sensing. To make this evolution possible, integrated sensing and communications (ISAC), has been recently proposed. As a key enabler for 6G and beyond, it tightly integrates sensing into communication systems within a single system through shared hardware, spectrum, waveforms, and signal processing frameworks. A core challenge in ISAC lies in target sensing, namely, estimating parameters such as range, velocity, and angle of arrival. This task becomes considerably more difficult in dynamic, rapidly time varying environments characterised by severe Doppler dispersion and off grid delay and Doppler components in the range Doppler map. Existing approaches often rely on high complexity techniques such as MUSIC, compressed sensing, and ESPRIT, which may be impractical for receivers operating under strict computational constraints.

To overcome these limitations, this project proposes an intelligent, machine learning based target sensing framework suitable for dynamic ISAC deployment scenarios. The key idea is to shift computationally intensive, high dimensional parameter estimation to an offline training stage, enabling low latency and low complexity inference during real time operation. In particular, the potential of generative adversarial networks will be explored, incorporating novel preprocessing schemes to enhance training stability. The machine learning model will be trained and optimised to achieve sensing accuracy approaching the Cramér–Rao bound. Furthermore, an adaptive transfer learning based online strategy will be developed to initialise the GAN with pretrained weights, facilitating rapid adaptation to evolving channel conditions.