This project aims to advance sensor array processing by combining traditional model-based methods with innovative data-driven techniques. While data-driven methods have revolutionized fields like speech processing, their impact on sensor array processing has been limited due to the distinct nature of physical systems governed by physical and statistical models. The project will focus on improving direction-of-arrival estimation using Sparse Bayesian Learning and maximum likelihood estimation. It will also develop novel beamformers for optimal signal power and waveform estimation, with finite-sample and asymptotic performance analysis. Additionally, the project explores data-driven generative modeling techniques like adversarial autoencoders and conditional GANs for denoising the sample covariance matrix in low SNR and limited snapshot conditions. The research will be tested in real-world applications and is supported by expert collaborators from top universities in US and Europe.

2025

2029