Unrolled Compressed Blind-Deconvolution

The problem of sparse multichannel blind deconvolution (S-MBD) arises frequently in many engineering applications such as radar/sonar/ultrasound imaging. To reduce its computational and implementation cost, we propose a compression method that enables recovery from much fewer measurements with respect to the full received signal time. proposed measures through filter followed by subsampling, allowing for significant reduction cost. We derive theoretical guarantees identifiability compressed measurements. Our results allow design wide class filters. We, then, data-driven unrolled learning framework learn solve S-MBD problem. encoder is recurrent inference network maps into an estimate demonstrate our more robust choices source shapes has better performance compared optimization-based methods. Finally, data-limited (fewshot learning), highlight superior generalization capability conventional deep learning.