A compressed-sensing sensor-on-chip incorporating statistics collection to improve reconstruction performance

Abstract

Reconstructing signals accurately is a critical aspect of compressed sensing. We propose a compressed-sensing sensor-on-chip that compresses and also extracts key statistics of the input signal at sampling time. These statistics can be used at the receiver to significantly improve the accuracy of reconstruction. When compared against a conventional compressed-sensing system, our experimental measured results demonstrate an improvement of as much as 9-18 dB in the signal-to-error (SER) of the reconstructed signal, depending on input data type and compression factor.