Parameter Estimation using Random 1 bit Streams (PEERS)

• Published in: 2021 55th Annual Conference on Information Sciences and Systems (CISS) pp. 1 - 4
• Main Authors: Zachariah, Nishant, Doxas, Isidoros, Molin, Jamal
• Format: Conference Proceeding
• Language: English
• Published: IEEE 24-03-2021
• Subjects: 1 bit compute; Encoding; Hardware; low power compute; matched filtering; neuromorphic hardware; Neuromorphics; Noise measurement; Parameter estimation; Power filters; Radar signal processing; random encoding
• DOI: 10.1109/CISS50987.2021.9400230

Matched filtering is used in radar signal processing for optimal detection of signals in highly noisy environments. In this paper, we demonstrate a novel method for matched matched filtering using 1 bit random encoding called PEERS (Parameter Estimation using Random 1bit streams). PEERS demonstrates three novel results. First, PEERS demonstrates robust parameter estimation with recovery errors (\Vert y-\hat{y}\Vert_{2}^{2}) that are as much as 50% lower than matched filtering in highly noisy environments. Secondly, we demonstrate that PEERS is resilient to grid jitter (instances in matched filtering where the signal is not contained in the library) with recovery errors that can be lower than matched filtering by as much as 88%. Finally, while recent work in this area has focused on noise reliance post encoding, we demonstrate a novel way of handling noise pre-encoding which gives rise to the superior estimation results described. Given the 1 bit arithmetic leveraged by PEERS, we demonstrate is efficacy on custom neuromorphic hardware where each PEERS operation consumes less than 20fJ/ operation for extremely low power edge computing.