A Training-Free One-Bit Compressed Sensing Framework for Wireless Neural Recording

Abstract

Signal compression is crucial for resource-constrained wireless neural recording applications with limited data bandwidth, and compressed sensing (CS) has successfully demonstrated its potential in this field. This letter proposes a training-free one-bit compressed sensing framework for wireless neural recording applications. By adopting the analysis model to enforce the signal sparsity and constructing a multi-order difference matrix as the analysis operator, it avoids the dictionary learning procedure and reduces the need for previously acquired data and computational complexity. Also, an analysis $\ell _{p}$ -minimization approach, which can be effectively solved by the analysis iteratively reweighted least squares algorithm, is proposed to recover signals from one-bit measurements. Experimental results on real data set reveal that the proposed approach not only drastically reduces the transmission bits but also outperforms the state-of-the-art CS-based methods in terms of both recovery accuracy and noise robustness.