Compact agile Tchebycheff transform variant for temporal compression of neural signals on brain-implantable microsystems

Abstract

This paper proposes a new approximate variant of the Tchebycheff transform for the compression of intra-cortically-recorded neural signals on brain-implantable microsystems. The proposed approximate transform is achieved using an innovative algorithm that guarantees ‘partial orthogonality’ of the basis functions corresponding to the coefficients conveying the majority of the signal energy. This significantly reduces inter-coefficient energy leakage in the proposed transform variant, which helps meaningfully enhance the data compression rate. While keeping the signal reconstruction accuracy within the acceptable range, the proposed Tchebycheff transform variant is maximally truncated in order to reduce the hardware cost of the associated implementation. Based on the resulting truncated approximate transform, a hardware-efficient 256-channel neural signal compressor is designed, prototyped, and tested using pre-recorded neural signals. Operated at a clock rate of 960 kHz, the prototyped neural signal compressor exhibits a true compression rate of 1880 with an RMS reconstruction error of 2.8%.