A wavelet-based compression for neural recording system

The power consumption of implanted multichannel system is dominated by the wireless transmission and as such can be significantly reduced by using on-chip data compressor. This paper proposes a wavelet-based data compressor suitable for multichannel implanted neural recording systems implemented as a System-on-a-Chip (SoC). Proposed algorithm exploits energy compactness features of DWT and self similarity across different scales to classify neural recordings as spike and non spike areas. Resulting DWT coefficients after dead-zone scalar quantization are encoded with a novel dedicated entropy encoder optimized for a very low entropy condition. To reduce power dissipation of the chip we apply distributed arithmetic to implement DWT, which reduces number of multiplications down to zero.