A low bandwidth pulse-based neural recording system

This research investigates a novel data reduction scheme using adaptive leaky refractory integrate-and-fire (ALRIF) neurons to generate pulses for an implanted neural recording system in wireless transmission applications. The wireless implanted multi-channel recording system imposes many constraints on the system but the major constraint is on low bandwidth. Other constraints including low bandwidth transmission, large dynamic range, low power consumption, small device size and noise robustness, though serious, can be more easily met. This proposed scheme promises to dramatically reduce the required communication bandwidth via three versatile neuron circuit strategies: adaptive, leaky and refractory neurons. This system consists of both front-end hardware and a back-end signal processing. Analog VLSI circuitry (AMI 0.6µm CMOS) was chosen to implement the front-end hardware to transform the signal to a pulse representation with ultra-low bandwidth. On the back-end, the system can either reconstruct the original signal and run traditional spike sorting or run spike sorting directly in the pulse domain. The ALRIF neuron circuit can reduce the bandwidth efficiently to support the pulse-based spike sorting. MATLAB simulation results for the neuron models are proof of concept. Circuit design for each building block has been presented and simulated in Cadence. Bench-top hardware system testing shows feasibility of in vivo neural recording applications.