A System-Level Analysis of a Wireless Low-Power Biosignal Recording Device
- Author(s): Chandler, Rodney James
- Advisor(s): Judy, Jack W
- et al.
Development of brain-machine interfaces and treatment of neurological diseases can benefit from analysis of recorded data from implanted electrodes. Existing wireless neural recording systems are often bulky, dissipate too much heat to be implanted, or only have a small number of channels. Furthermore, advances in micro-machined electrodes provide the possibility of high-density recordings, but the companion electronics do not provide enough simultaneous channels with low enough power, wireless telemetry, or a small form-factor. A system level view of wireless recording-circuitry which could overcome these deficiencies is described in this work. The overall system comprises of an analog front end (AFE), digital signal processing (DSP), and transmitter (TX). Each block is analyzed, and system-level specifications are derived. Based on these specifications, each block can be optimized for low power and small area. The analog front-end uses open-loop amplifiers to support lower voltage operation than previously published work. A prototype amplifier was also fabricated to measure performance in a 65-nm CMOS process that is needed for low-power digital signal processing. The amplifier performance was comparable to other recently published amplifiers with 2.5 μV noise in 10 kHz bandwidth while dissipating 17.2 μV from a low 1 V supply. The use of programmable bias currents in the amplifier, to exploit the trade-off between noise and power, was proposed to set each individual amplifier's noise level (and power) to meet requirements for accurate spike detection. Literature reviews of digital-signal processors and transmitters are used to construct approximate models of power versus performance. These models are then used to investigate the overall system power with different levels of digital processing. With a target application of neural spike recording, four modes (raw data, spike detection, feature extraction, and clustering) were analyzed. A system that uses feature extraction yields the lowest overall power, supports 400 channels with a practical wireless link, and consumes approximately 8 mW.