A Low-Noise, Wireless, Frequency-Shaping Neural Recorder

This paper presents a low-noise, wireless neural recorder that has a frequency dependent amplification to remove electrode offset and to attenuate motion artifacts. The recorder has 2.5 GQ and 50 MQ input impedance at 20 Hz and 1 kHz for recording local field potentials and extracellular spikes, respectively. To reduce the input-referred noise, we propose a low-noise frontend design with multiple novel noise suppression techniques. To reduce the power consumption, we have integrated an exponential component and polynomial component spike processor that automatically adjusts the recording bandwidth based on the signal contents. In bench-top measurement, the proposed neural recorder has 2.2-μV input-referred noise integrated from 300 Hz to 8 kHz and consumes 98-μW maximum power. In animal experiments, the output data of the neural signal processor are serialized and connected to a customized WiFi data link with up to 10 Mbps data rate. Through in vivo experiments, we find that the noise generated by the WiFi does not prevent brain recordings with microelectrodes and a clear interpretation of the neural signals; however, the noise can mask the weaker neural signals in nerve recordings with epineural electrodes.

I. Introduction

Neural recording circuits and systems are to acquire data from the nerve systems [1]–[8], where neural signals typically include multiple frequency components. For example, in the brain recording with penetrating probes, local field potentials (LFPs) appear at a low-frequency range from less than 1 Hz to 300 Hz while neural spikes are from 300 Hz to 8 kHz. The amplitude of neural signals depends on the animal preparations, the type of electrodes, and the recording locations, which can vary from a few (e.g. vagus nerve signals on epineural electrodes) to tens of mV (e.g. signals during a seizure). The neural signals are accompanied with various artifacts from muscles and organs, as well as interferences from ambient environment. Separating neural signals from artifacts and interferences requires a well characterized experimental environment and a high-precision data acquisition device.