Analysis and Reduction of Nonlinear Distortion in AC-Coupled CMOS Neural Amplifiers with Tunable Cutoff Frequencies

Beata Trzpil-Jurgielewicz,W Dąbrowski ,Paweł Hottowy

doi:10.3390/s21093116

Beata Trzpil-Jurgielewicz, W Dąbrowski + Show 1 more

Open Access

https://doi.org/10.3390/s21093116

Copy DOI

Abstract

Integrated CMOS neural amplifiers are key elements of modern large-scale neuroelectronic interfaces. The neural amplifiers are routinely AC-coupled to electrodes to remove the DC voltage. The large resistances required for the AC coupling circuit are usually realized using MOSFETs that are nonlinear. Specifically, designs with tunable cutoff frequency of the input high‑pass filter may suffer from excessive nonlinearity, since the gate-source voltages of the transistors forming the pseudoresistors vary following the signal being amplified. Consequently, the nonlinear distortion in such circuits may be high for signal frequencies close to the cutoff frequency of the input filter. Here we propose a simple modification of the architecture of a tunable AC-coupled amplifier, in which the bias voltages Vgs of the transistors forming the pseudoresistor are kept constant independently of the signal levels, what results in significantly improved linearity. Based on numerical simulations of the proposed circuit designed in 180 nm technology we analyze the Total Harmonic Distortion levels as a function of signal frequency and amplitude. We also investigate the impact of basic amplifier parameters—gain, cutoff frequency of the AC coupling circuit, and silicon area—on the distortion and noise performance. The post-layout simulations of the complete test ASIC show that the distortion is very significantly reduced at frequencies near the cutoff frequency, when compared to the commonly used circuits. The THD values are below 1.17% for signal frequencies 1 Hz–10 kHz and signal amplitudes up to 10 mV peak-to-peak. The preamplifier area is only 0.0046 mm2 and the noise is 8.3 µVrms in the 1 Hz–10 kHz range. To our knowledge this is the first report on a CMOS neural amplifier with systematic characterization of THD across complete range of frequencies and amplitudes of neuronal signals recorded by extracellular electrodes.

Highlights

Multielectrode neural interfaces are widely used in basic neuroscience research [1,2,3,4]and for development of advanced brain-computer interfaces and neuroelectronic prostheses [5,6,7]
SuilmatueldatTeHd DTHvDs. vssig. nsiaglnfarleqfrueeqnuceyncfoyrfoArCA-cCo-ucopulepdlendenuerualraalmamplpifiliefrie(rw(withith1 1HHz zcuctuotfoffffrferqeuquenencycy) )aannddfoforrvvaarrioi-us inpuotussiginnpaul tamsigpnliatluadmesp:li(tau)duessi:n(ag)tuhseinvgarthiaebvlea-rViagbslpe-sVegus dposeruesdiostroersi(sbto)ru(sbi)nugstihneg sthyemsmymetmriectfirixcefdix-eVdg-sVpgsspeuseduodroerseisstiostro. r. We note that these results clearly show that the variable-Vgs pseudoresistor configuration is not compatible with design requirements for the neural amplifier considered in this paper it may be a reasonable option for different applications
In order to confirm that the gate capacitances contribute significantly to the impedance of the feedback loop we show in Figure 4 the results of transient simulations for a sinewave signal of 2.5 Hz for three different sizes of transistors composing the pseudoresistor

Summary

Introduction

Multielectrode neural interfaces are widely used in basic neuroscience research [1,2,3,4]and for development of advanced brain-computer interfaces and neuroelectronic prostheses [5,6,7]. Taking advantage of large numbers of closely spaced microelectrodes, such systems make it possible to record activity of large neuronal populations with resolution of individual neurons, providing new insights into processing and coding of information in the brain circuits. Systems dedicated to large-scale recording of brain activity in human are being developed [11]. The neural signals acquired by extracellular electrodes are of two types. The action potentials (APs) can be recorded from individual neurons located close to the sensing electrode. An action potential is generated by a neuron when the total input signal received by this cell—either from sensory circuits of the central nervous system like the eyes or ears, or from other neurons—exceeds a specific threshold [12].

Objectives

Methods

Findings

Conclusion