Compact Low-Power Cortical Recording Architecture for Compressive Multichannel Data Acquisition

Abstract

This paper introduces an area- and power-efficient approach for compressive recording of cortical signals used in an implantable system prior to transmission. Recent research on compressive sensing has shown promising results for sub-Nyquist sampling of sparse biological signals. Still, any large-scale implementation of this technique faces critical issues caused by the increased hardware intensity. The cost of implementing compressive sensing in a multichannel system in terms of area usage can be significantly higher than a conventional data acquisition system without compression. To tackle this issue, a new multichannel compressive sensing scheme which exploits the spatial sparsity of the signals recorded from the electrodes of the sensor array is proposed. The analysis shows that using this method, the power efficiency is preserved to a great extent while the area overhead is significantly reduced resulting in an improved power-area product. The proposed circuit architecture is implemented in a UMC 0.18 [Formula: see text]m CMOS technology. Extensive performance analysis and design optimization has been done resulting in a low-noise, compact and power-efficient implementation. The results of simulations and subsequent reconstructions show the possibility of recovering fourfold compressed intracranial EEG signals with an SNR as high as 21.8 dB, while consuming 10.5 [Formula: see text]W of power within an effective area of 250 [Formula: see text]m × 250 [Formula: see text]m per channel.

Structured sampling and recovery of iEEG signals

“…The numerical results show that structured sampling, in combination with structured recovery, allows to more faithfully reconstruct the original signals, as compared with the traditional Bernoulli [4] or the multi-channel sampling [5] schemes.…”

“…Let X ∈ R N ×n be the signal matrix, each row containing the signal for one of the N channels. In order to promote group-sparsity, [5] proposed to use the 2,1 mixed norm, X 2,1 := n i=1 N j=1 X 2 i,j . In Figure 3 (top), we report the first 64 wavelet coefficients for the signals from two datasets, exhibiting the group structure among correlated channels.…”

“…The first, named BERN, uses the same random Bernoulli {±1} matrix to sample each channel independently [4]. The second, named Multi-Channel Sampling (MCS) [5], designed to be highly power-efficient, uses a Bernoulli {0, 1} matrix to sample across the channels at each time step. The compression achievable by this method depends on the number of samples taken at each time step, with a minimum of one sample, yielding a compression factor equal to the number of channels.…”

“…In order to reduce the power consumption of data transmission, many recent approaches (e.g., [3]- [5] and references therein) exploit compressive sensing (CS) [6], [7] to directly sample a compressed version of the signal. In a nutshell, CS consists in taking fewer linear samples than dictated by the Shannon-Nyquist theorem, while still allowing for robust offline signal reconstruction.…”

Learning-Based Near-Optimal Area-Power Trade-offs in Hardware Design for Neural Signal Acquisition

“…In order to reduce the power requirements of data transmission, compressive sensing (CS) [4,5] has been exploited in many recent approaches (e.g., [6,7,8] and references therein). In a nutshell, CS consists in taking fewer linear samples than dictated by the Shannon-Nyquist theorem, while still allowing robust off-line signal reconstruction.…”