2168-2356 (c) 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/MDAT.2015.2504367, IEEE
Design and Test
>IEEE Design & Test <
1
Abstract—Different types of multichannel wireless neural
recording architectures have been used for simultaneously
monitoring large populations of neurons. Yet it is not clear which
analog front-end (AFE) architecture would be most suitable for
systems with 100s of channels and beyond. In this paper,
theoretical analyses of three analog-to-digital conversion (ADC)
based and one analog-to-time conversion (ATC) based AFE
architectures have been presented, with particular emphasis on
the significant parameters that affect the design of neural
recording implants. In this theoretical analysis, comparisons are
made among models constructed in MATLAB for these neural
recording AFE architectures in terms of power consumption, area,
input-referred noise, crosstalk, and interference. Finally, the pros
and cons of these four architectures have been highlighted.
Index Terms—Brain-machine-interface (BMI), multichannel
neural recording, implantable microelectronic devices, low-power,
area-efficiency, neuroprostheses.
I. I
NTRODUCTION
eural recording systems are increasingly utilized for signal
acquisition, amplification, digitization, transmission, and
processing in invasive brain-machine-interfacing (BMI) app-
lications [1]-[3]. These are enabling tools for scientists to
explore the nervous system and to unlock the neural coding,
which is the basis of many neuroscience experiments. Another
application is in neural prostheses for the treatment of
neurological disorders. Since most functions of the nervous
system can only be well understood by simultaneously
observing the activities of large neuronal ensembles, high
channel-count recording is needed in BMI systems [1].
Over the past decades, the number of simultaneously
recorded neurons has been roughly doubling every 7 years [3],
[4]. Neural recording systems with several to hundreds of
channels have been reported [2], [4]-[16]. In design of
application specific integrated circuits (ASIC) for multichannel
neural recording, power consumption and size are key to reduce
surgical and thermal damage to the surrounding tissue [12],
[13]. Furthermore, maintaining low noise is necessary,
particularly in the analog front-end (AFE) to achieve sufficient
signal-to-noise ratio (SNR) throughout the system, while
Manuscript received May 29, 2015; revised October 19, 2015; accepted
November 23, 2015. This work was supported in part by the National Science
Foundation (ECCS-1408318) and the National Natural Science Foundation of
China (No. 61204029).
Xingyuan Tong is with the School of Electronic Engineering, Xi’an
University of Posts & Telecommunications, Xi’an 710121, China, and also
with the GT-Bionics Lab, School of Electrical and Computer Engineering,
Georgia Institute of Technology, Atlanta, GA 30308, USA. (mayxt@126.com)
Maysam Ghovanloo is with the GT-Bionics Lab, School of Electrical and
Computer Engineering, Georgia Institute of Technology, Atlanta, GA 30308,
USA. (e-mail: mgh@gatech.edu)
conditioning neural signals are within μV to mV range [14]. In
addition to optimizing individual integrated circuit blocks,
satisfying these requirements need detailed attention to the
architectural aspects of multichannel neural recording systems
in a way that power, size, input-referred noise, and crosstalk are
minimized.
Among architectures published in the literature, direct time
division multiplexing (TDM) of analog signals, analog-to-
digital conversion (ADC), and analog-to-time conversion
(ATC) are the main system-level AFE approaches for multi-
channel neural recording [4]-[19]. While analog TDM was
popular in early systems with lower number of channels due to
its simplicity and low power consumption, it has been
abandoned in recent designs due to crosstalk between adjacent
channels. The ADC-based structures can be found in three
types. The first type includes one ADC per channel, operating
at a low sampling rate, followed by TDM in the digital domain
[5]-[7]. The second type applies the TDM in the analog domain,
followed by a high speed ADC to take a sufficient number of
samples per channel [8]-[11]. In the third structure, the
recording channels are separated into several groups. These
recording channels in each group share one ADC through an
analog TDM block. The serial outputs of all group ADCs are
multiplexed again in the digital domain before transmission [4],
[12]-[14]. The ATC-based structure, encodes the amplitude of
the analog samples in the time domain using pulse-width-
modulation (PWM) in each AFE channel. The other half of the
conversion, i.e. time-to-digital conversion (TDC), is transferred
to the receiver (Rx) side, where power and size are not strictly
constrained [15], [16].
A survey of the recent literature revealed that despite high
performance implementations of multichannel wireless neural
recording systems, an analytical foundation for the optimal
choice of the AFE systematic architecture to guide designers on
the basis of the number of channels and characteristics of
different circuit blocks is missing. Especially, the theoretical
modeling and comparison between ADC-based and ATC-based
AFE architectures has not been presented. In a couple of
publications, [20], [21], the optimization of neural recording
AFE has been covered only for the ADC-based architecture. In
[22], introduction of multichannel neural recording system
architectures is provided, but with no theoretical calculation
and comparison.
In this paper, the abovementioned multichannel neural
recording AFE architectures have been discussed. Specifically,
the ADC-based and ATC-based AFE architectures are
compared intuitively by the MATLAB modeling results, and
theoretically analyzed with emphasis on the significant
parameters that affect the design of neural recording implants.
Multichannel Wireless Neural Recording AFE
Architectures Analysis, Modeling, and Tradeoffs
Xingyuan Tong, Member, IEEE, and Maysam Ghovanloo, Senior Member, IEEE