Introduction 介绍
Extracellular electrodes are a common tool for the recording and stimulation of electrically active tissue, with applications that go beyond the laboratory into medical implantable devices from pacemakers to deep brain stimulators. Biological neural tissue, and especially mammalian brains, possesses very large neural densities. To extract meaningful information from these networks, large electrode counts are required. In general the need to gain access to large neural populations has made multi-electrode arrays ubiquitous in neural research fields [1]–[3]. Such is the case for some of the ambitious brain–computer interfacing efforts for prosthetic applications [4], [5], in vivo applications [6]–[8], and in vitro applications such as our Micro Neural Interfacing System (
细胞外电极是记录和刺激电活性组织的常用工具,其应用不仅限于实验室,还扩展到从起搏器到脑深部刺激器的医用植入设备。生物神经组织,尤其是哺乳动物大脑,具有非常高的神经密度。为了从这些网络中提取有意义的信息,需要大量的电极。一般来说,获取大型神经群体的需求使多电极阵列在神经研究领域变得无处不在。这种情况在一些用于假肢应用的雄心勃勃的脑机接口研究中也是如此,用于体内应用以及体外应用,如我们的微神经接口系统(Micro Neural Interfacing System,NIS)在 3D 体外神经模型开发中的应用。
Microscale technologies have facilitated the commercialization of multi-electrode arrays (MEAs) that incorporate hundreds of electrodes [12]. Most existing commercial systems [Multi-Channel Systems (MCS), Plexon, Cyberkinetics, etc.], however, still use discrete components for the ancillary amplification and interfacing circuitry; this choice makes the transition to larger electrode counts cumbersome and costly. Several companies and researchers are addressing this problem by developing integrated circuits (ICs) with relatively large channel counts; these ICs reduce costs and facilitate significant size reductions for the interfacing hardware [8], [13]–[19].
微尺度技术促进了包含数百个电极的多电极阵列(MEA)的商业化。然而,大多数现有的商业系统(Multi-Channel Systems (MCS)、Plexon、Cyberkinetics 等)仍然使用离散组件进行辅助放大和接口电路;这种选择使得向更大电极数量的过渡变得繁琐且昂贵。为解决这一问题,几家公司和研究人员正在开发具有相对大量通道的集成电路(IC);这些 IC 降低了成本,并显著减小了接口硬件的尺寸。
Another problem, which has troubled electrophysiologists since early stimulation experiments, is the introduction of stimulation artifacts that obscure any neural activity near the stimulation site for tens or hundreds of milliseconds [20]. The stored electrode charge, which ultimately generates the artifact, introduces problems for long-term stimulation protocols, as it could cause ion migration or general recording system saturation. Although several investigators have tried to cancel the artifact after it has occurred [21], [22], and some groups have tried to eliminate it altogether [23], [24], there is little understanding of the artifact process and properties.
另一个自早期刺激实验以来一直困扰电生理学家的问题是刺激伪影的引入,这会在刺激部位附近数十或数百毫秒内掩盖任何神经活动 [20] 。最终生成伪影的存储电极电荷为长期刺激协议引入了问题,因为它可能导致离子迁移或记录系统的整体饱和。尽管几位研究人员试图在伪影发生后取消它 [21] , [22] ,一些研究组试图完全消除它 [23] , [24] ,但对伪影过程和性质的理解仍然很少。
In the present work we describe and characterize a stimulation, recording, and artifact elimination system (Fig. 1) developed around our custom ICs. Our artifact elimination system is built around a second-generation IC (Fig. 2) that introduces improvements in noise performance and stimulation circuitry with respect to the one described in [15]. This system was developed as part of a larger collaboration for the development of in-vitro 3-D neural systems under a Bioengineering Research Partnership grant.
在本工作中,我们描述并表征了围绕我们定制的集成电路( Fig. 1 )开发的刺激、记录和伪影消除系统。我们的伪影消除系统基于第二代集成电路( Fig. 2 ),相较于 [15] 中描述的系统,它在噪声性能和刺激电路方面引入了改进。该系统是作为生物工程研究合作伙伴计划下开发的体外三维神经系统更大合作的一部分而开发的。
Stimulation Artifacts 刺激伪影
Although further investigation is required to completely describe the stimulation artifact, we will present an overview of the principal factors that are relevant to this work. We focus on the use of a linear RC model of the electrode as a first-order approximation to the problem. In order to help standardize the study of stimulation artifacts and the factors that influence them, we provide our definitions while differentiating among the multiple possible artifact sources.
虽然需要进一步调查才能完全描述刺激伪影,但我们将概述与这项工作相关的主要因素。我们重点介绍将电极的线性 RC 模型作为问题的一阶近似。为了帮助标准化对刺激伪影及其影响因素的研究,我们在区分多种可能的伪影来源的同时提供我们的定义。
A. Artifact Origin A. 文物来源
The stimulation artifact is a direct consequence of the accumulated charge in the electrode–electrolyte interface during stimulation and of the effect that this charge has on the signal-chain filters and other elements after stimulation [25]. The problem is one of relative scales: stimulation signals are on the order of hundreds of millivolts, while recorded signals are on the order of tens of microvolts (which requires noise levels on the order of a few microvolts). Assuming that charge balancing is made part of the stimulation protocol, very small mismatches of 1% or less, which are common and acceptable in traditional circuit and analog signal processing designs, generate artifacts that would saturate the signal acquisition chain in extracellular recordings. The problem is further complicated by the redox reactions that take place at the electrode–electrolyte interface and make charge balancing difficult to accomplish. If no attempt at charge balancing is made, such saturation would be even more pronounced and longer lasting.
刺激伪迹是电极-电解质界面在刺激过程中累积电荷的直接后果,以及这种电荷在刺激后的信号链过滤器和其他元件上的影响。问题在于相对规模:刺激信号在数百毫伏的范围内,而记录信号在数十微伏的范围内(这需要噪声水平在几微伏的范围内)。假设在刺激协议中包含了电荷平衡,那么在传统电路和模拟信号处理设计中常见且可接受的 1%或更小的微小不匹配会产生伪迹,从而使细胞外记录中的信号采集链饱和。这个问题因电极-电解质界面发生的氧化还原反应而变得更加复杂,使电荷平衡难以实现。如果不尝试进行电荷平衡,那么这种饱和现象将更加明显且持续时间更长。
为了在刺激后允许信号的记录,刺激电极上的电荷必须被消散到允许放大电路正常运行的水平。对于典型的刺激协议,其中电极中可能残留 200 mV 或更多的电压,为了完全消除伪影(即将残留电压降至电极噪声水平),电极电荷必须被消散到十万分之一或更低。为了能够恢复信号,对于典型的信号放大链会在大约 2 mV 饱和,残留电极电荷必须被消散到百分之一或更低。使用 Fig. 3 [26] 中显示的线性电极模型和该模型预测的电极电荷的指数衰减,我们需要让电极的五个时间常数过去以恢复信号;为了消除伪影(即将电荷消散到十万分之一),至少需要经过十一时间常数。对于一个 40
A way to reduce the artifact duration would be to reduce the electrode discharge time constant, which can be achieved by connecting the electrode to its poststimulation stabilization voltage through a low impedance path. Such a connection would reduce the time constant from the 10 ms of our previous example to approximately 200
减少伪影持续时间的一种方法是减少电极放电时间常数,这可以通过通过低阻抗路径将电极连接到其刺激后稳定电压来实现。这种连接会将时间常数从我们之前例子中的 10 毫秒减少到大约 200 微秒,减少了 50 倍。这种减少将使伪影在 1 毫秒内进入记录电路的线性范围,并在 2.2 毫秒内进入噪声带;这些持续时间与预期的神经反应时间相称。由于电极的刺激后电压不一定提前知道,因此电极的刺激前电压可以用作合理的近似值。
The long-lasting artifact dependencies and nonlinear processes that result from redox reactions at the electrode–electrolyte interface are not captured by the linear electrode model; the interaction of these factors with the stimulation signal would be seen as temporary changes in the linearized electrode characteristics which could translate into additional artifact problems. Although more elaborate electrode models are readily available [27], [28], the linear model is sufficient to approximate most of the artifact behavior and, because of our use of continuous feedback, the behavior of the circuitry itself [25].
由于电极-电解质界面的氧化还原反应导致的持久性伪影依赖性和非线性过程,线性电极模型无法捕捉到这些现象;这些因素与刺激信号的相互作用会表现为线性化电极特性的暂时变化,这可能会转化为额外的伪影问题。尽管更复杂的电极模型随时可用 [27] , [28] ,但线性模型足以近似大多数伪影行为,并且由于我们使用连续反馈,电路本身的行为 [25] 。
It is worth pointing out that most existing designs attempt to cancel the artifact from the signal chain after it has been produced (the main exception being [23]). Our design seeks to eliminate the artifact from the electrode itself. Additionally, our design, by placing the electrode in a feedback loop and allowing for the continuous variation of system parameters, should be able to compensate for factors (e.g., nonlinearities in the electrode) that the existing, feed-forward, designs are not able to address.
值得指出的是,大多数现有设计试图在伪影产生后从信号链中消除它(主要例外是 [23] )。我们的设计旨在从电极本身消除伪影。此外,我们的设计通过将电极置于反馈回路中并允许系统参数的连续变化,应该能够补偿现有前馈设计无法解决的因素(例如电极中的非线性)。
B. Artifact Duration B. 工件持续时间
Given the nature of the discharge behavior, it is not really possible to specify objectively the point at which the artifact has dissipated because some long-lasting baseline shifts might be introduced; these shifts are at least partially due to electrode nonlinearities and can vary with the electrochemical environment and the stimulation history. Requiring the remaining artifact voltage to become comparable to the recording noise level after stimulation would be overly conservative for most applications and could introduce data artifacts due to dependance on the slow components of the phenomena that we are trying to measure.
鉴于放电行为的性质,实际上不可能客观地指定工件消散的确切时间点,因为可能会引入一些长期存在的基线漂移;这些漂移至少部分是由于电极的非线性引起的,并且会随着电化学环境和刺激历史而变化。要求在刺激后剩余的工件电压与记录噪声水平相当对于大多数应用来说过于保守,并且可能会由于我们试图测量的现象的慢成分而引入数据工件。
With regard to the recording range of the system (or, more restrictively, the linear range of the recording system) we can classify stimulation artifacts as nonsaturating or saturating. As long as the signal chain (i.e., all amplifiers and other elements in the recording path) is not saturated or overly distorted, the artifact can be eliminated using signal processing methods (SALPA [29], filtering, etc.). Such processing would be required before postprocessing algorithms, such as spike detection or spike sorting, can be applied to the signal. If the signal chain saturates, any signal present in that interval is lost and no amount of signal processing will be able to recover it, this saturating artifact, is what must be eliminated or at least reduced as much as possible. This saturating artifact is what traditionally makes recording impossible for tens or hundreds of milliseconds after stimulation.
关于系统的记录范围(或者更严格地说,记录系统的线性范围),我们可以将刺激伪影分类为非饱和或饱和。只要信号链(即记录路径中的所有放大器和其他元件)没有饱和或过度失真,就可以使用信号处理方法(SALPA [29] ,过滤等)来消除伪影。在应用诸如尖峰检测或尖峰分类等后处理算法之前,需要进行这样的处理。如果信号链饱和,任何存在于该时间间隔内的信号都将丢失,任何信号处理都无法恢复它,这种饱和伪影是必须消除或至少尽可能减少的。正是这种饱和伪影,传统上使得在刺激后几十或几百毫秒内的记录变得不可能。
To be able to make the distinction between nonsaturating and saturating stimulation artifacts, the whole signal chain must be taken into account; gains, filters, amplifiers, digital converter resolution, and all voltage ranges have to be considered (which will be different for each system and setup). Given typical resolutions and extracellular signal magnitudes for MEAs, ranges of approximately 2 mV are common. For the purpose of this paper, we chose to define artifact duration as the time from the end of stimulation to the time that the recording system returns to within 200
为了能够区分非饱和和饱和的刺激伪影,必须考虑整个信号链;增益、滤波器、放大器、数字转换器分辨率和所有电压范围都必须考虑(每个系统和设置会有所不同)。考虑到 MEA 的典型分辨率和细胞外信号幅度,大约 2 mV 的范围是常见的。为了本文的目的,我们选择将伪影持续时间定义为从刺激结束到记录系统恢复到刺激前电极电压的 200 μV 以内的时间。尽管这个值在推广我们的技术方面并不是一个好的选择,这样保守的阈值允许在非常不同的系统之间进行直接比较;并且它使得信号可以在剩余的伪影之上被观察到,而无需任何额外的处理。
C. Artifacts Resulting From Circuit Effects
C. 由于电路效应产生的伪影
Although the primary cause of stimulation artifacts is the charge stored in the electrode, the artifact can be aggravated by effects from the intervening circuitry. These effects are generally avoidable through careful circuit design and layout. Here we highlight how our system deals with circuit effects.
尽管刺激伪影的主要原因是电极中存储的电荷,干扰电路的影响也会加重伪影。这些影响通常可以通过精心的电路设计和布局来避免。在这里,我们重点介绍了我们的系统如何处理电路效应。
1. DC Rejection Passband 1. 直流抑制通带
Among the circuit effects, the highpass characteristics of the dc rejection filters in the recording circuitry can be the largest contributors to the stimulation artifact. When the system returns to recording mode, the transition caused by the stimulation-induced electrode offsets (and other offsets) will excite such filters. This voltage step can generate long-lasting and large magnitude effects. The best way to avoid this problem is to eliminate the charge-induced offset, before the signal reaches the filters in the recording path; this approach, however, is not always possible or convenient. An alternative is to use first order high pass elements and the highest highpass frequency acceptable for the application (which restricts the duration of such effects). Satisfying the conflicting requirements of signal bandwidth and artifact reduction can lead to undesirable compromises; we have avoided this tradeoff by varying the filter poles during artifact elimination (see Section III-D).
在电路效应中,记录电路中的直流抑制滤波器的高通特性可能是刺激伪影的最大贡献者。当系统返回到记录模式时,由刺激引起的电极偏移(及其他偏移)引起的转换会激发这些滤波器。这个电压阶跃会产生持久且幅度大的效应。避免此问题的最佳方法是在信号到达记录路径中的滤波器之前消除电荷引起的偏移;然而,这种方法并不总是可能或方便的。另一种方法是使用一阶高通元件和应用中可接受的最高高通频率(这限制了此类效应的持续时间)。满足信号带宽和伪影减少的冲突要求可能导致不理想的妥协;我们通过在伪影消除过程中改变滤波器极点来避免这种权衡(参见 Section III-D )。
2. Charge Injection 2. 电荷注入
Many early artifact suppression schemes were hampered by switching element charge injection [30], which is magnified by the required amplifier sensitivity. To reduce some of these effects, some designers amplify and limit the signal before introducing switching elements, thus limiting the introduced transients (e.g., the MCS MEA1060-BC preamplifiers and [23]). Our designs avoid this problem altogether by making all charge injection paths into common mode current bias paths [15]; that way, the existing common mode attenuation of differential amplifiers and the low impedance of the involved nodes drastically reduce any effects due to charge injection.
许多早期的伪影抑制方案受到开关元件电荷注入的阻碍,这种现象因所需放大器的灵敏度而放大。为了减少其中一些影响,一些设计师在引入开关元件之前对信号进行放大和限制,从而限制引入的瞬态(例如,MCS MEA1060-BC 前置放大器等)。我们的设计完全避免了这个问题,通过将所有电荷注入路径变为共模电流偏置路径;这样,差分放大器的现有共模衰减和相关节点的低阻抗大大减少了由于电荷注入引起的任何影响。
3. Circuit Mismatch 3. 电路不匹配
If we dynamically insert and remove elements from the signal path, or actively modify the characteristics of the elements themselves, any common-mode mismatch problem will cause transients on element switching. In our own design, due to the extremely low biases required, changing the highpass filter cutoff frequency introduces a mismatch transient that is a function of the filter's bias currents. The same is true for the interaction of the discharge path bias current with the electrode impedance and the remaining electrode charge. These mismatches can cause transients which might look similar to those generated by charge injection. A time-consuming circuitry redesign will inevitably face diminishing returns as such mismatches would be unavoidable without additional offset trimming [31] and it is physically impossible to alter the electrode discharge path interaction. Slowly changing any variable that is known to induce mismatch-related artifacts provides an alternative to reduce the transients caused by these phenomena. Our current firmware implementation can make use of this technique for the discharge path current.
如果我们动态地在信号路径中插入和移除元件,或者主动修改元件自身的特性,任何共模失配问题都会在元件切换时引起瞬态。在我们自己的设计中,由于需要极低的偏置,改变高通滤波器的截止频率会引入一个与滤波器偏置电流有关的失配瞬态。同样,放电路径偏置电流与电极阻抗和剩余电极电荷的相互作用也会如此。这些失配会引起类似于电荷注入产生的瞬态。一个耗时的电路重设计将不可避免地面临收益递减的问题,因为在没有额外的偏移调整的情况下,这种失配是不可避免的,并且在物理上不可能改变电极放电路径的相互作用。缓慢改变任何已知会引起失配相关伪影的变量提供了一种减少这些现象引起的瞬态的替代方法。我们当前的固件实现可以利用这一技术来处理放电路径电流。
4. Crosstalk 串扰
In any system with multiple channels in close proximity, signals can interfere with each other, a phenomenon known as crosstalk, signal coupling, or feed-through, depending on context. In a mixed-mode system (analog and digital), or in other systems in which large magnitude signals are present alongside small signals and amplification elements, the crosstalk problem can become more severe. For example, stimulation signals due to their large magnitude can introduce crosstalk into adjacent channels. This effect can be reduced through careful layout techniques or by adding design elements to increase the electrical separation between signals. Although our system can reduce external coupling to recording channels by providing a low-impedance path during stimulation, our current design has some layout oversights that introduce additional crosstalk problems (see Section V-A.3 and Fig. 15).
在任何具有多个相邻通道的系统中,信号之间会发生干扰,这种现象在不同的上下文中被称为串扰、信号耦合或馈通。在混合模式系统(模拟和数字)中,或者在存在大幅度信号和小信号以及放大元件的其他系统中,串扰问题可能会变得更加严重。例如,由于其大幅度,刺激信号可能会对相邻通道引入串扰。可以通过仔细的布局技术或添加设计元素以增加信号之间的电气隔离来减少这种效应。尽管我们的系统可以通过在刺激期间提供低阻抗路径来减少外部耦合到记录通道,但我们当前的设计存在一些布局上的疏忽,导致了额外的串扰问题(参见 Section V-A.3 和 Fig. 15 )。
IC Design 集成电路设计
The focus of this paper is on the design of the system built around our second generation 16 channel stimulation, artifact elimination, and recording IC and the evaluation of the system for its intended biological application. Most of the details of the design of the IC have been presented previously [15], the main design changes can be seen in Table I. In this section, we highlight the improvements with respect to the previous design and expand on details that are relevant to this work.
本文的重点是围绕我们的第二代 16 通道刺激、伪影消除和记录 IC 设计的系统及其在预期生物应用中的评估。有关 IC 设计的大部分细节已在先前的文献中介绍过 [15] ,主要的设计变化可以在 Table I 中看到。在本节中,我们强调了相对于先前设计的改进,并扩展了与本工作相关的细节。
A. Artifact Elimination A. 消除伪影
The artifact elimination circuitry was described previously [15]; for ease of reference we reproduce the circuitry during the discharge phase in Fig. 3, and an equivalent circuit model (valid for most of the discharge when the circuitry is out of saturation) can be seen in Fig. 4. As the equivalent model indicates, and given the subthreshold circuit elements, the discharge phase is roughly equivalent to connecting the electrode, through a controllable resistor given by
伪影消除电路之前已描述过 [15] ;为了便于参考,我们在 Fig. 3 中重现了放电阶段的电路,并且在 Fig. 4 中可以看到一个等效电路模型(在电路未饱和的大部分放电期间有效)。如等效模型所示,考虑到亚阈值电路元件,放电阶段大致相当于通过
Keep in mind that, besides a dependency on the involved ionic species and concentrations,
请记住,除了依赖所涉及的离子种类和浓度外,
We can see from the equivalent circuit of Fig. 4 that
从 Fig. 4 的等效电路中可以看出,
B. Stimulation Buffer B. 刺激缓冲液
The stimulation buffer, presented previously [15], had a significant current limitation due to its bias network. The biasing complexities of the circuit and the operating range of the design prompted us to simplify the circuitry (allowing for additional functionality). The new stimulator (Fig. 5) is a simple operational transconductance amplifier (OTA) in a follower configuration in which input signals and bias currents are switched between stimulation levels. This configuration allows for the application of an arbitrary externally generated stimulation waveform or for the use of current stimulation under some bias and electrode conditions (if the OTA's inputs differ by more than approximately 100 mV, it would behave as a constant current source). This circuitry can provide close to 100
先前介绍的刺激缓冲液 [15] 由于其偏置网络存在显著的电流限制。电路的偏置复杂性和设计的操作范围促使我们简化电路(以增加功能)。新的刺激器( Fig. 5 )是一个简单的跟随器配置的运算跨导放大器(OTA),其中输入信号和偏置电流在刺激水平之间切换。这种配置允许应用任意外部生成的刺激波形或在某些偏置和电极条件下使用电流刺激(如果 OTA 的输入差异超过约 100 mV,它将表现为恒定电流源)。该电路可以在
C. Recording Noise
As reported in [15], the noise level of our previous design was larger than expected; furthermore, subsequent evaluation revealed that the design noise specifications had to be further reduced. To overcome this problem without greatly increasing the amplifier size (or a risky redesign), a simple solution was carried out, which implied an increase of the voltage gain of the recording amplifier to 200 (from 50) by increasing its input capacitanceD. Pole Shifting D. 极移
Given that part of the stimulation artifact is due to the filter response of the recording circuitry, a recent work by DeMichele and Troyk [21] and some testing with our previous generation IC suggest the implementation of a temporary recording bandwidth modification, or pole shifting—that is, increasing the recording highpass pole in the stimulating channel to increase the recovery speed of the amplifiers. Given that we have dynamic control over the highpass pole of our main amplifier (through
鉴于部分刺激伪影是由于记录电路的滤波响应造成的,DeMichele 和 Troyk 最近的研究 [21] 以及我们上一代 IC 的一些测试表明,可以实施临时记录带宽修改或极点移动,即增加刺激通道中的记录高通极点以提高放大器的恢复速度。鉴于我们可以动态控制主放大器的高通极点(通过
E. IC Controls E. IC 控制
As a test design, the control of the multiple digital features of the IC was implemented through simple 16-bit latched shift registers, using one bit per channel. Each one of the five digital variables has its own dedicated shift-register, namely: Amplifier activation, pole shifting, stimulation buffer activation, stimulation phase, and discharge buffer activation.
作为测试设计,IC 的多个数字功能的控制是通过简单的 16 位锁存移位寄存器实现的,每个通道使用一位。每一个五个数字变量都有自己专用的移位寄存器,即:放大器激活、极点移位、刺激缓冲器激活、刺激相位和放电缓冲器激活。
Additionally, there are eight analog values to be controlled: highpass pole frequency, highpass pole shift frequency, stimulation high and low levels, stimulation high and low currents, discharge bias current, and amplifier reference level. Several of these analog input variables include resistive dividers that serve to increase the usable resolution of the external drive voltages while keeping the internal voltage inside a reasonable range. An external resistor, in combination with internal reference circuitry, sets the main amplifier bias current (25
另外,有八个模拟值需要控制:高通极点频率、高通极点移位频率、刺激高低水平、刺激高低电流、放电偏置电流和放大器参考电平。其中一些模拟输入变量包括电阻分压器,这些分压器在保持内部电压在合理范围内的同时,增加了外部驱动电压的可用分辨率。一个外部电阻与内部参考电路结合,设置主放大器偏置电流(25 μA)。
System Design 系统设计
The test system consists of a custom board, depicted in Fig. 6, with an isolated power supply for the circuitry, a PIC microcontroller (Microchip 18LF452) running at 30 MHz, twelve 12-bit digital to analog converter (DAC) channels (three Analog Devices DAC8420 ICs with four channels each), 32 discrete operational amplifiers (eight Texas Instruments LF347 with 2% 10K
测试系统由一个定制板组成,如 Fig. 6 所示,电路具有隔离电源,运行在 30 MHz 的 PIC 微控制器(Microchip 18LF452),十二个 12 位数字到模拟转换器(DAC)通道(三个 Analog Devices DAC8420 IC,每个有四个通道),32 个离散运算放大器(八个德州仪器 LF347,带有 2% 10K
The microcontroller firmware is programmed to sequence the signals for the ICs, so that the different biases are present, and the required circuit elements are turned on and off in the different recording, stimulation, and discharge phases. Switching in between phases can take 40
微控制器固件被编程为对 IC 的信号进行排序,以便存在不同的偏置,并且在不同的记录、刺激和放电阶段中所需的电路元件被打开和关闭。阶段之间的切换可以在 40 微秒或更短的时间内完成,具体取决于需要更改的信号数量和其他固件要求。系统的整体时间分辨率和抖动在 600 纳秒或更少的范围内,尽管接口要求将时间限制为 10 微秒的增量(固件校准常数用于补偿大多数延迟)。微控制器的附加光隔离触发输入可用于启动刺激序列。上述系统时间限制是由我们当前选择的接口(包括固件和 IC 接口)引起的。未来的实现可以通过固件修改、使用专用定制数字硬件,最终通过重新设计 IC 数字接口来消除所有这些限制。
Results 结果
我们已经描述了伪影消除系统,并在实际的神经刺激和记录应用中证明了定制 aVLSI 电路及其辅助板的有效性。由于伪影消除性能是本研究的主要特点,我们将重点放在评估设计和方法的这一方面(尽管为了完整性,我们在本节开始时提供了一些基本的放大器特性数据)。
A. IC Characterization A. IC 特性表征
To characterize the performance of the IC recording path, a Stanford Research Systems SR785 dynamic signal analyzer was successively connected to each channel under identical bias conditions, and the whole board gain, bandwidth and noise characteristics were measured in all 16 of the IC channels.
为了表征 IC 记录路径的性能,斯坦福研究系统 SR785 动态信号分析仪在相同偏置条件下依次连接到每个通道,并测量了所有 16 个 IC 通道的整个板增益、带宽和噪声特性。
1. Channel Uniformity 1. 渠道均匀性
In Fig. 7, the average response for a highpass filter setting of 10 Hz shows that at 500 Hz, an average gain of 3195, with a standard deviation of 26, was obtained (that is, 70
在 Fig. 7 ,高通滤波器设置为 10 Hz 的平均响应显示,在 500 Hz 时获得了 3195 的平均增益,标准偏差为 26(即 70
2. Input Referred Noise 2. 输入参考噪声
From Fig. 8, we can see that for a functional highpass setting of 10 Hz, the RMS noise in the 200Hz–3 kHz bandwidth of interest is of 3.0
从 Fig. 8 ,我们可以看到,对于 10 Hz 的高通滤波设置,感兴趣的 200Hz–3 kHz 带宽内的 RMS 噪声为 3.0
According to specifications, our noise is still three times that of the MCS MEA-1060 amplifier. To evaluate our noise figure, we compared our IC to the MEA-1060 connected to an active MEA under the same conditions and recording information from the same culture (see Section V-C.1). Defining SNR as peak recorded neural spike value divided by the RMS noise of the channel and averaging across 6 active electrode channels, we found that the MCS amplifier provided an SNR of 12.8 (
根据规格,我们的噪声仍然是 MCS MEA-1060 放大器的三倍。为了评估我们的噪声系数,我们将我们的 IC 与 MEA-1060 在相同条件下连接到一个活跃的 MEA 并记录同一培养物的信息进行了比较(见 Section V-C.1 )。将 SNR 定义为记录的神经尖峰峰值除以通道的 RMS 噪声,并对 6 个活跃电极通道取平均值,我们发现 MCS 放大器提供的 SNR 为 12.8(
3. Crosstalk 串扰
The present IC design has some layout oversights, mostly due to the addition of output buffers, that substantially increased the crosstalk in between channels in comparison with our previous-generation IC. Such crosstalk is particularly problematic during stimulation, given that it introduces artifacts in the recording path of adjacent nonstimulating channels. Although this is only noticeable thanks to our artifact removal architecture, we reduced this problem by modifying our stimulation protocol, turning off nonstimulating channels during the presence of the largest offending signals, namely stimulation and the initial phase of electrode discharge; we also have the possibility of activating the discharge circuitry on the recording electrodes during the stimulation phase.
目前的 IC 设计存在一些布局疏漏,主要是由于增加了输出缓冲器,与我们上一代 IC 相比,这大大增加了通道之间的串扰。这种串扰在刺激过程中尤其成问题,因为它在相邻的非刺激通道的记录路径中引入了伪影。尽管这仅仅是由于我们的伪影去除架构才得以察觉,但我们通过修改刺激协议来减少这个问题,在存在最大干扰信号(即刺激和电极放电的初始阶段)时关闭非刺激通道;我们还可以在刺激阶段激活记录电极上的放电电路。
B. Artifact Behavior B. 器物行为
To validate the circuitry, a 40
为了验证电路,一个 40
It is worth noting that, as a consequence of the design, the actual discharge current will be determined by the charge and voltage of the electrode itself; that said, for all discharge settings in this paper, as we use a symmetric biphasic voltage stimulus, the current to the electrode will have three phases: 1) a “precharge” (anodic) phase in which the electrode voltage is raised, thus increasing the available current for the next phase; 2) a “stimulation” (cathodic) phase in which the transition to a negative voltage generates the stimulation current in the MEA [32], which charges the electrode to a negative voltage (such symmetry in voltage causes asymmetry in current and charge); and finally, 3) a decaying anodic discharge current to drive the electrode to its resting potential, which will be present in the media and could affect neural response [35]. It is important to keep in mind that for some stimulation protocols the constraints on discharge current might impose some additional performance limitations; likewise, the use of different stimulation protocols (e.g., charge-balanced current or voltage stimulation) will improve on the reported performance.
值得注意的是,由于设计的原因,实际放电电流将由电极本身的电荷和电压决定;也就是说,对于本文中的所有放电设置,由于我们使用对称的双相电压刺激,电极的电流将有三个阶段:1)“预充电”(阳极)阶段,在此阶段电极电压升高,从而增加下一个阶段的可用电流;2)“刺激”(阴极)阶段,在此阶段过渡到负电压在 MEA 中产生刺激电流,这会使电极充电到负电压(这种电压对称性导致电流和电荷不对称);最后,3)衰减的阳极放电电流将电极驱动到其静息电位,这将在介质中存在并可能影响神经响应。需要记住的是,对于某些刺激协议,放电电流的限制可能会带来一些额外的性能限制;同样,使用不同的刺激协议(例如,电荷平衡电流或电压刺激)将改善报告的性能。
1. Artifact Duration 1. 器物持续时间
The recording range present in Fig. 9
Fig. 9
To show the meaning of the measures presented in subsequent figures, the data traces in Fig. 9 are presented alongside several artifact duration measurement thresholds, and the inset shows the effect of the choice of threshold on the final measurement. Note that the relation between the voltage threshold and the artifact duration is roughly linear, and that our choice of a lower threshold imposes a penalty of two to five milliseconds over a higher threshold that accounts for the recording range.
为了显示后续图中所示措施的意义, Fig. 9 中的数据轨迹与几个伪影持续时间测量阈值一起呈现,插图显示了阈值选择对最终测量的影响。请注意,电压阈值与伪影持续时间之间的关系大致呈线性,并且我们选择较低的阈值会比考虑记录范围的较高阈值多出两到五毫秒的惩罚。
2. Discharge Current and Time
2. 放电电流和时间
For clarity, and as a consequence of our design choices, we have divided the discharge process into two discharge periods. The initial one significantly lowers the charge of the electrode, while the second one allows us to better visualize the artifact. Fig. 10 shows the effect of the initial electrode discharge on the artifact duration. At the initiation of discharge, the main amplifier is turned on, and the large electrode offset drives it (and the discharge amplifier) out of its linear range (the initial spike seen in Fig. 9). During these few microseconds the equivalent circuit in Fig. 4 will not be valid, and the discharge rate will be directly proportional to the available discharge current
为了清晰起见,并作为我们设计选择的结果,我们将放电过程分为两个放电阶段。初始阶段显著降低电极的电荷,而第二阶段则使我们能够更好地观察伪影。 Fig. 10 显示了初始电极放电对伪影持续时间的影响。在放电开始时,主放大器被打开,大电极偏移将其(以及放电放大器)驱动出其线性范围( Fig. 9 中看到的初始尖峰)。在这几微秒内, Fig. 4 中的等效电路将无效,放电速率将与可用放电电流
To avoid such transients, a second discharge phase at a lower discharge current can be used to reduce overall discharge time. Fig. 11 shows the effect of the second discharge phase after the application of a 500-
为了避免这种瞬变,可以使用较低放电电流的第二放电阶段来减少总体放电时间。 Fig. 11 显示了在应用 500-
3. Soft Switching 3. 软切换
As large transients can affect the filter and the electrode itself, we have the option of switching smoothly between discharge values. Fig. 12 shows the effect of a smooth curve that reaches the second discharge phase with a zero time derivative (a voltage parabola distorted by the exponential voltage-to-current relationship of the bias circuitry). Note that the artifact duration has been further reduced by approximately 1 ms even though overall discharge time remains the same, the average discharge impedance is higher, and the transition at the end of discharge remains unchanged. Although we are not sure of the mechanism for such artifact reduction—and we found it while implementing a way to continue discharging the electrode during recording—we have found some evidence that the fast transients affect the electrode itself rather than the recording path (see Section V-B.5).
由于较大的瞬态会影响滤波器和电极本身,我们可以选择在放电值之间平滑切换。 Fig. 12 显示了一条平滑曲线的效果,该曲线以零时间导数达到第二个放电阶段(由偏置电路的指数电压-电流关系扭曲的电压抛物线)。请注意,即使总放电时间保持不变,平均放电阻抗更高,并且放电结束时的过渡保持不变,伪影的持续时间也减少了大约 1 毫秒。虽然我们不确定这种伪影减少的机制,并且我们是在实现一种在记录期间继续放电电极的方法时发现的,但我们发现了一些证据表明快速瞬态影响的是电极本身,而不是记录路径(见 Section V-B.5 )。
4. Pole Shifting 4. 极移
As was made clear by the design in [21], the frequency characteristics of the amplifier itself contribute to the artifact, so a way to further reduce the artifact duration is to modify the frequency response of the main amplifier for a short time period after discharge. Fig. 13 shows this effect. Due to the way the poles are controlled in our ICs (see Section II-C.3), this method will introduce some artifacts due to switching between offset levels (no soft switching has been implemented for this path yet). As long as the bandwidth is not considerably reduced, this mode of artifact reduction has the advantage that recording can take place during the period of pole shifting (as is also true, though harder to accomplish, for the second discharge phase), but we must keep in mind that the pole shift setting affects the recording chain itself and has no direct effect on the electrode.
如 [21] 的设计所明确指出的那样,放大器本身的频率特性会导致伪影,因此进一步减少伪影持续时间的方法是,在放电后的一段时间内修改主放大器的频率响应。 Fig. 13 显示了这一效果。由于我们 IC 中的极点控制方式(见 Section II-C.3 ),该方法会在偏移电平之间切换时引入一些伪影(此路径尚未实现软切换)。只要带宽没有显著减少,这种伪影减少模式的优点是极点转移期间可以进行记录(虽然在第二次放电阶段也是如此,但更难实现),但我们必须记住,极点转移设置会影响记录链本身,对电极没有直接影响。
5. Remaining Discharge 5. 剩余放电
In Fig. 13(a), we can see an apparently exponential decay after the main artifact—which is present, though not always evident, in the other figures—this decay has roughly a 4.4-ms time constant, which is slower than any time constant on the electronics path thus pointing to the electrode as the most likely source. All the curves reach the same exponential decay, which is to be expected as pole shifting only changes the recording path and not the electrode. From Fig. 12 a time constant of 4.8 ms can be extracted (which is within measurement error from the previous value), although interestingly the soft-switch curve is roughly 12
在 Fig. 13(a) 中,我们可以看到在主要伪影之后显然呈指数衰减——这种伪影在其他图中也存在,尽管并不总是明显的——这种衰减大约有 4.4 毫秒的时间常数,这比电子路径上的任何时间常数都要慢,因此指向电极作为最可能的来源。所有曲线都达到相同的指数衰减,这是可以预期的,因为极点移动只改变记录路径而不改变电极。从 Fig. 12 中可以提取出 4.8 毫秒的时间常数(这在测量误差范围内与前一个值一致),尽管有趣的是,软切换曲线大约比突然切换曲线低 12
Extracting time constants from the data in the other discharge curves (Figs. 10 and 11), we find that their values are not consistent, ranging from 15 down to 5 ms (and many of the traces have sign-inversion which suggests over-discharge). These measurements highlight some of the limitations of the linear model, as such variation cannot be explained with constant capacitance and resistance values. Nonetheless, the time constants are in the right order of magnitude, as required by the RC model of the electrode (approximately 10 ms). Interestingly, there seems to be a relationship between how fast the remaining artifact decays and how fast the discharge is, so that an undischarged electrode would generate not only the longest but also the slowest decaying artifact.
从其他放电曲线( Figs. 10 和 11 )中提取时间常数,我们发现它们的值并不一致,范围从 15 毫秒到 5 毫秒(许多曲线都有符号反转,这表明过放电)。这些测量结果突显了线性模型的一些局限性,因为这种变化不能用恒定的电容和电阻值来解释。尽管如此,这些时间常数的数量级是正确的,符合电极的 RC 模型(大约 10 毫秒)。有趣的是,剩余伪影的衰减速度与放电速度之间似乎存在某种关系,因此,未放电的电极不仅会产生最长的伪影,还会产生衰减最慢的伪影。
C. Biological Tests C. 生物测试
1. Methods 1. 方法
Dissociated E18 hippocampal neurons (Brain Bits1) were plated on an MEA (30
解离的 E18 海马神经元(Brain Bits 1 )被接种到 MEA(30
2. Results 2. 结果
Cultured hippocampal neurons were spontaneously active, and synchronized bursting activity could be routinely recorded through our IC. As indicated in Section V-A.2 the quality of the neural recording was comparable to the commercial preamplifier system.
培养的海马神经元自发活跃,并且可以通过我们的 IC 常规记录同步爆发活动。如 Section V-A.2 所示,神经记录的质量可与商业前置放大器系统相媲美。
Fig. 14 shows the stimulation and recording from the IC. When biphasic voltage pulses (positive–negative, 100 or 500mV) were applied, artifact-free recordings were possible after 3
Fig. 14 显示了来自 IC 的刺激和记录。当施加双相电压脉冲(正负 100 或 500mV)时,3
Fig. 15 demonstrates the capability of simultaneous multichannel recording and stimulation solely executed by our system. The stimulating channel (the topmost trace) recovered from the stimulation artifact as early as 4 ms and recorded diffusive time-locked responses around 10 ms. Nonstimulating channels recovered much earlier from the stimulation artifact (within 500
Fig. 15 展示了我们系统同时进行多通道记录和刺激的能力。刺激通道(最上面的轨迹)在刺激伪影后早至 4 毫秒恢复,并在大约 10 毫秒记录到扩散的时间锁定反应。非刺激通道在刺激伪影后更早恢复(在 500
Although we have made no effort in reducing the artifact in adjacent channels, our previous generation IC was able to record in less than 1 ms, without blanking such channels, after the stimulation (which compares favorably to the adjacent channel performance, with blanking, of the MEA1060BC from MCS). This IC lost some of that capability due to the introduction of crosstalk (the effects of which can be seen in Fig. 15), as the crosstalk from the remaining artifact spike shifts the baseline of the recording channels for 4 ms after stimulation; and it would produce a large stimulation artifact if the channels had not been blanked.
尽管我们没有努力减少相邻通道中的伪影,我们的上一代集成电路在刺激后不到 1 毫秒的时间内就能够记录下来,而不会空白这些通道(相比之下,MCS 的 MEA1060BC 在空白情况下的相邻通道性能)。由于引入了串扰,这款集成电路失去了一些这样的能力(其效果可见于 Fig. 15 ),因为剩余伪影尖峰的串扰在刺激后 4 毫秒内使记录通道的基线发生偏移;如果没有空白通道,它会产生一个大的刺激伪影。
Discussion 讨论
We have presented a scalable system built around a custom 16-channel IC that can stimulate and record within 3 ms of the stimulus on the stimulating channel and within 500
我们提出了一个可扩展的系统,该系统围绕一个定制的 16 通道集成电路构建,可以在刺激通道的刺激后 3 毫秒内以及相邻通道在 500 微秒内进行刺激和记录。我们还展示了我们的系统在神经培养应用中的一些能力。我们的系统通过一种新颖的反馈方案直接放电电极,并通过塑造这种反馈来优化电极行为来实现伪影消除;也就是说,我们预防问题而不是对抗后果。这种方法代表了与现有的伪影减少方案的一个突破,在现有方案中,积累的电极电荷没有得到管理,从而导致一些研究人员称之为培养皿饱和效应的现象。这种效应使得在快速刺激期间甚至快速刺激本身变得不可能。我们方法的另一个好处是电极寿命的增加,而未放电的电极在双极刺激几小时后会显示出其特性的变化,使用我们的伪影消除协议使我们能够连续数周运行相同的电极而没有任何明显的退化。
The main discharge speed limitation is given by the electrode impedance characteristics [as predicted by Section III-A and (2)]; within this limitation, however, our circuitry is able to reliably control the artifact on the stimulating electrode to the point that relatively simple signal processing (e.g., averaging and subtraction or simple filtering) can obtain responses within 3 ms or less. The addition of dynamic bandwidth control further facilitates the achievement of our artifact elimination goal. We have shown how each of the characteristics of the IC interplay with the artifact, and how each of the functions can be used to tune the desired response. Furthermore, we have shown that our design directly affects the electrode by removing the artifact at the source. This effect introduces other design possibilities, such as using the system as a stand-alone stimulator, or independently optimizing the recording and artifact elimination signal paths.
主要的放电速度限制由电极阻抗特性决定 [如 Section III-A 和 (2) 所预测的];然而,在这一限制内,我们的电路能够可靠地控制刺激电极上的伪影,以至于相对简单的信号处理(例如,平均和减法或简单滤波)可以在 3 毫秒或更短的时间内获得响应。动态带宽控制的加入进一步促进了我们消除伪影目标的实现。我们展示了 IC 的每个特性如何与伪影相互作用,以及如何使用每个功能来调整所需的响应。此外,我们还展示了我们的设计如何通过在源头上消除伪影直接影响电极。这一效果引入了其他设计可能性,例如将系统用作独立的刺激器,或独立优化记录和伪影消除信号路径。
Although we have shown conservative results, and we have not undertaken a careful optimization of our system parameters, we fully expect that better performance will be possible by modifying some of our protocols. Furthermore, given the large number of factors involved, we have strived to provide as much information as possible about the conditions under which artifact elimination is achieved; and to be as precise as possible with our definitions and terminology thus enabling any future comparisons.
尽管我们展示了保守的结果,并且没有对系统参数进行仔细优化,但我们完全相信,通过修改我们的一些协议,可以实现更好的性能。此外,考虑到涉及的因素众多,我们努力尽可能多地提供有关消除伪影所需条件的信息;并尽可能精确地定义和术语,从而使未来的任何比较成为可能。
Future Directions 未来方向
The system is now in use by ourselves and our collaborators in two different laboratories, and the elimination of electrode charge has made possible fast sequential stimulation and recording of neural signals, something that was not possible with existing commercial and custom stimulators. We have shown one example of its use in this paper and we expect other examples to follow as part of additional research. Further work in artifact modeling and characterization is also under way.
该系统目前由我们自己和我们的合作者在两个不同的实验室中使用,电极电荷的消除使得快速的连续刺激和神经信号记录成为可能,这是现有的商业和定制刺激器所无法实现的。我们在本文中展示了一个使用示例,并期望在后续研究中出现其他示例。关于伪影建模和表征的进一步工作也在进行中。
Although our system has significantly reduced the artifact duration, we are exploring ways to further improve the performance of our circuitry for the stimulating electrode as well as in the adjacent channels. Through further improvements of our technology, we expect to enable the study of the direct neural responses which have been suggested by [36] and by Fig. 15. Besides correcting some of the shortcomings that have been introduced in the design of the current IC, one such improvement is the use of a negative resistance for
尽管我们的系统显著减少了伪影持续时间,我们正在探索进一步提高刺激电极电路性能以及相邻通道性能的方法。通过进一步改进我们的技术,我们期望能够研究由 [36] 和 Fig. 15 提出的直接神经反应。除了纠正当前 IC 设计中引入的一些缺陷之外,其中一个改进是使用负电阻
We are in the process of implementing a new IC which will incorporate several improvements in terms of noise performance, bandwidth control, better matching, faster discharge circuitry, and more control for current stimulation. This IC will also incorporate better interfacing characteristics with the full system in mind.
我们正在实施一种新的集成电路(IC),它将在噪声性能、带宽控制、更好的匹配、更快的放电电路和电流刺激的更多控制方面进行多项改进。该 IC 还将结合更好的接口特性,以适应整个系统。
ACKNOWLEDGMENT 确认
The authors would like to thank N. Reddy for the data gathering and processing that made this paper possible; S. Buscemi, K. Gosrani, and S. Das for the layout and implementation of the system and support boards; and Dr. S. Potter (Georgia Institute of Technology, Atlanta) and his group for the multiple suggestions on dealing with electrode arrays. MOSIS provided the IC fabrication for this work.
作者感谢 N. Reddy 收集和处理数据,使本文成为可能;感谢 S. Buscemi、K. Gosrani 和 S. Das 对系统和支持板的布局和实现;感谢 S. Potter 博士(亚特兰大乔治亚理工学院)及其团队对电极阵列处理的多次建议。此外,感谢 MOSIS 提供了本工作的集成电路制造。