Collaborative Research: Large-Scale Wireless RF Networks of Microchip Sensors

我们周围的世界越来越多地被电子传感器包围。对于可穿戴和可植入的生物医学传感器等应用，特别需要和有机会使用不显眼的微型设备，这些设备作为大型集合自主操作，以绘制人体感兴趣区域的生理活动。一个挑战是如何构建一个无线网络，在这个网络中，来自微传感器群体的数据被发送、接收和解码，以解开数据，比如来自1000个传感器的数据。一个粗略的类比是一组普通的射频标签，这些标签必须由一个收发机一次读取--只是每个传感器位置的信号现在在时间和幅度上都会有所不同。在这种情况下，脑机接口提供了一个范例：如何通过一组植入大脑的自主微型传感器以高分辨率捕获神经元信号。目前在世界各地实验室开发脑机接口的研究集中在一些方案上，在这些方案中，寻求进入大脑皮层的数千个点，以将大脑计算转换为有用的电子命令，例如用于预期的语音。神经技术的问题有三个：以不引人注意的方式记录来自大脑的电信号，将数据无线传输到身体外部接收器，以及实时破译大量信号。动态环境的分布式感知的许多情况都以事件稀疏性为特征，无论是在自然系统中还是在人造系统中，大脑中的神经元就是一个例子。拟议的事件驱动通信策略能够利用大脑作为灵感，在由数千个无线微传感器组成的网络中高效传输、准确检索和解释稀疏事件。提出的工作集中在构建大规模无线微传感器射频网络的一体化方法上。外部收发器在向传感器提供无线电力的同时收集数据。每个传感器都是一个亚毫米大小的硅片上系统，带有定制电路，专为“事件检测”而设计，其中时变的传感器输入被编码为一系列短的“尖峰”。最近，在所谓的动态视觉摄像机中出现了一种受大脑启发的对稀疏事件的数据进行编码的方法。脉冲序列数据被转换为片上数字形式，并传输到一个公共接收器。由于只有事件驱动的尖峰信号通过网络传输，通信系统的带宽可以非常有效地利用，从而使大量的传感器整合到网络中。该团队建议在实验室建立一个微传感器系统，并使用制造的微芯片演示低错误率和高效的异步编码无线传输，并通过模拟展示对数千个节点的扩展适用性。重要的是，事件感知检测和无线通信方法非常适合用于分析多传感器数据的神经形态计算方法；该项目的第三个关键元素。该团队将展示如何从一个由多达8000个微型传感器组成的假想植入物中解码实际的大脑数据(从其他地方合成的实际灵长类大脑记录)。就效率和短延迟而言，神经形态计算似乎特别适合解码基于事件的数据。使用灵长类运动皮质的可用数据，该团队计划展示如何解码来自数千个神经元的无线信号，以预测计划的手臂和手的移动。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The world around us is increasingly surrounded by electronic sensors. For applications such as wearable and implantable biomedical sensors there is a particular need and opportunity for unobtrusive microdevices which operate autonomously as large ensembles to map physiological activity across a body area of interest. A challenge is how to construct a wireless network whereby data from a microsensor population is transmitted, received, and decoded to unravel data, say from 1000 individual sensors. A rough analogy is that of a population of common radio-frequency tags which must be read at once by a single transceiver - with the twist that signals at each sensor location will now vary both in time and in magnitude. A brain-computer interface suggests a paradigm in this context: how to capture neuronal signals at high resolution by a population of autonomous brain implanted microsensors. Ongoing research for development of brain-machine interfaces in laboratories worldwide is focused on a number of schemes where access to thousands of points in the cortex is sought to translate brain computations to useful electronic commands e.g. for intended speech. The neurotechnology problem is three-fold: to record electrical signals from the brain unobtrusively, to transmit the data wirelessly to a body external receiver, and to decipher the multitude of signals in real time. Many cases of distributed sensing of a dynamical environment are characterized by sparsity of events whether in nature or man-made systems, neurons in the brain being an example. The proposed event-driven communication strategy enables the efficient transmission, accurate retrieval, and interpretation of sparse events across a network of thousands of wireless microsensors – using the brain as an inspiration. The proposed work is focused on an all-in-one approach to build a large scale wireless microsensor radio-frequency network. An external transceiver collects data while supplying wireless power to the sensors. Each sensor is a sub-millimeter size silicon system-on-microchip with custom circuitry designed for ‘event detection’ where time-varying sensor inputs are encoded as a series of short ‘spikes’. The brain-inspired method of encoding data from sparse events has emerged recently in so- called dynamic vision cameras. Spike train data are converted into digital form on chip and transmitted to one common receiver. Since only the event-driven spikes are transmitted through the network, the bandwidth of the communication system can be utilized very efficiently enabling a large population of sensors to be incorporated into the network.The team proposes to build a microsensor system and demonstrate low-error rate and efficient asynchronous, encoded wireless transmission in the laboratory using fabricated microchips, and to show extended applicability to thousands of nodes though simulations.Importantly, the event-sensing detection and wireless communication approach is quite suited for a neuromorphic computational approach for analyzing multisensory data; the third key element in the project. The team will show how to decode actual brain data (synthesized elsewhere from actual primate brain recordings) from a hypothetical implant composed of up to 8000 microsensors. Neuromorphic computing appears particularly suited decoding event-based data in terms of efficiency and short latency. Using available data from the primate motor cortex the team plans to show how to decode wireless signals from thousands of neurons for the prediction of planned arm and hand movement.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.