Effective data reduction for wireless transmission of neural activity (EDnA)

多通道信号的神经系统推导在生物技术和神经科学中具有重要意义。无论是在基础研究中，还是在假肢控制或脑机接口方面，这些植入式系统都是一个关键的研究平台。记录仪的无线操作对于体内使用至关重要：在行为科学中，允许在尽可能不受干扰的环境中进行测试；在人体植入中，由于接受度更高，感染风险更低。尽管对这些系统的研究已经进行了十多年，但一个主要的未解决的挑战仍然存在：用于评估和使用的原始数据的无线传输速度超过10Mbps，例如分类。由于数据速率高，数据通信所需的功耗大得不切实际，这限制了当前系统要么在少数信道上，要么迫使它们以数据量换取数据质量。该项目提出了解决方案，通过提供数据减少而没有（重大）信息损失的方式来解决这一瓶颈的挑战：o一方面，原始数据的频谱分离允许利用-超过一个数量级-与单个神经元的峰值相比，低频局部场电位的不同动态范围。频带的分离允许在频谱上限制整个记录信号的大动态范围，从而减少数据量。o此外，将采取压缩策略，在几乎不损失信息的情况下减少数字数据流。首先，考虑了delta压缩和压缩感知，因为这两种方法都受益于相当低的神经波形活动。为了获得数据压缩质量的阈值，将使用Spikesorting算法，它允许将原始数据的分类与数据压缩和恢复后获得的分类进行比较。这确保了信号不仅可以定性地恢复，而且可以定量地恢复。o估计该方法所需的电路复杂性和功耗。这特别与无线通信中的数据减少所节省的功率有关。该研究项目的结果是允许多通道神经记录仪无线原始数据传输的方法，这反过来又导致了多通道可植入神经记录仪的可能实现，具有外部可用的完整信号质量-神经科学和主动假肢的工具。

Derivation of neural systems for multi-channel signals are of major importance in biotechnology and neuroscience. Both in basic research, as well as for prosthetic control or brain-machine interfaces, these implantable systems are a key research platform. A wireless operation of the recorder is essential for in vivo use: in the behavioral sciences to allow the tests in an as undisturbed environment as possible, and in human implantation due to higher acceptance and reduced risk of infection.Although research is being conducted for more than a decade on these systems, a major unsolved challenge remains: the wireless transmission of raw data in the order of more than 10Mbps for the evaluation and use, e.g. such as classification. Due to the high data rate, the required power consumption for data communication is unrealistically large, which limits current systems either on a small number of channels, or forces them to trade data quantity against quality. The project proposes solutions to this challenge to this bottleneck by means of providing data reduction without (significant) loss of information:o On the one hand a spectral separation of the raw data allows to take advantage of the - by more than an order of magnitude - different dynamic range of low-frequency local field potentials in contrast to the spikes of individual neurons. A separation of the bands allows to limit the large dynamic range of the entire recorded signal spectrally, and thereby reduce the amount of data.o In addition, compression strategies will be pursued to reduce the digital data stream almost without loss of information. At first, delta compression and compressed sensing are considered, since both methods benefit from the rather low activity of neural waveforms. In order to obtain a threshold value for the quality of the data compression, Spikesorting algorithms will be used, which allow a comparison of the classification of the raw data with that which is obtained after compression and recovery of the data. This ensures that the signals are not only recovered qualitatively, but can also be recovered quantitatively.o The required circuit complexity and power consumption of the method is estimated. This is set in particular in relation to the saved power that can be saved by a data reduction in wireless communication.The result of this research project are methods that allow wireless raw data transmission for multi-channel neuro-recorders, which in turn leads to the possible implementation of multi-channel, implantable neuro-recorders with externally available, full signal quality - a tool for neuroscience and active prostheses.