SHF: Small: Bio-inspired ultra-broadband RF scene analysis

在嘈杂和混乱的环境中检测和分析结构化信号是无线电通信、图像处理和语音识别等领域的一个基本问题。经过数百万年的进化，生物感官系统已经得到了优化，以极高的精度和效率解决了这个问题；人造通信和信号处理系统在任何地方都无法达到相同的性能水平，甚至无法共享相似的基本设计原则。这个项目将试图通过理解听觉系统用于分析自然声音的通用信息处理原理，然后将其应用于分析人造射频（RF）信号，来弥合这一差距。特别是，它将专注于开发电子和算法，以模仿生物耳蜗（内耳）和听觉通路的一些惊人功能。研究生和本科生，包括代表性不足的群体的成员，将作为这项研究的一部分接受培训，从而扩大未来受过技术培训的劳动力。这个项目的生物启发方法是由两个观察激发的。首先，从算法的角度来看，听觉系统从耳蜗开始分析声音的精细时频内容的过程是非常精确和高效的。其次，音频和射频场景是由类似的物理（波传播、吸收、散射、衍射和干涉）产生的，尽管相关的速度和时间延迟相差约一百万倍。因此，音频和射频场景具有许多相同的特征，这使得考虑在更高频率下运行的耳蜗力学、信号转导和听觉编码模型变得有趣。第一个研究目标是建立一个单芯片耳蜗模型，分析GHz范围内的射频信号，并将频率、幅度和相位信息编码成类似于听觉神经纤维的并行事件驱动输出。第二个目标是通过开发一个健壮的编码框架来创建耳蜗输出的压缩表示，从而允许从输入信号中有效地提取更高级别的属性，例如源位置和类别。

The detection and analysis of structured signals in noisy and cluttered environments is a fundamental problem in areas ranging from radio communications to image processing and speech recognition. Biological sensory systems have been optimized by millions of years of evolution to solve this problem with exquisite precision and efficiency; man-made communication and signal processing systems do not achieve anywhere near the same level of performance or even share similar fundamental design principles. This project will try to bridge this gap by understanding the universal information processing principles used by the auditory system to analyze natural sounds, and then adapting them to analyze man-made radio frequency (RF) signals. In particular, it will focus on developing electronics and algorithms that emulate some of the amazing capabilities of the biological cochlea (inner ear) and auditory pathway. Graduate and undergraduate students including members of underrepresented groups will be trained as part of this research, thus enlarging the technologically trained workforce of the future.The bio-inspired approach of this project was motivated by two observations. Firstly, the process by which the auditory system, beginning with the cochlea, analyzes the fine time-frequency content of sounds is both extremely precise and also highly efficient from an algorithmic viewpoint. Secondly, audio and RF scenes are generated by similar physics (wave propagation, absorption, scattering, diffraction, and interference), even though the relevant velocities and time delays differ by a factor of about a million. Thus audio and RF scenes share many of the same characteristics, which makes it interesting to consider models of cochlear mechanics, signal transduction, and auditory coding that are scaled to operate at much higher frequencies. The first research goal is to build a single-chip cochlear model that analyzes RF signals in the GHz range and encodes frequency, amplitude, and phase information into parallel event-driven outputs that are analogous to auditory nerve fibers. The second goal is to allow higher-level properties, such as source locations and categories, to be efficiently extracted from input signals by developing a robust coding framework to create compressed representations of the cochlear outputs.