CAREER: Heterogeneous Non-orthogonal Wireless Communications Enhanced by Unified SDR - Deep Learning Architectures

点击翻译按钮获取中文摘要

Enabling energy-efficient, ubiquitous connectivity is a critical task for the highly anticipated concept of pervasive Internet-of-Things, where massively deployed heterogeneous Internet-of-Things devices are seamlessly interconnected without the need of manual network management. The outcomes of this research can expedite the vision of trillions of Internet-of-Things devices and enable heterogeneous classes of new applications such as energy-harvesting and wirelessly-connected smart-dust devices, autonomous aerial and ground vehicles with ultra-reliable low latency communications, and intelligent automated factories with deep learning-assisted collaborative networks in highly congested channels. The proposed research targets orders-of-magnitude reduction in power consumption and complexity for wireless connectivity to realize ultra-low cost, ultra-small, disposable, and ubiquitous wireless Internet-of-Things devices. Leading into the realistic world of energy-autonomous Internet-of-Things platforms, this program investigates new ultra-low power wireless connectivity solutions assisted by novel digital signal processor architectures optimized for software-defined radio processing and machine learning. The proposed interdisciplinary research spans a wide range of topics including digital communication, low power integrated circuits, machine learning, and processor architectures to explore cross-layer approaches that are indispensable to tackle challenges in heterogenous classes of energy-efficient and versatile communication systems.Wireless communication is often the dominant source of latency and power consumption for the majority of mission-critical and energy-constrained Internet-of-Things applications. One main objective of the proposed research is to deliver a truly energy-autonomous, fully self-contained wireless communication system that optimally utilizes the scarce harvested energy and dramatically enhances ultra-low power analog circuits performance via novel digital signal processing. A new non-orthogonal modulation and multiple access scheme that exploits sparsity of novel hyper-dimensional modulation is proposed to efficiently eliminate significant power and complexity overhead imposed on orthogonal modulation and multiple access schemes that require explicit synchronization. The research scope includes 1) new non-orthogonal modulation and multiple access schemes for energy-autonomous, disposable IoT sensor nodes without a power-hungry phase-lock-loop circuit and its frequency reference crystal, 2) novel hyper-dimensional modulation to replace or complement conventional orthogonal modulation and error correction codes for ultra-reliable low latency communication, and 3) unified software-defined radio and deep learning processor architectures to enable dynamic communication-computation tradeoffs, energy-efficient execution of baseband modem processing, systolic array based belief propagation algorithm acceleration, and deep neural network assisted signal processing. This program includes experimental researches to fabricate transceiver integrated circuits and design real-time prototype systems to demonstrate the proposed concept of non-orthogonal modulation and multiple access wireless communication throughout extensive field trials rendering realistic scenarios.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.