CAREER: Removing Energy Barrier Towards Capacity-Approaching Information Transmission and Storage

项目摘要：消除接近容量的信息传输和存储的能量障碍本研究的目的是减少实现接近香农容量的信息传输和存储的总能量。虽然通常使用信道编码来节省传输能量以实现可靠的传输，但由于电路复杂性和超过传输能量，一些最佳容量接近代码的解码能量可能令人望而却步。许多接近容量的代码也不能保证很好地工作，因为它们经常在低错误率时放弃其有效性，需要额外的传输能量来维持其性能。由于接近最终信道容量的操作成本不断上升，传输和解码的能量不断上升，对未来的高性能应用（如光学和有线通信）构成了障碍。低能耗应用，如电池供电的便携式设备和传感器，在获得更强大的通信和存储能力的道路上，遇到了更严重的障碍。本研究通过解决近容量低密度奇偶校验（LDPC）码的两个研究前沿，推进了最先进的编码算法和技术：(1)LDPC码的内在弱点，表现为错误层；(2)由于互连复杂性和存储效率低下，解码能量高。这些方法涵盖了编码理论和集成电路，它们的新颖之处有三个：(1)创造了一种通用的、低成本的局部搜索算法来克服错误层；(2)建立瞬态存储器，利用低能量的数据访问模式；(3)提高非二进制LDPC码的可实现性能，同时降低互连复杂性的新技术。该项目对学生进行编码理论和集成电路的培训，并吸引学生、专业同事和工业合作伙伴广泛参与合作研究和教育。

Project AbstractRemoving Energy Barrier towards Capacity-Approaching Information Transmission and StorageThe objective of this research is to cut the total energy of achieving near Shannon capacity information transmission and storage. While it is common to use channel coding to save transmit energy for a reliable delivery, the decode energy of some of the best capacity-approaching codes can be prohibitive due to circuit complexities and overtake transmit energy. Many capacity-approaching codes are not guaranteed to work well either, as they often give up their effectiveness at low error rates, requiring extra transmit energy to maintain their performance. The rising energy for both transmit and decode poses obstacles to future high-performance applications, such as optical and wireline communications, due to the escalating cost of operating closer to the ultimate channel capacity. More severe impediments are placed on low-energy applications, such as battery-powered portable devices and sensors, in their path of acquiring more robust communication and storage capabilities.This research advances state-of-the-art coding algorithms and techniques by addressing two research frontiers of near-capacity low-density parity-check (LDPC) codes: (1) the intrinsic weakness of LDPC codes that is manifested as error floors; and (2) the high decode energy due to interconnection complexity and memory inefficiency. The approaches span coding theory and integrated circuits, and their novelties are threefold: (1) creation of a universal, low-cost local search algorithm to overcome error floors; (2) creation of transient memory to exploit data access pattern for low energy; and (3) new techniques to improve the achievable performance of non-binary LDPC codes while reducing interconnection complexity. This project trains students in both coding theory and integrated circuits and draws broad participation of students, professional colleagues, and industrial partners in collaborative research and education.