报告题目：BMST: A Flexible and Universal Approach to Several Coding Problems
报告时间：2025年11月27日 下午15:00
报告地点：犀浦校区X9406
报告人：马啸 教授，中山大学
邀请人：唐小虎 教授
摘要：In this paper, we make an attempt to integrate source coding, channel coding and joint source-channel coding (JSCC) into a single pair of encoder and decoder (with input length fixed but output length tunable according to the source entropy and the channel capacity). We introduce a special class of time-varying convolutional codes, called convolutional sampling and superposition (Conv-SaS) codes. We prove that, under blockwise maximum a posteriori (MAP) decoding, the Conv-SaS code ensemble achieves the limit of minimum transmission ratio (LMTR) when transmitting a Bernoulli source over memoryless binary-input output-symmetric (MBIOS) channels, also known as binary memoryless symmetric (BMS) channels, in terms of frame-error rate (FER). The main theorem is then adapted to the source coding, channel coding, JSCC, and (noiseless or noisy) Slepian-Wolf (SW) coding problems. In the simulations, the Conv-SaS code ensemble is expurgated as the convolutional repetition and superposition (Conv-RaS) code ensemble, a generalized block Markov superposition transmission (BMST) code ensemble. The construction of BMST codes is flexible with code rates (defined as the ratio of the output length to the input length) ranging from less than one (typically for source codes) to greater than one (typically for channel codes). More importantly, the BMST codes with moderately large sub-blocks can be decoded iteratively by a sliding window message passing algorithm, performing universally well for the JSCC and SW coding problems, as confirmed by simulations. Without complicated optimizations, the BMST codes can be implemented with a pair of encoder and decoder of essentially the same architecture but adjustable code rates.
报告人简介：马啸，博士，中山大学计算机学院教授，现任中国电子学会信息论分会副主任委员，《电子学报》编委等。2007年入选教育部新世纪优秀人才计划；2008年成为广东省高校“千百十工程” 省级培养对象； 2016年获得广东特支计划“百千万工程领军人才”项目资助； 2021中国电子学会科学技术奖自然科学三等奖；获2022年高等教育国家级教学成果二等奖。曾主持国家973计划项目课题、国家重点研发计划项目课题、国家自然科学基金重大研究计划培育项目、联合基金（重点） 项目及多项产学研合作项目等。主要研究信息理论与信源信道编码技术等。近五年，在《IEEE信息论汇刊》等国际顶级刊物、 IEEE信息论年会等国际旗舰会议及《电子学报》等国内重要刊物上发表论文百余篇，谷歌学术引用三千余次，授权美国发明专利两项，中国发明专利二十余项，其中一项研究成果被国外教授大篇幅写进专业教科书，出版专著《多元LDPC码及其应用》、 译著《信道编码：经典与现代》。