Low-Complexity Interval Passing Algorithm and VLSI Architecture for Binary Compressed Sensing

• Published in: IEEE transactions on very large scale integration (VLSI) systems Vol. 28; no. 5; pp. 1283 - 1291
• Main Authors: Kalipatnapu, Shantharam, Chakrabarti, Indrajit
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.05.2020; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Application specific integrated circuits; Approximation algorithms; Binary compressed sensing (BCS); Complexity; Complexity theory; Computer architecture; Decoding; Euclidean geometry low-density parity check (EG-LDPC); Fault detection; Field programmable gate arrays; Integrated circuits; interval passing algorithm (IPA); LDPC codes; Manganese; Parity check codes; Reconstruction; Sparse matrices; Very large scale integration; VLSI; Wireless sensor networks
• ISSN: 1063-8210; 1557-9999
• DOI: 10.1109/TVLSI.2020.2967477

Binary compressed sensing (BCS), in which signals of interest have binary values, finds applications in areas including fault detection and wireless sensor networks. In this article, a low-complexity VLSI architecture for BCS based on interval passing algorithm is proposed. Moreover, the algorithm is modified in order to reduce its complexity without significant loss in performance, and its corresponding VLSI architecture is proposed. Binary low-density parity check (LDPC) matrices based on finite geometry have been used as measurement matrices. The proposed VLSI architectures have been synthesized in both ASIC and field-programmable gate array (FPGA) platforms. The hardware consumption of the proposed designs is independent of sparsity values. Moreover, the proposed architectures offer high frequency of operation and low reconstruction time when compared to the state-of-the-art designs. Specifically, the 65-nm ASIC realization operates at a maximum frequency of 500 and 666.67 MHz and offer a reconstruction time of 6.3 and 4.7 ns, respectively, for a <inline-formula> <tex-math notation="LaTeX">64\times 256 </tex-math></inline-formula> deterministic measurement matrix.