Neural Network Detection for Bandwidth-Limited Non-Orthogonal Multiband CAP UVLC System

• Main Authors: Chen, J. (Author), Chi, N. (Author), Li, Z. (Author), Shen, C. (Author), Wang, Z. (Author), Zhang, J. (Author), Zhao, Y. (Author)
• Format: Article
• Language: English
• Published: Institute of Electrical and Electronics Engineers Inc. 2022
• Subjects: Amplitude and phase modulations; Bandwidth; Carrier-less; Complex networks; Equalizers; Independent component analysis; Intersymbol interference; Least mean squares; Light; Modulation; Multi band; neural network; Neural-networks; Non-orthogonal; Non-orthogonal multi-band carrierless ampli-tude and phase; Sparse data; Sub-carriers; underwater visible light communication; Underwater visible light communication; Visible light communication
• Online Access: View Fulltext in Publisher

 In this paper, we propose a novel sparse data-to-symbol neural network (SDSNN) receiver for bandwidth-limited underwater visible light communication (UVLC) based on non-orthogonal multi-band carrierless amplitude and phase modulation (NM-CAP). Bandwidth limited NM-CAP signals usually carry severe inter-symbol interference (ISI) and inter-band interference (IBI). The SDSNN receiver directly converts the received NM-CAP data with ISI and IBI into quadrature amplitude modulation symbols without distortion for each sub-band. In contrast, the conventional receiver requires the least mean square (LMS) equalizer to cancel ISI, and the subcarrier component extraction with complex independent component analysis (SCE-ICA) to cancel IBI, respectively. SDSNN provides a novel receiving structure to replace post-equalization, matched filtering, and SCE-ICA. A blue-LED based UVLC system has been demonstrated utilizing NM-CAP16 with 3 sub-bands. The experimental results show that NM-CAP with the SDSNN receiver case reaches the highest spectral efficiency, where an enhancement of 43%, 20%, 6% has been measured over the orthogonal multi-band CAP case, NM-CAP with LMS equalizer case, and NM-CAP with joint LMS equalizer and SCE-ICA case, respectively. Compared with joint LMS equalizer and SCE-ICA case, the proposed SDSNN receiver can achieve 98% reduction of computational complexity. © 2009-2012 IEEE.