2021/07/14 by Bharath Keshavamurthy, Keshavamurthy, Bharath, Nicolò Michelusi +1
Computer Science · Engineering · #Advanced MIMO Systems Optimization #Cognitive Radio Networks and Spectrum Sensing #Distributed Sensor Networks and Detection Algorithms #FOS: Computer and information sciences #FOS: Electrical engineering #Machine Learning (cs.LG) #Signal Processing (eess.SP) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2107.07049
openalex publication_date 2021/07/14 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
A novel LEarning-based Spectrum Sensing and Access (LESSA) framework is\nproposed, wherein a cognitive radio (CR) learns a time-frequency correlation\nmodel underlying spectrum occupancy of licensed users (LUs) in a radio\necosystem; concurrently, it devises an approximately optimal spectrum sensing\nand access policy under sensing constraints. A Baum-Welch algorithm is proposed\nto learn a parametric Markov transition model of LU spectrum occupancy based on\nnoisy spectrum measurements. Spectrum sensing and access are cast as a\nPartially-Observable Markov Decision Process, approximately optimized via\nrandomized point-based value iteration. Fragmentation, Hamming-distance state\nfilters and Monte-Carlo methods are proposed to alleviate the inherent\ncomputational complexity, and a weighted reward metric to regulate the\ntrade-off between CR throughput and LU interference. Numerical evaluations\ndemonstrate that LESSA performs within 5 percent of a genie-aided upper bound\nwith foreknowledge of LU spectrum occupancy, and outperforms state-of-the-art\nalgorithms across the entire trade-off region: 71 percent over\ncorrelation-based clustering, 26 percent over Neyman-Pearson detection, 6\npercent over the Viterbi algorithm, and 9 percent over an adaptive Deep\nQ-Network. LESSA is then extended to a distributed Multi-Agent setting\n(MA-LESSA), by proposing novel neighbor discovery and channel access rank\nallocation. MA-LESSA improves CR throughput by 43 percent over cooperative\nTD-SARSA, 84 percent over cooperative greedy distributed learning, and 3x over\nnon-cooperative learning via g-statistics and ACKs. Finally, MA-LESSA is\nimplemented on the DARPA SC2 platform, manifesting superior performance over\ncompetitors in a real-world TDWR-UNII WLAN emulation; its implementation\nfeasibility is further validated on a testbed of ESP32 radios, exhibiting 96\npercent success probability.\n