All Dates/Times are Australian Eastern Standard Time (AEST)

Technical Program

Paper Detail

Paper IDD2-S7-T3.3
Paper Title Model-based Neural Networks for Massive and Sporadic Connectivity
Authors Jeremy Johnston, Xiaodong Wang, Columbia University, United States
Session D2-S7-T3: Neural Networks II
Chaired Session: Wednesday, 14 July, 00:00 - 00:20
Engagement Session: Wednesday, 14 July, 00:20 - 00:40
Abstract We present two model-based neural network architectures purposed for sporadic user activity detection and channel estimation in the massive connectivity regime. In the considered scenario, the set of pilot sequences assigned to users is linearly dependent; but assuming user activity is sporadic, the detection/estimation problem is amenable to sparse recovery algorithms. We apply the deep unfolding framework to unroll two such algorithms, (1) linearized alternating direction method of multipliers and (2) vector approximate message passing, into a set of custom neural network layers. The networks thus inherit domain knowledge encapsulated in the signal model, plus suitable layer operations informed by the algorithms. The networks, trained on randomly generated data, offer improved complexity and accuracy relative to their iterative counterparts, and are a potential boon to cell-free massive MIMO systems.