Deep Learning-based Uplink Data Detection in User-Centric Cell-Free mMIMO Systems

This paper tackles the problem of uplink data detection in a user-centric cell-free massive multi-input multi-output (UC-CF-mMIMO). First of all, we show that the problem of uplink data detection in UC-CF-mMIMO with large-scale fading decoding (LSFD) can be cast as a classical MIMO detection problem. Next, we develop a new detection structure, called LMDPIC, which combines linear minimum mean square error (LMMSE) and deep-learning-based parallel interference cancellation (DeepPIC) detectors for symbol detection. Simulation results demonstrate that LMDPIC outperforms other state-of-the-art MIMO detection schemes for BPSK and QPSK modulation schemes, both in the case of the perfect and in that of imperfect channel state information (CSI) available at the APs. We also propose two heuristic pilot assignment schemes to improve the quality of CSI acquisition during uplink training. The performance of LMDPIC is also evaluated for several power-control strategies, including half-power, max-min, and full-power transmission. Our results show that for 5\% UEs with the highest pairwise symbol error rate, the LMDPIC with half power transmission outperforms the LMMSE with full power transmission. Finally the paper also shows that the deep neural network (DNN) used in the LMDPIC structure is robust against CSI time variations.