Unsupervised Learning for Energy Efficiency Optimization Over CF-mMIMO Under URLLC

This letter investigates the energy efficiency (EE) of cell-free massive multiple-input multiple-output (CF-mMIMO) systems under ultra-reliable low-latency communication (URLLC) constraints. To improve the EE and satisfy the reliability of each user equipment (UE), UEs are classified into power-constrained UEs and power-tolerant UEs. Accordingly, an unsupervised deep neural network (UNSNet) is proposed, which consists of three sub-modules for extracting the channel characteristics of the power-constrained UEs, the power-tolerant UEs, and all the UEs, respectively. The UNSNet achieves reliability improvement for power-tolerant UEs with minimal impact on EE and enhances EE for power-constrained UEs while maintaining reliability. To accommodate dynamic communication environments, UNSNet integrates online learning techniques, further enhancing the robustness of the network while ensuring that the training process is label-independent to achieve low computational complexity. Numerical results show that the proposed method achieves the trade-off between EE and reliability and has a faster processing speed than traditional iterative methods.