Data-driven cell-free scheduler

Efficient scheduling is essential in cell-free (CF) networks, where user equipments (UEs) communicate with multiple distributed transceivers (radio units (RUs)) linked to a centralized base station (BS) that coordinates and processes the received or transmitted signals. Unlike traditional cellular networks, CF networks operate without cell boundaries, allowing UEs to seamlessly connect to multiple RUs, and thus eliminating the conventional necessity for handoffs between transceivers. In this paper, we introduce a novel CF scheduler designed to enhance data quality of service (QoS) parameters, including throughput, and latency. The scheduler employs a neural network (NN) algorithm to autonomously manage interactions with users across a distributed network of transceivers. This approach utilizes both model and data driven methods to optimize user communication. To mitigate the high computational complexity of traditional model-driven algorithms, we propose a supervised NN that learns from the model-driven approach. We assess its performance using simulated data from orthogonal frequency division multiple access (OFDMA) waveforms in frequency, time, space, and polarization (e.g., resource blocks, OFDM symbols, beam ID), within multi-transceiver RU environments. Our results indicate that the model-driven algorithms exhibit competitive performance compared to the exhaustive search method, while the supervised NN demonstrates comparable efficiency after offline learning. Consequently, our NN-based scheduler emerges as a viable, efficient solution for optimizing CF network scheduling.