Resource Block-Granularity Precoding Optimization and Compression for Cell-Free Mobile Networks

The forward-precoding scheme could significantly reduce downlink fronthaul traffic in cell-free mobile networks by compressing resource-block (RB)-granularity precoding matrices. Nevertheless, this scheme would result in a sum-rate degradation due to the compression distortion and the significant frequency-selective fading. To address this issue, the compressed RB-granularity precoding optimization is first formulated as a non-convex stochastic problem and then its lower bound is derived. By maximizing the lower bound, the stochastic non-convex problem is transformed into a deterministic RB-granularity precoding optimization problem and a matrix compression problem. For the deterministic RB-granularity precoding optimization problem, its stationary point is proved to be a linear combination of frequency channel matrices. Based on this property, an RB-granularity weighted minimum mean-square-error (RB-WMMSE) precoding algorithm is designed. For the matrix compression problem, a transformer-based vector-quantized variational autoencoder (TVQ-VAE) algorithm is designed to achieve a high ratio compression. Simulation results show that the proposed RB-WMMSE algorithm could improve the sum-rate by a maximum of 101% compared to the traditional RB-granularity precoding algorithm. When the compression distortion of precoding matrix is considered, the proposed TVQ-VAE algorithm could improve the sum-rate by a maximum of 106% compared to the traditional autoencoder algorithm.