Precoder Design in Multi-User FDD Systems with VQ-VAE and GNN
Srikar Allaparapu, Michael Baur, Benedikt Böck, Michael Joham, Wolfgang Utschick
TL;DR
This paper tackles the challenge of CSI feedback overhead in multi-user FDD systems by proposing an end-to-end framework that jointly learns the pilot matrix, VQ-VAE-based CSI feedback, and a GNN-based precoder. The VQ-VAE replaces Gaussian mixture priors to provide a compact, discrete latent representation, enabling end-to-end training that optimizes both feedback and precoding with pilot optimization. A two-stage training scheme (pre-training and fine-tuning) leverages structural insights like block-Toeplitz channel covariance to reduce model complexity while maintaining performance, demonstrated on real-world measurement data. Results show the proposed VQ-VAE+GNN with learned pilots outperforms conventional DFT pilots and iterative precoding schemes, particularly in low-pilot, low-feedback regimes, highlighting the practical impact for 6G-like systems with stringent overhead constraints.
Abstract
Robust precoding is efficiently feasible in frequency division duplex (FDD) systems by incorporating the learnt statistics of the propagation environment through a generative model. We build on previous work that successfully designed site-specific precoders based on a combination of Gaussian mixture models (GMMs) and graph neural networks (GNNs). In this paper, by utilizing a vector quantized-variational autoencoder (VQ-VAE), we circumvent one of the key drawbacks of GMMs, i.e., the number of GMM components scales exponentially to the feedback bits. In addition, the deep learning architecture of the VQ-VAE allows us to jointly train the GNN together with VQ-VAE along with pilot optimization forming an end-to-end (E2E) model, resulting in considerable performance gains in sum rate for multi-user wireless systems. Simulations demonstrate the superiority of the proposed frameworks over the conventional methods involving the sub-discrete Fourier transform (DFT) pilot matrix and iterative precoder algorithms enabling the deployment of systems characterized by fewer pilots or feedback bits.