Data-driven Cross-band Channel Feedback Prediction for MIMO Networks

Accurate channel state information (CSI) is essential for the performance of modern wireless com- munication systems, particularly those leveraging multiple-input multiple-output (MIMO) archi- tectures. However, acquiring CSI over wide bandwidths introduces significant feedback overhead, especially in Frequency Division Duplex (FDD) systems. This thesis proposes a data-driven frame- work for downlink-to-downlink cross-band channel prediction, enabling the estimation of wideband CSI using narrower-band measurements. Using simulated data generated through a ray-tracing model of an indoor environment, we investigate how a neural network can infer 80 MHz channel frequency response (CFR) from corre- sponding 40 MHz CFRs. The dataset spans multiple bandwidths, carrier frequencies, and antenna configurations, and is preprocessed with careful outlier removal and spatial train-test separation to ensure model generalization. A feedforward neural network (MLP) is trained to learn the mapping between input and target CFRs, showing promising performance across a range of test conditions. Results demonstrate that the model effectively captures the spectral structure of the wideband channel, achieving low mean squared error and high consistency with ground truth. The proposed approach offers a viable path toward reducing CSI feedback requirements in next-generation wire- less networks, contributing to more scalable, efficient, and adaptive MIMO systems.