Keysight Technologies, Inc. collaborated with Qualcomm Technologies, Inc. to demonstrate machine learning (ML)-based Channel State Information (CSI) compression to enhance link adaptation efficiency in advanced Multiple-Input Multiple-Output (MIMO) systems at Mobile World Congress (MWC) Barcelona 2026. In a controlled lab validation, the ML-based CSI feedback method achieved more than 40% downlink throughput improvement compared to 3GPP eType II CSI reporting in four-layer (rank-4) operation. As 5G-Advanced networks use more antennas and wider channels, CSI becomes increasingly important in how the network steers beams and selects the right transmission settings.
However, sending more detailed CSI can increase uplink reporting overhead, especially for higher-layer MIMO, creating a tradeoff between performance and signaling efficiency. To address this, Keysight and Qualcomm Technologies tested a mobile test platform powered by a Qualcomm 5G Modem-RF, together with Keysight?s network emulation solutions in a repeatable lab environment. Under fixed CSI feedback constraints, the ML-based compression method provided a more efficient channel representation than the standardized eType II method while maintaining the information needed for accurate beamforming and link decisions.
The result supports scalable advanced MIMO configurations and contributes to industry work exploring AI-native physical-layer enhancements for future 6G wireless systems.
