Rohde & Schwarz has successful validated machine learning-based channel-state information (CSI) feedback compression for 5G-Advanced networks, and will demonstrate this at next week’s Mobile World Congress in Barcelona.

With support from Qualcomm, it demonstrated a significant increase in throughput compared with conventional methods. This breakthrough confirms the feasibility of cross-vendor AI implementation in wireless communications with the aim of enhancing network performance.

In the evolving 5G landscape, CSI is essential for optimising network performance and user capacity. CSI enables efficient channel-dependent scheduling and adaptive modulation, ensuring robust high-speed communications between basestation and mobile device. AI and ML-driven CSI enhancements promise even greater efficiency, reduced overhead and improved user experience in 5G-Advanced and future 6G networks. However, implementation across vendors will be difficult.

Rohde & Schwarz and Qualcomm have achieved a milestone by demonstrating cross-vendor interoperability of ML-based CSI feedback enhancements, to be showcased at MWC in Barcelona.

The two companies achieved interoperability between ML models running on a mobile form factor reference design powered by a Qualcomm 5G Modem-RF and the CMX500 5G one-box signalling tester from Rohde & Schwarz, implementing enhanced CSI feedback mechanisms studied in 3GPP Releases 18 and 19. The setup enables efficient compression of the channel state based on CSI reference signal (CSI-RS) measurements, optimising massive MIMO operations critical for 5G networks. This validation demonstrated the throughput performance improved by 51% compared with type-one feedback followed by wideband precoding, as defined in 3GPP Release 15.

Both parties employed separate training approaches for AI models on the network and device side. Compatibility was achieved through specified reference models. Using autoencoder architecture, Qualcomm implemented a proprietary device encoder, while Rohde & Schwarz developed a decoder for its network emulator. The CMX500 one-box tester supports flexible ML model integration through the Open Neural Network eXchange (ONNX) format, enabling users to implement and validate their own AI architectures for wireless testing scenarios.

The successful interoperability proves two-sided ML-based air interface enhancements can be effectively implemented and tested across equipment coming from different vendors. It is a significant step towards implementing AI-enhanced wireless communications, providing a framework for testing and verification essential for the commercial deployment of 5G-Advanced features. It also serves as a milestone for the upcoming standardisation for a future 6G standard, where AI is expected to be natively integrated into the air interface design from the beginning. Hence, cross-vendor AI interoperability will be an essential foundation for future wireless systems.

“The capability of our CMX500 to implement and validate ML-based signal processing highlights the need for evolving test and measurement alongside wireless innovation,” said Christoph Pointner, senior vice president at Rohde & Schwarz. “Thanks to ONNX support, customers can integrate their own ML architectures, making the CMX500 a versatile platform for AI-enhanced wireless testing. Validating interoperability with Qualcomm’s encoder confirms our commitment to establishing essential verification frameworks for AI-driven communications, from 5G-Advanced to future 6G systems.”

John Smee, senior vice president at Qualcomm, added: “In this new 5G-Advanced era of connectivity, and as we look towards 6G, AI is even more critical not just for the best user experiences but also for network performance. The joint research between Qualcomm and Rohde & Schwarz validates that AI-based CSI will help ensure these performance enhancements.”

Rohde & Schwarz (www.rohde-schwarz.com) will present the CSI feedback enhancements demonstration with Qualcomm (www.qualcomm.com) live at MWC at Fira Gran Via in Barcelona in Hall 5, stand 5A80. A video preview of the demonstration is available at www.rohde-schwarz.com/_251220-1545553.html.