Wireless communications testing | AI and ML for 6G networks
#ThinkSix Tech Talk: AI in wireless communications – Neural receivers based on AI/ML
AI-driven network optimization
Recently, Rohde & Schwarz and Qualcomm Technologies demonstrated a groundbreaking industry-first execution of “cross-node” AI/ML. This involved two separately developed models collaborating to enhance downlink throughput by over 50% in a complex 5G MIMO scenario.
The key elements
Channel state information (CSI) feedback is vital for the functioning of massive MIMO antenna systems, as it allows for precise beamforming for high-performance transmission. AI/ML is supposed to increase system efficiency, reduce overhead, and enhance the user experience in both 5G-Advanced and eventually 6G networks.
However, several factors make the enhancement of CSI feedback through ML particularly challenging. First, two models are required for it to function – one operating on the network side and the other on the user device. This means that different vendors develop each model, and the two models must work together closely. Therefore, cross-vendor interoperability is critical to achieving the full benefits. ML-based CSI feedback deserves special attention as it is currently the only cross-node or “two-sided” AI pilot scenario considered by 3GPP.
The collaboration of the models
The work of the two AI/ML models can be compared to the encoding and decoding processes used in high-definition broadcasting: a complex image is compressed into a smaller data package for transmission and then reassembled, using the appropriate encoders and decoders on each side of the transmission.
In this case, Rohde & Schwarz designed a decoder powered by ML for its CMX500 5G one-box signaling tester – this emulated the network side. Qualcomm Technologies, on the other hand, created a device-based ML-powered encoder. Both companies used different methods of training their models. The models were trained to be compatible by using predefined reference models as the basis for their training.
After the training of the models, they were applied together in a 5G-Advanced scenario with 8×4 MIMO using the CMX500, which transferred the scenario to the Qualcomm test device. The smartphone model performed the calculations, compressed the results, and sent them back to the CMX500. The network-side model then used this data to fine-tune the beamforming in the downlink.
ML-based CSI-RS feedback enhancements
Watch the discussion between Rohde & Schwarz and Qualcomm Technologies experts about the joint project to validate ML-enhanced channel-state information (CSI) feedback for 5G Advanced.
The results
What was the outcome? A significant throughput improvement of 51% compared to standard 5G! Thus, this cooperation not only demonstrated the feasibility of cross-vendor AI/ML implementations to enhance radio performance, but it also proved that AI/ML-driven solutions can be effectively integrated and tested across different vendors. This marks an important step toward the commercialization of AI-driven solutions.
It also underscored the level of partnership and teamwork necessary at this stage of AI development, to create a working AI solution for complex radio systems.
It was the first time that two industry players have done this together: training an ML algorithm, implementing it, and seeing it work. This is the foundation for two-sided models and paves the way for 6G, when an AI-native air interface will be available.
Want to discuss your specific AI/ML test cases with our experts?
