AI for Wireless: Evaluating Neural Receivers with Test and Measurement and Digital Twins

At this year’s RCR Wireless' 6G Forum, Taro Eichler from Rohde & Schwarz (R&S) explored how artificial intelligence is becoming an essential part of next generation wireless systems. His session focused on three areas that are rapidly moving from research into practical engineering. These are AI enhanced channel state information feedback, neural receivers and the use of digital twins to bring realistic environments into the lab. His talk provided a useful snapshot of where the industry stands as standardisation work for 6G scales up.

Taro began by setting the scene with the ongoing 3GPP timeline. Work on Release 20 has started and AI already features across several study items. The ambition for 6G is to make AI intrinsic to the physical layer, rather than an add on. This requires a framework that builds on 5G Advanced but goes further with new use cases that rely on machine learning models running inside transceivers.

The first technical area he examined was channel state information (CSI) feedback. As multiple antenna systems grow in size, the need for accurate channel information becomes more demanding. Conventional CSI feedback places a heavy signalling burden on the system. The approach shown in the talk used an industry standard format to exchange AI model information between a Qualcomm device and the Rohde & Schwarz base station simulator. With both ends running aligned models, they were able to demonstrate significant throughput improvements. It reflected a broader trend in which AI is used not simply to optimise classical signal processing but to replace it.

He then moved to neural receivers. Traditional receivers break the processing chain into stages such as channel estimation, equalisation and demodulation. A neural receiver combines these functions into a single learned model. Under a range of channel conditions, the AI based receiver achieved performance gains that would be difficult to match with classical techniques. It also opened the door to using custom constellation designs which can operate without the pilot signals needed for conventional schemes. Removing these pilots directly reduces overhead and frees up resources for data.

The final section of the talk focused on digital twins. A recurring challenge in AI training is the need for realistic datasets. Standardised channel models help with benchmarking but struggle to capture the nuances of real environments. R&S addressed this by creating detailed ray tracing models of actual sites, including their own headquarters in Munich. Measurements were taken to refine material characteristics so the digital representation aligned closely with physical behaviour. These models can then be loaded into a channel generator to recreate dynamic scenarios in real time. One example showed a vehicle and a drone receiving signals from a base station mounted on a building, allowing the entire scene to be replayed in the lab. This creates a bridge between synthetic training data and the real world, improving the reliability of AI based receivers.

Taro’s session highlighted how test and measurement companies are preparing for a future where AI models form part of the radio chain. Accurate validation, cross vendor testing and reproducible environments will be essential as AI moves from promising research to deployable features in commercial networks. His examples demonstrated that this future is already emerging through work on CSI compression, neural receivers and the creation of wireless digital twins.

The video of his talk is embedded below:

Related Posts:

Posted by on

Comments

Post a Comment