Keysight Demonstrates 6G Neural Receiver Design Flow in Collaboration with NVIDIA

External channel models simulated by NVIDIA Sionna are emulated using Keysight equipment to create custom end-to-end system

Keysight Technologies, Inc has collaborated with NVIDIA to create a complete design flow for training and validating neural receivers that will be shown at Mobile World Congress Barcelona 2024. To be presented at Keysight’s booth, Hall 5 Stand 5E12, the demonstration will feature an Open RAN testbed that has been augmented by a multi-user MIMO neural receiver.

While 5G integrates artificial intelligence (AI) to enhance specific components in wireless networks, 6G will be the first generation of wireless technology that is AI-native. A key goal is to develop site-specific neural receivers to replace the entire human-designed receive chain of the physical layer. Yet, the data required to train these neural receivers is limited, and validating their performance in end-to-end systems presents a challenge. Before neural receivers can be deployed in commercial networks, they need to be adequately trained, demonstrated to surpass traditional receivers in performance and shown to robustly handle the channel conditions of real-world networks.

Recommended AI News: Introducing Gemini 1.5: The Ultimate Game-Changer in Next-Generation Models

This demonstration shows how Keysight solutions enable the design and validation of a neural receiver. With the NVIDIA Sionna library used to train the neural receiver, ray-traced channels allow for site-specific training data generation and hence the creation of digital twins of real-world systems. As such, the neural receiver can be optimized for any intended environment.

When the training is complete, the neural receiver is deployed in an Open RAN testbed using Keysight’s equipment connected to a FlexFi commercial radio unit from LITEON Technology. To emulate the site-specific channel, the Keysight PROPSIM channel emulator is used. It allows for the seamless import of ray-traced channel impulse responses. The trained neural receiver then demodulates the signal. Measurements of the block error rate of the end-to-end system provide insights into the neural receiver’s performance.

Recommended AI News: Schneider Electric Launches Digitally Enabled SureSeT Medium Voltage Switchgear in Canada

Giampaolo Tardioli, Vice President, 6G and Next Generation Technology at Keysight, said: “By collaborating with industry leaders like NVIDIA, Keysight is helping provide the tools and insights needed to allow artificial intelligence to move into the mainstream for wireless communications. Through use of our simulated and measured data for training digital twins of Open RAN networks and the validation of AI performance, we are able to provide the complete end-to-end design environment that is needed to develop neural receivers and other AI components for 6G.”

Ronnie Vasishta, Senior Vice President of Telecom at NVIDIA, said: “Software-defined RAN enables the native integration of AI across the entire protocol stack, ultimately leading to 6G systems that can optimally adapt to any environment. Keysight’s ability to generate and capture real-world data for training and evaluation, bolstered by the open-source NVIDIA Sionna library for 6G physical-layer research, is a key element helping to drive the adoption of AI in wireless networks.”

Recommended AI News: Deutsche Telekom and Brain.ai Unveil Revolutionary App-less Phone at Mobile World Congress

[To share your insights with us as part of editorial or sponsored content, please write to sghosh@martechseries.com]