An Evaluation of AI-RAN Techniques based on Network Digital Twin with Hardware-in-the-Loop

We present a testbed to experimentally evaluate the use of artificial intelligence (AI) and machine learning techniques in the physical layer of future mobile networks. The testbed relies on the use of a digital twin of the propagation environment which, via ray-tracing, is used to generate realistic channel responses tied to a particular (digital) deployment site. Arbitrary waveform generators are used to create standard-compliant 5G radiofrequency (RF) signals whose convolution with the channel is emulated with Keysight's radio channel emulator PROPSIM. After channel convolution, the signal is processed by a commercial O-RAN radio unit by LiteOn, which delivers the digitized samples to a server for AI processing. With this setup, we illustrate two use cases: 1) a neural receiver developed by NVIDIA, capable of multi-user MIMO detection and decoding, and 2) an AI system for activity detection and classification using RF signals, developed by InterDigital.