AI-empowered Networking

The WMNL is investigating AI/ML-empowered approaches to various networking problems. For example, we are re-visiting the WLAN-5G coexistence scenario to adopt AI techniques for autonomous network reconfiguration. In the WLAN-5G heterogeneous network (HetNet), an asymmetry can arise such that Wi-Fi is exposed to 5G NR-U, but NR-U is hidden from Wi-Fi, since Wi-Fi’s CCA threshold is -62 dBm whereas NR-U’s is -72 dBm, as directed by 3GPP. Although we have already proposed a series of analytical frameworks to model such asymmetry (in our TMC’18 and Access’19 papers), we are trying to develop a more effective mechanism that can be applied to more diverse coexistence scenarios. For this, we consider to utilize AI-based techniques likes Deep Reinforcement Learning (DRL) and Generative adversarial network (GAN) as illustrated below.

AI_HetNet

 

We are also looking at various technology enablers of 6G, including RAN agnostic transceivers, extreme URLLC, AI inspired air interfaces, etc. Among them, we currently focus on the next-generation RAN agnostic transceiver design, powered by AI-based signal examination, so as to make transceivers less restricted by the given structure of wireless standards.