Towards 6G: fast and self-adaptive dynamic bandwidth allocation for next-generation mobile fronthaul

6G networks will deliver dynamic and immersive applications that bridge the real and digital worlds. The next-generation passive optical access network is a potential optical transport solution for the fronthaul of open radio access networks. With this solution, uplink bandwidth is shared, and uplink latency performance is thus highly dependent on how bandwidth is allocated. Compounding this issue is that future mobile fronthaul (MFH) is expected to support a range of applications that could vary in terms of traffic patterns, bit rates, etc. In view of dynamic network conditions, we present in this paper a machine learning driven dynamic bandwidth allocation scheme that rapidly learns to optimize bandwidth allocation decisions to satisfy uplink latency requirements. The principle of operation of the scheme detailing the reinforcement and transfer learning framework is first described. Performance evaluation results implemented on a target empirical network are then presented. Results show that self-adaptive bandwidth decisions can be rapidly achieved in response to different traffic patterns, network loads, and line rates, consolidating the potential of the dynamic allocation scheme in supporting diverse applications in future MFH networks.