Resource Allocation Optimization with Cluster Based Auction in Wireless Virtual Reality Network

Virtual reality (VR) has shown great potential in various fields to provide novel experiences and enhance industrial working efficiencies. The characteristics of high resolution for VR streaming lead to the high requirement of bandwidth during VR streaming, while the proactive streaming in wireless VR communication provides users with an immersive experience under strict latency constraints through viewport prediction. As a large amount of user data is required to train a strong-applicable prediction model and these data are collected in a distributed manner, the federated learning (FL) framework can be utilized to gather local training weights into global weights without occupying limited bandwidth for data transmission. Considering users may have varying individual requirements while participating in global FL training, an auction-based user clustering algorithm is proposed in this paper to group users based on their preferences for user utility. The simulated annealing (SA) scheme is utilized to select the number of users in each cluster for FL participation and a cluster-based resource allocation approach is presented to allocate resources to users based on characteristics of clusters. By considering the utility of both the small base station (SBS) and users, the proposed approach can provide users with enhanced utility and SBS with sustainable bandwidth usage.