Data compression for low-latency, high-fidelity, and ultra-reliable AR/VR in bandwidth constrained environments

An innovative software-defined machine learning (ML)-powered network architecture will be developed to enable high performance within the globally distributed Synthetic Training Environment (STE), the High-Performance STE Overlay Network (HP-SON). Our HP-SON architecture utilizes software-defined networking (SDN) concepts combined with machine learning (ML) concepts to provide a virtualized overlay network architecture capable of autonomously learning network paths between distributed STE components and intelligently adapting to optimize user experience. Deep learning neural networks are utilized to learn and characterize network paths between STE components within the globally distributed STE network. HP-SON edge nodes use this learning to perform intelligent adaptation of network behavior and treatment of traffic based on network characteristics, user-driven policy, application behaviors and characteristics, and network loading conditions. Specifically, HP-SON implements dynamic application, performance, and traffic-aware 1) path selection, 2) protocol optimization, and 3) data compression, all enabled by recent advances in neural network ML and SDN techniques. The result is efficient usage of network resources and an improved overall user experience. While initially intended for application to the STE, HP-SON will be network agnostic. Consequently, this approach is believed to have wide commercial application to any IP network as a network performance acceleration appliance