A Novel Federated Computation Approach for Artificial Intelligence Applications in Delay and Disruption Tolerant Networks

Communication between the Earth and distant spacecraft is challenging, since there are frequent link outages due to orbital trajectories, loss of signal due to noise or antenna pointing problems, and long and highly variable, one-way light time delays due to the extreme orbits and distances involved. Deep space link reliability has traditionally required manual intervention by Mission Operations teams, and new Solar System Internetworking protocols known as Delay and Disruption-Tolerant Networking (DTN) have been developed to automate space communications. This automation now provides a mechanism for space networking and link performance optimization using artificial intelligence (AI) and machine learning (ML) methods which we present in this paper. We discuss Federated Learning (FL) and our application of FL to monitoring and optimization of DTN networks. Federated learning is a distributed learning framework that enables machine learning to be performed on decentralized data with improved protection for each collaborator’s data privacy and communication efficiency. This work outlines a novel approach to advancing current AI applications running over delay and disruption tolerant networks by leveraging two forms of federated learning for space applications, and the core DTN Bundle Protocol to produce smarter ‘on-edge device’ machine learning models. Our work demonstrates that the network can operate with more link bandwidth efficiency and less resource consumption, while ensuring privacy and quality constraints. Experiments on a popular image dataset show that DTN-ML can have similar or better performance when compared to existing alternatives, but with much less communication overhead.