Machine learning for space communications service management tasks

Currently, NASA space communications links are individually scheduled by each mission's operations personnel coordinating with the network service providers. The scheduling of communications services typically takes place many days in advance of when the service is needed. This suffices because there are only several dozen mission platforms, using point-to-point communications, and generally in nominal operating modes. In the future, with potentially many more active flight platforms, more complex relaying or internetworking, and more emphasis on quality of service for different types of data, network service management will increase in difficulty. Scheduling and other service management activities could grow more labor intensive and costly for both mission operations and communication service provider staff. In order to enable scale up the communications services, while reducing human involvement, this paper describes our work applying machine learning techniques to implement intelligent routing that addresses space communications service management challenges. There are precedents for similar problems in terrestrial networking, and the main contribution of this paper is in extending to the unique aspects of space communications. Successful application of machine learning can assist in automation of both current space communications and future space internetworking service management activities, including pre-service planning, provisioning of acquisition data, in-service performance monitoring, real-time service control, and identification of anomalies or other contingency modes. This paper includes description of some relevant problems, existing machine learning approaches to similar problems, and description of initial evaluations using network emulation.