Deep Learning Cognitive Cooperative Data Scheduling Protocol for Small Spacecraft Swarms

A small spacecraft swarm remotely sensing the earth is considered. Sensed observations are to be delivered to an earth station either directly or through geo-stationary earth orbit (GEO) satellite relays with minimal delay. A cognitive cooperative data scheduling (CCDS) protocol is proposed for spacecrafts to leverage node cooperation and deep learning predictions to fully utilize available link capacities in order to minimize data delivery delay. Either periodically, or on an event-driven basis, each satellite predicts its expected sensing data payload and the link capacities to both Earth and GEO stations. Each spacecraft implements a recurrent neural network (RNN) that continuously learns from observed data to predict the amount of expected data over the scheduling horizon. Nodes with extra link capacity than anticipated data, called the relay nodes, broadcast their availability to act as relays to those with more expected data than their link capacities (called the source nodes). Based on neighbor requests, each relay node computes a data acceptance schedule to minimize the overall data delivery delay of the swarm. During the execution phase, source nodes forward their data to the relay nodes over the satellite crosslinks who then forward their payloads to the desired destination. Performance evaluations on a simulated system based on low-earth orbit (LEO) satellite clusters show the promise of the proposed cognitive cooperative data scheduling protocol to significantly reduce the overall data delivery delay of the swarm and to improve utilization of available link capacities. This CCDS protocol can easily be adapted for swarms of small spacecrafts deployed to the Moon, Mars, near-Earth objects (NEOs), and other distant locations for remote sensing missions.