2013.B.2.1 Robust Communications for CubeSat Cluster using Network Coding

Author(s)

Gokul Bhat (1), Obulapathi Challa (1), Paul Muri (1), Janise McNair (1)

University of Florida, USA

Session

B.2

Keywords

Reliability, network coding, retransmissions, latency, raspberry pi, encoding time

Abstract

Miniaturized satellites (e.g. CubeSats) have stringent weight, volume, power, and geometry constraints that severely limit their processing, memory, and communication capabilities. A typical CubeSat mission suffers from low-speed data communication along with a highly unpredictable downlink communication channel with high packet erasure rates. Current solutions split the data into small chunks and downlink them to corresponding ground stations from different CubeSats simultaneously. However, these links are characterized by their unreliable nature and high link error rates that may lead to severe packet losses. Majority of current implementations handle lost packets by performing retransmissions which incur a significant added transmission delay leading to timeouts and lower throughput.

Our contributions in this paper are two-fold: we propose to handle packet losses by sending coded packets (instead of packet chunks) created by linear combinations of the chunks, also known as random linear network coding. We then demonstrate the utility of this approach for tail-end and real-time packet delivery scenarios. We also evaluate our algorithm on hardware test-bed comprising of Intel Core 2 Duo and Raspberry pi processors for varying cluster sizes and packet erasure rates by determining total transmission time and current drawn while encoding packet chunks. For instance, with a cluster size of 10, a packet erasure rate of 10% and a timeout period (which is the waiting time before the sender decides to retransmit in case of a packet loss) of 3 seconds, the total transmission time for a packet chunk of 100 bytes using traditional re-transmission scheme was 3.17 seconds and the traditional redundant packet transmission scheme was 3.13 seconds. On the other hand, the network coding approach resulted in a total transmission time of 195 ms. We also determined that the network coding algorithm gives the lowest encoding time (4 ms) for a cluster size of 5 and the worst case encoding time (43 ms) for a cluster size of 25. For an average cluster size of 10, the encoding time was 10 ms, which is insignificant compared to the large transmission times on the order of a few seconds involved in the traditional retransmission scheme. Our numerical results show that for high packet erasure links, sending network coded packets (i) reduces successful packet transmission time by 95% compared to existing approaches for CubeSat to ground station communication and (ii) draws negligible amount of current for the only memory intensive process of encoding the packet chunks.