Network Coding for Mobile Devices - Systematic Binary Random Rateless Codes

Abstract

In this work we consider the implementation of Random Linear Network Coding (RLNC) on battery constrained mobile devices with low computational capabilities such as; sensors, mobile phones and Personal Digital Assistants (PDAs). It is non-trivial to create an efficient implementation of RLNC which is needed to ensure high throughput, low computational requirements and energy consumption. As a consequence there does not, to the best of our knowledge, exist any such implemen- tation for mobile device that allow for throughput close to what can be achieved in e.g. Wireless Local Area Network (WLAN). In this paper we propose to base RLNC on the binary Galois field and to use a systematic code. We have implemented this approach in C++ and Symbian C++ and achieve synthetic encoding/decoding throughput of up to 40/30 MB/s on a Nokia N95-8GB mobile phone and 1.5/1.0 GB/s on a high end laptop. Index Terms—Mobile devices, Network coding, Reliable Mul- ticast. I. INTRODUCTION A large body of existing literature (1) treats the theoretical benefits of Network Coding (NC). However, the costs of implementing NC in terms computational overhead, memory consumption or network usage is often not considered. In this work we consider the implementation of RLNC on mobile bat- tery constrained devices with low computational capabilities, such as sensors, mobile phones or PDAs. The computations performed using RLNC is based on finite fields arithmetic also known as Galois fields. From a coding perspective the field size, q, used should be large to ensure that coded packets are linearly independent, additionally increasing the size of the field elements is advantageous as this reduces the number of