Improving fountain codes for short message lengths by adding memory

We consider applications in which a larger chunk of data needs to be distributed from a centralized, high-powered base station to a set of distributed sensor nodes. Fountain codes offer attractive features for this kind of application, since they require no feedback from receivers and they work over a wide variety of channel characteristics. However, existing work has shown that fountain codes have a relatively high overhead especially in cases when the message length is only small to medium-sized, which is a typical setting in wireless sensor networks. In this paper we present a scheme that improves the efficiency of fountain codes and thus allows receiving sensor nodes to switch off their transceivers earlier. In contrast to earlier schemes that aim to improve the efficiency of fountain or LT codes, our scheme is the first one that introduces memory into the encoding process, i.e. the encoding of the current packet depends on the encoding of previous packets.