Orchestrating Feedback for Hybrid Rateless Codes

Abstract:
In several scenarios, such as in training and education, it is frequently important to distribute large files to many clients, e.g., operating system boot images or raw data files. To carry out such distribution efficiently, we use wireless broadcast and Fountain Codes with low density random encoding to tolerate network packet losses. In particular, we use a hybrid coding technique that combines forward coding using weak LT Codes with a feedback phase at the end, which allows concluding the process faster with lower computing cost than traditional LT codes and without the precoding of Raptor Codes. The novelty of our approach consists of using a scheduler to orchestrate the clients’ feedback. The scheduler is scalable in the number of clients and allows the server to know which clients are active and in which phase, and control the coding process adequately. We show early experimental results from simulation and from a practical testbed that confirm the effectiveness of our proposed approach. With a feedback phase triggered when just 2% symbols are missing, we reduced the codewords overhead by 2.5% with respect to traditional LT Codes. We also achieved a high reduction in computational overhead associated to weak LT Codes. Moreover, this approach gives the server full control over the transfer, avoiding producing codewords beyond those strictly needed. Using 802.11g in ad-hoc mode and exploiting TXOPs, we transferred a 100MB file in 21.77s with up to 10% losses.