Reducing the memory for iteration-exchanged information and border future metrics in the HomePlug AV turbo decoder implementation

摘要

HomePlug AV is the most successful standard for in home power line communications. To combat non-ideality of the power line channel it includes a double binary turbo forward error correcting scheme. Unfortunately, it is known that the memory required by double binary turbo decoders for iteration-exchanged information is roughly three times the memory required for binary turbo codes. Moreover, high throughput implementations based on border state metric inheritance, require additional memories to store border state metrics from an iteration to the next one. This work faces these two aspects by analyzing compression techniques to reduce the amount of memory required to store both iteration-exchanged information and border future metrics. Experimental simulations show that non-uniform quantization and least significant bits dropping allow for a significant memory reduction (up to 30%) with a bit error rate performance loss of about 0.1 dB and a negligible logic gates overhead.