14. Juni 2017;

On Monday, June 19, doctoral student Sebastian Cammerer will be giving a talk "On Deep Learning-based Channel Decoding" at a meeting on Deep Learning and Coding as the Tel Aviv University.

In this talk, we revisit the possibilities and the practical limits of deep neural network-based one-shot channel decoding. Therefore, we analyze two code families: random codes and structured codes, such as polar codes. Although it is possible to achieve maximum a posteriori (MAP) bit error rate (BER) performance for both code families and for short codeword lengths, we observe that (i) structured codes are easier to learn and (ii) the neural network is able to generalize to codewords that it has never seen during training for structured, but not for random codes. These results provide some evidence that deep neural networks (DNN) can learn a form of decoding algorithm, rather than only a simple classifier. Although it is proven that the memory capacity of the DNN scales exponentially, the training complexity of deep learning-based channel decoders scales exponentially with the codebook size and therefore with the number of information bits. As alternative approach, we show that the conventional iterative decoding algorithm for polar codes can be enhanced when sub-blocks of the decoder are replaced by DNN-based components. Thus, we partition the encoding graph into smaller sub-blocks and train them individually, closely approaching MAP performance per sub-block, while keeping the training complexity in a feasible region. The resulting decoding algorithm is non-iterative and inherently enables a high level of parallelization, while showing a competitive BER performance.

-On Deep Learning-Based Channel Decoding [arXiv:1701.07738]
Tobias Gruber, Sebastian Cammerer, Jakob Hoydis, Stephan ten Brink
-Scaling Deep Learning-based Decoding of Polar Codes via Partitioning [arXiv:1702.06901]
Sebastian Cammerer, Tobias Gruber, Jakob Hoydis, Stephan ten Brink

He will then head to Haifa to attend the 7th Annual International TCE Conference on Coding for Storage and Information Systems.

Links:  The 7th Annual International TCE Conference on Coding for Storage and Information Systems