Trained Neural Decoder for Convolution Code Signals Over AWGN Channel

This release contains pre-trained models for 1/2 - RSC Neural Decoder discussed in the paper "Communication Algorithms via Deep Learning" (ICLR 2018).

• BiGRU2-400_bkl100_snr0_12k.hdf5 : pre-trained 2 Bidirectional GRU (400 hidden units) Neural Decoder. This model is trained on 12,000 inputs with block_length=100 at signal to noise (SNR) 0.0.
• BiGRU2-400_bkl100_snr0_120k.hdf5 : pre-trained 2 Bidirectional GRU (400 hidden units) Neural Decoder. This model is trained on 120,000 inputs with block_length=100 at signal to noise (SNR) 0.0.

A weekly report, including README.md and notebooks, is available under reports/week2