Skip to content

Trained Neural Decoder for Convolution Code Signals Over AWGN Channel

Latest
Latest
Compare
Choose a tag to compare
@datlife datlife released this 24 Jun 02:18
· 63 commits to master since this release
week2
49becb7

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

Assets 4