Machine Learned Wireless Channel Codes
Overview:
We use autoencoder neural networks to generate aribtrary rate, single user codes. Arbitrary rates are particularly useful for multiple access channels where the achievable rate regions often require highly irrational rates. Conventional codes, which are typically simple rational rates leave a large "gap" between the actual rate and the achievable capacity. We demonstrate the effectiveness of these codes by using a modified Successive Interference Cancellation decoder.
Publications:
- 2022 IEEE Latincom Design and Analysis of Neural-Network-based, Single-User Codes for Multiuser Channels [pdf]
