Learned Autoencoder Physical Layer
End-to-End Deep Learning Wireless Transceiver Replacing Standard Blocks
AI system where transmitter encoding and receiver decoding are trained end-to-end as an interconnected deep neural network autoencoder.
Technical Explanation
Traditional wireless systems are composed of modular hand-crafted blocks: source encoder, channel encoder, modulator, channel estimator, equalizer, demapper, and channel decoder. The Learned Autoencoder Physical Layer replaces all of these blocks with a single deep neural network. The transmitter acts as a learned encoder outputting complex I/Q symbols, the wireless propagation channel acts as an unalterable stochastic layer, and the receiver acts as a learned decoder, discovering novel waveforms and constellations superior to human engineering.
Key Functions
- Replaces legacy modular transceivers with an end-to-end learned neural autoencoder
- Discovers unconventional multi-dimensional geometric constellations and codes
- Optimizes communication performance jointly over non-linear hardware distortions
- Adapts transmission strategies autonomously to specific physical environments