Learning Asymmetric Cross-Layer Encryption for Wireless Communication

The tremendous growth of wireless services will make the spectrum congested, competitive, and vulnerable to malicious intents of strong adversaries. This compels us to rethink wireless security for a cross-layer solution that solves encryption, key distribution, and modulation as a joint problem. We propose a novel neural network security algorithm where the trusted transmitter encodes a secret message using a public key to generate a secure waveform. The cipher waveform remains deciphered by the adversary, given the prior knowledge of the public key, while the trusted receiver decodes the secret information from the cipher waveform using both the public and private keys. To ensure integrity, we design a key generation network that generates the public key from the private key with minimal information leakage. Cooperative learning is introduced to lean secure waveform generation between trusted partners over erroneous channels in the presence of an adversary. Our results indicate that the learned cipher algorithm doesn't suffer from the cliff-effect and is reliable over erroneous channels compared to traditional encryption methods.