We tackle the problem of secret key generation for underwater acoustic channels, where two users, Alice and Bob, aim at generating a secret key from their estimated features of the underwater channel (e.g., the channel impulse response). We focus on the advantage distillation step, where Alice and Bob extract bits from the channel feature. However, in a wireless channel, any malicious user sufficiently close to either Alice or Bob, namely Eve, can get partial information on the key. Typically, Eve’s knowledge about the key is reduced in the last step, called privacy amplification.

This paper proposes a strategy to increase Alice’s and Bob’s secrecy capacity during the advantage distillation step. Here, each user designs a quantizer, whose thresholds are chosen by exploiting the statistical distribution of the chosen acoustic channel feature. Next, Alice broadcasts over an authenticated public channel a correction, that allows Bob to improve the correlation between Alice’s and Bob’s extracted sequence. We show that a proper design of the quantizer allows Bob’s information about the sequence extracted by Alice to be greater than the one obtained by Eve. Performance are tested using both simulated and experimental dataset, drawn from a sea experiment, using several channel features.