Source Polarization-Adjusted Convolutional Codes

Motivated by applications to low-latency secret key generation in physical-layer security, we study Polarization-Adjusted Convolutional (PAC) codes for source coding with side information. Source PAC codes operate in a dual manner to channel PAC codes by introducing a rate-one convolutional code after the polarization transform. The decoding of source PAC codes requires a careful scheduling of the successive cancellation decoder and a careful optimization of the rate profiling. Our empirical results demonstrate the improved performance of source PAC codes over regular polar codes using Successive Cancellation List (SCL) decoding. We illustrate the performance in terms for key generation rate in a secret-key generation setup over an Additive White Gaussian Noise (AWGN) channel, suggesting that PAC codes could improve the performance of physical-layer security schemes at short blocklength.