IEEE International Symposium on Information Theory (ISIT)
Large Kernel Construction Based on Generalized Polarization Transforms
Anatolii Ashikhmin, Peter Trifonov
Abstract
A novel method for constructing polarization transforms is presented. It can be employed for constructing large kernels. We show that the polarization transforms of polar codes with large kernels and convolutional polar codes can be constructed by employing the proposed framework.A new kernel search algorithm is presented. It takes a given kernel and tries to find kernels with better polarization properties of lower processing complexity, applying elementary row operations to close rows.Novel kernels of size 64 with improved rate of polarization and processing complexity are obtained.