Codeword extension for polar decoding and CNN-aided adaptation

Abstract

Decoder of polar coded bits requires the successive cancellation (SC). Its capability is known to be limited when the code length is shorter. Encoding block size for the polar coding is predetermined and the block is divided into two types: frozen bits that contain fixed zeros and information bits that contain data. This letter proposes a new decoding scheme that artificially extends the block length by concatenating the copied codeword at the receiver side, focusing on the polar codes’ generator matrix characteristics. Since it simply doubles the computation load, we additionally apply a convolutional neural network (CNN) to determine its applicability by observing the received signal power spectrum. Computer simulations verifies its effectiveness; improved bit error rate and reduced computation complexity.