2021/05/11 by Mathieu Léonardon, Léonardon, Mathieu, Vincent Gripon +1
Computer Science · Engineering · #Advanced Wireless Communication Techniques #Algorithms and Data Compression #Error Correcting Code Techniques #FOS: Computer and information sciences #Information Theory (cs.IT) #Machine Learning (cs.LG)
paper · pdf · doi:10.48550/arxiv.2105.04922
openalex publication_date 2021/05/11 · openalex created_date 2021/08/02 · openalex updated_date 2026/07/28
Polar codes can theoretically achieve very competitive Frame Error Rates. In\npractice, their performance may depend on the chosen decoding procedure, as\nwell as other parameters of the communication system they are deployed upon. As\na consequence, designing efficient polar codes for a specific context can\nquickly become challenging. In this paper, we introduce a methodology that\nconsists in training deep neural networks to predict the frame error rate of\npolar codes based on their frozen bit construction sequence. We introduce an\nalgorithm based on Projected Gradient Descent that leverages the gradient of\nthe neural network function to generate promising frozen bit sequences. We\nshowcase on generated datasets the ability of the proposed methodology to\nproduce codes more efficient than those used to train the neural networks, even\nwhen the latter are selected among the most efficient ones.\n