This thesis proposes a novel technique for implementing a rateless coding scheme by employing intelligent methods, where the agent learns to decide the corresponding rate given a channel capacity. The main concepts behind reinforcement learning (RL)-based rateless coding are (i) learning capability of the decoder and (ii) learning capability of rate determination to satisfy the Shannon channel coding theorem. This thesis integrates both a transfer learning (TL) framework and a reinforcement learning framework to address this concept.
This thesis: (i) studies machine learning (ML) structure for box-plus operation as an element of future error correction based on artificial intelligence (AI) using soft information processing with log-likelihood ratio (LLR) values, (ii) investigates the best structure of neurons in ML to deal with box-plus operation, (iii) utilizes a TL approach to learn a generalized message-passing algorithm for quasi-cyclic low-density parity-check (QC- LDPC) codes, by replacing