Deep Learning-Based Detector With Modulation Parameter-Independent Structure for OFDM With Index Modulation

Abstract

Every existing deep learning (DL)-based detector for orthogonal frequency division multiplexing with index modulation (OFDM-IM) is optimized for one instance for modulation parameters (MPs) such as the modulation orders and the number of active subcarriers but does not operate well for other sets of MPs. In this paper, we propose a novel index-bit-detection neural network with modulation parameter-independent structure (IBDNN-MPI) for OFDM-IM, which aims to operate with various MPs. IBDNN-MPI adopts a concatenation of Inception module and fully-connected layers to detect index bits effectively. It is shown via simulations that IBDNN-MPI implementations show better bit error rate (BER) performance and less complexity than existing DL-based detectors optimized for each set of MPs. They have BER performances close to the maximum likelihood and the log-likelihood ratio detectors. Moreover, they can be applicable to dual-mode OFDM-IM at slightly increased computational complexity and cover the scenarios with imperfect channel state information.

Keywords

• Deep learning
• Inception module
• orthogonal frequency division multiplexing with index modulation (OFDM-IM)
• log-likelihood ratio (LLR) detector
• maximum likelihood (ML) detector
• parameter-independent detector