Intelligence-aided channel estimation and detection for multi-dimensional index modulation

Abstract

Multiple-Input Multiple-Output (MIMO) technology is a cornerstone of the Fifth Generation Mobile Systems (5G) wireless communication systems. Although it offers significant gains in Energy Efficiency (EE), Spectral Efficiency (SE), performance and throughput, supporting the escalate increasing and dynamically changing demands of large wireless environments remains a significant challenge. In this case, a prominent technique in next-generation wireless communications known as Index Modulation (IM) building on the MIMO system, develops its unique advantages. The concept of index modulation (IM) has been actively researched due to its flexible performance, SE, EE and hardware trade-offs. In order to further improve the SE, Compressed Sensing (CS) and Multi-dimensional Index Modulation (MIM) have been investigated. However, the flexibility and performance gains achieved with IM come at the cost of increased detection complexity. Channel Estimation (CE) for IM systems presents a challenging problem due to the more complex transmission conditions and the pilot allocation requirements. Conventional detection such as Maximum Likelihood (ML) could achieve near-capacity performance, while the detection complexity of computation will exponentially increase with the increase of the degrees of freedom of the IM. Nowadays, machine learning technology has been widely used in wireless network solving massive data problem by its robust computing capability. With the success of many attempts to combine the complex MIMO system and machine learning to detect the signal, it has been proved that learning-aided detection could significantly reduce the detection complexity. Hence, this thesis focusses on designing composite IM system, which may employ CS and learning based reduced complexity detector. First, we propose Deep Learning (DL) based detection for CS-MIM, where both Hard-Decision (HD) as well as Soft-Decision (SD) detection combined with iterative decoding are conceived. We propose two novel neural network aided methods for Iterative Soft Detection (ISD), where iterations are carried out between the CS-MIM detector and a channel decoder. In contrast to the conventional detection of CS-MIM system, which critically relies on the knowledge of Channel State Information (CSI) at the receiver, the proposed learning-aided methods are capable of eliminating the overhead and complexity of CE, which results in an improved transmission rate. Our simulation results demonstrate that the proposed learning techniques conceived for SD CS-MIM combined with iterative detection are capable of achieving near-capacity performance at a reduced complexity compared to the conventional model-based SD relying on CSI acquisition. However, having accurate CSI is essential for reliable MIM, which requires high pilot overhead. Hence, Joint Channel Estimation and data Detection (JCED) is harnessed to reduce the pilot overhead and improve the detection performance at a modestly increased estimation complexity. Then this thesis circumvents this by proposing Deep Learning (DL) based JCED for CS-MIM for significantly reducing the complexity, despite reducing the pilot overhead needed for Channel Estimation (CE). Our simulation results confirm a Deep Neural Network (DNN) is capable of near-capacity JCED of CS-MIM at a reduced pilot overhead and reduced complexity both for Hard-Decision (HD) and SD detection. On the other hand, while MIM uncovers the possibility of combination of IM in different dimensions, roughly assemble of different dimension IM does not fully exploit the flexibility of MIM. In this thesis, we propose Joint Multi-dimensional Index Modulation (JMIM) that can utilize the time-, space- and frequency-dimensions in order to increase the IM mapping design flexibility. Explicitly, this thesis develops a jointly designed MIM architecture combined with CS. Three different JMIM mapping methods are proposed for a space- and frequency-domain aided JMIM system, which can attain different throughput and diversity gains. Then, this thesis extends the proposed JMIM design to three dimensions by combining it with the time domain. To circumvent the high detection complexity of the proposed CS-aided JMIM design, a large fraction of the overall computational complexity is relegated to the offline model training phase. Furthermore, we integrate the different communication tasks, such as CE, decoding and detection into a simple neural network model. Additionally, this thesis investigates the adaptive design of the proposed CS-aided JMIM system, where a learning-based adaptive modulation configuration method is applied. Our simulation results demonstrate that the proposed CS-aided JMIM (CS-JMIM) is capable of outperforming its CS-aided separate-domain MIM counterpart. Furthermore, the learning-aided adaptive scheme is capable of increasing the throughput while maintaining the required error probability target.

More information

Identifiers

ORCID for Xinyu Feng: orcid.org/0009-0006-8363-4771

ORCID for Mohammed El-Hajjar: orcid.org/0000-0002-7987-1401

ORCID for Chao Xu: orcid.org/0000-0002-8423-0342

ORCID for Lajos Hanzo: orcid.org/0000-0002-2636-5214

Catalogue record

Export record