Codebook design and multi-user detection algorithms for the massive machine type communication with uplink grant-free NOMA

Hasan, Shah Mahdi

Title: Codebook design and multi-user detection algorithms for the massive machine type communication with uplink grant-free NOMA
Creator: Hasan, Shah Mahdi
Relation: University of Newcastle Research Higher Degree Thesis
Resource Type: thesis
Date: 2022
Description: Research Doctorate - Doctor of Philosophy (PhD)
Description: Massive Machine Type Communication (mMTC) is a wireless network paradigm which aims for providing connectivity among a very large number of automated agents (e.g., sensors, smart meters). The mMTC systems fundamentally diﬀer from the human-agent based wireless networks in many ways. This prohibits the direct adoption of oﬀ-the-shelf technologies that are tailor-made for the legacy wireless systems (e.g.: 4G LTE-A). Moreover, due to the stochastic nature of the operation and scarce frequency resources, the mMTC systems pose a set of unique challenges to be addressed. To tackle some of those challenges, Grant-Free Non Orthogonal Multiple Access (NOMA) framework has been proposed. A grant-free system envisions lowering the signalling overhead that is commonly required by the legacy grant-access based machine type networks (i.e., Narrow-Band Internet of Things (NB-IoT) and Long Term Evolution-MTC (LTE-M)). The NOMA addresses the problem of scarce resource allocation by relaxing the requirement of orthogonality. This is can be achieved by, for example, assigning unique Spreading Sequences (SS) to each User Equipment (UE). The collection of the spreading sequences is known as the codebook. The design of the codebook plays a pivotal role in the performance of the NOMA systems, in terms of robustness and scalability. Together, the grantfree NOMA framework provides two-fold support for realizing the mMTC systems. Firstly, it provides supports to enable massive connectivity while improving the spectral efciency. Secondly, the reduced signalling overhead allows the inclusion of the battery-constrained, low-rate UEs to be a part of the mMTC system. In this thesis we focus on two diﬀerent aspects of the grant-free NOMA in an UL mMTC system. Hence, the contribution in this thesis is categorized into two parts, namely Part II: Designing Robust and Scalable Sparse Code Multiple Access Codebooks, and Part III: Developing Multi-user Detection Algorithms for Uplink Grant-free NOMA. In Part 2, we propose novel codebook designing methods for Sparse Code Multiple Access (SCMA), which is an emerging subclass of the NOMA schemes. More specifcally, we propose two novel uplink (UL) SCMA codebook optimization algorithms. We show how jointly optimizing the UL SCMA codebooks for correlated fading channels for a given Eucledian Distnace Matrix (EDM) can be benefcial. We demonstrate how the matrix-lifting semidefnite relaxation (MSDR) can be used to tackle the non-convex constraints which arise from the average power constraints. Furthermore, we also propose using the complex Equiangular Tight Frames (ETF) as SCMA codebooks. We demonstrate a constrained gradient based alternating projection algorithm to construct the SCMA codebooks. By conducting elaborate numerical simulations we demonstrate the efcacy of the proposed codebook designs. In Part 3 we focus on developing novel Multi-user Detection (MUD) algorithms for UL grant-free NOMA systems. We show that, by intelligently engineering the spreading sequences, the active UE detection (AUD) problem can be formulated as a frequency estimation problem. By exploiting the resultant Vandermonde structure in the signal model, we demonstrate that the non-iterative subspace algorithms can outperform the Compressed Sensing (CS) based MUD algorithms (CS-MUD) in terms of both error performance and algorithm runtime. Moreover, these subspace estimation algorithms do not require any prior information about the UE activity and the channel/noise statistics. We also consider the problem of dynamic Random Access (RA) where the UE can enter and exit within an RA opportunity independently. By combining sparse covariance estimation techniques (SPICE) and log-likelihood ratio (LLR) based hypothesis testing, we propose a MUD algorithm that is capable of not only estimating the set of active UEs, but also their activity/inactivity across the RA opportunity. Furthermore, we develop a hierarchical MUD algorithm to address the poor AUD performance demonstrated by both CS-MUD and the subspace estimation based MUD in highly overloaded mMTC systems. In this method, we propose a two-phase data transmission strategy where the UEs are randomly assigned to some pre-defned clusters. Therefore, the MUD algorithm successively estimates the active clusters, followed by the detection of the active UEs. We demonstrate that the proposed hierarchical MUD combines the strength of both CS-MUD and subspaceMUD and deliver promising error performance in highly overloaded scenarios. Up until this point, all of the works have considered ﬂat Rayleigh fading system models. Finally, we explore the utility of the deep neural network (DNN) based solutions for frequency selective Rayleigh fading models. We formulate the AUD problem as a multi-label classifcation problem and propose using model compression techniques like Knowledge Distillation (KD) for approximating the ensembles of trained DNN based MUD to enable fast AUD. We also demonstrate how the inter-resource correlation can be used to estimate the complex channel vectors which is an ill-posed problem otherwise.
Subject: 5g; NOMA; IoT; massive machine type communication; subspace estimation; compressed sensing; knowledge distillation; grant-free access; channel estimation
Identifier: http://hdl.handle.net/1959.13/1514227
Identifier: uon:56827
Language: eng

Hits: 9
Visitors: 9
Downloads: 0