Data-Driven Based Methods for Physical Layer Detection and Estimation in 5G and Beyond Wireless Communication Systems

• Author / Creator

  Mehrabi, Mehrtash

• The fifth-generation (5G) mobile network is growing rapidly and is set to revolutionize the way we communicate, work and live. It offers faster speeds, lower latency, and greater capacity than previous generations of mobile networks. Three main use cases have been defined for fifth-generation (5G) including Enhanced Mobile Broadband (eMBB), ultra reliable & low latency (uRLLC), and massive machine-type communications (mMTC).
  One way to achieve high data rate, low latency, and also massive connectivity is by using multiple-input multiple-output (MIMO) systems. MIMO systems are able to accommodate a high volume of data traffic through multiplexing and also reduce the error rate by diversity. Although, MIMO systems can increase the data rate and improve the performance at the same time, the current explosion in data traffic makes existing detection and estimation algorithms in the physical layer less effective.
  Recently, data-driven methods such as deep learning (DL) have been proposed in various types of problems. In communication system design, in the absence of an accurate mathematical model or when finding the analytical solutions is overly complicated, DL methods can be employed. This thesis focuses on data-driven methods on physical layer problems in MIMO communications. We study the application of DL for detection and estimation problems where the optimal algorithms are costly and existing sub-optimal methods are built on inaccurate mathematical models.
  In the first part of this thesis, we study the problem of channel estimation (CE) in highly dynamic systems. We propose the use of deep neural network (DNN) for k-step channel prediction for Space-time block code (STBC)s, and show that DL-based decision-directed (DD)-CE can remove the need for Doppler spread estimation in fast time-varying flat fading channels.
  Another problem investigated in this thesis is link adaptation for MIMO communication systems. We develop a DL-based link adaptation algorithm for highly dynamic communication links, where adaptive transmission parameters are decided for l > 1 forward time steps without a priori knowledge on channel statistics. Compared to conventional solutions, our approach reduces the feedback requirements from the receiver to the transmitter by a factor of l which significantly reduces the complexity. This achievement comes at no additional cost on the data rate and/or bit error rate.
  The next part of this thesis studies the detection problems in MIMO communication systems. We propose a data-driven-based sphere decoding algorithm, where we model the probability density function (PDF) of the radiuses and use them to select the best radius based on its statistical properties. The performance achieved by the proposed algorithm is very close to the optimal maximum likelihood decoding (MLD), while the computational complexity, compared to existing sphere decoding variants, is significantly reduced. It is shown that the number of lattice points inside the decoding hypersphere is drastically smaller in the proposed algorithm compared to conventional sphere decoding methods. We also study the application of DL in the radius selection and we show that the proposed data-driven-based method has less computational complexity than the DL-based method.
  The last problem studied in this thesis is activity detection (AD) in massive Internet of Things (IoT) networks in grant-free non-orthogonal multiple access (NOMA) systems. Some studies propose compressive sensing (CS)-based method for AD where the high level of message sparsity is needed. In order to remove this need and exploit the statistical properties of the channels we propose a convolutional neural network (CNN)-based method to detect active IoT devices. Our proposed CNN-based method can achieve higher performance compared to the existing non-Bayesian greedy-based methods, while they need to know the activity rate of IoT devices, and our method works for unknown and even time-varying activity rates.

• Subjects / Keywords

  • Wireless Communications
  • Data-Driven
  • Signal Detection
  • Channel Estimation
  • Physical Layer
  • Deep Learning

• Graduation date

  Fall 2023

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/r3-cavd-h620

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.