Loading...

Beamforming for 5G Wireless Communication Systems using Machine Learning Methods

Sadeghi, Ali | 2023

17 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56672 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Behroozi, Hamid; Fakharzadeh, Mohammad
  7. Abstract:
  8. The significant growth in wireless data traffic over the past few years has necessitated the use of new frequency bands for wireless communications. With the advent of the fifth-generation mobile network (5G), millimeter wave band will be widely employed for the purpose of wireless data transmission, increasing in channel bandwidth and data rate. Due to high propagation loss, severe vulnerability to blockage and poor reflection in millimeter wave frequencies, beamforming for antenna arrays is being considered a key enabler for millimeter wave applications in 5G. Beamforming contributes significantly to maintaining channel stability and capacity through enhancing Signal-to-Noise-Ratio (SNR) and eliminating interference. Using high frequencies, the dimensions of wireless equipment such as antenna size and spacing decrease, making it possible to implement low-cost wireless communications beamforming systems suited for commercial purposes. Antenna size reduction also allows preserving a greater number of antennas in a single package. The greater the number of antennas, the higher the directivity. Hence, millimeter wave band is suitable for producing directional beams, which creates an opportunity for developing techniques to enhance signal power at the receiver. When enabling beamforming for 5G, profound attention should be paid to practical conditions such as the need for low delay and non-ideal characteristics of real-world instruments including phase shifters, needs that might not be met using conventional approaches. In this thesis, a beamforming method for a phased array system with practical phase shifters is proposed, using machine learning and neural network algorithms to take advantage of their swiftness in performing beamforming computations as well as their capabilities in providing robustness against non-ideal conditions of instruments and the environment. In the beginning, examples of conventional beamforming methods are mentioned such as eigenvalue-based and iterative algorithms, and their limitations are reminded, necessity of prior knowledge about the impinging signals and high computational complexity to name a few. A beamforming method with practical phase shifters using a tailored Deep Neural Network (DNN) architecture is then proposed in order to overcome the limitations mentioned above. The proposed system, uses the signals received by the antenna array, needless of any further knowledge, to estimate optimum control voltages for the phase shifters so that the output signal reaches its highest possible amplitude. Finally, the proposed system is compared to the state-of-the-art algorithms from different aspects such as array gain, noise robustness and side lobe level. The most notable novelties of this work can be listed as follows: Employing Gray code to train and represent the neural network output to improve its performance, solving the beamforming problem concerning non-ideal factors, and generating the required data as well as designing an appropriate structure for the neural network concerning the practical issues of the problem
  9. Keywords:
  10. Beamforming ; Phased Array ; Phase Shifter Circuit ; Deep Learning ; Neural Network ; Machine Learning ; Fifth Generation of Mobile Networks

 Digital Object List

 Bookmark

No TOC