Performance analysis of deep learning and polarized antenna based hybrid beamforming system in mmwave band

Abstract

This thesis aims to analyze the performance analysis of deep learning and polarized antenna based hybrid beamforming (HB) system in millimeter wave (mmWave) band. Considering the trade-off between cost and performance, the HB system is an essential technology in mmWave communication. Therefore, this thesis proposes a deep neural network (DNN), which is the core of artificial intelligence technology, and polarized antennas based HB system. Then, supervised learning of deep learning was applied to HB system using polarized antenna. First, we propose HB algorithms based on dual polarization array antenna (DPAA). In mmWave communications, channel estimation is difficult due to high path-loss. Thus, optimal beamforming weights are determined using codebook based beam training. The first HB algorithm based on DPAA proposed in this thesis is a method to find the optimal beam codewords from the codebooks without considering the beam training complexity. The second DPAA based HB algorithm reduces the beam training complexity of the first proposed scheme with maintaining the performance. The third DPAA based HB algorithm reduces the beam training complexity of the first proposed scheme and robust to cross polarization channel compared to the second proposed scheme. Simulation results showed that the proposed methods achieve higher spectral efficiency than the conventional HB system based on the single polarization array antenna (SPAA). Second, we propose a DNN based HB system using SPAA. In the proposed DNN based HB system using SPAA, supervised learning is applied to classify beam codewords optimized for a given channel environment in a codebook. Then, optimal beam codewords can be inferred without comparing all the beam codewords in the codebook. Due to these reasons, the proposed DNN based HB system using SPAA can drastically reduce the complexity of beam training compared to existing methods. Then, it was investigated that the proposed DNN based HB system using SPAA can achieve almost the same performance as the conventional scheme with very low beam training complexity through simulation. Finally, a DNN based HB system using DPAA is proposed. The proposed DNN for HB using DPAA was designed considering DPAA characteristics. Through the proposed DNN for HB using DPAA, it is possible to predict all beam codewords of all polarization array antenna. Therefore, beam training complexity of the proposed system is much lower than that of the conventional scheme. In this thesis, it was investigated that the performance of the proposed DNN based HB system using DPAA is higher than that of conventional scheme with reduced beam training complexity through simulation.