Machine learning algorithms for 5G PUCCH Channel estimation

Abstract

The arrival of 5G technology marks a new era in the field of mobile communications, which not only greatly improves data transmission speeds and network responsiveness, but also opens up a rich palette for emerging applications such as the Internet of Things, industrial automation, autonomous driving, and augmented reality. As these advanced applications continue to emerge, 5G networks face unprecedented technical challenges, especially in terms of more stringent requirements for maintaining efficient and stable mobile connectivity. Among these techniques, channel estimation is one of the most relevant procedures that should be continuously updated using novel approaches beyond current state-of-the-art solutions. Effective channel estimation algorithms can accurately capture and analyze the propagation characteristics of wireless signals in complex environments, which is crucial for the optimization of 5G networks to give an effective solution to the different kinds of services that have been identified. Traditional channel estimation algorithms based on Least Square optimization methods do not always provide sufficient performance when facing high-speed, dynamic and complex transmission environments in the context of 5G, especially when the wireless channel conditions are poor. In this context, the introduction of Convolutional Neural Networks (CNN) has been lately considered as a novel approach to revolutionize the way in which channel estimation is implemented.

Based on this, three CNN-based channel estimation algorithms have been studied in detail in this MSc thesis, with the aim to improve the channel estimation performance of PUCCH channels. For this purpose, a PUCCH link-level simulator based on the Matlab 5G Toolbox has been first built and its accuracy has been verified by comparing the obtained performance results in the form or BLER curves with the ones reported in reliable sources found in the literature. Based on this link simulator, channel estimation has been performed using three CNN based algorithms, namely "SRIR", "SR", "LI+SR". By testing the developed CNN-based algorithms for channel estimation under different working conditions, it can be concluded that the use of CNN for channel estimation is not recommended for Line-of-Sight channels, as its performance gain is not sufficient to justify the additional complexity required to implement CNN in 5G networks. On the contrary, for Non-Line-of-Sight channels, especially in channel conditions with high Doppler shift and high delay spread for high mobility users in Urban Macro Cell scenarios, the CNN-based channel estimation algorithm is able to achieve a performance gain of 2-3 dB.