Optimization of CSI Precoding Matrix Generation in 5G New Radio
Virta, Aleksi (2025)
2025
Sähkötekniikan DI-ohjelma - Master's Programme in Electrical Engineering
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2025-05-22
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202505225971
https://urn.fi/URN:NBN:fi:tuni-202505225971
Tiivistelmä
Precoding is an integral part of the transmitter signal processing chain in wireless communication, allowing to improve signal quality and capacity in both modern and future generations of telecommunication. The 3rd Generation Partnership Project (3GPP) technical specification TS 38.214 [1] provides definitions for Type I single panel and multi-panel codebooks. The codebooks consist of precoding matrices, which are used for precoding the vector representing a cyclic prefix orthogonal frequency-division multiplexed waveform linearly using the matrix product. The choice of a precoding matrix is carried out according to precoding matrix indicator, which is a part of channel state information (CSI) reporting.
This thesis aims to propose an efficient technique to implement 5G Type I CSI single panel precoding matrices using 4 or 8 antenna ports. This proposal is of interest for a practical codebook generation, simplifying it in terms of the memory needed to store the precoding matrices. The proposed algorithm has been validated and compared against a so called direct implementation, where the precoding matrices are stored as they are defined in the standard. While 4 and 8 antenna ports cases have been studied more thoroughly, the proposed algorithm can be extended into cases of more antenna ports. The definition of NR codebooks presented in the 3GPP standard is revised and analyzed without loss of generality from signal processing point of view. Optional mathematical expressions for the codebooks are presented alongside with proofs for their derivation. The proposed schemes demonstrate that the required memory can be reduced by approximately 95 % compared to the direct storing of each precoder. Our proposals thus bring fundamental benefit to precoder generation systems. Furthermore, with respect to direct precoding using a single matrix multiplication, the analysis carried out in this thesis indicates that precoding in parts shows a reduction of up to 87.5 % in number of needed real additions, and a reduction of up to 37.5 % of needed real multiplications associated with complex arithmetics. Lastly, further development ideas are presented to gain prospective of the impact and potential ideas for extending the algorithm to more complex cases.
This thesis aims to propose an efficient technique to implement 5G Type I CSI single panel precoding matrices using 4 or 8 antenna ports. This proposal is of interest for a practical codebook generation, simplifying it in terms of the memory needed to store the precoding matrices. The proposed algorithm has been validated and compared against a so called direct implementation, where the precoding matrices are stored as they are defined in the standard. While 4 and 8 antenna ports cases have been studied more thoroughly, the proposed algorithm can be extended into cases of more antenna ports. The definition of NR codebooks presented in the 3GPP standard is revised and analyzed without loss of generality from signal processing point of view. Optional mathematical expressions for the codebooks are presented alongside with proofs for their derivation. The proposed schemes demonstrate that the required memory can be reduced by approximately 95 % compared to the direct storing of each precoder. Our proposals thus bring fundamental benefit to precoder generation systems. Furthermore, with respect to direct precoding using a single matrix multiplication, the analysis carried out in this thesis indicates that precoding in parts shows a reduction of up to 87.5 % in number of needed real additions, and a reduction of up to 37.5 % of needed real multiplications associated with complex arithmetics. Lastly, further development ideas are presented to gain prospective of the impact and potential ideas for extending the algorithm to more complex cases.