CSI compression and subspace downlink beamforming for massive MIMO
Amjad, Mubeen (2026)
2026
Tietotekniikan DI-ohjelma - Master's Programme in Information Technology
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
Hyväksymispäivämäärä
2026-05-07
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202605075145
https://urn.fi/URN:NBN:fi:tuni-202605075145
Tiivistelmä
Massive Multiple-Input Multiple-Output (MIMO) systems use large number of antennas at the base station, to serve multiple users simultaneously. This system can utilize beamforming techniques to improve cell coverage and spectral efficiency (SE). In a traditional radio access network (RAN) architecture, the beamforming weights are computed at the distributed unit (DU) , based on the channel state information (CSI) received at the base station. These beamforming weights need to be transferred to the remote unit (RU), which introduces additional load on the fronthaul.
The thesis aims to provide a practical solution to fronthaul congestion due to beamforming weight transfer, that may be a challenge for beamforming with large antenna arrays. The representation of the channel, in the subspace spanned by the eigen vectors of the channel’s covariance, is known to be the most compact representation of the channel. By utilizing large uniform rectangular array (URA), under certain channel conditions the channel eigen vectors can be approximated by appropriately selected 2-dimensional discrete Fourier transform (2D-DFT) vectors. The benefit of using 2D-DFT vectors is that they can be pre-defined and stored in memory at RU. DU needs to send only the vector indexes of the selected 2D-DFT vectors to RU. The algorithm proposed in the thesis uses 2D-DFT vectors to approximate the dominant eigen modes from the channel covariance matrix (CoMa), that are used to transform the channel estimates. Transformed channel estimates are then used to compute regularized zero-forcing (RZF) beamforming weights with reduced dimensions, thus decreasing the data transfer load over the interface between DU and RU. Spectral efficiency and mean throughputs are compared between the uncompressed weight-vectors and the compressed weight-vectors based 3D-beamforming via computer simulations. Moreover, the computational complexities are computed, and the fronthaul load reduction rates of each algorithm have been evaluated.
The thesis aims to provide a practical solution to fronthaul congestion due to beamforming weight transfer, that may be a challenge for beamforming with large antenna arrays. The representation of the channel, in the subspace spanned by the eigen vectors of the channel’s covariance, is known to be the most compact representation of the channel. By utilizing large uniform rectangular array (URA), under certain channel conditions the channel eigen vectors can be approximated by appropriately selected 2-dimensional discrete Fourier transform (2D-DFT) vectors. The benefit of using 2D-DFT vectors is that they can be pre-defined and stored in memory at RU. DU needs to send only the vector indexes of the selected 2D-DFT vectors to RU. The algorithm proposed in the thesis uses 2D-DFT vectors to approximate the dominant eigen modes from the channel covariance matrix (CoMa), that are used to transform the channel estimates. Transformed channel estimates are then used to compute regularized zero-forcing (RZF) beamforming weights with reduced dimensions, thus decreasing the data transfer load over the interface between DU and RU. Spectral efficiency and mean throughputs are compared between the uncompressed weight-vectors and the compressed weight-vectors based 3D-beamforming via computer simulations. Moreover, the computational complexities are computed, and the fronthaul load reduction rates of each algorithm have been evaluated.