Speaker
Description
A widely studied configuration within Cell-Free massive Multiple-Input Multiple-Output (CF mMIMO) networks is the centralized architecture, where all Access Points (APs) are connected to a Central Processing Unit (CPU) that performs Linear Minimum Mean Square Error (LMMSE)-based uplink signal estimation. As an alternative, sequential uplink signal processing refines users' signal estimates incrementally through a daisy-chain of APs. While theoretical analysis predicts equal performance as centralized processing, the effect of real-world constraints on this equivalency has not yet been evaluated.
This thesis presents an over-the-air implementation of sequential uplink processing for CF mMIMO systems using GNU Radio and PlutoSDRs. A small-scale prototype with two users and two APs is developed to evaluate sequential signal estimation. Unlike many studies in the literature, the evaluation accounts for real-world signal propagation and processing constraints. Those include propagation through a wireless channel and impairments introduced by the radio transceiver, as well as limited precision in digital signal processing, imperfect synchronization, and measured channel statistics. As a foundation, a state-of-the-art Orthogonal Frequency Division Multiplexing (OFDM) transmitter in GNU Radio is extended for flexible multi-user operation. A custom-designed receiver is introduced, incorporating support for LMMSE-based signal and channel estimation, multi-user and multi-AP configurations and coarse time synchronization via a Zadoff-Chu sequence preamble.
To enable LMMSE channel and signal estimation, a method to derive real-world channel statistics through experimental measurements is proposed and applied. The results reveal key deviations from idealized assumptions with most notably channel statistics that vary across measurements, likely caused by fine time synchronization errors in the PlutoSDRs. Nonetheless, since the channel variance consistently exceeds the noise variance, the estimation method remains robust to these variations.
Experimental results confirm that the implemented sequential and centralized architectures produce nearly identical uplink signal estimates when applied to the same dataset, with discrepancies on the order of 10-3 likely due to limited numerical precision in Python.
| Talk Length | 15 Minutes |
|---|
