Large scale SDR networking testbed-- Prototyping, data collection and processing, applications

The explosion of wireless mobile devices and services demands more spectrum efficiency, higher data rate, larger capacity for the next generation of wireless network. Cognitive radio and massive MIMO are candidates of disruptive technologies tackling the challenges. Building a flexible and multi-functional large scale networking testbed is required for the experimental validation and real data collection for the theoretical research.;In this dissertation, the system architecture and detailed software design of a large scale networking testbed are presented. The testbed is based on the state-of-the-art software-defined radio (SDR) platform. The concept of cognitive radio as sensors is demonstrated by several novel applications and experiments performed on this testbed, including through tree detection, passive target localization with machine learning method, weighted centroid localization, etc.;A multi-functional, flexible SDR networking platform is made possible to support 70 SDR front-ends and 30 high performance PCs, by integrating functionalists including physical data collecting, MAC protocols (TDMA), control plane, distributed and parallel computing. We reshape our large scale SDR networking testbed to a massive MIMO system. With the measurement of the large channel matrix, it is for the first time that the feasibility of Vandermonde model is demonstrated by the experimental data. We gain insight from the experimental work that the Vandermonde channel matrix is more realistic than the Gaussian channel matrix in modeling the massive MIMO system. Our massive MIMO system testbed naturally becomes a big data collecting system. We applied variant random matrix theoretical models on the wireless network data corresponded to different scenarios. The encouraging potential of random matrix based big data analysis is revealed. Random matrix theory provides a new approach for wireless big data analysis.