Software-based Baseband Processing for Massive MIMO
Master of Science
Large-Scale multiple-input multiple-output (MIMO) is a key technology for improving spectral efficiency. However, it requires massive, real-time computation. All existing solutions are based on dedicated, specialized hardware, e.g., FPGA, that is expensive, inflexible and difficult to program. This thesis investigates a software-only solution that exploits recent CPU development supporting many cores and architectural extensions for fine-grained parallelism. We present a high-performance framework for real-time, large-scale baseband processing on a many-core server. To achieve the high data rate and low latency promised by 5G, the framework utilizes data parallelism and exploits architecture features, including memory hierarchy and SIMD extensions, to accelerate computations and data movements. We report a prototype on a 36-core server and evaluate its performance.
baseband processing; massive MIMO; parallel computing; 5G