Optimized Polar Encoding on an NVIDIA GPU : An Evaluation of Polar Codes Adhering to the 5G Standard

Sammanfattning

Global mobile data traffic is continuously increasing, placing greater demands on faster, more reliable communication systems. Channel coding is the process of adding controlled redundancy that enables the receiver to detect and correct transmission errors. To support higher data rates in 5G mobile networks while maintaining reliable communication, it is important to minimize the computational time of channel coding. This thesis investigates how polar encoding can be parallelized to reduce encoding latency, as polar codes are the channel coding technique used for control information in 5G networks. To achieve this, two functionally equivalent polar encoding algorithms were developed: a sequential implementation for execution on a CPU and a partially parallel implementation for execution on a GPU. The implementations were evaluated by comparing execution times across different code lengths and varying numbers of concurrent codewords. The results show that GPU parallelism can reduce the computational time of polar encoding, provided the workload is sufficiently large to offset GPU execution overhead.