Efficient Design of GPU-based Reconfigurable Software Modem in HetNet

Abstract

This dissertation presents an efficient design of a software modem to be used in HetNet for software defined radio (SDR) system using graphic processing unit (GPU) as its modem processor for a high-speed data processing. To verify the possibility of using GPU for SDR system, this dissertation confirms the real-time processing of software modem using GPU, and checks its reconfiguration operation. Because GPU contains many arithmetic logic unit (ALU) that enable very high-speed parallel processing, GPU has been widely used for signal processing applications that need the identical calculation repeatedly for a large amount of data. Compute unified device architecture (CUDA) provides a C-based programming environment where a large number of ALUs contained in NVIDIA’s GPUs can be easily utilized GPU. In order to fully exploit the performance of parallel processing in a GPU, both a proper algorithm for parallel processing and an efficient method of accessing the layered memory architecture should be taken into consideration. Also, to verify the reconfigurability of GPU-based software modem, the implemented system developed in accordance with ETSI RRS standards. The software modem based reconfigurable MD standardized by ETSI TC RRS is that next generation MD which is available to use multiple wireless communication standards simultaneously in a same hardware platform. To achieve this, this dissertation introduces the MD categories according to the reconfiguration capability defined by ETSI TC RRS. Also, this dissertation represents the software architecture and key interfaces of reconfigurable MD which is applicable the multiple software modem simultaneously. Form the experimental results, it has been confirmed that the implemented system using GPU as its modem processor can operate real-time communication system. Also, it has been confirmed the reconfigurability of software modem through a dual-mode controller as a CR engine which is able to automatically search for a best appropriate waveform and its frequency band. Although the implemented system adopts for the signal processing of the physical layer only, it is possible to develop the commercial LTE and WiMAX system expanding the implemented GPU based software modem in this dissertation.