Design Of Real-time Spectrum Processing Software Based On GPU Parallel Computing

Benefitting from the powerful ability of calculation and practicability,Graphics Processing Unit(GPU)has been universally concerned in digital signal processing domain.GPU is provided with great ability of intensive data processing,which instantaneously implements multiple threads via numerous yet simple Arithmetic Logical Units(ALUs)to realize the parallel computing.Real-time spectrum analysis device generally have the characteristics of wide-band,high sampling rate and large data throughput.By virtue of the efficient parallel computing capacity of GPU,a large number of frequency domain operations,statistical processing and other transactions are deployed to GPU for completion,which can greatly improve the spectrum processing efficiency.Dissertation designs a real-time spectrum analyzer software based on GPU heterogeneous computing system and OpenCL software programming framework.The spectrum analyzer adopts throughput-oriented parallel processing method in signal processing,which is characterized by real-time eff-iciency.The main contents of this dissertation are as follows:(1)Real-time spectrum software requirement analysis and software architecture design.The main controlling module,test module,graphic display module and algorithm processing module are designed for real-time spectrum analysis.(2)The application of GPU parallel computing in real-time spectrum software.The parallel computing capacity of GPU and OpenCL standard for heterogeneous systems are utilized to implement overlapping,windowing,Fourier transform,three-dimensional spectrum density statistics and other real-time signal processing for signal flow in time domain.(3)The realization of GPU-based graphic display technology.It consists of logarithmic compression,statistical interpolation and digital afterglow algorithm in 3-D spectrum density graph,waterfall graph and single spectrum graph.(4)Research on the coordination mechanism between GPU high-speed computing and software operation.In order to improve the execution eff-iciency of real-time spectrum software and coordinate the speed matching problem between GPU operation and software,this software adopts the synchronization mechanism of multi-thread communication and realizes the synchronization between threads by means of mutual exclusion and semaphore.This dissertation realize the GPU-based real-time spectrum software design.We exploit“CPU + GPU”joint execution mode for spectrum analysis and calculation and deploy the intensive computation on the GPU,finally solving the problem that CPU can hardly satisfy the calculation needs of large amount of data,improving the efficiency of the frequency spectrum analysis and realizing each modules.The experimental results show that the GPU-based real-time spectrum processing software performs outstandingly.The parallel framework OpenCL was used for spectrum processing at the same time with CPU and GPU,and the spectrum analysis efficiency of GPU was 1.39 times of the CPU when 512K data volume was recorded and the computational refresh rate of the 3D spectrum reaches 30 frames/s.