Development Of Real-Time Signal Parallel Processing Software Based On Multi-Core Framework

With the throughput of real-time signal processing becoming faster and higher,parallel processing is imperative.The development of multi-core processors and the continuous optimization of real-time operating systems make real-time signal processing in multi-core,multi-threaded parallel operations a higher performance.This paper mainly studies the parallel program from three aspects to improve the real-time signal processing performance:functional thread performance optimization,memory optimal allocation and thread optimization scheduling.This paper first introduces the task requirements and the specific processing flow of real-time signal processing.Then the paper introduces and analyzes the main function modules of signal processing.After that the paper extracts the typical data processing links,such as matrix operation and Fourier transform,which have large computational complexity.Then the article analyzes the processing performance of these processes,and puts forward the possible optimization scheme.Then this paper optimizes the real-time signal processing of the functional modules and thread the module:The optimization of the scheme and the optimization of the MKL function library are carried out for each data processing section;According to the performance evaluation of the two programs,the paper selects the better performance of the program optimization function modules;On the basis of the module time-consuming,throughput and mutual relations,the paper redistributes,combines and threads the modules.And then this paper gives the optimization of the use of memory between threads:Analyze the various implementations of thread locking;According to the characteristics of multi-threaded shared memory,select the better performance of the lock to solve the thread safety issues;Use preallocation for shared memory to improve the performance of threads;Optimize memory management with loop queue optimizes shared memory preallocation for high quality multi-threaded parallel operation.Finally,the paper gives the optimization of thread scheduling between threads:The use of threading to further enhance the real-time signal processing performance based on the NUMA architecture;According to the task of the system load requirements,use static scheduling based on scheduling domain and dynamic thread migration based on Fixed Threshold Method to balance the system load;The article evaluates and compares performance of the entire multithreaded program over time and throughput.Through the operation of the program before and after the optimization test dealing randomly generated data signals for processing.The test results show that the optimization of real-time signal processing program in this paper satisfies the requirement of high-throughput and high-speed real-time signal processing.Multi-threaded parallel architecture based on multi-core framework can achieve real-time signal processing.