A Study Of Parallel Implementation Of Segmented Mismatched Filter Based On CUDA

With receiving and transmitting equipment deploying in different places, the passive radar system has strong anti-stealth capability and it is quite safe. On the other hand, it does not transmit electromagnetic waves by itself, while making use of the existing ones from the natural environment. Much manufacturing cost has been saved without launching system. In addition, a variety of emission sources are used to achieve fusion location in multi-station, enhancing the ability of detecting the target. Therefore, much attention has been drawn from experts, scholars and researchers. And research work has also been gradually carried out by universities, research institutes of nations.As a non-cooperative system, passive radar can use third-party non-cooperative external radiation source signals, such as the analog TV signal, FM radio signal, and mobile phone communication signal to detect the target. Among them, the coverage of the analog TV signal is very wide and its transmit power is very large, which is often used as a third-party opportunity radiation source. Since these third-party external radiation source signals are non-cooperative, and not specifically designed for the radar modulation, the ambiguity function shape is not clear thumbtack and the sidelobes may be high or even equal to the main lobe, which makes the target unable to be detected correctly. When the sidelobe is higher than the main lobe, false or missing alarms could occur. At this point, a mismatch filter is designed to suppress the height of the sidelobe so that the main lobe is more prominent, and easily detected.However, for the passive radar system, sometimes the echo signal energy of the received signal is low in the actual project. Therefore, it takes a long time to accumulate and the length of the filter is very long, which leads to a particularly large calculation. To overcome the shortcoming of costing computing time, segmented mismatched filtering algorithm with parallel processing is adopted in this paper, and the hardware advantages of many-core and multi-threaded are employed to shorten the processing time.Using the CUDA(Compute Unified Device Architecture) to call a large number of threads, the advantage of massively parallel processing by GPU(Graphics Processor Unit) are obtained, and data-intensive computing operations are processed. On the other hand, CPU (Central Processing Unit) is responsible for logic control, cooperating with GPU, writing parallel code and achieving the parallel segmented mismatched filtering algorithm implemented in software.