Research And Implementation Of Intelligent Communication Signal Analyzer

With the development of information technology,the electromagnetic spectrum has gradually developed into a new operational field of modern information warfare,and electromagnetic spectrum equipment is playing an increasingly important role.The current combat technology is becoming more and more advanced,the electromagnetic environment is becoming more and more complex,and the modern electromagnetic battlefield has higher and higher requirements for the intelligence of electromagnetic spectrum equipment.Intelligent combat equipment must not only have the monitoring and interference capabilities of traditional equipment,but also be capable of intelligent monitoring,independent decision-making,and intelligent interference based on the actual electromagnetic environment.Therefore,this paper develops a communication signal intelligent analyzer based on artificial intelligence technology,and studies the key communication signal detection algorithm and embedded lightweight deployment technology of the algorithm.The main research contents of this paper are as follows:According to the demand for intelligent electromagnetic combat equipment,the design,development and platform construction of intelligent analyzers for communication signals are carried out.First,analyze the system requirements and clarify the system design points.Secondly,the hardware platform is designed and built based on the "AD9361+ZYNQ+GPU" heterogeneous architecture scheme.Then,we designed the software algorithm flow and analyzed the key algorithms based on the "sensing-decision-action" loop.Finally,a simple software system is designed based on the system functions.To address the problem that it is difficult for traditional detection algorithms to effectively detect and estimate multiple overlapping signals in the time-frequency domain in non-cooperative scenarios,an intelligent detection algorithm based on time-frequency map and target detection network is proposed.First,the simulated signal time-frequency map dataset is constructed and the signal detection process is derived.On this basis,the YOLOv5-based target detection network is studied,and its attention fusion and anchor frame clustering optimization are improved according to the characteristics of the time-frequency map signals.The simulation results show that the detection probability of the signal is 94.74%,the false alarm probability is 2.86%,and the average error of signal parameter estimation is2.15% in the signal-to-noise ratio range from 0 to 20 d B.A complete model lightweighting and deployment inference acceleration scheme is studied for embedded devices with limited power consumption storage space and arithmetic power,and deep learning models are difficult to deploy and run efficiently on them.The test results show that Slim-YOLOv5 s has 55.7% less parameters,59.6% less volume,and 38.8%faster inference than the benchmark model.The inference speed is increased by 38.8% and the average accuracy is reduced by only 0.4%;based on this,the Tensor RT-FP16 deployment solution increases the inference speed by 140.7% and the average accuracy is reduced by only0.7% compared with the original Pytorch framework.Finally,this paper builds a real test environment and completes the system technical index test.The test results show that the measured system metrics are consistent with the theory and meet the design requirements.