Dynamic intelligent measurement of multiple chirped signals of different types based on the optical computing STFT and the YOLOv3 neural network

We propose a simultaneous measurement system for multiple signals of different types which combines the optical computing short-time Fourier transform (STFT) and You Only Look Once (YOLOv3) neural network. Through the system, the analytical expressions of multiple broadband signals of different types can be obtained in real time with high-frequency resolution. Experimentally, the accuracy of the signal type in the detection results can almost reach 100%. Additionally, the parameter measurement errors for the bandwidth (BW), pulse width (PW), center frequency (CF), and time of arrival (TOA) of each linear frequency-modulated (LFM) or quadratic frequency-modulated (QFM) signal are within ±30 MHz, ±20 ns, ±15 MHz, and ±20 ns, respectively. The frequency resolution can reach 60 MHz. Factors affecting the performance of the measurement system, such as the quantity of the signal and the number of the category, are discussed.