کاهش سطح گلبرگ‌های فرعی در فشرده‌سازی پالس رادار برای آشکار‌سازی اهداف همسان با استفاده از شبکه عصبی

We use a multilayer back propagation neural network whose training is based on Levenberg-Marquardt optimization method and apply polyphase codes to its input layer for radar pulse compression in environments with similar targets. The advantage of using polyphase codes (in contrast to binary codes) is lower sidelobe levels and much better Doppler tolerance. The use of Levenberg-Marquardt optimization method in our scheme significantly increases the speed of convergence. The results indicated that after 100 iterations, the main sidelobe for P3 code (N=30) is reduced to -238.66 dB and for P4 code (N=45), it is reduced to -240.51 dB. Besides, the proposed method is robust against noise and shows much higher Doppler tolerance.