Parameter estimation of sub-sampling LFM signal based on convolutional neural network

This paper presents a novel parameter estimation method for sub-sampling Linear Frequency Modulated (LFM) signals in low SNR environments using a Convolutional Neural Network (CNN). LFM signals are widely used in various applications like radar and sonar due to their large time-bandwidth product and high range resolution. Traditional parameter estimation methods such as maximum likelihood estimation and time-frequency analysis have limitations like high computational costs and sensitivity to noise. The proposed method leverages the feature learning capability of CNN to effectively extract discriminant features from sub-sampling LFM signals and maps these features to the corresponding parameters. The key contributions include designing a new CNN model with modules like dilated convolutions and multi-scale feature fusion to enhance feature extraction and introducing a data augmentation strategy to improve the model's generalization capability. Experimental results show that the proposed method achieves higher estimation accuracy and better noise resistance compared to traditional methods and existing CNN-based approaches, even in low SNR scenarios. However, the work also has limitations such as the requirement for a large amount of training data and the current model being primarily designed for single-component LFM signals. Future research directions include exploring advanced data augmentation techniques, extending the method to handle multi-component LFM signals, and improving performance in extremely low SNR scenarios.