Channel Estimation Methods for Frequency Hopping System Based on Machine Learning

—Frequency Hopping (FH) spread spectrum system is extensively used in military and civilian fields due to its robustness against interference and efficiency in confronting radio jamming. Channel estimation is a crucial part of the FH system. However, signal processing-based channel estimation methods have some constraints, such as high computational complexity, sensitivity to noise level, and excessive overhead. To alleviate these issues, we propose a Machine Learning (ML) model for precisely estimating Narrow Band (NB) multipath fading channel parameters for a Slow Frequency Hopping (SFH) spread spectrum system. In the proposed model, we employed a Neural Network (NN) with three layers consisting of an input layer that interprets the signal's fundamental patterns, a hidden layer to extract the correlation found in the time scene, and an output layer that utilizes a linear activation function to provide the flexibility required to address the dynamic relationship between channel gain and time delay. Without prior experience, leveraging a synthetic dataset rich in complex temporal variations and channel gain nuances, the NN architecture, characterized by multiple dense layers, effectively captures complex temporal relationships. Following rigorous training and validation utilizing the Mean-Square Error (MSE) loss function, the model significantly reduced loss, emphasizing its proficiency for an accurate delay and gain estimation. A computer simulation comparison between the performance of the proposed model and previous classical models was included in this paper. Based on simulation results, the proposed ML-based estimator model significantly outperforms many classical subspace-based methods in terms of MSE, the performance improvement appears over several Signal-to-Noise Ratios (SNR). Furthermore, the proposed model provided a reasonable tradeoff between complexity and performance.