Deep Learning Based AI Model for Improving Carrier Frequency Offset Estimation in Sensor Driven Communication Systems

Carrier frequency offset (CFO) represents one of the most significant issues when it comes to wireless networks and sensor systems (communication) that occur due to mismatches of oscillators and the Doppler shifts that arise because of user mobility. These variations can disrupt signal synchronization, so communication and security in wireless ad hoc and sensor networks can be seriously affected. Determining the CFO accurately and reliably is the foundation of the safety of the whole system. This work digitally announces a neural network-based method (deep learning) that merges Bidirectional Long Short-Term Memory (BiLSTM) with Convolutional Neural Networks (CNN) in order to estimate CFO from the corrupted Zadoff-Chu (ZC) sequences. The proposed network can make an estimation of the CFO with a very small error rate, even under signal conditions with a low signal-to-noise ratio (SNR). In the experiments, the proposed scheme raises the level of performance beyond the classical techniques. The proposed model's Mean Absolute Error (MAE) at various SNR levels is lower than that of traditional methods. At 10 dB, MAE is 13.02 Hz for the proposed model as compared to 17.44 Hz for CNN, and 21.22 Hz for LSTM. Besides this, the model at low SNR conditions (0-5 dB) dramatically reduces the error margin, hence it is very quiet in a noisy environment. The main performance metrics for this work are Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and regression correlation coefficient (R). All these parameters reflect the superior performance of the proposed method when implemented in real-time for sensor-driven networks.