Abstract : Channel estimation is a pivotal component in determining the performance of wireless networks. Recent advancements in deep learning have significantly enhanced communication reliability and reduced computational complexity in 5G and future wireless networks. While least squares (LS) estimation remains widely used due to its simplicity and lack of requirement for prior statistical knowledge about the channel, it often suffers from relatively high estimation errors in digital communication. This paper introduces a novel channel estimation framework that leverages deep learning hybrid models to enhance the accuracy of channel estimates traditionally obtained through the least squares (LS) method. Our approach is validated using a Multiple-Input Multiple-Output (MIMO) system, incorporating a multi-path channel profile, and simulating scenarios in 5G and beyond networks under various mobility conditions characterized by Doppler effects. The system model is designed to accommodate any number of transceiver antennas, and the machine learning component is versatile, allowing the use of various neural network architectures. Numerical results show that the proposed deep learning-based channel estimation structure outperforms traditional methods commonly used in previous studies. Furthermore, our analysis indicates that bidirectional long short-term memory (LSTM) networks achieve the maximum channel estimation accuracy and the lowest bit error rate among the evaluated artificial neural network architectures.