Real-Time Machine Learning–Based Channel Estimation Technique for Intelligent Mobility Services

This paper proposes a model architecture emphasizing real-time capability to support intelligent mobility services. The proposed model performs channel estimation with reduced computational time, thereby maintaining stable communication quality and enabling channel-based positioning. To this end, Least Squares (LS) estimation is applied to the input, while the output is replaced with the Channel Impulse Response (CIR). Since the proposed technique is independent of the type of reference signal, it reduces memory requirements, and maintains estimation performance while lowering the total number of model parameters by reducting output size. The CIR approach allows channel utilization over all subcarriers, which can also be effectively applied to channel-based positioning. The proposed method evaluates performance using the Mean Squared Error (MSE) metric and compares it with conventional channel estimation techniques. In addition to accuracy, computational time of the proposed model and traditional methods is analyzed to demonstrate the advantages of the machine learning–based approach. As a result, the proposed architecture achieves higher estimation accuracy while maintaining low computational time compared to conventional methods, thereby validating the effectiveness of machine learning–based channel estimation techniques for realizing intelligent wireless access networks.