RSSI-Based Indoor Localization Using Deep Learning with A Custom Loss Function

This paper proposes a novel deep-learning approach for RSSI-based wireless indoor localization by introducing a custom loss function designed to address the unique challenges of RSSI data. The proposed loss function combines logarithmic distance and spatial correlation components, enhancing the model’s ability to predict indoor positions with improved accuracy. The experiment was conducted using generated RSSI data from different anchor nodes, processed in MATLAB. The RSSI values were then trained using neural networks with the proposed loss functions. Experimental results demonstrate the model’s superior performance, confirming the effectiveness of the approach in enhancing localization accuracy and robustness. Key contributions include the development of a specialized loss function and the validation of an advanced deep learning model, providing a foundation for future research to refine the model and test it in real-world environments.