GRIDLoc: A Gradient Blending and Deep Learning-Based Localization Approach Combining RSS and CSI

Received Signal Strength (RSS) and Channel State Information (CSI) are two commonly used fingerprints in fingerprint-based localization systems. Combining RSS and CSI has the potential to enhance the precision of indoor localization systems. Therefore, it is preferable to combine these two fingerprints to build robust localization systems. This letter proposes GRIDLoc, a method for indoor localization based on gradient blending (GB) and deep learning (DL). We extract location-related features with smaller dimensions from the original data using Convolutional Neural Networks (CNNs) and concatenate the features for localization utilizing feature-based fusion. Then, GB is leveraged to avoid the overfitting phenomenon in the fusion network, thereby improving localization accuracy. Experimental results indicate that GRIDLoc achieves an average Localization Error (ALE) of
$1.42{m}$
, representing a reduction of 19.3%, 59.1%, 34.6%, and 53.6%, compared to RSS-only method based on CNN, RSS-only method based on K Nearest Neighbors (KNN), CSI-only method, and Data concatenation method, respectively.