Stacked Sparse Autoencoder-Based WLAN Fingerprinting Localization

Abstract

Currently, deep learning algorithms have been used in a variety of fields because their performance usually outperforms traditional machine learning algorithms. In this study, we present a stacked sparse autoencoder (SSAE)-based wireless local area network (WLAN) fingerprinting localization algorithm. The hidden layer dimension of a basic autoencoder (AE) is limited by its input dimension, which could affect its performance. By contrast, the hidden layer dimension of a sparse autoencoder can be set to be greater than its input dimension, so we train two sparse autoencoders in turn for unsupervised received signal strength indicator (RSSI) feature learning. Then we stack the encoders and also add a linear regression layer as an output layer to the SSAE structure. We obtain the SSAE fingerprinting localization algorithm after supervised fine-tuning. We evaluate our presented SSAE localization algorithm in an actual indoor office scenario. The localization results show that our presented SSAE fingerprinting algorithm achieves a better performance compared with some other popular fingerprinting localization algorithms.KeywordsWLANLocalizationDeep learningStacked sparse autoencoder