A CNN-LSTM Quantifier for Single Access Point CSI Indoor Localization

TL;DR

This work tackles single-AP WiFi indoor localization by turning CSI fingerprints into a quantifiable position estimate rather than a fixed-class classification. It introduces a CNN-LSTM framework that first extracts spatial CSI features from amplitude images and then models temporal correlations across steps to produce trajectory-aware location estimates, with a two-stage training regime and a median-filtered, min-max normalized CSI input. Large-scale autonomous data collection across Nexus 5 and Intel 5300 NIC devices demonstrates robust performance in realistic environments, achieving an average error of $2.5\,\mathrm{m}$ and $80\%$ of errors below $4\,\mathrm{m}$, outperforming competing methods by roughly 50%. The findings underscore the value of incorporating temporal context for single-AP fingerprinting and establish a strong baseline dataset for future single-AP localization research. Practical impact includes enabling accurate indoor localization in spaces with limited infrastructure, while providing a framework adaptable to additional APs.

Abstract

This paper proposes a combined network structure between convolutional neural network (CNN) and long-short term memory (LSTM) quantifier for WiFi fingerprinting indoor localization. In contrast to conventional methods that utilize only spatial data with classification models, our CNN-LSTM network extracts both space and time features of the received channel state information (CSI) from a single router. Furthermore, the proposed network builds a quantification model rather than a limited classification model as in most of the literature work, which enables the estimation of testing points that are not identical to the reference points. We analyze the instability of CSI and demonstrate a mitigation solution using a comprehensive filter and normalization scheme. The localization accuracy is investigated through extensive on-site experiments with several mobile devices including mobile phone (Nexus 5) and laptop (Intel 5300 NIC) on hundreds of testing locations. Using only a single WiFi router, our structure achieves an average localization error of 2.5~m with $\mathrm{80\%}$ of the errors under 4~m, which outperforms the other reported algorithms by approximately $\mathrm{50\%}$ under the same test environment.

Figures (10)

• Figure 1: (a) Localization process of the proposed CNN-LSTM system. (b) Proposed CNN-LSTM model. (c) CSI images before and after applying filter and normalization.
• Figure 2: (a) Diagram of the data collection points. The solid blue arrows indicate the walking direction of the testing route. (b) Heat map of AP RSSI signal collected from Intel 5300 NIC and Nexus 5 phone. (c) CSI waterfall plot from Intel 5300 PCIe card at location (0,0) in 2 different time. (d) CSI waterfall plot from Nexus 5 phone at location (0,0) in 2 different time. (e) Correlation coefficient of the collected CSI images at location (0,0) along 7 hours.
• Figure 3: CNN Layer Learning Curve (a) Intel 5300 NIC. (b) Nexus 5 Phone.
• Figure 4: Correlation coefficient of original CSI images before CNN and output spatial features after CNN with Intel 5300 NIC dataset.
• Figure 5: LSTM Layer Learning Curve (a) Intel 5300 NIC. (b) Nexus 5 Phone.