Multi-channel convolutional neural network with attention mechanism using dual-band WiFi signals for indoor positioning systems in smart buildings

One of the most crucial Internet of Things (IoT) services for smart buildings is the indoor positioning service, which enables the detection of the exact location of any object within a closed area. Indoor localization, a significant aspect of Internet of Things, often relies on Received Signal Strength Indicator (RSSI) values from WiFi access points due to their ubiquity. However, the indoor localization systems face challenges like RSSI variance, device diversity, and fingerprint similarities. To address these challenges, most existing methods utilize machine learning and deep learning techniques. However, most existing methods introduce additional overhead through pre-processing steps such as filtering or signal transformation. Moreover, they commonly use the same feature space for different frequency bands, which causes ignores the specific statistical correlations of different frequency bands. To mitigate these issues and enhance accuracy without extra hardware, this paper proposes a dual-band approach utilizing 2.4 GHz and 5 GHz WiFi signals by using a multi-channel convolutional neural network regression with attention mechanism (MC-ACNNR). It aims to capture high-level correlations for each band data, and to fuse high-level patterns by combining two features map coming from different channels. The proposed method is tested on signal maps from four buildings across two datasets: UTMInDualSymFi and SODIndoorLoc. The results show that the proposed method achieves higher positioning performance compared to existing methods in the literature. •An end-to-end deep learning method for WiFi-based indoor positioning.•A multi-channel model focusing on a separate frequency band of dual-band data in each channel.•A deep regression architecture that outputs both floor and location coordinates.•The presented approach does not require additional hardware.•Experimental results are provided using four different buildings.