An Indoor Localization System Using Residual Learning with Channel State Information.

Area of Science:

• Artificial Intelligence
• Wireless Communication
• Computer Engineering

Background:

• Deep neural networks (NNs) are crucial for accurate indoor localization but suffer from degradation and gradient vanishing.
• Existing NN models face challenges with performance degradation and vanishing gradients in deep architectures.

Purpose of the Study:

• To develop an advanced indoor localization system that overcomes the limitations of traditional deep neural networks.
• To enhance the accuracy and robustness of indoor localization using channel state information (CSI).

Main Methods:

• A novel system combining a denoising neural network and a residual network (ResNet) for indoor localization.
• Implementation of stochastic residual blocks and long-range stochastic shortcut connections (LRSSC) in ResNet to prevent overfitting and mitigate degradation.
• Utilization of dilated convolutions to achieve a large receptive field without information loss.

Main Results:

• The proposed system effectively predicts the location of moving objects using channel state information (CSI).
• Stochastic elements and LRSSC successfully addressed degradation and gradient vanishing issues in the ResNet.
• The system demonstrated superior performance compared to state-of-the-art methods in experimental indoor environments.

Conclusions:

• The novel indoor localization system offers a robust solution for accurate positioning.
• The integration of denoising NNs, stochastic ResNets, and dilated convolutions significantly improves localization accuracy.
• This approach provides a promising advancement for location-based services in indoor environments.