The Usage of ANN for Regression Analysis in Visible Light Positioning Systems.

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Traditional indoor positioning systems often rely on simplified models like line-of-sight, limiting accuracy in complex environments.
• Multipath propagation in visible light positioning (VLP) systems introduces significant challenges for accurate estimation.
• Artificial Neural Networks (ANNs) offer a potential solution for robust positioning in the presence of channel impairments.

Purpose of the Study:

• To investigate the design aspects of an indoor visible light positioning (VLP) system utilizing artificial neural networks (ANNs) for enhanced positioning estimation.
• To evaluate the performance of different ANN algorithms and optimization parameters under multipath channel conditions and varying noise levels.
• To compare the proposed ANN-based VLP approach against traditional received signal strength (RSS) techniques.

Main Methods:

• Employed three distinct ANN algorithms: Levenberg-Marquardt, Bayesian regularization, and scaled conjugate gradient.
• Optimized ANN design by varying the number of hidden layer neurons, training epochs, and training dataset size.
• Introduced signal-to-noise ratio (SNR) and positioning error (εp) as key performance metrics.
• Considered the impact of multipath channels and noise on positioning accuracy.

Main Results:

• The ANN with Bayesian regularization demonstrated superior performance over traditional non-linear least square estimation for RSS across all SNRs.
• Significant improvements in positioning accuracy were observed, particularly in the inner region (up to 55%) and outer region (up to 57%) of the receiving plane.
• Achieved a minimum positioning error (εp) of 2 cm at 30 dB SNR with an optimized ANN design and random data selection.
• Demonstrated low positioning errors even at lower SNRs, with εp values of 2, 11, and 44 cm for SNRs of 30, 20, and 10 dB, respectively.

Conclusions:

• ANNs, particularly with Bayesian regularization, provide a robust and accurate solution for indoor VLP systems operating in multipath environments.
• The proposed ANN-based VLP system significantly outperforms conventional methods, offering improved positioning accuracy.
• Optimizing ANN parameters like network architecture and training data size is crucial for achieving centimeter-level positioning precision.