Complexity Reduction in Machine Learning-Based Wireless Positioning: Minimum Description Features

TL;DR

The paper tackles the high data dimensionality of deep learning-based wireless positioning (WP) by introducing a minimum description length (MDL) feature framework that uses the $F$ largest PDP measurements and their temporal locations. A dual-branch CNN (P-NN) processes energy and index information, followed by fully connected layers to output $N_z$ zone classifications, achieving strong performance with significantly reduced feature size. An adaptive feature-size selection criterion combines log-likelihood gains, information acquisition probability, and inter-zone KL divergence to choose $F^ abla$. Numerical results show that P-NN attains PDP-based performance while using less than about 20% of the feature dimensions, offering a practical, scalable solution for mobile WP applications.

Abstract

A recent line of research has been investigating deep learning approaches to wireless positioning (WP). Although these WP algorithms have demonstrated high accuracy and robust performance against diverse channel conditions, they also have a major drawback: they require processing high-dimensional features, which can be prohibitive for mobile applications. In this work, we design a positioning neural network (P-NN) that substantially reduces the complexity of deep learning-based WP through carefully crafted minimum description features. Our feature selection is based on maximum power measurements and their temporal locations to convey information needed to conduct WP. We also develop a novel methodology for adaptively selecting the size of feature space, which optimizes over balancing the expected amount of useful information and classification capability, quantified using information-theoretic measures on the signal bin selection. Numerical results show that P-NN achieves a significant advantage in performance-complexity tradeoff over deep learning baselines that leverage the full power delay profile (PDP).

Figures (8)

• Figure 1: Illustrations of positioning spaces (left) and channel propagation (right).
• Figure 2: An overall diagram on wireless positioning.
• Figure 3: Zone layouts with $N_\mathsf{z}=8$ (left) and $N_\mathsf{z}=32$ (right). Red circles indicate sensor positions.
• Figure 4: Architecture of our positioning neural network (P-NN).
• Figure 5: An illustration of training (left) and testing (right) sets in a 2D plane. For the training set, same color implies the same classification zone. For the testing set, redder color indicates lower classification accuracy.