Heteroscedastic Neural Networks for Path Loss Prediction with Link-Specific Uncertainty

2025-11-28

AI Summary

Traditional path loss models assume homoscedastic prediction uncertainty, failing to capture link-specific variability. To address this limitation, we propose a heteroscedastic neural network that jointly estimates the mean path loss and link-level predictive variance—using Gaussian negative log-likelihood as the loss function. The model supports critical radio planning tasks, including coverage margin optimization, RF network planning, and self-diagnostic model calibration. Employing a shared-parameter architecture, it is trained and validated on large-scale public drive-test data. Blind testing yields an RMSE of 7.4 dB, a 95% prediction interval coverage rate of 95.1% (near the ideal), and an average interval width of 29.6 dB—demonstrating excellent calibration and sharpness. This approach significantly enhances both predictive reliability and engineering applicability.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Planning, Routing, and Scheduling: Planning with Language Models
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Social Networks and Social Media: Influence propagation, information diffusion, and the prediction on networks
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Traditional and modern machine learning-based path loss models typically assume a constant prediction variance. We propose a neural network that jointly predicts the mean and link-specific variance by minimizing a Gaussian negative log-likelihood, enabling heteroscedastic uncertainty estimates. We compare shared, partially shared, and independent-parameter architectures using accuracy, calibration, and sharpness metrics on blind test sets from large public RF drive-test datasets. The shared-parameter architecture performs best, achieving an RMSE of 7.4 dB, 95.1 percent coverage for 95 percent prediction intervals, and a mean interval width of 29.6 dB. These uncertainty estimates further support link-specific coverage margins, improve RF planning and interference analyses, and provide effective self-diagnostics of model weaknesses.

Problem

• Predicting path loss with link-specific uncertainty estimates
• Overcoming constant variance assumption in machine learning models
• Evaluating neural network architectures for improved RF planning

Innovation

• Neural network predicts mean and link-specific variance
• Minimizes Gaussian negative log-likelihood for uncertainty estimates
• Shared-parameter architecture achieves best calibration and accuracy