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

TL;DR

This paper introduces a neural network model that predicts both path loss and link-specific uncertainty, improving accuracy and calibration in RF planning and interference analysis.

Contribution

It proposes a heteroscedastic neural network architecture that jointly predicts mean and variance, with the shared-parameter design outperforming alternatives.

Findings

Achieved an RMSE of 7.4 dB on test data.

Attained 95.1% coverage for 95% prediction intervals.

Provided effective link-specific uncertainty estimates for RF 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.