Reciprocity-Aware Convolutional Neural Networks for Map-Based Path Loss Prediction
Ryan G. Dempsey, Jonathan Ethier, Halim Yanikomeroglu
TL;DR
The paper investigates reciprocity in map-based path loss predictions and proposes data augmentation to generalize a CNN-based model from downlink drive-test data to uplink and backhaul scenarios. By applying Identity and Reflection transformations to Ofcom data and evaluating with CRC backhaul data, the authors show that adding a small number of synthetic uplink samples dramatically improves uplink RMSE (from ~16 dB to ~7.8 dB) while preserving or slightly improving downlink performance. The work demonstrates that limited synthetic data can induce reciprocity-aware predictions, enabling a single model to handle multiple link configurations. This approach offers a practical path to more versatile path loss models for real-world wireless planning and interference management.
Abstract
Path loss modeling is a widely used technique for estimating point-to-point losses along a communications link from transmitter (Tx) to receiver (Rx). Accurate path loss predictions can optimize use of the radio frequency spectrum and minimize unwanted interference. Modern path loss modeling often leverages data-driven approaches, using machine learning to train models on drive test measurement datasets. Drive tests primarily represent downlink scenarios, where the Tx is located on a building and the Rx is located on a moving vehicle. Consequently, trained models are frequently reserved for downlink coverage estimation, lacking representation of uplink scenarios. In this paper, we demonstrate that data augmentation can be used to train a path loss model that is generalized to uplink, downlink, and backhaul scenarios, training using only downlink drive test measurements. By adding a small number of synthetic samples representing uplink scenarios to the training set, root mean squared error is reduced by > 8 dB on uplink examples in the test set.