Machine Learning-Based Urban Canyon Path Loss Prediction Using 28 GHz Manhattan Measurements

Large bandwidth at mm-wave is crucial for 5G and beyond but the high path loss (PL) requires highly accurate PL prediction network planning optimization. Statistical models with slope-intercept fit fall short in capturing large variations seen urban canyons, whereas ray-tracing, capable of characterizing site-specific features, faces challenges describing foliage street clutter associated reflection/diffraction ray calculation. Machine learning (ML) promising three key prediction: 1) insufficient measurement data; 2) lack extrapolation to new streets; 3) overwhelmingly complex features/models. We propose an ML-based canyon model based on extensive 28 GHz measurements from Manhattan where clutters are modeled via a LiDAR point cloud dataset buildings by mesh-grid building dataset. extract expert knowledge-driven features aggressively compress 3D-building information using convolutional-autoencoder. Using street-by-street training testing procedure improve generalizability, proposed both achieves error (RMSE) $4.8 \pm 1.1$ dB compared $10.6 4.4$ $6.5 2.0$ 3GPP LOS prediction, respectively, standard deviation indicates variation. By only four most influential RMSE $5.5\pm achieved.