Coverage prediction models for 60 GHz links in urban areas

Accurate RF planning is key to successful deployment of 60 GHz systems, especially, given the uncertainty about higher propagation losses at higher frequencies. Robust and reliable propagation models are essential and not available. Accuracy and robustness are obtained with a massive and statistically significant measurement data set. We present a comprehensive analysis of directional measurements collected from over 1,070 links at 60 GHz, aimed at reliably characterizing 90% coverage in urban street canyon environments using a rooftop base station (BS) under three relevant propagation conditions: same-street coverage with unobstructed antennas, around-the-corner coverage, and coverage with BS hidden from street view by a rooftop parapet. We find that the classical "slope-and-intercept" accurately predicts path gain for same street coverage with the antenna unobstructed by rooftop elements, achieving root mean square errors (RMSE) of 2.7 dB for line-of-sight (LOS) and 4.5 dB for non-line-of-sight (NLOS) links. This contrasts with RMSEs of 3.1 dB for LOS and 13.7 for NLOS of the 3GPP models. For NLOS links around street corners, a single-slope diffraction-based model is shown to predict path-losses with RMSEs of 4.3 dB or less. We present a Knife-Edge Diffraction model to characterize the channel when antennas are obstructed by building parapets, resulting in an RMSE of 3.1 dB. Antenna gain degradation due to multipath for a 10° half-power-beam width antenna was found to be significant for obstructed paths, requiring a 5.8 dB margin to cover 90% of users for around-the-corner links. Our study also shows that, even when a narrow angular spread causes the directive antenna to achieve nearly nominal gain, small-scale spatial fades can exceed 10 dB for 10% of users. This phenomenon is explained by angular spectra composed of multiple similar-power wave fronts that interact constructively and destructively. Finally, we characterize the penalty incurred by selecting a limited number of best-on-average fixed directions instead of the best direction over the full azimuth at each location. Selecting a single direction along the street with a 10° half-power beamwidth antenna results in a penalty of 7 dB if the aim is 90% user coverage in that street. This loss occurs because, even under these conditions, not all users will be on the antenna's boresight. In contrast, losses can exceed 15 dB for 90% coverage in our NLOS scenarios when scanning is limited to the three best average directions.