The Efficacy of Path Loss Models for Fixed Rural Wireless Links Caleb Phillips 1 , Scott Raynel 2 , Jamie Curtis 2 , Sam Bartels 2 , Douglas Sicker 1 , Dirk Grunwald 1 , and Tony McGregor 2 1 Computer Science Department University of Colorado, Boulder, USA 2 Computer Science Department University of Waikato, New Zealand Abstract. In this paper we make use of a large set of measurements from a production wireless network in rural New Zealand to analyze the performance of 28 path loss prediction models, published over the course of 60 years. We propose five metrics to determine the performance of each model. We show that the state of the art, even for the “simple” case of rural environments, is surprisingly ill-equipped to make accurate predictions. After combining the best elements of the best models and hand-tuning their parameters, we are unable to achieve an accuracy of better than 12 dB root mean squared error (RMSE)—four orders of magnitude away from ground truth. 1 Introduction Modeling the propagation of a wireless transmitter in a complex environment has entertained scientists for at least sixty years. The result is a staggering num- ber of proposals of just about every shape, size, and approach imaginable. The basis for this level of interest is solid—predicting the attenuation of transmit- ted signals with high precision has very important applications in the design, trouble-shooting, and simulation of wireless systems. Despite the large quantity of work done, we recognize an important shortcom- ing: there have been relatively few comparative evaluations of path loss predic- tion models using a sufficiently representative dataset as a basis for evaluation. Those studies that do exist make comparisons between a small number of similar models. And, where there has been substantial work of serious rigor done, for in- stance in the VHF bands where solid work in the 1960’s produced well validated results for analog television (TV) propagation, it is not clear how well these models work for making predictions outside their intended coverage (i.e., fre- quency, distance, environment type, etc.). The result is that wireless researchers are left without proper guidance in picking among dozens of propagation models from which it is not clear which is best or what the penalty is of using a model outside of its intended coverage. This work provides a first step towards solving that problem.