Prediction of Digital Terrestrial Television Coverage Using Machine Learning Regression

Appropriate coverage prediction is a fundamental task for an operator during the dimensioning process and planning of a digital terrestrial television (DTT) system because it allows offering a satisfactory quality of service to end users. Accordingly, several prediction methods based on propagation path loss estimation and traditional statistical models have been proposed. However, the choice of model depends on many factors, such as the presence of obstacles (buildings, trees, and so on) and propagation paths. This fact leads to increasing the error gap between the predicted and real value, which varies from one propagation model to the next. Therefore, novel techniques are required to achieve a high accuracy in the prediction of the signal strength based on few local measurements over the zone of interest. A machine learning regression algorithm is a novel approach that improves the accuracy of DTT coverage prediction regardless of the aforementioned constraints. To this end, we propose an approach based on clustering and machine regression algorithms, such as random forest regression, AdaBoost regression, and
${K}$
-nearest neighbors regression, where we choose the best algorithm for our approach. We use real measurements in terms of electric field strength corresponding to eight DTT channels operating in the city of Quito, Ecuador. Furthermore, we display the coverage results in Google Maps. We perform extensively simulation analysis based on the tenfold cross validation method to evaluate the performance of the machine learning regressor algorithms and compare the results in three error metrics with support vector regression, lasso regression, multilayer perceptron regression, and ordinary kriging technique. Satisfactorily, the results using random forest regression depict a considerable improvement in the accuracy of coverage prediction under a low computational load.