From MFN to SFN: Performance Prediction Through Machine Learning

Abstract

In the last decade, the transition of digital terrestrial television (DTT) systems from multi-frequency networks (MFNs) to single-frequency networks (SFNs) has become a reality. SFN offers multiple advantages concerning MFN, such as more efficient management of the radioelectric spectrum, homogenizing the network parameters, and a potential SFN gain. However, the transition process can be cumbersome for operators due to the multiple measurement campaigns and required finetuning of the final SFN system to ensure the desired quality of service. To avoid time-consuming field measurements and reduce the costs associated with the SFN implementation, this paper aims to predict the performance of an SFN system from the legacy MFN and position data through machine learning (ML) algorithms. It is proposed a ML concatenated structure based on classification and regression to predict SFN electric-field strength, modulation error ratio, and gain. The model's training and test process are performed with a dataset from an SFN/MFN trial in Ghent, Belgium. Multiple algorithms have been tuned and compared to extract the data patterns and select the most accurate algorithms. The best performance to predict the SFN electric-field strength is obtained with a coefficient of determination (R-2) of 0.93, modulation error ratio of 0.98, and SFN gain of 0.89 starting from MFN parameters and position data. The proposed method allows classifying the data points according to positive or negative SFN gain with an accuracy of 0.97.