A Broadcast Map Constructing Method Based on the LSTM and Assimilation Theory

Abstract

Frequency modulation (FM) broadcasting is a robust and widely applied technology that offers unparalleled advantages over other broadcasting methods in challenging environments. In order to achieve high accuracy in constructing broadcasting maps for scenarios with uneven and sparse distribution of measurement data, we introduce the concept of FM broadcasting maps and propose a novel methodology for their construction. This paper utilizes the Long Short-Term Memory (LSTM) model to assimilate predictions from the ITU-R models for modeling purposes. To begin, we analyzed critical environmental parameters influencing radio wave propagation. Based on this analysis, we identified the foundational input features for the LSTM model. Subsequently, predictions from the ITU-R P.1546 and 2001 models were assimilated as features and input into the LSTM model for training, resulting in assimilation modeling. Finally, a broadcast map is constructed using the parameter construction method based on the proposed model. The results indicate that the relative error between the measurements and the proposed models, ITU-R P.1546 and ITU-R P.2001, are 3.14%, 6.48%, and 9.89%, respectively. The prediction accuracy of the proposed model surpasses that of the ITU-R models, and stability is significantly improved compared to models solely based on LSTM. The broadcast map in this paper provides an objective reflection of measured field strength values across multiple dimensions, including frequency, distance, various terrains, and error distribution. It demonstrates notable advantages in scenarios characterized by sparse and unevenly distributed sampling points.