A Study of the Influence of Data Normalization on 3D REM Reconstruction Methods

The need for fast and accurate spectrum utilization characterization in three dimensional (3D) space, has been established as a significant topic in modern wireless communications. A critical consideration in this regard, is the reconstruction of radio environment maps (REMs) from a set of measurements, which is ordinarily much smaller than the number of samples necessary to achieve high fidelity. Consequently, 3D interpolation methods are widely utilized. This study compares the performance of established statistical and deep learning (DL) approaches for 3D REM reconstruction for three types of simulated and real-world measured spectrum data. More specifically, the effect of the received signal strength (RSS) range between non-normalized and normalized values, is assessed. Root mean square error (RMSE) of as low as -8 dB is achieved for sampling ratio of 10 %. The results establish the normalization of the RSS data improves the accuracy for applications that do not require prediction of the exact RSS values such as identifying areas with under-utilized spectrum.

• Antoni Ivanov;
• Vladimir Poulkov;
• Krasimir Tonchev;
• Atanas Vlahov;
• Yue Xiao;
• Ping Yang

2026 Joint International Conference on Digital Arts, Media and Technology with ECTI Northern Section Conference on Electrical, Electronics, Computer and Telecommunication Engineering (ECTI DAMT & NCON)

https://ieeexplore.ieee.org/abstract/document/11511867