andreajohanacv/DeepREM

DeepREM: Deep-Learning-Based Radio Environment Map Estimation from Sparse Measurements

This app lets you to estimate complete REMs in urban scenarios based on two DL architectures: U-Net and Conditional GAN of two parameters, RSRP and BS coverage.

RSRP: Received power (equivalent to the reference signal received power (RSRP) in 4G cellular networks)

BS coverage: Base station (BS) coverage, defined as an integer that indicates which BS is serving each location in the map

The DL-based models were trained using keras from tensorflow with a dataset of 1800 urban maps (link: https://doi.org/10.5281/zenodo.7091115 ). To obtain a model able to generalize the propagation estimations we use areas of different cities (200 maps per area) with different topographical elevations and buildings.

The estimation of REMs is performed from known sparse measurements in an array of 256 x 256 with a resolution of 10 m per pixel. The app includes some sample REMs among the options as well as allows you to use completely new maps with the option (upload a map). Depending on the coverage index, the app lets select the following inputs to the two models

BS coverage : a map with 5% of known coverage samples (The values must be integers between {1,2,3,4} to determine the BS connection, 0 in case the sample is not known.)

RSRP: a map with some known received power indexes (The values must be continuous in the range [-200, 0] dBm determining the power strength at that position, -200 in case the measurement is not known). For this specific case, the input maps can be as follows:

• —5% of known samples
• —2.5% of known samples
• —unknown region

Upload a map: a new map selected by the user

Requirements

keras==2.9.0 matplotlib==3.2.2 numpy==1.24.2 pandas==1.5.3 scikit_learn==1.2.1 seaborn==0.12.2 streamlit==1.17.0 tensorflow==2.9.2

How to cite this material

10.5281/zenodo.7839404