This repository contains the code from the work detailed in the paper submitted to IEEE Access

@article{Thrane2020,
    author       = {Jakob Thrane, Darko Zibar, Henrik L. Christiansen},
    title        = {{Model-aided Deep Learning Method for Path Loss Prediction in Mobile Communication Systems at 2.6 GHz}},
    month        = Jan,
    year         = 2020,
    publisher    = {IEEE},
    journal      = {IEEE Access}
}

Previous work is detailed in:

@article{Thrane2018,
    author      = {Thrane, Jakob and Artuso, Matteo and Zibar, Darko and Christiansen, Henrik L},
    journal     = {VTC 2018 Fall},
    publisher   = {IEEE}
    title       = {{Drive test minimization using Deep Learning with Bayesian approximation}},
    year        = {2018}
}

1. Download the dataset from https://ieee-dataport.org/open-access/mobile-communication-system-measurements-and-satellite-images

2. Put the raw data into the raw_data folder

   • such that the data is located in:
   • raw_data\feature_matrix.csv
   • raw_data\output_matrix.csv
   • raw_data\mapbox_api\*.png

3. Generate the test and training set using generate_training_test.py

4. Run the training of the model by train.py, see the script for commandline arguments

This directory formats SigCap data in exactly the same way as formatted in Thrane's paper, i.e.

,Longitude,Latitude,Speed,Distance,Distance_x,Distance_y,PCI,PCI_3,PCI_4,PCI_6,PCI_7,PCI_20,PCI_40,PCI_184,PCI_237

The SigCap data is collected using an Android phone.

The formatting of the last few columns (from the first PCI_i onwards to the right) depends on the PCI values available.

The exe file for geckodriver from https://github.com/mozilla/geckodriver/releases should be in venv/bin. Depending on the environment, geckodriver may need to be in another directory. Refer to https://github.com/thewati and https://medium.com/@watipasomulwafu for more information.

The data/data_{date} directory must be inside the formatted_data_gen directory.

Run getData_Michael.py to produce the csv file with formatted data. Run getImages_Michael.py to produce the satellite images based on the (latitude, longitude) pairs in the data/data_{date} file.

This directory produces height maps for data collected in DTU (Technical University of Denmark). The DSM_6188_720_2x2 directory mustw be present because it contains tiff files detailing the height information of DTU. grid_data stores the height maps and the feature_matrix which are generated using (latitude, longitude) values that are arranged in a grid.

grid_pattern_feature_gen.py generates the features in exactly the same format as in Thrane's paper based on gridded locations.

grid_pattern_height_map_gen.py generates the height maps in exactly the same format as Thrane's satellite images based on gridded locations.

heatmap_prototyping tests code for plotting a heat map of signal strength.

height_map_gen_orig_raw_data.py generates the height maps in exaclty the same format as Thrane's satellite images based on Thrane's (latitude, longitude) measurements provided in IEEE's data portal.

The directory raw_data must be present in the parent directory of DTU-processing-maps.