Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Radio Map Estimation -- An Open Dataset with Directive Transmitter Antennas and Initial Experiments

Fabian Jaensch, Giuseppe Caire, Begüm Demir

TL;DR

A publicly available dataset of simulated path loss radio maps together with realistic city maps from real-world locations and aerial images from open datasources is released and the code is made available.

Abstract

Over the last years, several works have explored the application of deep learning algorithms to determine the large-scale signal fading (also referred to as ``path loss'') between transmitter and receiver pairs in urban communication networks. The central idea is to replace costly measurement campaigns, inaccurate statistical models or computationally expensive ray-tracing simulations by machine learning models which, once trained, produce accurate predictions almost instantly. Although the topic has attracted attention from many researchers, there are few open benchmark datasets and codebases that would allow everyone to test and compare the developed methods and algorithms. We take a step towards filling this gap by releasing a publicly available dataset of simulated path loss radio maps together with realistic city maps from real-world locations and aerial images from open datasources. Initial experiments regarding model architectures, input feature design and estimation of radio maps from aerial images are presented and the code is made available.

Radio Map Estimation -- An Open Dataset with Directive Transmitter Antennas and Initial Experiments

TL;DR

A publicly available dataset of simulated path loss radio maps together with realistic city maps from real-world locations and aerial images from open datasources is released and the code is made available.

Abstract

Over the last years, several works have explored the application of deep learning algorithms to determine the large-scale signal fading (also referred to as ``path loss'') between transmitter and receiver pairs in urban communication networks. The central idea is to replace costly measurement campaigns, inaccurate statistical models or computationally expensive ray-tracing simulations by machine learning models which, once trained, produce accurate predictions almost instantly. Although the topic has attracted attention from many researchers, there are few open benchmark datasets and codebases that would allow everyone to test and compare the developed methods and algorithms. We take a step towards filling this gap by releasing a publicly available dataset of simulated path loss radio maps together with realistic city maps from real-world locations and aerial images from open datasources. Initial experiments regarding model architectures, input feature design and estimation of radio maps from aerial images are presented and the code is made available.
Paper Structure (16 sections, 3 equations, 9 figures, 7 tables)

This paper contains 16 sections, 3 equations, 9 figures, 7 tables.

Table of Contents

  1. INTRODUCTION
  2. PROBLEM FORMULATION
  3. GENERAL APPROACH AND RELATED WORKS
  4. OUR CONTRIBUTION
  5. SIMULATION SETUP AND DATASET
  6. CITY MAPS
  7. RADIO MAP SIMULATIONS
  8. OTHER DATA
  9. EXPERIMENT DESIGN
  10. CNN-ARCHITECTURES
  11. INPUT FEATURES
  12. ENVIRONMENT ENCODING AND TX LOCATION
  13. TX ANTENNA PATTERN AND ORIENTATION
  14. LINE-OF-SIGHT
  15. NUMERICAL EXPERIMENTS
  16. ...and 1 more sections

Figures (9)

  • Figure 1: Example of sample from the dataset -- \ref{['fig:maps:img']} aerial image \ref{['fig:maps:gis']} nDSMs of buildings (red) and vegetation (green) overlayed with darker colors corresponding to larger height values \ref{['fig:maps:3d']} 3D model in simulation with Tx (green cube) and different orientations of the antenna main lobe (red lines) \ref{['fig:maps:rm']} example of a radio map
  • Figure 2: 3D plot of antenna radiation patterns. The values in bracket indicate the half power beam width and first null beam width, respectively.
  • Figure 3: Illustration (inspired by dcn) of sampling points in standard, dilated and deformable convolution with kernel size $3\times3$.
  • Figure 4: Example radio map \ref{['fig:input_features:target']} and (normalized) input features. Baseline Tx and city geometry encoding: \ref{['fig:input_features:tx_one_hot']} Tx location, \ref{['fig:input_features:build_ndsm']} building nDSM, \ref{['fig:input_features:veg_ndsm']} vegetation nDSM. Cylindrical coordinates: \ref{['fig:input_features:dist2d']} 2D distance to Tx, \ref{['fig:input_features:azimuth']} azimuth angle, \ref{['fig:input_features:build_rel']} building height relative to Tx, \ref{['fig:input_features:veg_rel']} vegetation height relative to Tx, \ref{['fig:input_features:floor_rel']} ground height relative to Tx. Tx antenna pattern: \ref{['fig:input_features:gain_floor']} pattern projected onto the ground, \ref{['fig:input_features:gain_top']} pattern projected onto building top. Line-of-sight (LoS) information: \ref{['fig:input_features:los_floor']} binary LoS ground, \ref{['fig:input_features:los_top']} binary LoS building top, \ref{['fig:input_features:los_min']} ours (for relative heights from cylindrical coordinates).
  • Figure 5: Illustration of our LoS encoding and elevation angles -- vertical cut through a simple city environment with buildings in red, Tx in green with antenna pattern and main direction in orange, location of interest in yellow, height of the building at this location (z) and smallest height value for which LoS is not obstructed (min) in blue, elevation angles in purple (corresponding to the building in the location of interest and the left corner of the building in the middle).
  • ...and 4 more figures