Graph Learning and Augmentation Based Interpolation of Signal Strength for Location-aware Communications

Authors: Hong-Ming Chiu, Carrson C. Fung, Antonio Ortega

Link to Paper

Code regarding GMRF Graph Learning referenced from: https://github.com/STAC-USC/Graph_Learning

Cite

Please cite our paper if you use this code in your own work:

@inproceedings{chiu2020gla,
  title={Graph Learning and Augmentation Based Interpolation of Signal Strength for Location-aware Communicationsr},
  author={Hong-Ming Chiu and Carrson C. Fung and Antonio Ortega},
  booktitle={European Signal Processing Conference (EUSIPCO)},
  year={2020}
}

Table of Contents

• Intorduction
• Directory Tree
• Setup and Usage
• Reference
• Author

Intorduction

This MATLAB program contains the code for the paper "Graph learning and augmentation based interpolation of signal strength for location-aware communications" and code to perform Monta Carlo simulation. The code for the interpolator using Gaussian Process [2] and Kriging [3] are also included for comparison.

Directory Tree

• /:
  • GLA_interpolator.m: main function for Graph Learning and Augmentation based interpolator [1].
  • GP_interpolator.m: main function for Gaussian Process based interpolator [2].
  • LS_interpolator.m: main function for Kriging based interpolator [3].
  • AggregatedMSETest.m: function for Monta Carlo simulation for aggregated MSE [1].
  • RegionalMSETest.m: function for Monta Carlo simulation for Regonal MSE [1].
  • SNRTest.m: function for Monta Carlo simulation for SNR v.s.MSE.
  • startup.m: startup file for setting search path and directory, executed automatically at startup.
  • cleanup.m: clean all .mat file in Data/ directory.
• Function/:
  • MakeGroundTrue.m: function for generating ground true power map [4].
  • MakeObservation.m: function for generating for observed power map.
  • MakeSample.m: function for generating for sample points.
  • DataStatistic.m: function for generating data statistic.
  • GMRF.m: function for GMRF graph learning [5].
  • LearnParameter.m: function for learning paramters for graph augmentation.
• BCD_Algorithm/ [5] :
  • EstimateGGL.m: function for GGL graph learning.
  • EstimateDDGL.m: function for DDGL graph learning.
  • EstimateCGL.m: function for CGL graph learning.
  • nonnegative_qp_solver.m: function for solving inner QP subproblem.
  • update_sherman_morrison_diag.m: funciton for updating diagonal elements in DDGL graph learning.
• MonteCarlo/ :
  • AggregatedMSETest/: directory for storing outputs of AggregatedMSETest.m.
  • RegionalMSETest/: directory for storing outputs of RegionalMSETest.m.
  • SNRTest/: directory for storing outputs of SNRTest.m.
  • PlotAggregatedMSETest.m: function for ploting results of AggregatedMSETest.m.
  • PlotRegionalMSETest.m: function for ploting results of RegionalMSETest.m.
  • PlotSNRTest.m: function for ploting results of SNRTest.m.
• Data/ :
  • GP/: directory for storing results of GP_interpolator.m.
  • GraphLearningData/: directory for storing results of DataStatistic.m, GMRF.m and LearnParameter.m.
  • GroundTrueData/: directory for storing results of MakeGroundTrue.m.
  • Sample/: directory for storing results of MakeSample.m.
• Figure/ :
  • TruePowerMap.fig: true power map generated from MakeGroundTrue.m.
  • ObservedPowerMap.fig: noisy observation of power map plotted in TruePowerMap.fig.
  • InterpolatedPowerMap(GLA).fig: interpolated result of GLA_interpolator.m.
  • InterpolatedPowerMap(GP).fig: interpolated result of GP_interpolator.m.
  • InterpolatedPowerMap(LS).fig: interpolated result of LS_interpolator.m.
  • AggregatedMSE.fig: result of AggregatedMSETest.m.
  • RegionalMSE.fig: result of RegionalMSETest.m.
  • SNR.fig: result of SNRTest.m.

Setup and Usage

Requirements

• MATLAB version R2019a or later
• CVX [6] installed (Optional), you can use the solver in BCD_Algorithm/ to run the code.

Usage

1. Run startup to setup search paths and required directories.

   >>startup
   

2. Run GLA interpolator

   >>GLA_interpolator
   

   Refer to the comment section in GLA_interpolator.m for detailed simulations, parameters and solvers setting.
3. This program stores the outputs and learned parameters in Data/ to speed up implementation, this might take some space in your computer, you can run cleanup to cleanup those data

   >>cleanup
   

Reference

[1] Hong-Ming Chiu, Carrson C. Fung, and Antonio Ortega, "Graph learning and augmentation based interpolation of signal strength for location-aware communications," European Signal Processing Conference (EUSIPCO) 2020.

[2] Ferris Brian, Hähnel Dirk and Fox Dieter, "Gaussian Processes for Signal Strength-Based Location Estimation," 2006.

[3] A. Serrano, B. Girault, and A. Ortega, "Geostatistical Data Interpolation using Graph Signal Spectral Prior," 2019.

[4] R. Di Taranto et al., "Location-aware communicationsfor 5g networks," IEEE Signal Processing Magazine, vol. 31, no. 6, pp. 102?112, 2014.

[5] H.E. Egilmez, E. Pavez, and A. Ortega, "Graph learn-ing from data under laplacian and structural constraints," IEEE Journal of Selected Topics in Signal processing, vol. 11, no. 6, pp. 825?841, 2017.

[6] Michael Grant and Stephen Boyd. CVX: Matlab software for disciplined convex programming, version 2.0 beta. http://cvxr.com/cvx, September 2013.

Author

Name : Hong-Ming Chiu

Email : hmchiu2 [at] illinois.edu

Website : https://hong-ming.github.io

License

MIT License