Using the Power Delay Profile to accelerate the training of neural network-based classifiers for the identification of LOS and NLOS UWB propagation conditions

This repository contains the code needed to replicate the experiments described in the article : "Using the Power Delay Profile to accelerate the training of neural network-based classifiers for the identification of LOS and NLOS UWB propagation conditions". The code is divided into two parts, one in Matlab to pre-process the samples and extract the PDP, and another part in Python+Tensorflow to train and test the LOS-NLOS classifier.

Pre-processing in Matlab

To generate measurement sets be used later in Tensorflow, it is necessary to pre-process the original data. This requires the following steps:

• Clone the repository https://github.com/ewine-project/UWB-LOS-NLOS-Data-Set inside the "./Matlab/Measurements/ directory.

• Run the file "./Matlab/parseNLOSClassificationData.m". This will generate a file called "./Measurements/External/RangingWithCIRData3_v5.mat".

• Run the file "./Matlab/extractFeaturesFromCir.m". This will generate four new files with the PDP samples and the rest of the features.

• Run the file "./Matlab/Export_external_dataset_to_csv_train_test_random.m". This will generate random sets of training and testing. The number of generated sets is configured within the script in the variable "numReps".

• Finally, the generated ".csv" files must be copied into the "./ExternalDatasetWithPDP_v5" folder

Classification

To run the simulation, the "main.py" file must be executed. It is necessary to have previously installed the next python libraries:

• tensorflow
• tensorflow_addons
• pydotplus
• pandas
• pydot
• matplotlib

They can be installed used pip.

After executing the "main.py" script, the results will be stored in "./Results_v5". Different figures can be obtained using the scripts "./plot_results_multi_pdp.py" and "plot_results.py".

References

This work uses the UWB measurements datataset related with the paper:

Klemen Bregar, Andrej Hrovat, Mihael Mohorčič, "NLOS Channel Detection with Multilayer Perceptron in Low-Rate Personal Area Networks for Indoor Localization Accuracy Improvement". Proceedings of the 8th Jožef Stefan International Postgraduate School Students’ Conference, Ljubljana, Slovenia, May 31-June 1, 2016.