Deep Neuronal Filter (DNF)

A closed-loop deep neural network (DNN) is used to filter the noise from the Delta and Alpha regions of EEG signals. The EEG signals were obtained from 12 subjects using a novel smart 3D printed electrode consisting of a central and a ring electrodes.

Prerequisites Libraries and packages:

1. Install the Fir1 library from https://github.com/berndporr/fir1 by running:

sudo add-apt-repository ppa:berndporr/dsp

sudo apt-get install fir1

sudo apt-get install fir1-dev

2. Install openCV library by running:

sudo apt install libopencv-dev -y

3. Install boost library by running:

sudo apt-get install libboost-all-dev

4. And make sure you have cmake installed.

How to compile and run the program:

1. Clone this repository:

git clone https://github.com/berndporr/deepNeuronalFilter.git

2. Navigate to the folder:

cd Deep_Neuronal_Filter

3. execute the shell file:

./run_all_stat.sh

This will create the necessary folders, build the program and the DNF library, and will run the executable. An interactive GUI will appear as below.

Conditioning the Inner and Outer signals:

1. To pre-filter the Inner and Outer signals:
   • Use the provided python script pyFiles/generateFirFilters.py to generate the pyFiles/forInner.dat and pyFiles/forOuter.dat files before running the program.
   • In parameters.h file make sure that doOuterPreFilter and doInnerPreFilter are defined.
   • Insert the max length of filter buffers in the parameters.h file as the #define maxFilterLength variable.

Delays (T):

1. Delay line for the Outer signal (small Ts in the diagram above):
   • In parameters.h file make sure that doOuterDelayLine is defined
   • In the same file, use define outerDelayLineLength to specify the length of the delay line.
   • This defines the number of inputs to the DNF.
2. Delay the Inner signal (the large T in the diagram):
   • In parameters.h file make sure that doInnerDelay is defined
   • In the same file, use define innerDelayLineLength to specify the duration of the delay
   • This synchronises the correlation of the signals for optimised learning

Internal parameters of the network:

1. Number of layers and Neurons in DNF:
   • In parameters.h use #define NLAYERS i to set the number of layers to i
   • Make sure the same number of #define Ni x are provided to set the number of neurons x in each layer
   • In main.cpp change the line int nNeurons[NLAYERS]={..., N3, N2, N1}; to add or remove Ni
2. Learning rates:
   • The learning rate for updating the weights and biases are defined in the form of Eta.10^(power). Use the Eta and power sliders to change the two learning rates.

Closing the program:

Use the Esc key on the interactive window to close the program safely and avoid corrupted files.

Final results

The output data will be saved in the cppData folder for all subjects. You can then use the provided python scripts to plot and analyse the final results.