fast-ica

C implementation of FastICA algorithm: the program generates three signals (a sine wave, a square wave and a sawtooth wave), mixes them and tries to reconstruct the original signals using FastICA.

Build:

mkdir build  && cd build
cmake ..
make

Usage:

./fast_ica [STRATEGY [G_FUNCTION [N_SAMPLES [SAMPLING_WINDOW_SIZE [THRESHOLD [ MAX_ITER [VERBOSE]]]]]]

where:

• strategy can be 0 (Parallel, default) or 1 (Deflation);
• g_function can be 0 (LogCosh, default), 1 (Exp) or 2 (Cube);
• n_samples must be a non-negative integer defining the number of samples in each time signal;
• sampling_window_size must be a non-negative float defining the range in which the samples are taken;
• threshold must be a non-negative float defining the convergence threshold for FastICA;
• threshold must be a non-negative integer defining the maximum number of iteration of FastICA.

The program writes the original signals, the mixtures and the reconstructed signals in three binary files (S.bin, X.bin and S_res.bin, respectively). Such binary files can be read and plotted using the pyutils/plot_signals.py Python scripts, which accepts an argument describing the precision of the data type, either float (default) for 32 bit or double for 64 bit.

Moreover, in the pyutils folder there's a Jupyter notebook and a fast_ica.py modules which implement the same algorithms in Python.