Parallel Power Spectral Density Estimate Using Welch’s Method

Welch’s method is a power spectral density estimation method that is widely used in spectral analysis for discrete signals. Magnetoencephalography (MEG) is a neuroimaging technique which records magnetic fields inside human brains as discrete signals. Measurements are done by hundreds of channels of magnetometers mapping to different regions of a human brain. To help find out what is going on in an experiment subject’s brain, it is often necessary to obtain the strength of frequency components (i.e. the spectral density) of the magnetic fields. In current mainstream desktop computers, running MATLAB’s pwelch() function on data recorded by all channels in a 7-minute long experiment could take several minutes, with most computational time spent on fast Fourier transform. This project explores the possibility to parallelize Welch’s method in C language with different FFT libraries and reduce the total time running Welch’s method on hundreds of streams of signals. The two FFT libraries used are FFTW3 and cuFFT. OpenMP is used for parallelization.