5.7.5 Filter

The purpose of this step is to apply a Fourier low-pass filter to the (detrended) data and add back the trend that was removed previously by the Detrend Module (Sec. 5.7.4).

 

The control panel for this processor module is show below. Its main component is a depiction of the Fourier power spectrum of the input data. This plot can be zoomed in by dragging the cursor on top of it, and reset by clicking on the Zoom reset button. The WCS view port can be further customized by clicking the mouse on any one of its corners (see Sec. 2.5).

 

Two choices are available for the filter form:

 

 

The quadratic log power function is the optimal filter for a pure Gaussian signal. Thus the Brault & White filter should work at its best on spectrograms comprised of pure Gaussian components of the same width, plus white noise. Spectrograms that deviate from this ideal model will show poorer filter performance.

 

The cutoff frequency can be input by the user manually; just uncheck the Auto cutoff check box and type the desired value in the Cutoff frequency text field. If left in auto mode, the cutoff frequency will be defined as the first point in the power spectrum where the power falls below the noise level. Again, this works well for the idealized spectrograms, but often results in too low values for the cutoff frequency when applied to more complex data.

 
 

This module acts on a single spectrogram only; thus the Coadd Module (Sec. 5.7.3) must be run before it. It is also advisable to run the Detrend Module (Sec. 5.7.4) as well, to minimize zero-frequency effects. If the Detrend Module was run before the filter, the filter module will take care of putting the trend back on the filtered data.

Bug: due to synchronization problems within the graphics thread, sometimes the blue lines that depict the quadratic component and the linear noise are plotted in black, not blue, making them harder to spot on the plot.