The input is an image. The input image can have any number of dimensions. The input can have the following data type: "byte" (BITPIX=8), "short" (BITPIX=16), "long" (BITPIX=32), "float" (BITPIX=-32), and "double" (BITPIX=-64).

Currently, there MUST be a real part to the data, though the imaginary part can be "none".

The image can be a virtual image as defined by the datamodel, e.g. "my_file.fits[EVENTS][bin x=1:100:1,y=1:100:1]".

A table array must contain at least two columns. The first column, used to determine the Fourier frequencies, must be evenly spaced, and cannot contain any data gaps. In practice, roundoff errors in generating time bins may make this problematic, and the time column may have to be replaced with a uniformly spaced grid. This can be achieved, for example, via dmtcalc using the following command:

dmtcalc infile=my_ltc.fits outfile=my_fixed_ltc.fits expression="INT_TIME((long)#row)"

The new column, INT_TIME, now contains the integer value of the row number from the lightcurve, which is guaranteed to be evenly spaced (although information about data gaps is not retained). One can then create the psd via:

apowerspectrum infilereal="my_fixed_ltc.fits[cols in_time,total_counts]" outfile=my_psd.fits

The output file my_psd.fits contains the frequency column INV_INT_TIME, and the psd is found in the DATA column.

The output data file is a "float" image with the power spectrum computed. The output is scaled according to the scale parameter.

The output format is controlled by the kernel parameter. The output will either be a FITS image (kernel="fits"), an IRAF .imh/.pix file (kernel="iraf"), or will default to whatever format the input is in (kernel="default").

The autoname filename suffix is "_pow". To completely specify the file name one should specify the file and extension name like: myfile.fits[foo] -- for fits or ./myfile.imh -- for iraf.

For the case of a table array input, the output file contains two columns, INV_[COLUMN1] giving the Fourier frequencies, and DATA, giving the psd. The latter is calculated as

psd = 1/N * (SUM_{x=0}^{N-1} f(x) exp[-j 2 \pi ux/N])^2

where f(x) is the discrete function of the column 1 independent variable, x, u is the Fourier frequency, N is the number of data points, and j is the imaginary unit. Thus, comparing to IDL v.5, for example, the psd here is equivalent to the square of the IDL forward transform times N, or the square of the IDL reverse transform divided by N.

The center parameter controls whether the zero frequency point is at pixel=1 or pixel=N/2. Note: both center and crop cannot be set to "yes".

The scale parameter can take the following values.

• linear : linear scaling: output = [absolute value of FFT(a)]**2
• log : log (base 10) scaling: output = log(abs.[FFT(a)]**2)
• db : 10 * log() scaling: output = 10*log(abs.[FFT(a)]**2)

Note: For scale=log and scale=db, IEEE NaN's may be generated in the output files, log(0) = NaN. NB: NaN means "not a number".

The crop parameter controls whether the entire power spectrum is output or just up to the Nyquist frequency (N/2). Note: both center and crop cannot be set to "yes".

The output format is controlled by the kernel parameter. The output will either be a FITS image (kernel="fits"), an IRAF .imh/.pix file (kernel="iraf"), or will default to whatever format the input is in (kernel="default").