FT_FREQDESCRIPTIVES computes descriptive univariate statistics of
 the frequency or time-frequency decomposition of the EEG/MEG signal,
 thus the powerspectrum and its standard error.

 Use as
   [freq] = ft_freqdescriptives(cfg, freq)
   [freq] = ft_freqdescriptives(cfg, freqmvar)

 The data in freq should be organised in a structure as obtained from
 from the FT_FREQANALYSIS or FT_MVARANALYSIS function. The output structure is comparable
 to the input structure and can be used in most functions that require
 a freq input.

 The configuration options are
   cfg.variance      = 'yes' or 'no', estimate standard error in the standard way (default = 'no')
   cfg.jackknife     = 'yes' or 'no', estimate standard error by means of the jack-knife (default = 'no')
   cfg.keeptrials    = 'yes' or 'no', estimate single trial power (useful for fourier data) (default = 'no')
   cfg.channel       = Nx1 cell-array with selection of channels (default = 'all'),
                       see FT_CHANNELSELECTION for details
   cfg.trials        = 'all' or a selection given as a 1xN vector (default = 'all')
   cfg.frequency     = [fmin fmax] or 'all', to specify a subset of frequencies (default = 'all')
   cfg.latency       = [tmin tmax] or 'all', to specify a subset of latencies (default = 'all')

 A variance estimate can only be computed if results from trials and/or
 tapers have been kept.

 Descriptive statistics of bivariate metrics is not computed by this function anymore. To this end you

 To facilitate data-handling and distributed computing you can use
   cfg.inputfile   =  ...
   cfg.outputfile  =  ...
 If you specify one of these (or both) the input data will be read from a *.mat
 file on disk and/or the output data will be written to a *.mat file. These mat
 files should contain only a single variable, corresponding with the
 input/output structure.