FT_FREQGRANDAVERAGE computes the average powerspectrum or time-frequency spectrum
 over multiple subjects

 Use as
   [grandavg] = ft_freqgrandaverage(cfg, freq1, freq2, freq3...)

 The input data freq1..N are obtained from either FT_FREQANALYSIS with
 keeptrials=no or from FT_FREQDESCRIPTIVES. The configuration structure
 can contain
   cfg.keepindividual = 'yes' or 'no' (default = 'no')
   cfg.foilim         = [fmin fmax] or 'all', to specify a subset of frequencies (default = 'all')
   cfg.toilim         = [tmin tmax] or 'all', to specify a subset of latencies (default = 'all')
   cfg.channel        = Nx1 cell-array with selection of channels (default = 'all'),
                        see FT_CHANNELSELECTION for details
   cfg.parameter      = string or cell-array of strings indicating which
                        parameter(s) to average. default is set to
                        'powspctrm', if it is present in the data.

 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. For this particular function, the input should be
 specified as a cell-array.