How to deal with dpss errors when you don't have the Signal Processing Toolbox?

FieldTrip’s multitaper spectral estimation (ft_freqanalysis with cfg.method = 'mtmfft' or 'mtmconvol' and the default cfg.taper = 'dpss') uses Slepian tapers computed by the dpss function from the MATLAB Signal Processing Toolbox. Without that toolbox, or under GNU Octave, the analysis stops with a message such as

Undefined function 'dpss' for input arguments of type 'double'.

or, under Octave,

error: 'dpss' undefined

You have three options.

Use a taper that does not need the toolbox

Set cfg.taper = 'hanning'. FieldTrip ships its own hanning in external/signal (see here for details), so this works everywhere. It gives a single-taper estimate: there is no spectral smoothing across frequencies, so cfg.tapsmofrq is not used. This is the usual choice for low-frequency spectral estimates, and the tutorials that use multitapers for high-frequency (gamma) activity can be run with 'hanning' at the cost of a noisier high-frequency estimate.

Use sine tapers

Set cfg.taper = 'sine' and specify cfg.tapsmofrq as you would for 'dpss'. Sine tapers (see Riedel and Sidorenko, 1995) are an orthogonal family that approximates the Slepian sequences; FieldTrip implements them itself, so no toolbox is needed. They give the spectral smoothing of the multitaper method with slightly worse sidelobe suppression than true Slepian tapers.

The dpss_hack (not recommended)

The directory external/signal/dpss_hack contains a dpss replacement that interpolates precomputed Slepian tapers. It was made for a workshop in 2013, is only an approximation for settings close to the precomputed ones, and it cannot be used with the 'mtmfft' method at all, because that method also needs the taper concentrations that the hack does not return (see issue 2614). Prefer one of the two options above.

Under GNU Octave, the first two options are the ones that work; see also Can I use Octave instead of MATLAB?