tfr_array_morlet issues with wavelet

[RainyMeadows]

Steps to reproduce

import numpy as np
from mne.time_frequency import tfr_array_morlet

freqs=np.arange(1, 50, 1)
power = tfr_array_morlet(Input_data, 1000, freqs, n_cycles=7.0, output='avg_power') # Input data is a 70 * 128 * 7000 array

Link to data

I can provide actual data of a single (or couple) channel(s), if necessary.

Expected results

An array of shape (n_chans, n_freqs, n_times)

Actual results

ValueError: At least one of the wavelets is longer than the signal. Use a longer signal or shorter wavelets.

The function works ONLY for n_cycles <= 4.

Additional information

Platform: Windows-10-10.0.22621-SP0
Python: 3.10.11 (tags/v3.10.11:7d4cc5a, Apr 5 2023, 00:38:17) [MSC v.1929 64 bit (AMD64)]
Executable: C:\Users\user\AppData\Local\Programs\Python\Python310\python.exe
CPU: Intel64 Family 6 Model 167 Stepping 1, GenuineIntel: 12 cores
Memory: 15.7 GB

mne: 1.3.1
numpy: 1.24.2 {unknown linalg bindings (threadpoolctl module not found: No module named 'threadpoolctl')}
scipy: 1.10.1

[welcome]

Hello! 👋 Thanks for opening your first issue here! ❤️ We will try to get back to you soon. 🚴

[agramfort]

your minimum frequency is too low. 6 cycles at 1 Hz is longer than your signal given the way the Morlet wavelet is clipped temporally. Message ID: ***@***.***>

…

[RainyMeadows]

Ah, I was assuming that at lowest (1 Hz), 7 cycles equate to 7s of signal!

It appears that the minimum length, as obtained from len([mne.time_frequency.morlet(1000, np.arange(1, 50, 1), n_cycles=7.0)][0][0]) = 11141 Maybe, the error message can be a bit more suggestive in mentioning the maximum wave number feasible given the current choice of wave number, min freq, and time sample?

[agramfort]

look at the code. I don't know on the top of my head how the clipping is done.

…

Message ID: ***@***.***>

[RainyMeadows]

See my (edited) comment; l I did take a look!

[agramfort]

@RainyMeadows would you be up for submitting a PR to update our doc? that would be much appreciated. thanks

[drammock]

@RainyMeadows In the code there's a comment that I think would resolve your confusion:

mne-python/mne/time_frequency/tfr.py

Lines 153 to 155 in 57f5ce3

	# time vector. We go 5 standard deviations out to make sure we're
	# *very* close to zero at the ends. We also make sure that there's a
	# sample at exactly t=0

In other words, the wavelet will always be 10 * sigma - 1 samples long (after converting sigma into a number of samples rather than seconds).

(If sigma isn't specified it defaults to n_cycles / (2.0 * np.pi * f))