Amplitude envelope correlation methods: Old and new (2023) versions

Discussions
1

Hi everyone,

I would like to inquire about the differences between the old and new versions of amplitude envelope correlation (2023) methods. I've reviewed previous discussions regarding the disparities between the old and the preceding new versions (2021; Amplitude envelope correlation: methods). According to the discussion, there is no distinction between these versions when the same trial length and Hilbert transform are applied to the same data. However, after applying both methods to 50 participants using identical parameters, I observed variations in the values between methods for each participant.

While the old AEC exhibited a range from 0 to 0.10, the new AEC (2023) demonstrated a range from 0.16 to 0.25 across participants. I would greatly appreciate it if you could provide any insights into the reasons behind these discrepancies.

Best regards,
Ryoki Sasaki

2

The main changes are:

  • The newer process (HENV) Envelope correlation [2023] allows to compute the amplitude envelope using the (complex-Morlet) wavelet transform approach, not only Hilbert transform as in the previous process (AEC) Amplitude envelop correlation.

  • Symmetry in the connectivity matrix. When orthogonalization is used the amplitude envelope correlation is not symmetric. In the previous version (AEC) the correlations A×B and B×A were averaged with their sign. According to @hossein27en, this can result in strong correlations cancelling each other. For this reason HENV, obtains the absolute value of the correlations before averaging them. See:

  • In a recent update, August 2023, the orthogonalization and correlation were improved in HENV top avoid numerical errors. PR #632
3

Hi Raymundo.Cassani,

Thank you so much for the help.

Best regards,
Ryoki Sasaki