[FieldTrip] Normalization of LCMV source power

[Schoffelen, J.M. (Jan Mathijs)]

Hi Gabriel,

Given that according to Parseval’s theorem the sum across the full spectrum is equal to the total variance, and since you ask for theoretical justification, division by a measure of total variance ( ~source.avg.pow ) would  be the way forward. Whether or not this is more or less interpretable than a division with an integral of the estimated power in a more constrained bandwidth I don’t know.

Best wishes,

Jan-Mathijs




On 11 Sep 2026, at 11:54, Gabriel Costa via fieldtrip <fieldtrip at science.ru.nl> wrote:

Dear Fieldtrip community,
I came across an issue regarding normalization of resting-state EEG power in source space using beamforming. I was hoping to get your opinion on this. I am wondering if using the avg.pow output from LCMV source analysis is a valid way of getting band-specific source power relative to total power. Using the following LCMV configuration for the source analysis:
     cfg = []; ...
    cfg.method = 'lcmv';
    cfg.lcmv.projectnoise = 'yes';
    cfg.lcmv.keepfilter   = 'yes';
    cfg.lcmv.weightnorm   = 'nai';
    source = ft_sourceanalysis(cfg, data_timelock);

Fieldtrip returns "source" with a common filter and, I assume, an estimate of broadband source power "source.avg.pow". After reconstructing source time series (source_ts) using the common filter, I perform frequency analysis for each source/dipole within a given frequency band, for example:

    cfg.foi        = [8 12];
    cfg.taper      = 'hanning';
    Pow_alpha = ft_freqanalysis(cfg, source_ts);

From here, the two possible ways of normalizing are:
(a)  Get a broadband source power estimate to normalize the alpha power.
    cfg.output     = 'pow';
    cfg.method     = 'mtmfft';
    cfg.foi        = [1 90]; % same as  high and low pass filters of data
    cfg.taper      = 'hanning';
    FA_source    = ft_freqanalysis(cfg, source_ts);
    Pow_alphanorm  =  Pow_alpha.powspctrm / sum(FA_source.powscptrm,2)
(b)  Use the estimate from LCMV ft_sourceanalysis
    Pow_alphanorm  =  Pow_alpha.powspctrm / source.avg.pow;

My question is whether (b) is a valid approach to normalizing the spectrum/FOI power, or would normalization as in (a) be more interpretable?
I noticed that the two approaches produce different results but not dramatically so. I would greatly appreciate help on deciding which is more theoretically justified.
Thanks,

Gabriel Costa
_______________________________________________
fieldtrip mailing list
https://mailman.science.ru.nl/mailman/listinfo/fieldtrip
https://doi.org/10.1371/journal.pcbi.1002202

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.science.ru.nl/pipermail/fieldtrip/attachments/20260923/f2bb75ca/attachment.htm>