phase-locked versus time locked

[José Luis Blues]

Hello Mike,

I just started to read your book, and I must to say is very illuminating,
I have a very basic question, you will forgive but I am just a little bit confused by the difference between phase-locked and time-locked activity,
I don't see this distinction for instance in Tallon-Baudry & Bertrand 1999 (I have the feeling that there induced activity is non-phase-locked and non-time-locked), I wonder if you could reference me to a couple of papers where I can see this difference,

thanks!
José Luis

[Mike X Cohen]

Dear Jose Luis,

Your confusion is understandable, and reflects that the literature hasn't settled on consistent terminology. In the literature, the term 'induced' has used to refer either to non-phase-locked or to total time-frequency power (including both phase-locked and non-phase-locked), the latter of which is indicated in the Tallon-Baudry 99 paper. I prefer the term non-phase-locked because it is more a description of the analysis than an interpretation of the result. 

An EEG response is time-locked simply if it manifests the same pattern at roughly the same time on each trial after the stimulus onset (or whatever is the time=0 event). It doesn't matter whether the EEG pattern is in power, phase, cross-frequency-coupling, or anything else. In contrast, non-time-locked activity means that the change in EEG signal occurs after a variable time on different trials. Whether you can measure non-time-locked activity with time-frequency-based analyses depends on how non-time-locked it is (that is, how variable the response is over trials). 

Phase-locked is a more stringent feature. A signal is phase-locked only if it takes the same phase angle on each trial. An EEG response has to be strongly time-locked in order to be phase-locked. Non-phase-locked, on the other hand, simply means that the time=0 event doesn't affect the phase characteristics of the signal. Imagine that the stimulus increases the amplitude of an ongoing theta oscillation. Because the theta phase will be random on different trials, this response will be non-phase-locked, and will tend towards zero in time-domain averaging (the ERP). The ERP reflects the phase-locked and time-locked part of the EEG signal. See figure 5.2 (page 56) and figure 20.1 (the color version Plate 11 in the middle of the book). 

Separating the phase-locked from the non-phase-locked signal is another area where the literature hasn't settled on a consistent solution. Nonetheless, the approach I used in the book is the most common one. The total signal is simply the time-frequency decomposition of each trial and then the average over trials. To compute the non-phase-locked signal, you first compute the ERP and then subtract the ERP from each trial, and then perform time-frequency decomposition of the residual (the assumption is that the ERP is the phase-locked signal, and thus by subtracting the phase-locked from the total, you are left with the non-phase-locked part). After computing the total and the non-phase-locked time-frequency power, subtract the non-phase-locked from the total. This gives you the phase-locked power. You might think that you should just take the time-frequency decomposition of the ERP, but this generally produces results that are very difficult to interpret. 

Well, that may be a longer answer than you needed, but I hope it's clearer now. We recently applied the above reasoning to study the difference between the ERP and non-phase-locked theta power during cognitive control: http://www.ncbi.nlm.nih.gov/pubmed/24068756.

Mike

--
You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.
To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimes...@googlegroups.com.
Visit this group at http://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.

--

Mike X Cohen, PhD
mikexcohen.com

[José Luis Blues]

Dear Mike,

I do really appreciate the throughout response, it concern other questions I have as well,
For the sake of simplicity, if I may, I would like to re-ask you in relation to the responses you give me,

Your confusion is understandable, and reflects that the literature hasn't settled on consistent terminology. In the literature, the term 'induced' has used to refer either to non-phase-locked or to total time-frequency power (including both phase-locked and non-phase-locked), the latter of which is indicated in the Tallon-Baudry 99 paper. I prefer the term non-phase-locked because it is more a description of the analysis than an interpretation of the result.

To my understanding, the term 'induced' is opposed to 'evoked', so I'm not sure that we could find studies where the term 'induced' is applied to total TF power (where there is a phase-locked component there), do we? I guess the term 'induced' may be time-locked or not, but above all is not phase-locked, I'm wrong?, please correct me,
 

An EEG response is time-locked simply if it manifests the same pattern at roughly the same time on each trial after the stimulus onset (or whatever is the time=0 event). It doesn't matter whether the EEG pattern is in power, phase, cross-frequency-coupling, or anything else. In contrast, non-time-locked activity means that the change in EEG signal occurs after a variable time on different trials. Whether you can measure non-time-locked activity with time-frequency-based analyses depends on how non-time-locked it is (that is, how variable the response is over trials).

As a practical matter, I guess when you analyse your data it does not really matter if the activity is time-locked or non time-locked, right? I mean ERP does not guide TF analyses,  so when look at your TF charts, if the activity is too much spread it can difficult to interpret,
 

Phase-locked is a more stringent feature. A signal is phase-locked only if it takes the same phase angle on each trial. An EEG response has to be strongly time-locked in order to be phase-locked. Non-phase-locked, on the other hand, simply means that the time=0 event doesn't affect the phase characteristics of the signal. Imagine that the stimulus increases the amplitude of an ongoing theta oscillation. Because the theta phase will be random on different trials, this response will be non-phase-locked, and will tend towards zero in time-domain averaging (the ERP). The ERP reflects the phase-locked and time-locked part of the EEG signal. See figure 5.2 (page 56) and figure 20.1 (the color version Plate 11 in the middle of the book). 

Plate 11 is great, is very clear,
My question now is in relation to the way researchers do TF analyses,

In "Gamma oscillations distinguish mere exposure from other likability effects" (2014) (http://www.ncbi.nlm.nih.gov/pubmed/24389505)

The authors do the following,
"Preprocessing for the time–frequency analysis was similar to that of ERP, except that a 0.5–101 Hz band-pass filter was used and trials marked as artifacts during ERP processing were not included.To analyze gamma oscillation, time–frequency representation(TFR) was calculated using the open-source software FieldTrip. TFR was obtained from EEG activity -200–800 ms relative to stimulus onset and from frequency ranges of 8–100Hz, in 2-Hz frequency steps by means of a 7-cycle complex Morlet wavelet, and was baseline corrected relative to activity between -200 and 0 ms(activeperiod/baseline)(e.g. Luft,Nolte,&Bhattacharya,2013)".

This is what you call the direct TF descomposition from ERP signal, and thus the "D) ERP power" in plate 11?

Now, in "EEG oscillatory patterns are associated with error prediction during music performance and are altered in musician's dystonia" (2011) (http://www.ncbi.nlm.nih.gov/pubmed/21195188)

The authors do:
"For that purpose, we computed the wavelet-based time–frequency representations (TFR) of the EEG signals corresponding to the brain responses triggered by actions leading to performance errors and to correct notes. A complex Morlet wavelet was used to extract time–frequency complex phases (...) and amplitudes (...) of the EEG signal x(t). (...). To study changes in the spectral power, we used the wavelet energy, which was computed as the average across epochs of the squared norm of the complex wavelet transform (...)".
This is "A) total power" in plate 11, right?

This is all the most confusing,
I can only think that distinct studies reporting let's say total power, ERP power... cannot uncover the same phenomenon, but anyway they all talk of I don't know gamma activity and draw conclusions of that basis,

I appreciate the last paragraph, that is what I have read as to most standard method, and I'll take a look at the paper for sure,

José Luis

[Mike X Cohen]

Dear Jose Luis, 

See below...

On Thu, Apr 3, 2014 at 7:22 PM, José Luis Blues <joselu...@gmail.com> wrote:

Dear Mike,

I do really appreciate the throughout response, it concern other questions I have as well,
For the sake of simplicity, if I may, I would like to re-ask you in relation to the responses you give me,

Your confusion is understandable, and reflects that the literature hasn't settled on consistent terminology. In the literature, the term 'induced' has used to refer either to non-phase-locked or to total time-frequency power (including both phase-locked and non-phase-locked), the latter of which is indicated in the Tallon-Baudry 99 paper. I prefer the term non-phase-locked because it is more a description of the analysis than an interpretation of the result.

To my understanding, the term 'induced' is opposed to 'evoked', so I'm not sure that we could find studies where the term 'induced' is applied to total TF power (where there is a phase-locked component there), do we? I guess the term 'induced' may be time-locked or not, but above all is not phase-locked, I'm wrong?, please correct me,
 

Yes, induced is opposed to evoked. It is largely used to refer to non-phase-locked activity, but you will occasionally see papers where they use the term induced to refer to the total signal (that is, phase-locked plus non-phase-locked). Unfortunately, the terminology is not always consistent, so it's important to read the Methods section of a paper carefully, if this distinction is important. 

 

An EEG response is time-locked simply if it manifests the same pattern at roughly the same time on each trial after the stimulus onset (or whatever is the time=0 event). It doesn't matter whether the EEG pattern is in power, phase, cross-frequency-coupling, or anything else. In contrast, non-time-locked activity means that the change in EEG signal occurs after a variable time on different trials. Whether you can measure non-time-locked activity with time-frequency-based analyses depends on how non-time-locked it is (that is, how variable the response is over trials).

As a practical matter, I guess when you analyse your data it does not really matter if the activity is time-locked or non time-locked, right? I mean ERP does not guide TF analyses,  so when look at your TF charts, if the activity is too much spread it can difficult to interpret,
 

That depends on how non-time-locked the activity is. Imagine that some oscillatory response has a duration of 300 ms and onsets variously between 400 and 600 ms post-stimulus over trials. In this case, the non-time-locking is not so strong, so it wouldn't really affect the TF power (although it would misrepresent the duration and the amplitude would be decreased). On the other hand, imagine that an oscillatory response has a duration of 50 ms (e.g., a gamma burst) and onsets variously between 200 and 2000 ms post-stimulus over trials. In this case, there is extreme non-time-locking and it is likely that trial averaging will simply eliminate the response all together. 

 

Phase-locked is a more stringent feature. A signal is phase-locked only if it takes the same phase angle on each trial. An EEG response has to be strongly time-locked in order to be phase-locked. Non-phase-locked, on the other hand, simply means that the time=0 event doesn't affect the phase characteristics of the signal. Imagine that the stimulus increases the amplitude of an ongoing theta oscillation. Because the theta phase will be random on different trials, this response will be non-phase-locked, and will tend towards zero in time-domain averaging (the ERP). The ERP reflects the phase-locked and time-locked part of the EEG signal. See figure 5.2 (page 56) and figure 20.1 (the color version Plate 11 in the middle of the book). 

Plate 11 is great, is very clear,
My question now is in relation to the way researchers do TF analyses,

In "Gamma oscillations distinguish mere exposure from other likability effects" (2014) (http://www.ncbi.nlm.nih.gov/pubmed/24389505)

The authors do the following,
"Preprocessing for the time–frequency analysis was similar to that of ERP, except that a 0.5–101 Hz band-pass filter was used and trials marked as artifacts during ERP processing were not included.To analyze gamma oscillation, time–frequency representation(TFR) was calculated using the open-source software FieldTrip. TFR was obtained from EEG activity -200–800 ms relative to stimulus onset and from frequency ranges of 8–100Hz, in 2-Hz frequency steps by means of a 7-cycle complex Morlet wavelet, and was baseline corrected relative to activity between -200 and 0 ms(activeperiod/baseline)(e.g. Luft,Nolte,&Bhattacharya,2013)".

This is what you call the direct TF descomposition from ERP signal, and thus the "D) ERP power" in plate 11?

I am not familiar with this paper, but I'm pretty sure it was *not* the ERP power, but rather the total power (that is, wavelet convolution on each trial, extract power, and then average the power over trials). It is generally very difficult to get gamma activity in the ERP power, because gamma power is generally non-phase-locked. 

 

Now, in "EEG oscillatory patterns are associated with error prediction during music performance and are altered in musician's dystonia" (2011) (http://www.ncbi.nlm.nih.gov/pubmed/21195188)

The authors do:
"For that purpose, we computed the wavelet-based time–frequency representations (TFR) of the EEG signals corresponding to the brain responses triggered by actions leading to performance errors and to correct notes. A complex Morlet wavelet was used to extract time–frequency complex phases (...) and amplitudes (...) of the EEG signal x(t). (...). To study changes in the spectral power, we used the wavelet energy, which was computed as the average across epochs of the squared norm of the complex wavelet transform (...)".
This is "A) total power" in plate 11, right?

Yes, most likely. 

 

This is all the most confusing,
I can only think that distinct studies reporting let's say total power, ERP power... cannot uncover the same phenomenon, but anyway they all talk of I don't know gamma activity and draw conclusions of that basis,

It is indeed unfortunate that the literature has not settled on consistent terminology. Very few researchers actually compute the ERP power (although it is occasionally done). You are correct that the ERP power and total power can show very different features of the data. In general, you should compute and interpret the total power, unless there is a specific reason otherwise (usually a methodological reason). 

 

I appreciate the last paragraph, that is what I have read as to most standard method, and I'll take a look at the paper for sure,

José Luis

--

You received this message because you are subscribed to the Google Groups "AnalyzingNeuralTimeSeriesData" group.
To unsubscribe from this group and stop receiving emails from it, send an email to analyzingneuraltimes...@googlegroups.com.
Visit this group at http://groups.google.com/group/analyzingneuraltimeseriesdata.
For more options, visit https://groups.google.com/d/optout.