Let's make an ISF design (Another infra low thread)


So i became really interested in infra low training. Been doing NFB since 2013, ostly on myself and family. Have a few amps, one of them is a OpenBCI which should be able to go infra low since it is DC coupled. I know there is another thread in thus subforum, but i don't want to hijack it and want this thread to be more focused on ISF (Infra-Slow Ffluctuations). I know it is going to be mostly experimental and the method itself is proprietary. And i know that one should assume responsibility for negative effects this kind of training may entail.

Now, there is an article by the inventor of the method in which he uncovers some details. From reading that article, here is the gist of the method that i was able to distill:

  • ISF uses a low-passed signal cut in the range at 0.002 to 0.012
  • The filter being used is a 1st order butterworth with some special features.
  • The low-passed signal is being compared with a dampened average trace of this same signal and when the original signal goes over or under the average signal the feedback sound is produced. Only two feedback sounds are used - for higher than average and for lower than average.
  • Practitioner is to find the optimal reward frequency.
  • There are several inhibits across the regular EEG spectrum for these bands: 1–3,
    4–7, 8–12, 12–15, 15–20, 20–30, and 30–40 Hz. Each is being inhibited at 3%.

  • It's a bipolar montage. Although, when doing ISF one can combine bipolar ISF training with referential synchrony training.

It was very easy to create a design in BioEra that does just that - calculates the difference between averaged and original band-passed signal. One may then modulate feedback based on that difference - changing pitch of a tone, for example.

Here is the simplest form of what i was able to create in BioEra:

The difference signal looks like this:

I know that this is probably too naive to think that it could be so simple and that probably it is a bit more complicated in real life, but maybe there are those who have some experience and do their own variations of this training, i would like to ask a few questions:

  • Is the filter choice correct?
  • How do you calculate the average? A TimeTransform element in BioEra with Average or Long Average? Which averaging window?
  • Should one compare the averaged and original band-passed signal or its amplitude in the expression evaluator?
  • It is stated that electrode choice is crucial. Mine are Ag/AgCl sintered and i apply them through a cap with a gel. How can i check if i am getting the minimum necessary quality?

What do you guys think in general?


  • stellarpowerstellarpower Scotland

    This reminds me of how I implemented things in the conventional EEG spectrum. I use a continual Z-score: we take the band amplitude using some method, then find the mean and standard deviation over the last three minutes (or however long). Therefore, at any given timepoint in the session we have the band amplitude expressed in terms of the number of standard deviations above the mean. This value can of course be anywhere between plus or minus infinity - it won't, but obviously it can - and so it then goes into a sigmoid function to provide smooth tapering. This then means that we can control screen brightness, with 0 for being exactly on the mean, in the range (0, +1) for being above the mean and in the range (-1, 0) for being below it. Hence we can use that with something that is inherently clamped, like brightness. I only did this just cause it seemed conceptually simpler than all this thresholding and reward percentages. Maybe that is better, I have no idea, I'm just yet to have read an explanation as to why it's done that way.

    Although recently I have been implementing ILF protocols, and after some time fiddling with things I realised I was being stupid, and what I have done instead is control the hue - because it's a circle. We can thus increase or decrease as much as we want and simply mod it and go round the circle again. So no need to worry about scaling the feedback to be within a certain range - it's the speed the colours change that indicate the rate of change of the signal and thus are the feedback. Then we just have a multiplying factor to adjust the sensitivity and I can tune that until I get the impression it's changing enough for me to see it but slowly enough it's meaningful variations in the original signal.

Sign In or Register to comment.