Quick background:
I've been analyzing EEG data from consumer headbands for several years. When I started digging into my own data, I found something I wasn't expecting. Earth's naturally occurring Schumann Resonance oscillates at ~7.8 Hz with harmonics at roughly 14, 20, 26, and 32 Hz. All of these overlap with canonical EEG bands. That overlap has been noted before but generally treated as coincidence.

My research suggests it isn't coincidence. I found that brain oscillation peaks can align with golden ratio (φ = 1.618) precision, anchored near the same ~7.8 Hz fundamental. I tested this across 1M+ peaks from multiple independent datasets. Less than 2% error. (And yes, I would be skeptical too

Here's the paper:

Golden Ratio Architecture of Human Neural Oscillations (preprint)
https://doi.org/10.5281/zenodo.18244908

The research potentially validates ideas proposed about golden ratio organization of EEG bands in 2010 by Pletzer, Kerschbaum, and Klimesch at Universität Salzburg:

When frequencies never synchronize: The golden mean and the resting EEG
https://doi.org/10.1016/j.brainres.2010.03.074

Signal processing pipeline:
The app runs spectral parameterization (FOOOF) on each channel to separate genuine oscillatory peaks from the aperiodic 1/f background, then scores how precisely detected peaks match predicted golden ratio frequency positions across three bands. It simultaneously computes coherence, phase-locking value, and bicoherence across all channel pairs. When precision, amplitude, coherence, and PLV all pass threshold simultaneously for a minimum duration, it flags a Schumann Ignition Event and plays an audio tone through Web Audio API. Everything runs client-side in the browser. No backend signal processing.

Where OpenBCI comes in:
The biggest limitation right now is channel count. With 4 channels I can compute 6 channel pairs for coherence and PLV. With 8 I get 28 pairs. With 16 I get 120. The detection algorithm should scale well because more channel pairs means better separation of genuine cross-regional coherence from noise, and better spatial resolution for mapping where ignition events originate.

I'd also like to add OpenBCI support. For Cyton/Ganglion over Web Serial or Web Bluetooth, has anyone here built browser-based streaming from these boards? Curious about latency, packet parsing, and whether the throughput is sufficient for real-time spectral analysis at 250 Hz across 8-16 channels.

How to try it:
Open Chrome/Edge/Opera on desktop, go to https://resonate.neurokinetikz.com, pair a supported device or use demo mode without hardware. No signup needed.

Research code is open: https://github.com/neurokinetikz/schumann
Would love feedback on the detection approach, the signal processing, or ideas for OpenBCI integration.

1. Hyperscanning support
Does the All-in-One Biosensing Kit support EEG hyperscanning (i.e., simultaneous recording from two separate systems)? If yes, what is the recommended setup for running two devices together?

2. Synchronization between two EEG systems
How is synchronization between the two EEG systems typically achieved? Do they support hardware-based synchronization (e.g., shared clock, sync cable) or software-based synchronization?
What is the expected synchronization accuracy or latency between the two systems?

3. Trigger markers and event synchronization

Do the systems support external and/or internal trigger markers?
Can the same trigger be sent simultaneously to both EEG systems during hyperscanning tasks?
Are trigger inputs available via TTL, USB, or software-based event markers (e.g., via Lab Streaming Layer-LSL)?

4. Data visualization and alignment
Is it possible to view or stream data from both EEG systems simultaneously during acquisition?
What is the recommended workflow for aligning and analyzing the two EEG datasets post-acquisition for inter-brain synchrony analysis?

5. Adequacy of the All-in-One Biosensing Kit
Based on your experience, is the All-in-One Biosensing Kit sufficient for EEG hyperscanning studies focused on inter-brain synchrony (IBS)? Or would you recommend any additional hardware for improved timing precision?

6. Training, documentation, and support
Do they provide remote training sessions after purchase, specifically for multi-device or hyperscanning use cases? Are there any manuals, tutorials, example pipelines, or documentation focused on EEG hyperscanning, multi-device synchronization, or dual-stream data acquisition?

Our proposed study is a mixed-method EEG hyperscanning project examining inter-brain synchrony during cooperative social tasks, with a focus on parent and adolescent dyads. Given the complexity of synchronization and trigger alignment in such designs, your guidance will be extremely valuable in helping us finalize the most appropriate setup.

We would greatly appreciate any technical notes, references, or recommendations you can share to support hyperscanning research using OpenBCI systems.

Thank you for your time and support. I look forward to your response.

I am a 53 year old man who has progressive MS that has led to my normal speech being virtually unintelligible. I am practically bedbound (I live in a care home) and I cannot control my left arm or leg so sign language is not an option for me. My typing with my right hand is done with only one finger and so standard text to speech software applications are far too slow for me to have a normal-speed verbal conversation.
I have participated in research into EEG and ,although specialist equipment was used, I have seen that it is possible to use an EEG headset to interpret brain activity and generate text on a screen.
I am very keen to use a bluetooth eeg to record brain activity whilst I am listening to an audiobook. Both files can then be compared using ai.(possibly ChatGPT)
I would then record my brain activity whilst reading text from an old school exercise book that has never been published on the internet or even typed before. If the ai analysis of the first files can be used to interpret the second brain activity then this should prove the principle and it should be possible to generate real-time speech from measured brain activity.

I that realise real-time speech generation is considered to be the holy grail of speech generation but I genuinely believe that the technology exists to make it possible.
If I could have the use of a bluetooth eeg headset I will use the methods described above to produce a real-time speech generation solution.

All of the methods described above can be seen on many published research documents however there is one aspect of my procedure that ensures that the solution development is entirely unique.

I would be very interested in all comments on this matter.

The research I want to do will probably require more than a basic 4-band EEG. Are there any discounts? What are my options?

Secondly, when I remove the EOG signal using detrending, a 4th order Butterworth filter (0.5 Hz to 45 Hz), a 50 Hz notch filter, and ICA, the low frequency energy is slightly reduced, but still high. Am I missing some processing step?

I need to use the EEG signal for fatigue detection, and I cannot filter out the delta and theta bands because the energy of the delta and theta bands may increase when I am tired. But now the energy at the low frequencies is so high that it is difficult for me to analyze.
Can you tell me if this is normal? Is there something wrong with my process steps?

My goal is to process this data further to compute aEEG (amplitude-integrated EEG), so it’s important for me to ensure that the signal units are correct. I’ve read that there are manual ways to convert the raw values to microvolts using gain and scale factors, but I’m not confident whether these conversions are accurate or consistent across sessions/devices.

Can someone confirm if the data acquired via BrainFlow from the Ganglion board is in µV by default or if it requires manual conversion?

If manual conversion is required, what is the most reliable way to do it?

Is there any recommended method to validate that the converted signal is truly in microvolts (e.g., comparing with reference data or using a test signal)?

Any guidance or references would be greatly appreciated.

Looking forward to your insights!

Adrian

We're planning to use OpenBCI hardware (Cyton + Daisy) to run a visual perception study and we have some doubts about which would be the best approach to synchronize EEG stream data (ideally 16 channels) + EEG markers (stimuli + button responses). Our goal is to present visual stimuli to participants in a go/no-go task (e.g., green/red lights) using external software (e.g., PsychoPy, Psychtoolbox, or EEG-ExPy) and record the participants' responses. For that, we need to mark the EEG stream with the stimulus onset and with the button response.

So far, I have read different entries on this forum from the past years and I have come up with two different potential ways of pursuing the study that I would rather double-check with you (so, here we go!):

(1) ANALOG APPROACH: Use the OpenBCI GUI to record the EEG data + an external button attached to the analog input of the Cyton board + a photosensor attached to the screen to detect the stimulus changes and also plugged to the analog input of the board (as shown in this example and here). This option will save directly the data from the OpenBCI GUI in our preferred (.txt) format. Easy peasy.

(2) DIGITAL APPROACH: Use EEG-ExPy for on-screen stimuli presentation + BrainFlow Python + LSL to be able to use two different markers (stimulus + response) + LabRecorder to connect to both streams (they will be synchronized and merged into one XDF file). Here, I would need to do more research on whether EEG-ExPy/Psytoolbox allows for the recollection of responses (e.g., via keyboard) and it can all be done in digital mode, or whether we need the external button for it. Also, in the use of BrainFlow Pyton and/or LSL.

QUESTIONS
- Regarding the response time accuracy: Is it correct that the external trigger approach (1) more accurate (vs digital) since the trigger data recorded in the Cyton 'Aux' channels is sampled at the same time as the channel data (i.e., 250 Hz in the case of Cyton 8 channel alone or 125 Hz in the case of Cyton + Daisy)?
- Restrictions of channels (Cyton + Daisy): In the analog approach (1) can we use Cyton + Daisy or we can only use the Cyton because we are using the analog mode? In the digital approach (2), do we have this channel restriction?
- Best approach for computer/s use: Would it be ideal to use one computer for stimulus presentation and another for EEG recording? Or that would depend on the approach we decide to use?
- Digital approach (2): BrainFlow Python + LSL + LabRecorder is better recommended than OpenBCI GUI Networking + LSL widget, right (see previous forum entry)?

Is there a third approach more suitable for our needs that we are not considering?
What do you think the best approach would be in our case?

Thank you beforehand and keep rocking & rolling with OpenBCI!
Irene Vigué-Guix

I work with children diagnosed with the most extreme cases of traumatic brain injury and brain-based disorders.

All the kids at our school are non-verbal, and they typically have trouble even making most basic coordinated movements with their hands. Many of them can't control the movement of their eyes.

AND YET - many of them understand speech and attempt to communicate.

I'd like to explore the possibility of creating the following research project:

I want to attempt to scan the EEG and maybe other signals (EMG, EKG - honestly, I don't know what I need to scan) - associated with words of English language thought by a test subject.

My hypothesis is that when a person thinks a word, there have to be some kind of readable micro-signals from the brain, and possibly facial and other muscles, that can be recognized as a certain pattern. If that pattern is reasonably consistent for the test subject, we may be able to pinpoint such consistent pattern for something like 1,000 words - and that would allow our test subject to communicate by controlling the display of these words on the computer screen, or for example having those words spoken by a computer voice.

So, to rephrase, I want to experiment to try to scan verbal thoughts via BCI into computer, essentially enabling non-verbal kids to communicate verbally with this assistive technology.

So, my questions are: What device / devices would you recommend I need to purchase in order to be able to conduct this type of experiment at quality? Do you know if any attempts had been made using the technology in this way (perhaps there are some documents I could get familiar with)? What else could you share that you feel might help? Thank you in advance!

def read_data(file_path):
datax = mne.io.read_raw_edf(file_path, preload=True)
datax.drop_channels(['Fp2', 'O1', 'F4'])
print("Channel names before applying montage:", datax.ch_names)
datax.crop(tmax=27)
datax.filter(l_freq=LowPass_Freq, h_freq=HighPass_Freq)
datax.resample(sfreq=Resampled_Frequency)
datax.set_eeg_reference()
datax.set_montage(montage)
epochs = mne.make_fixed_length_epochs(datax, duration=Epoch_Duration, overlap=OverLap)
return epochs

My first question is do I need to set any channels as references? I meant this part in my code
datax.set_eeg_reference()
Also, I dropped 3 channels since the stream data through GUI is not railed and does not have a good signal-to-noise ratio.
Moreover, I am using following code to calculate PSD
Epochs[3].compute_psd().plot()

but once want to hardcode it and use the frequencies, the range is somehow like this:
[1.37151917e-11, 1.47513447e-11, 5.01511143e-12]

Also used 10*np.log10(psd_data) and the data would be in range of
[ -78.16904615, -75.18809093, -75.52985457,...]
which is weird.
so, how should I convert the PSD to dB to match that plot?

Thanks in advance

Best,

Check out https://umfragen.uni-paderborn.de/index.php/671226?lang=en for further information.

Please check if you're from an eligible country beforehand: https://www.gdpradvisor.co.uk/gdpr-countries

As a token of our appreciation for your time and effort, we offer a compensation of 3.50€ for completing the full survey (can be declined). Please consider participating and sharing this opportunity with others who might be interested!

There is another thing I'd like to ask. Is there any relevant publication/documentation related to the Focus Widget's implementation? From my understanding so far, the concentration/relaxation state seems to be mainly related to the alpha and beta bands, from which the power is calculated. But I'd like to learn more about the specifics, if it's all right.

Thank you!

I would assume classifiers ought to be at least somewhat board-specific, though I could be wrong here.

Is there any online library with pretrained brainflow classifiers + explanations for the training methdologoy ?

I am currently tinkering with OpenBCI integration with Unity as part of my academic research. I've got some basic experience with Unity and EEG signal processing alike (using SciPy and existing datasets), though this is the first time I attempt to implement BCI on Unity (using brainflow), so I'm... a little lost.

(What I aim to do is to control Flappy Bird-like game using BCI, but I won't delve too much into it for now. Instead, I want to ask about the "general" workflow on BCI-Unity integration.)

I've successfully connected the OpenBCI to Unity using brainflow and get some data from it, so that's a start. However, it's the next steps that I'm still confused at. So for now, here are a few questions I'd like to ask:
1. Should I also integrate the training session into the game (i.e. also implement it in Unity)? If that's the case, then does it mean I should also do the training data recording on Unity as well (including the classification model training), or should I write a separate console C# program for that and only use the trained model in Unity? (especially since I'm not sure about ML implementation in Unity yet)
2. Say that with the existing model, I want to retrieve the signal (in real-time/online) and classify it into a relevant input. The thing is that since Unity calls on Update() and FixedUpdate() on periodic basis, how am I supposed to handle this? Should I let the signal processing take place in every Update()/FixedUpdate() call, or should I set a timer/countdown and only do the signal processing and classification when that countdown runs out?

I'd like to learn more, so any pointer is appreciated. Thank you!

Thank you for helping, hope somebody could answer cause I'm gonna hit my head against the wall xD

FFT

Engagement index

Some more results from Daniel using AI analysis of EEG explained here: https://danielexhibit.web.app/
If you would like to contact me, you can use steffan@meditationmonitor.com