File Formats

EEG data comes from many places, recorded by many different amplifiers, saved in many different formats. Getting all of that into a consistent internal representation is the unglamorous first step that everything else depends on. If the loading step gets something wrong—channel labels swapped, units misinterpreted, events misaligned—every analysis downstream inherits that error silently. A power spectrum computed from data scaled in volts instead of microvolts will be off by twelve orders of magnitude. A connectivity matrix with transposed channel names will show the wrong regions talking to each other.

The Coherence Workstation handles format detection and normalization automatically. But understanding what formats exist, where complexity hides, and what to watch for makes you a better judge of whether your data loaded correctly.

What the Workstation Accepts

Three format families cover the recordings you’re most likely to work with.

NFX—Neurofield Q20/Q21 Recordings

The native format of the Neurofield headset. Raw data at 256 Hz, 19 EEG channels in the standard international 10-20 montage, plus a photoplethysmography (PPG) channel for heart rate variability. This is the Coherence Workstation’s most thoroughly tested import path—we’ve processed hundreds of recordings in this format and the import is robust.

Two internal variants of the NFX format exist: a newer XDF-wrapped version (an open standard for multi-stream physiological data) and an older legacy binary format. The workstation detects which variant you have and handles both transparently. You don’t need to know which one your files use.

SET—EEGLAB-Format Files

If you’ve preprocessed your data in EEGLAB—or if you used the older NeuroPipe/Neurofield processing pipeline—the resulting .set files load directly into the Coherence Workstation. These are MATLAB-format files (either standard or HDF5) containing the EEG data along with channel information, event markers, and preprocessing history.

The workstation looks for files matching the *_Clean.set naming convention, which indicates data that has already been through a preprocessing pipeline. Channel names, events, and metadata are preserved from the EEGLAB structure. If your .set files came from a different pipeline or don’t follow the _Clean naming convention, they may still load—but this is the best-tested path.

EDF—European Data Format

The universal exchange format for EEG data. If your amplifier exports EDF (or its extended variant, EDF+), the Coherence Workstation can import it. This is the path for Mitsar recordings and for any other system that doesn’t have a native import—BrainMaster, Deymed, Compumedics, or anything else that offers EDF export.

EDF support is solid, but it’s the format where the variability described in the next section matters most. Every manufacturer writes EDF files slightly differently, and the Coherence Workstation has to make assumptions about channel naming, scaling, and event encoding that may not hold for every system we haven’t seen yet.

Why Loading EEG Data Is Harder Than It Looks

If you’ve ever tried to open a colleague’s EEG recording in your own software and gotten something that looked nothing like what they described, you’ve already encountered this problem. EEG file formats are not as standardized as they appear.

Every amplifier manufacturer makes different choices about how to encode the same fundamental data. Channel naming is the most visible example: the electrode position you know as T3 in the original 10-20 nomenclature is labeled T7 in the revised 10-10 system. Some systems use A1 and A2 for the earlobes; others use M1 and M2. Some include the reference channel in the data file; others record it implicitly and don’t export it. A file that says it has 19 channels might actually have 21 if the system includes reference and ground electrodes in the export.

Sampling rates vary. The Neurofield Q20/Q21 records at 256 Hz. Many clinical systems run at 500 Hz. Research-grade amplifiers may sample at 1000 Hz or higher. The Coherence Workstation resamples when necessary, but resampling introduces subtle effects on high-frequency content that are worth being aware of if you’re comparing recordings made at different native rates.

Amplitude units are another source of invisible disagreement. Most EEG software displays data in microvolts, but the raw values stored in the file might be in volts, millivolts, or arbitrary digital units with a scaling factor buried in the header. If the workstation misinterprets the scaling, your spectra will look dramatically wrong—power values orders of magnitude too high or too low. The numbers will be internally consistent, but they won’t match what you see in your amplifier’s native software. This is usually the first sign of an import issue.

Recording reference schemes are baked into the data in ways that aren’t always explicit. Some systems record against linked ears, some against Cz, some against a proprietary reference electrode. The Coherence Workstation detects the recording reference when it can and re-references to average—but if the reference detection fails for an unfamiliar format, the resulting data will have a systematic bias that affects every channel.

Event markers for ERP paradigms add another layer. Some formats embed stimulus and response events directly in the data stream. Others store them in separate files with their own timing conventions. Some use numeric event codes; others use text labels. The Coherence Workstation’s event detection works reliably for Neurofield’s ERP paradigm and for standard EEGLAB event structures, but non-standard event encoding from other systems may require attention.

None of this is to discourage you from trying recordings from different systems. It’s to explain why “just open the file” is more complex than it sounds—and why, when something looks wrong, the import step is the first place to check.

What We’ve Tested

Neurofield Q20/Q21 is our primary development platform. Both NFX variants (XDF and legacy binary) are thoroughly tested. If you’re recording with a Neurofield system, the import path is well-worn and reliable.

Mitsar recordings via EDF export are our second most-tested path. Channel naming, scaling, and event handling for Mitsar systems have been verified against multiple recordings.

Other platforms—we’ve tested with recordings from several other amplifier systems exported as EDF. Results have been good in the cases we’ve seen. But EDF is a format where manufacturer-specific choices can surface, and we haven’t seen every amplifier’s version of it.

If your spectra look unusual, if channel labels seem wrong, or if the data just doesn’t look like what you see in your amplifier’s native software, it may be a loading issue rather than a genuine finding in the data. We genuinely want to hear about it. Every new amplifier we support well makes the tool more useful for everyone, and we can usually diagnose import issues quickly when we can see a sample file.

What Happens During Import

When you create a new session and point it to a folder, the Coherence Workstation scans for recognized file types and classifies what it finds. The discovery process identifies recording conditions—resting eyes-open, resting eyes-closed, ERP task data—based on file names, directory structure, and embedded metadata.

The application checks channel counts against known montages (19-channel standard 10-20, 37-channel extended) and validates signal scaling to catch obvious import problems before you commit to a full processing run. You see a summary of what was discovered, and you can correct any misclassifications before proceeding.

This discovery step is intentionally transparent. You should be able to look at the summary and confirm that what the workstation found matches what you know about the recording. If something is missing, misclassified, or flagged with a warning, you can address it before the pipeline runs—rather than discovering the problem twenty minutes later in the analysis output.

If Something Doesn’t Look Right

Start with the basics. Check the channel count—19 for a standard 10-20 montage, 37 for an extended montage. If the count is different from what your amplifier recorded, something went wrong during import.

Look at the raw signal in the Visual Inspection step after processing. Does the EEG look like EEG you recognize? Are the amplitudes in a reasonable range (typically 10–100 µV for resting EEG)? Do the channel labels correspond to the electrode positions you expect?

Check the resting spectra. A healthy resting EEG with eyes closed should show a posterior alpha peak somewhere between 8 and 13 Hz. If you see no alpha peak, or if the power values are orders of magnitude off from what you’d expect, the issue is likely in how the data was scaled during import.

If something doesn’t match, contact us with the details and—if you can—a sample recording file. We can usually identify the issue quickly when we can see the raw data alongside your amplifier’s specifications. Import problems are solvable problems, and solving them for your system improves the workstation for every clinician using the same hardware.