Thermo Fisher OMNIC .spa / .spg

Thermo Fisher OMNIC (Nicolet FT-IR spectrometers) saves one spectrum per .spa file, a group of spectra per .spg file and series as .srs files. OpenReadout returns the spectra as traces, with the sample and background interferograms when OMNIC kept them. OMNIC Atlµs map files (.map) are not read.

Derived from SpectroChemPy’s OMNIC reader (CeCILL-B, a BSD-style licence; read as prior art and run as a reference reader), hex dumps of public files and the CSV files OMNIC exported from them. Provenance: docs/provenance/thermo-omnic.md. Format id thermo-omnic, family spectroscopy, crate openreadout-spectro (OmnicReader, OMNIC_FORMAT_ID).

.spa holds one spectrum (with the sample and background interferograms when OMNIC kept them); .spg a group of spectra. Detection: the file starts with Spectral Data File (.spa, .spg) or Spectral Exte File (.srs series: below).

Layout

offset	type	meaning
0	18 bytes	`Spectral Data File`
30	256 bytes	title (the spectrum’s title; the group’s file name in `.spg`)
294	u16	number of key records
296	u32	timestamp: seconds since 1899-12-31T00:00:00Z (UTC)
304	16 × n	key records: key byte, u32 offset at +2, u32 length at +6, u16 spectrum number at +10 (`.spg`)

Keys used: 2 spectrum header, 3 intensities (float32), 4 comment, 27 history text, 102/103 sample/background interferogram (float32), 106 acquisition parameters, 107 spectrum title (256 bytes) and timestamp (u32 at +256) in groups, 130 experiment information (first byte 0x79: fixed text slots — experiment path, title, description, accessory). A spectrum of a group starts at its key 2; its other records carry the same spectrum number.

Spectrum header (key 2), offsets from its start: +4 points (u32), +8 x-unit code, +12 y-unit code, +16 x of the first stored point (f32), +20 x of the last point (f32), +28 scan points, +32 interferogram peak position, +36 sample scans, +44 FFT points, +52 background scans (u32), +56 background gain, +68 collection time in 1/100 s (u32), +80 laser frequency (cm⁻¹, f32), +84 sample spacing, +92 aperture, +96 Raman excitation (cm⁻¹), +188 optical velocity.

x-unit codes: 1 wavenumber (1/cm), 2 points (interferograms), 3 wavelength (nm), 4 wavelength (µm), 32 raman_shift (1/cm); others x.
y-unit codes (channel name, unit): 11 reflectance (%), 12 log_inverse_reflectance, 15 single_beam, 16 transmittance (%), 17 absorbance (AU), 20 kubelka_munk, 21 reflectance, 22 detector_signal (V), 26 photoacoustic, 31 raman_intensity; others intensity. Codes 17 and 22 are confirmed by the corpus (history “Final format: Absorbance / Volts”); the others follow SpectroChemPy.

The x axis runs evenly from the header’s first to last value (x_at(i) = first + i × (last − first)/(n − 1)); OMNIC’s CSV exports confirm it point by point (they list the same points in ascending x and add one row with value 0 one step below the range).

Traces

.spa: trace 0 is the spectrum (name = the title), then sample interferogram and background interferogram when stored (x in points, y detector_signal in V).
.spg: spectra sharing points, axis ends and units form one trace with one sweep per spectrum, in file order; a group with different axes gives one trace per axis. Table spectra (one row per sweep: spectrum number, acquired_unix_s, elapsed_s) and extra.spectrum_titles list them.

`traces[].extra` (our vocabulary)

key	from
`axis`, `data_type`, `y_quantity`	header codes (`data_type`: `INFRARED SPECTRUM`, `INFRARED INTERFEROGRAM`, `RAMAN SPECTRUM`, `UV/VIS SPECTRUM`)
`scans`, `background_scans`	header +36, +52
`laser_wavenumber_cm1`	header +80 (the reference laser)
`laser_wavelength_nm`, `raman_excitation_cm1`	header +96, Raman spectra only
`optical_velocity`, `aperture`, `background_gain`, `sample_spacing`	header
`collection_time_s`	header +68 / 100
`scan_points`, `peak_position`, `fft_points`	header
`resolution_cm1`	history text `Resolution:` / `Résolution:` (decimal comma accepted)
`final_format`, `instrument_serial`	history text `Final format:`, `Bench Serial Number:`
`history`	key 27
`title`, `acquired_at`	offset 30, offset 296 (`.spa`); key 107 (`.spg`, first spectrum)
`spectrum_titles`	key 107 of each spectrum (`.spg`)
`x_units_code`, `y_units_code`	header +8, +12
`spectrum_role`	`sample` / `background` on interferogram traces

Experiment. sample.id = the title (.spa), instrument (Thermo Fisher Scientific, OMNIC, serial from the history), method.name = the experiment title (key 130) or the group title, method.parameters: resolution, scans, background_scans, laser_wavenumber, laser_wavelength (Raman), accessory (key 130), final_format; acquisition.started_at (UTC), acquisition.comment (key 4).

Listing and vendor tree. info --view structure lists every key record; info --view full holds the title, timestamp, key table, comments, the experiment-information records and the acquisition-parameter blocks (key 106: digitizer bits, high/low-pass filters, sample gain, optical velocity).

Validation

SpectroChemPy 1.0.0 read_omnic on 14 files (8 Toffolo .SPA, Orange’s sample1.spa, and five held SpectroChemPy test files: three .spg groups of 2, 19 and 55 spectra, an interferogram and a spectrum): every value bit for bit; scans, background scans, laser frequency; x axis within SpectroChemPy’s 3-decimal rounding.
OMNIC’s CSV exports of 144 Toffolo spectra agree point by point to their 7 significant digits (8 of them in the automated corpus test).

Series (`.srs`)

Rapid-scan, high-speed real-time, GC-IR and TGA-IR series. The file keeps the .spa header (title at 30, record count at 294, timestamp at 296) but its key records at 304 are 22 bytes: u16 key, u64 offset, u32 length, u32 set (1 the series, 0 the backgrounds, 2 and 3 processing), u32 index (which background, or packed codes in key 130). The reader follows the key table; it does not search for byte signatures, so key tables in another order read the same.

Key 301, the series: a 140-byte spectrum header (the key-2 layout), a 56-byte acquisition block, then per spectrum 16 bytes (u32, then the spectrum’s time in 1/100 s), an 84-byte record starting with its name (Linked spectrum at 0.083 min.) and points float32 values. The record must be exactly 196 + n × (100 + 4 × points) bytes (else exit 4); n is checked against the series information.
Key 325, series information: title text at +2, first time, last time and time step in minutes (f32) at +66, +70, +74, the spectrum count (u32) at +90.
Backgrounds: each set-0 key-2 header with its key-3 values (index from the record).
Axis order: series and background values are stored from the smaller x to the larger (a TGA-IR series read that way shows CO₂ at 2349 and 667 1/cm and water at 1500–1700, read the other way it would not), unlike .spa; extra.axis runs ascending. Interferogram series (x code 2) run over points 0 … n − 1.

Traces: 0 = the series (named by the series title; one sweep per spectrum; extra: series_title, first_time_min, last_time_min, time_step_min, spectrum_names (first and last), scans, background_scans, laser_wavenumber_cm1, x_units_code, y_units_code), then background 0, background 1, … (extra.spectrum_role = background, title). Table series: spectrum, time_min (from each spectrum’s own time). Experiment: method.name = the series title, acquisition.comment = the series history (key 27), acquisition.started_at = the timestamp at 296 only when the series header repeats it (at +836, +368 or +828; reprocessed and GC files hold none). Keys 110, 111 (profiles over time: u32 count, u32 m, m × count float32), 112, 130, 146 and 300 are listed, not decoded (a note; assurance undecoded without scope).

Validation (series): SpectroChemPy read_srs on six held test files (GC-IR, TGA-IR, rapid scan as interferograms and reprocessed to absorbance, high-speed with two backgrounds, a TGA-IR file whose records are in another order): every spectrum of every series and the first background, bit for bit, and the axis ends.

Known gaps

.srs files with a licence confirmed for redistribution were not found: the series layout rests on SpectroChemPy’s held test files. The profile records of a series (Gram–Schmidt, chemigrams) are not decoded. OMNIC Atlµs map files (.map) are not recognised. Resolution and the serial number are only known when the history text records them.

Vocabulary (public API of `openreadout-spectro`)

identifier	meaning
`OmnicReader`	the reader (`FormatReader`): detection by the first bytes, `open`/`open_input`
`OMNIC_FORMAT_ID`	the format id, `thermo-omnic`
`SpectroDataset`	an opened file (the `Dataset` the four spectroscopy readers share): traces, tables, map images and attachments read lazily

Everything else — trace names, channel names, extra keys and their values — is listed in the tables above in our own words.

How this reader was derived, file by file: provenance log.