Data Products
Formats and contents of the EPIC pixel spectrograms, imaging sessions, FITS image cubes, daily digest and video stream.
Retaining every 80 ms all-sky image across the full station bandwidth would require approximately 110 TB per day (Reddy et al. 2024). The data products therefore combine two kinds of output:
- high-cadence, full-resolution data for a selected set of sources;
- lower-cadence data for the full sky.
| Product | Resolution | Coverage | Storage |
|---|---|---|---|
| Pixel spectrograms | 80 ms; 25 kHz; four polarization products | 5 × 5 pixels around each watched source | PostgreSQL, epic_pixels |
| Imaging sessions | One record per tuning | Metadata | PostgreSQL, epic_img_metadata2 |
| Image cubes | 1.28 s; all channels; four products | Full sky | FITS and PNG files |
| Daily digest | 1.28 s | Stokes I and V spectra of watched sources | PostgreSQL, epic_daily_digest |
| Video | 80 ms (12.5 frames s⁻¹); one channel; Stokes I | Full sky | EPIC TV |
Watch list
Pixel data are recorded only for sources on the watch list, which is maintained by the watchdog service. The list contains two kinds of entry:
- Persistent entries, such as the Sun, Jupiter, bright pulsars and the A-team sources.
- Time-limited entries, created in response to external alerts such as CHIME/FRB VOEvents.
Each entry records the event time, position, observing window, reason and, where applicable, the originating VOEvent. Sources below an elevation of 10° are not extracted.
Pixel spectrograms (epic_pixels)
Each row contains one pixel of one source in one 80 ms image.
| Column | Type | Description |
|---|---|---|
img_time | timestamp | Mid-point of the integration (UTC) |
chan0 | integer | First channel of the tuning; kHz |
source_name | text | Watch-list identifier |
pixel_coord | point | Pixel coordinates |
pix_offset | point | Offset from the source pixel; denotes the central pixel |
pixel_lm | point | Direction cosines |
pixel_values | bytea | Spectra of the four polarization products |
The field pixel_values contains little-endian 32-bit floating-point values, ordered by channel, with the four products interleaved:
[ XX*(ν₀), YY*(ν₀), Re XY*(ν₀), Im XY*(ν₀), XX*(ν₁), YY*(ν₁), … ] length 4 n_chan
A row may be decoded as follows:
import numpy as np
v = np.frombuffer(row.pixel_values, dtype='<f4').reshape(n_chan, 4)
xx, yy, xy_re, xy_im = v.T
stokes_i = xx + yy # uncalibrated pseudo-Stokes I
freqs = (chan0 + np.arange(n_chan)) * 25e3 # Hz
The dynamic spectrum of a source is formed from the central pixel, pix_offset = (0,0), ordered by img_time. The surrounding pixels of the patch provide a local reference. Interference and ionospheric effects tend to appear across the patch, whereas emission from an unresolved source is concentrated in the central pixel.
The table accumulates of order rows per day. It is distributed across the database cluster, partitioned in time, and retained for a finite period. Spectrograms are retrieved through the Waterfalls application or its API, in intervals of up to 120 s.
Imaging sessions (epic_img_metadata2)
An imaging session is a contiguous interval at a fixed tuning. A new session, with a new session_id, is opened whenever the station is retuned. Each session record contains:
session_startandsession_endchan0,n_chan,n_pol(four) andchan_bw_hz- the integration time
int_time - the image size
- the patch size
npix_kernel - the git revision of the imager,
epic_version - the names of the watched sources
The Waterfalls application uses this table to enumerate the sessions available for a source within a given time range.
Image cubes (FITS)
An accumulated image cube is written every --nimg_accum images (1.28 s at LWA-SV). The file format is as follows:
- File name.
EPIC_<unix_time>_<cfreq>MHz.fits, accompanied by a_QL.pngquick-look image. - Axes.
(stokes, frequency, y, x). The four planes contain , , and . - Projection.
RA---SIN/DEC--SIN, with the reference position at the zenith at the epoch of the image, in FK5 J2000. - Units.
BUNIT = UNCALIB(see Calibration). - Further keywords. Integration time, center frequency, grid size and grid sampling.
Each file is registered in the table epic_files_metadata, together with its imaging parameters.
Daily digest (epic_daily_digest)
For each accumulated cube, two spectra are recorded for every watched source:
- Stokes I: the sum of over the source patch.
- Stokes V: a spectrum derived from .
Both are tagged with the center frequency. The Daily Digest view presents these spectra as day-long dynamic spectra, which serve as an index to intervals of activity in the 80 ms data.
Video stream
The EPIC TV stream presents Stokes I of a single channel at 80 ms cadence (12.5 frames s⁻¹), with a perceptual color map, a horizon mask and a frequency and time annotation. Each frame is normalized independently, so the stream represents relative rather than absolute brightness. The most prominent features are:
- the Sun and the A-team sources traversing the sky;
- ionospheric scintillation;
- interference near the horizon.
During station health checks the stream displays a placeholder frame.