IRAS Explanatory Supplement
IX. The Low-Resolution Spectra
C. Data Processing

Chapter Contents | Authors | References
Table of Contents | Index | Previous Section | Next Section
  1. The Database
  2. Processing the Individual Spectra
    1. Despiking
    2. Conversion to a Linear Scale
    3. Interpolation to a Standard Wavelength Grid
    4. Correction for Wavelength-Dependent Responsivity
    5. Correction for Cross-Scan-Dependent Responsivity
    6. Overall Responsivity
    7. Joining of the Two Spectrum Halves
  3. Averaging Spectra, Quality Checks
  4. Final Selection of Spectra

C.1 The Database

The spectrometer data consist of three types of data: (i) uncorrected spectra with header information, (ii) calibration tables, and (iii) administrative files. The spectra were extracted out of the data stream whenever an hours-confirmed source with a signal-to-noise ratio greater than 25 or a source specified as "known source" (Section V.D.4), crossed the spectrometer aperture. Extractions were also made for designated calibration sources. The spectra were linked to survey sources using index association records produced by the hours-and weeks- confirmation processors (Section V.D.7).

There were three types of calibration tables: (a) the responses of the five detectors to the internal reference source flashes and the intensities of the flashes as derived by the survey calibration processor; (b) correction tables for the relative responsivity as a function of position across the five detectors; (c) correction tables for the relative responsivity as a function of wavelength. The cross-scan and wavelength dependent responsivity tables were derived from special observations (see Section IX.B).

C.2 Processing the Individual Spectra

Processing of the data always started with the raw data. This allowed the correction procedures to be improved continuously up to the time of production of the catalog. The major processing steps are discussed below.

C.2.a Despiking

Single sample spikes with an amplitude greater than 8% caused by multiplexer errors (Section IX.B) were removed and replaced by an interpolated value using a simple algorithm operating on the raw data. Multiple-sample spikes were not removed but their presence was noted so that the spectrum-half would be rejected later (see Section IX.C.3).

C.2.b Conversion to a Linear Scale

A lookup table of 256 entries was used to convert the raw data to voltages on a linear scale. A standard reconstruction of the input-voltage to the high-pass filter from the measured output voltage was carried out. An offset correction was reset to zero whenever the sample voltage dropped below a specified threshold because of the effects of the zero clamping (Section IX.A.4).

C.2.c Interpolation to a Standard Wavelength Grid

For ease of processing, an interpolation was carried out to a standard regular grid of angular positions in the dispersion direction. Because the dispersion of the spectrograph changed rapidly as a function of the angular position (Fig. IX.A.1), the wavelength values corresponding to the standard sample values were not equidistant. Before the interpolation, allowance was made for variations in the scan speed of the telescope. For the spectra of the brighter sources (signal-to-noise ratio greater than 10) the well-defined in-scan detector edges were in-scan to center the spectrum. The centering correction reduced in-scan errors to approximately 2", corresponding to approximately 0.03 µm in wavelength.

C.2.d Correction for Wavelength-Dependent Responsivity

The interpolated samples were multiplied by a responsivity table sampled at the same standard grid. There was a table for each of the five detectors derived from observations of -Tau (Fig. IX.B.2). Although the software allowed selecting a different table for each of 16 regularly spaced cross-scan positions on the detector, the evidence for a cross-scan variation of the wavelength-dependent gain was too weak to justify using this option.

C.2.e Correction for Cross-Scan-Dependent Responsivity

Depending on the nominal cross-scan position of the source, a correction (Fig. IX.B.1) was applied for the decrease of responsivity towards the edges of the detectors. This correction was the weakest link in the calibration process because of the relatively large uncertainty in the cross-scan position. The correction applied is uncertain by up to 20%, although this uncertainty was decreased by the process of joining the two spectrum-halves together (see Section IX.C.2.g).

C.2.f Overall Responsivity

The overall responsivity depended on the individual detector, on the time and/or on the sky position. To account for these variations a correction was derived from the voltage responses to the two internal reference source flashes bracketing the time of observation. After applying the responsivity correction to a large sample of spectra the integrated fluxes in the spectra were compared to the fluxes measured by the survey array. Systematic factors of 0.75 and 1.00 were applied to the integrated spectrometer fluxes to bring them in line with the survey observations.

C.2.g Joining of the Two Spectrum Halves

The two spectrum-halves (8-12 µm and 11-25 µm) were treated independently until this point. Because of uncertainties in the cross-scan position of the scan path over the detectors and therefore in the nominal cross-scan responsivity correction, the two spectrum-halves often differed by up to 15 or 20% after subtraction of a linear baseline. The overlapping portion of the spectrum-halves was used to determine another correction factor. In doing so, the nominal relative cross-scan positions were used to determine which half of the spectrum to change by the largest amount. If either half had been observed by the central part of a detector, it was considered reliably calibrated, and that portion was not changed by the joining process, and the half of the spectrum observed near the edge of a detector was shifted up or down towards the other half. If both halves were considered equally reliable, then each was scaled by the square root of the ratio between the overlapping sections. This joining process reduced the overall error in the responsivity correction to less than 10%.

C.3 Averaging Spectral Quality Checks

Before spectra were averaged, a number of quality checks were performed on the individual measurements of the two halves of a source's spectrum. First, all measurements made within 18" of the edge of any detector were flagged. Measurements were rejected:
  1. if they contained multiple-sample spikes (Section IX.C.2);
  2. if the join-factor obtained before (Section IX.C.2) was outside the range 0.30 to 3.3;
  3. if the measurement was confused by neighboring sources; this was considered to occur when the measurement met one of two criteria: (i) the central portion was below the baseline determined from signal-free parts of the spectrum-half; (ii) the baseline at the low wavelength end of the spectrum-half differed from that at the high wavelength end by more than 20% of the signal in the 8-12 µm band or by 10% in the 11-25 µm band. The lower limit of these thresholds was 2.5 times the sample noise.
  4. if the measurement did not correlate with the "reference measurement", defined as the measurement with the smallest number of check-flags. This choice gave preference to measurements passing over the central part of the detector. Any 8-12 µm measurement for which the correlation coefficient with the reference was below 60% or any 11-25 µm measurement for which it was below 50% was rejected. For line spectra without a continuum in the 8-12 µm region the first criterion was waived.

At least 80% of the spectra in the catalog had correlation coefficients above 70 and 60% in the short and long wavelength halves, respectively. Some 40% correlated internally with coefficients better than 80% in both spectrum-halves (see Section IX.C.4).

The spectrum-halves passing through all of the above tests were averaged using the inverse of the square of the noise as a weighting factor. At least two measurements in each of the two spectrum-halves (8-13 and 11-25 µm) had to be accepted before the spectrum could be averaged and included in the spectral catalog.

After averaging the two spectrum-halves were rejoined, giving both halves equal weight (see Section IX.C.1). Generally the join factors differed from 1.00 by only a few percent.

The averaged spectrum was convolved with the 12 µm survey passband. The integrated flux thus obtained, was compared to the average 12 µm survey flux of the source. The ratio between the two fluxes is given in the low resolution spectrometer catalog record and has a 1 dispersion around unity of about 15%. Exceptions to this rule will be spectra with sharp lines (classes 8 and 9; see Section IX.D.2) or small 12 µm fluxes.

C.4 Final Selection of Spectra

Three selection criteria were applied for inclusion in the Catalog of Low Resolution Spectra.
  1. The source is contained in the IRAS point source catalog.
  2. The entire spectrum must have been observed at least twice and the individual measurements should be mutually consistent. (individual spectra must pass all the checks mentioned in Section IX.C.3 and must have a minimum correlation coefficient of 50% between any two measurements of the source. The large majority of spectra had however, much higher correlation coefficients (see below).
  3. The source must pass a subjective visual inspection. About 2.5% of all sources were rejected by this process, mostly because they showed non-point source characteristics or confusion with other sources.

Four samples of sources were selected for inclusion in the catalog.

  1. Sources whose 12 µm survey flux density was larger than 25 Jy or whose 25 µm survey flux density was larger than 50 Jy. Individual spectrum-halves were required to correlate with each other with a correlation coefficient of 80% in either spectrum-half. This sample contains about 2150 sources.
  2. Sources with 12 µm flux densities larger than 1 Jy or 25  µm flux densities larger than 2 Jy but not contained in Sample a. The vast majority of these sources proved to be brighter than approximately 5 Jy at 12 µm or 10 Jy at 25 µm. The correlation coefficients were required to exceed 70% in the 8-12 µm band and 60% in the 11-25 µm band. This sample contains about 2450 sources.
  3. Sources in the same flux density range as sample b but with lower correlation coefficients: between 60 and 70% in the 11-22  µm spectrum-half or between 50 and 60% in the 11-25 µm hall; respectively. There are roughly 850 sources in this sample.
  4. Sources with minimum survey flux densities of 1 Jy at 12  µm or 2 Jy at 25 µm. Of these sources only the 11-25 µm spectrum-half was required to have been measured consistently with a minimum correlation coefficient of 50% between individual measurements. Out of this sample only sources with specified spectral lines and not contained in sample a b, c were selected for the catalog. The selection was carried out by the classification program: only classes 8 and 9 (see Section IX.D.2 and Table IX.D.1) were kept. There are approximately 40 spectra with lines in this sample.

Samples a b, and c contain continuum sources and approximately 75% of the line sources of the catalog; sample d contains the remainder of the line sources.

Chapter Contents | Authors | References
Table of Contents | Index | Previous Section | Next Section