NICMOS3 Detector for Spectroscopy

L.Vanzi (1), A.Marconi (1) and S.Gennari (2)

(1) Università degli Studi di Firenze - Dipartimento di Astronomia, Largo E.Fermi 5, 50125 Firenze, Italy

(2) Osservatorio Astrofisico di Arcetri, Largo E.Fermi 5, 50125 Firenze, Italy


We present some results about the use of an engineering grade NICMOS3 array for medium resolution spectroscopy. All the work described here was obtained at the Infrared Laboratory of the Osservatorio Astrofisico di Arcetri in Florence and at the Italian Telescope TIRGO (Citterio, 1978). The detector has been mounted on LonGSp (Longslit Gornergrat Spectrometer), an instrument designed and realized in Arcetri and described in Gennari & Vanzi 1994.

This paper deals with the characterization of engineering grade arrays and with some problems in the reduction of infrared spectra.

Characterization of the Array

Our spectrometer uses a subsection of an engineering grade NICMOS3 array as detector, our purpose is to understand if this kind of array is useful for application to spectroscopy and, meanwhile, to have a good characterization to choose the best available subsection.

Laboratory Tests

The characterization of the array consisted of evaluating the bad pixels percentage, the read out noise and the dark current rate.

We acquired dark and flatfield frames, the former without any illumination of the detector and the latter with the uniform illumination of a led diode inside the dewar. The dark frames were taken at three different temperatures (55, 65 and 78 K) with different integration times (stacks of ten frames at 1,2,5,10,20,30,60,120 and 180 sec).

The dark frames were obtained in three different conditions: the first series after a complete reset of the array, the second and the third after a weak and a strong illumination of the detector to test for memory effects.

For each series of measurements stacks of ten frames were taken with the same integration times used for darks.

The badpixels were found out from normalized dark and flatfield frames by dividing each quadrant in 4x4 subareas and defining the badpixels as those exceeding by 4 sigma the median value in the subarea. In all of the following analysis the badpixels were not considered and their value was substituted with the median over an 11x11 box.

The mean and the standard deviation were computed for each stack (pixel by pixel) after having renormalized the frames to get the same median value in a selected region of the first quadrant. Therefore, the averages over 16x16 subareas were taken both for pixel stack mean and standard deviation.

The readout noise was determined as the mean standard deviation of each pixel in the stacks at 1 and 2 sec and then averaging over areas of 32x32.

The dark current rate and the bias were derived by a linear fit, pixel by pixel, of the stack mean versus integration time, with the error on the mean given by the pixel standard deviation in each stack.


We characterized two engineering grade arrays and compared the results with the science grade chip presently mounted in the large field camera ARNICA (ARcetri Near Infrared CAmera) at the TIRGO telescope (Lisi et al. 1993). Fig. 1 compares the flatfields obtained with the two engineering grade arrays.

In tab.1 the results of our measurements over the best 100x100 pixels subsection of the two engineering grade arrays and over the whole scientific array are presented.

                          TAB. 1   Measured characteristics of the arrays

                           1 best 100x100   2 best 100x100    Scientific  

Bad Pix (%)                     3.0              1.7             1.0      
Read Out Noise (electrons)      240               45              45     
Dark Current (electrons/sec)     40                2               1    

In fig. 2 the read-out noise is plotted for each subarea as a function of the detector temperature. An indicative value for RON is 2.5 ADU (about 45 electrons).

Fig.2 - Readout noise (RON) in counts (ADU) of the new engineering grade array as a function of the detector temperature.

In fig. 3 the dark rate is plotted as a function of temperature. The three different symbols at 78 K indicate frames taken after complete reset (circle), after normal illumination (triangle, like measurements at lower T) and after strong illumination (squares). In the latter case one can notice an increase of the dark current rate by about 20%. Indicative values for the dark rate are 0.1 ADU/s (2 electrons/sec).

Fig.3 - Dark rate (ADU/sec) of the new engineering grade array as a function of the detector temperature and in three different conditions: after a full reset (circles), after normal (triangles) and strong illumination (squares).

Data Reduction

Data reduction is performed with the ESO package MIDAS in the context IRSPEC. The context has been optimized for reduction of IRSPEC data but can be used for LONGSP since the principles of data reduction are quite general and not specific to a particular instrument.

Data from Observations

Each observation includes object and sky frames with the same integration time taken with the ABBA sequence as well as the spectrum of a standard star and, if necessary, a wavelength calibration lamp. Flatfield and dark frames should be taken, for instance, at the beginning or at the end of the night or every few hours according to the stability of the detector.

Bad Pixels Correction

This correction can be performed once a mask of badpixels is given. Bad pixels are substituted by the average value over the neighboring pixel in the x or y direction. The chosen direction is that with the lower gradient. This method works only with fairly isolated bad pixels and cannot handle cases when the bad pixels are clustered. In such a situation one may use techniques similar to those for cosmic rays removal in CCD frames.

Flat Field and Dark Current

The original flat frame cleaned by badpixels is divided by the average of those rows where there will be the object spectrum. After the normalization, all of those pixels that are lower than a threshold (for distance 0.5) are considered vignetted and set to a fixed high value (i.e. 100) so that after flat division the same pixels in the object frame will not be considered. The flat image is usually a measurement of the halogen lamp with counts level as close as possible -and well within a factor of 2- to those in the astronomical frames.

Sky Subtraction

The greatest problem in NIR observation is the subtraction of OH sky lines which are very bright and variable on time scale of a minute.

A simple object-sky subtraction is not always enough to guarantee a complete removal of the emission lines as the sky spectra in the two frames (source and sky) may be different, because of a shift in the wavelength direction due to mechanical flexures in the instrument and of variation of line intensity.

The used technique is to consider the difference between object and sky frames and choose an area of the frame where there are bad subtracted sky lines. The extra noise there is mainly due to residuals of OH lines. The factor to rescale the sky frame and the shift to apply are fixed by minimizing the noise in the previously chosen area. In the spectral region where the background is not too high one also need a dark frame. The procedure can be summarized in the following equation:

where A is the factor by which OH line varied (and this correction must not be applied to the dark frame!) and D is the shift due to grating movements. A and D are determined by minimizing the noise on the background.

This procedure is not able to correct grating shifts > 0.1 pixels during the integration, in that case the problem is still open ....

This procedure of sky subtraction becomes less and less efficient as the integration time increases, in particular one can notice how the noise gets greater than simple fluctuations of the background when integrating longer than 200 sec (fig. 4).

This problem limits the integration time and increases the limiting flux (about 10e-13 erg s-1 cm-2 arcsec-2 micron m-1),preventing us from reaching background limited performance in the J and K bands.

Fig.4 - Noise (ADU) of the sky subtracted images as a function of the integration time. The lines represent the poissonian component of the background noise, note the high noise due to bad subtraction of OH sky lines after 200 sec of on-chip integration.

Rectification of the Frames

The slit images at the various wavelengths ("spectral lines") are tilted as a consequence of the off-axis mount of the grating, and the angle by which they are tilted varies with the position of the grating, i.e. with the wavelength. The tilt angle is computed analytically from the instrumental calibration parameters (on line central wavelength etc) so the correction does not present particular problems.

Wavelength Calibration

The wavelength dispersion on the array is linear within a small and totally negligible fraction of the pixel size. Hence, wavelength calibrating simply means modifying the x-start and x-step values (descriptor) of the image. Another advantage is that one can very precisely compute (analytically) the pixel size -in wavelength- once the central wavelength of the frame is known. A quite precise, usually within 1 pixel, estimate of this quantity is available on-line at the instrument ("mechanical" calibration). One can directly use this information and determine the "mechanical" wavelength calibration. It is possible to determine more precisely the central wavelength of the frame, and hence to obtain a very accurate wavelength calibration, if one has a frame containing lines with known wavelengths, up to 2.3 microns the OH lines in the sky frames are a very convenient calibrator (Oliva & Origlia 1992).

Calibration with standard star

The calibration procedures up to wavelength calibration followed for the object must be repeated also for the standard star. Having flux calibrated the spectrum of the standard star with known photometric data one divides the object frame by the standard star, pixel by pixel, to achieve the cancellation of atmospheric absorption lines and flux calibrate the spectrum. Possibly there might be small shifts and different widths between absorption features in object and star spectra. In that case one determines by hand the shift and the smoothing to apply to the star to better remove the features.


Engineering grade arrays have quite large subsections with performance comparable to a scientific grade array and can be a cheap solution if a large field is not required. Doing infrared spectroscopy at medium resolution with these arrays allows the use of quite long integration times and the limits on the signal to noise ratio are mainly posed by the subtraction of sky lines.


Citterio, 1978, Mem.S.A.It., 49, 57

Gennari S. and Vanzi L., 1994, InfraRed Astronomy with Arrays, ed. I. McLean, p.351

Lisi F., Baffa C., Hunt L.K., 1993, Comunication 1946.58 at the International Symposium of SPIE, The international Society for Optical Engineering, Orlando (USA) 12-16 April 1993.

Oliva E. and Origlia L., 1992. A&A 254, 466

Last update 4 April 96

LonGSp Start Page       LonGSp Publications Start Page