5 Towards an operational DELTA camera

This section introduces the engineering characteristics of a prototype presently under construction.

5.1 Choice of a CCD

The characteristics of the linear CCDs used as targets for projections determinate the maximum photon rate, as well as the spatial and temporal resolutions of the camera. The choice of these CCDs is therefore crucial.

The readout frequency of the linear CCDs must be as high as possible to reduce the number of photons per frame at high photon flux and therefore the probability of cross-photons, and to increase the time resolution.

We have chosen the recently released Thomson "Mega Speed'' TH7809A. It is a 1024 pixels linear CCD, each pixel being $10\times 20\, \mu$m in size, with a $10\, \mu$m pitch. The TH7809A maximum readout frequency is 400 million pixels per second, thanks to the 16 parallel outputs of an integrated shift register. This shift register allows frame integration during the previous frame readout. As the frame transfer requires less than one clock period, i.e. less than the afterglow of spots caught by the CCDs, no spot will be lost by the frame transfer. The prevention of detecting the same photon in two successive frames is discussed in the next section. The specifications of this CCD chip can be found on the "CCD products'' data book from Thomson CSF (1996).

5.2 Optical setup

Projections in the DELTA camera are made by three identical optical trains, each one projecting onto a given axis ($\xi$in Fig. 6). In the proposed scheme, (L₁) is a spherical collimating lens. (L₂) and (L₃) are cylindric lenses. While (L₂) images in the $\xi$ direction the spot from the front intensifier output onto the linear detector, lens (L₃) images the pupil formed by (L₂). The intensified photon is imaged as a small segment, perpendicular to the CCD line. The image location on the CCD is independent of the spot position along $\eta$ (orthogonal to $\xi$), and proportional to the spot position along $\xi$. With dimensions corresponding to commercial grade lenses, the illuminance of one spot onto the CCDs has been evaluated to $4.1\linebreak 10^{-2}$ ph/$\mu{\rm m}^2$=8 ph/pixel (considering a Lambertian emission from the head intensifier output). An auxiliary image intensifier must be placed before each CCD to raise the illuminance over 7.5 ph/$\mu{\rm m}^2$ (the TH7809A readout noise is 300 electrons/pixel, and its quantum efficiency is about 20% in the wavelengths concerned. Actually, only one auxiliary image intensifier is required if the optical trains are mounted close enough to each other. In this case, the three projections fit in the field of a 25 mm diameter photocathode.

  

Figure 6: Scheme of one (among three) optical trains projecting the field onto the CCDs (the other two optical trains are rotated by $\pm 120^\circ$ with regard to this one): a) case of two spots separated horizontally in the field (the beam impacts on the CCD are separated), b) case of two spots separated vertically in the field (their impacts on this CCD are superposed)

To match the temporal resolution given by the linear CCDs, the image intensifiers (head and auxiliary) must feature fast decay outputs. The phosphor type P-46 is the most suitable for the auxiliary intensifier, as it provides a 100 ns decay time (from 100% to 10%), and a spectral emission matching the spectral response of the TH7809A better than other fast decay phosphors (P-47 or P-90).

The DELTA camera short frame time and the phosphor afterglow may cause some photo-events occurring at the end of a frame to be still present at the beginning of the next frame. Two solutions can be used to prevent that. The first one is a dead time between frame integrations. This dead time span is a trade-off between the maximum tolerated quantity of photo-events covering two frames, and the minimum desired quantum efficiency. The second solution is to eliminate from a frame the photons having the same coordinates as one in the previous frame. As the probability of having two photons within the same pixel in two consecutive frames is very low (evaluated to $8.8\ 10^{-5}$ for $\bar N=4$), no significant artifact such as those affecting the CP40 would be generated. The same remark can be made for the probability of having two photons within the same pixel in a single frame, evaluated to less than $3.6\ 10^{-5}$.

5.3 Photon image analysis

Once projection lists are established for a frame, the next step is the null-sum test. Considering that lists $L_{\rm A}$,$L_{\rm B}$,$L_{\rm C}$have the same number N of elements. One could think that scanning all the possible triplets requires a time proportional to N³. In fact, this process takes a time proportional to N²: for one of the axes (for example C), a binarized image of the corresponding CCD line is stored. For each couple (a,b) of elements in $L_{\rm A}$ and $L_{\rm B}$, the address a+b in the buffer is probed. If it contains a 1, photon coordinates $\left(x=a,y=1.155(a/2+b)\right)$ are generated. Figure 7 describes the whole process, from CCD acquisition to spatiotemporal photon coordinates. FIFOs and buffer swapping allow a pipe-lined data flow.

  

Figure 7: Flow chart of the process yielding Cartesian (x,y,t) photon coordinates from the linear CCDs. Dashed segments indicate the possible states for switches

With such a design, the prototype should have a 2.6 $\mu$s temporal resolution, allowing photon flux up to 1.5 million per second with a good quantum efficiency (85% of the quantum efficiency of a Gen I intensifier) at an average of $\bar N=4$ photons per frame. A micro-computer (300 Mips or more) is the simplest way for converting directly digitized CCD signals into photon coordinates.

5.4 Tuning the camera

Wrong projection directions and/or displacements of the linear CCDs will cause non-hexagonal field and distorted images. To test the tuning procedures and optical alignment requirements, the software simulator takes into account these parameters. Figure 8a shows the effect of a misdirected projection: one axis was rotated by only $1^\circ$. A similar phenomenon (Fig. 8b) is produced when the three axes have correct angles, but with an angular difference between one axis and its corresponding detector.

  

Figure 8: Effect of optical misalignments. Flat field images with: a) $1^\circ$ misdirected projection, b) $10^\circ$ rotation between one axis and its corresponding CCD. The null test tolerance $\varepsilon$ is 4 pixels in these tests. In both cases the size of the field is reduced

Increasing $\varepsilon$ to values higher than 2 is not recommended, since it reduces the resolution, may cause an important loss of quantum efficiency and does not correct image distortion. Using the simulator, we found a simple method for tuning the optical setup, analyzing photon coordinates from a flat-field. This tuning method does not require specific test charts, and can be operated when the camera is docked to a telescope or an interferometer.