The parallelism test employs an identical optical arrangement to that for astronomical imaging, with the exception of masks added at the focal and pupil planes. Figure 1 shows the main components of the system. Our basic method is to irradiate the TTF with a monochromatic source. In the focal plane, we use a slit which projects to a few rows of the CCD. The slit mask ensures that only a thin strip of monochromatic light is imaged at the centre of the CCD detector. The tunable filter is located in a collimated beam and quadrant-shaped masks placed in the pupil plane isolate a quarter of the TTF plate area at a time for testing.
The long-slit spectra used for determining parallelism are obtained in the following manner. An exposure is taken, at the end of which the shutter is closed, the TTF plates shift to the next value of plate spacing and the current frame in the CCD is shifted by a small amount. The shutter is then re-opened for the next exposure, sampling the light source at the new spacing. The CCD is multiply-exposed 100-200 times before it is eventually read out. This produces a long-slit spectrum similar in appearance to that obtained from a grating spectrograph. The method can be sped up by keeping the shutter open and shuffling the charge during the exposure. But a significant time constant, which determines the settle time of the plates, can lead to a spurious phase offset.
Figure 2 shows an example of a long-slit spectrum of a Cu-Ar lamp obtained using the red TTF and charge shuffling with the Tektronix CCD. The spectrum covers -788 nm and comprises 80 discrete shuffle exposures, only covering the top half of the CCD. This is because a fraction (n-1)/(2n-1) of the detector area must be sacrificed in a shuffle of n rows to allow space for the spectrum to be shuffled along as it is built. The lines are narrow because the TTF was set to a large gap (12 m). The curvature of the lines is due to the combined effects of radial phase change and scan increment. It is emphasised at large gap (higher orders of interference) or when small increments in spacing are used between successive shuffles. Once parallelism has been established at large spacing, the plates retain their alignment irrespective of how TTF is tuned thereafter. As shuffling is ordered top-to-bottom, the row number on the image is linearly related to increasing gap spacing between the plates. However, this does not necessarily imply that wavelength is strictly increasing over a full shuffle. Occasionally wavelength wrap-around occurs between adjacent orders of interference. To remove this confusion, we typically observe through two orders of interference.