Narrowband ghost images

While doing some experiments investigating the diffraction spikes around stars I took a 10 minute exposure of Vega.
I used a Hydrogen Alpha filter. A 10 minute shot of Vega is not done often because the star is so bright.
This is the image:

I noticed some ghosting so did a screen stretch:

The shadow of the secondary mirror and supporting vanes shows up in a regular array.
The absence of circular symmetry indicates that this effect is not due to the primary or secondary mirrors, or the coma corrector.
It must be some artifact produced by the camera (QSI 683 wsg8)

I posed the question over on PAIG in the technical section and 'Cosmophoton' suggested that the effect is produced
by reflection between the ccd and the underside of the filter. I struggled to see why a rectangular array was
produced rather than a circular effect until I realised that the multiple pixel cells on the ccd surface could be acting as a diffraction grating.
Rectangular pixels would give a grating with north-south and east-west grating lines.

The QSI camera has an 8300 sensor and this has pixels 5.4 microns square.
The hydrogen alpha filter transmits light at 656 nm wavelength with a bandwidth of 5nm.
Light between the filter and ccd is therefore virtually monochromatic.

Diffraction grating theory says that the angle theta at which the light is reflected will be governed by SIN(theta)=order x wavelength / d
where d is the spacing between lines.
A calculation on a spreadsheet shows that the first order angle should reflect at 6.98 degrees,
and for this to produce ghosts spaced by 963 pixels, the distance from ccd to filter should be 21.252mm

The QSI data says that the distance is in fact 17.53mm, but the filter has a plastic surround so the actual distance will be greater than 17.53mm.

Similar calculations predict that an Oxygen III filter at 501nm wavelength, should produce ghosts with a spacing of 733 pixels.
A Sulphur II filter at 672nm would have spacing of 987 pixels.

The next clear night was used to take images with these filters. The results strongly support the diffraction grating theory.
The Oxygen filter had a spacing of 729 pixels and the Sulphur was 977 pixels.

This animated Gif shows all three filters:

Ghost spacing for Ha and SII is similar, and OIII clearly has ghosts much closer together.

These ghosts provide an easy way of determining which filter is which.
Some time ago I forgot which filters were in filterwheel slots 6 and 7, and transposed the OIII and SII filters in Maxim.
I only spotted the mistake when I tried a Hubble pallette image and the colours were most peculiar.

A colleague made a similar error, and to make matters worse, his filters had no labels on them.

I carefully measured the ghost spacing, and in agreement with the diffraction grating theory, there is a
linear relationship between filter wavelength and the spacing.
So, a quick check of which filter is which can easily be made.