Did some calcs to see how some cooled mono cameras might do for lucky imaging. Evaluated only for broadband signal-to-noise ratio.
assumed conditions: 21 mag/arcsec^2 target against a 20m sky. Scope was C9.25 at f/10 for the initial calc and the total exposure time for all was 1 hour.
The results depend on how the cameras are assessed - the following are 3 different methods.
Method 1: bolt the cameras onto the C9.25 scope and view the resulting images at native (raw) sampling scale with no resizing of the images.
The big pixels of the 174 catch >4x as many photons as those in the the small-pixel chips, so the SNR is much better in the 174 raw images. The 1600 pixels are intermediate in size, as is the SNR. The cameras with small pixels are way down, since they collect fewer photons.
These results may explain why cameras that have theoretical advantages may not seem to perform well when they are compared with other cameras. Bolting alternative cameras onto the same telescope and then eyeballing the raw images is not an informative way to compare cameras.
Method 2: Software resample the raw images from method 1 so that the the pixel scales match (in this case, to the same scale as the 174 at about 0.5 arcseconds per pixel). The resampling algorithm is assumed to give a "partial binning" capability - eg results for 2x downsampling will match 2x2 software binning.
Software resampling significantly improves the SNR from the small-pixel sensors, because data from many pixels is combined into each pixel in the resampled image. This brings the small-pixel cameras back to about the same level of performance as the 174.
Method 3: Vary the focal length of the telescope so that the native sampling is 0.5 arcsec/pixel on each camera. Of course, this cannot be done easily with a C9.25, but the test is really intended to illustrate the advantage of getting the sampling right in the first place. The aperture of the modelled telescope was kept at 9.25 inches, but the (hypothetical scope) focal lengths for the cameras were 2350mm for the 174 (f/10), 1550mm for the 1600 (f/6.6) and 1000mm for the 178 and 183 (f/4).
Getting the sampling right may provide an additional and very significant improvement in SNR with short subs, when compared to adjusting the sampling in software. This conclusion applies in any situation where read noise is dominant (eg it could also apply to narrowband conventional imaging with a slow scope). It is way better to adjust the sampling before detection rather than oversample and then downsample the image after detection.
Getting the sampling right is very advantageous for the small-pixel cameras and the 178 is the best of the group, The 183 and the 1600 both do relatively well, but the slightly higher read noise of the 183 causes a roll-off in performance at 1 second (and shorter). The lower QE of the 1600 works against it, but it's low read noise is helpful with short subs. The 174 is the worst performer overall, particularly with short subs - a complete reversal from the simplistic test 1.
One further consideration is that the maximum scope size will be limited by the small pixels in the 183/178. eg if the required resolution is 0.5 arcseconds, the focal length for these chips will be 1m and the biggest practical aperture will be be about a 250mm (f/4), since true binning to increase the effective pixel size is not available in CMOS (although software binning can be used with lower efficiency). Larger scopes could of course be used if finer resolution was required. However, the 174 may be used on much larger scopes for any given resolution and this may be a consideration - this possibility has not been included in the analysis, but if you happen to have a 20 inch f4, forget the small-pixel chips..
Some things to consider maybe:
- the native scale images straight off the chip (on the same scope) will look way better from the 174, even though the other chips outperform it when optical sampling is adjusted - eyeballing raw images can be a misleading way of assessing sensors .
- the 178 is clearly the best of this bunch when used with short subs and appropriate (optical) sampling - and at apertures up to about 250mm.
- for lucky imaging, software resampling (including software binning) is, at best. a halfway house between the raw images and the images you could get with proper optical sampling - it is way better to try to get the sampling right in the first place, rather than oversample and then adjust it in software..
Thanks for looking - would welcome discussion. Cheers Ray
Edited by Shiraz, 27 June 2018 - 05:37 AM.