ELPH 115 IS uses a CCD?

Gonini wrote:

Whats wrong if it uses a CCD?.... Quite a few P/S still use CCD.

It's not so much that it's wrong to use a CCD; they might be an older technology, but still can make excellent sensors with high QE (quantum efficiency) and no rolling shutter effects. However, CCDs tend to be more expensive to fab and it is harder to do things like live view and video.

A CCD is an array of analog shift registers in which you generally have to read every cell to read any cell in a line. In contrast, CMOS sensors easily could have individual pixels (sensels) directly addressable. The live view on the camera is nowhere near the (16MP) sensor resolution, so the refresh rate can be increased dramatically by only sampling a tiny fraction of the sensor pixels (sensels) for it. The same is typically true of video -- you don't need 16MP for 720 HD video, and very few cameras shooting video actually look at all the pixels and scale the image down.  Judging by the performance of the raw sensor system as exposed using CHDK, the A4000 IS and ELPH 115 IS apparently are not capable of filling the sensor's raw buffer at the live view or video frame rates. Further, video and live view image capture go through very different code paths from still capture in the camera's firmware and live view doesn't overwrite the sensor's raw frame buffer (and there isn't space for a second full-resolution raw frame buffer in the camera's memory)... which strongly suggests that only a tiny fraction of the sensels on the  sensor are being sampled for live view -- again, something CCDs normally can't do.

In sum, if it really is a CCD, there must be some interesting readout logic somewhere. Perhaps the readout is fast enough that just skipping lines works, but that would imply still reading an entire line when you only needed a tiny fraction of its values. I believe that's usually been the trick, but that's also why a lot of early cameras had terrible live view refresh rates. Maybe line readout is now fast enough? I don't see image quality signs of live view and video being computationally scaled down -- scaling would improve S/N ratio, but noise levels appear no better than if only a sparse sampling of sensels was performed, so perhaps digital logic is only saving the values for the desired pixels?

I'm using these cameras in a computer engineering course I'm teaching (largely using CHDK) about the computing aspects of cameras, so I'd really like to know what is going on at this low level. Obviously, from an ordinary user perspective, you shouldn't much care what sensor technology is used as long as the camera does its job well enough....  Perhaps I should have posted this in the tech forum or on the CHDK wiki?