FS700 sensor read out?

[David Heath]

Originally Posted by Alister Chapman

The read out of the high speed modes is very different to the standard speed modes whether raw or otherwise. In the high speed modes the camera combines pixels or pixel skips depending on the frame rate, hence the aliasing.

There's no debate about that, fully agreed. At 240fps, it's evident that the basic resolution unit is a 4x4 bayer block, hence real detail turning into aliases at 540lpph.

I'm inclined to think it must be skipping to give the alias levels it does at 240fps. (So reading 2x2 blocks with skipped rows/lines in between) If it binned the 16 values in the 4x4 block to give an R, G, and B value, the spatial averaging due to the binning would give a similar effect to an OLPF.

Aliasing is dependant on the sensor resolution. ......... if you read the sensor either directly or via the DSP by combining pixels 2x2 then in effect the sensor resolution is now 2K so strong aliasing would be expected at 2K and HD, but we don't see strong aliasing in the HD modes.

This is where I disagree, at least in so far as "strong" aliasing goes. If you do the "2x2" readout, then the limiting system resolution will be 1080lpph - but the much finer photosite structure will tend to iron out or average out the stronger aliasing effects that you refer to - as long as no skipping takes place. Look at it this way - even at 1080lpph, vertical lines will still straddle two complete sensor columns. That effectively leads to a spatial averaging until you get to much finer detail, closer to when you get a single line per column.

And as physical proof I'd come back to the C500. It must have a 4K optimised OLPF, otherwise it wouldn't work at 4K. But it uses the same (2x2) read as the C300 for HD - and manages to give a good result without excessive aliasing. This doesn't surprise me for the reasons in the paragraph above. What I refer to as "spatial averaging" acts (in HD mode) in a comparable manner to an OLPF for frequencies a little greater than HD - the true OLPF then being effective at the higher frequencies.

All that said, then it is true that the image IS being sampled at 1080lpph (if 2x2 decoding), so I would expect to see some aliasing due to that - just not very much. And I come back to the chart by Adam Wilt that I referenced before, and the converging vertical lines. That they stop converging, and start to diverge, is a clear sign of *some* aliasing, and it starts happening around 850lpph - at the 1000 figure they are clearly diverging - the results are aliases. (And before anybody says it, not strong enough to give much effect on real pictures, but academically they are very significant for giving a clue into the inner workings.)

OK, from the observed results it is CONCEIVABLE that the signal is being fully deBayered, is being subsampled at a rate corresponding to about 850lpph electronically (hence generating the aliases we see), then upscaled to normal HD. That's conceivable (would explain the observed results), but would be one of the craziest bits of camera design it's possible to imagine - hugely complicated and when a simpler approach would give a better result much more easily. I don't believe it.