Re: [Openexr-devel] Query: Image Based Lighting and HDRI-enabled cameras

[Brian Willoughby]

[  12 bit may require that the imager be cooled, else noise may
[  overwhelm.  I have seen reports or a website or something at least
[  a couple of years back where a peltier cooler was used to improve
[  a ccd's noise floor enough to allow a better bit- depth.  IIRC
[  the pwoer requirements were too much for use in many portable
[  applications, notably including the consumer & prosumer markets.
[
[  If the power requirements can be controlled I am sure 12 to 16
[  bit imagers will be feasible.

I'm venturing outside my experience here, but I presume that the noise is only  
a problem with the current technology.  It seems that input devices in the  
next generation could be designed to reach deeper bit accuracy without being  
required to undo the high level noise inherent in the current technology.  That 
 
is, unless there is some law of physics which sets a rather absolute limit on  
how small of a signal can be detected.  It seems that in terms of sensitivity  
to individual photons, CCD chips already surpass optical film by a slight  
margin.


[  Getting floats directly from the imager will be an engineering
[  feat.  Are there *any* ADCs in the wild that sample into floats?
[
[  Perhaps sampling the log of what is currently sampled would do?

Many A/D chips employ successive approximation techniques which could  
potentially be altered to compute floating point readings.  This would require  
floating point D/A chips, since the existing successive approximation circuits  
I am familiar with employ linear D/A chips in the feedback.

Early audio D/A chips for CD had only 14 bits of accuracy.  Techniques were  
used to adjust the scale of the D/A such that the 14 bits of accuracy could be  
shifted to where they were needed based on the signal.  The upper 2 bits were  
used to determine the range of voltage output from the D/A.  Low-level signals  
had effectively 16-bits of accuracy, but as the signal rose above 1/4  
full-scale, the D/A shifted into coarser zones with 15 or 14 bits of accuracy.  
 
I believe it was YAMAHA who employed this technique.

It should be possible to use similar techniques to develop a float D/A, and  
therefore a successive approximation float A/D.

But this is all merely potential future technology.  My original point was  
that HDRI, in its preferred half-float format, will probably not be coming  
directly off the CCD chips.  Then again, I suppose there may not be much  
difference in practice between that and a camera which internally converts to  
half-float using DSP before output.

Brian Willoughby
Sound Consulting