[opendtv] Re: Precision

Hi, 

I'll give Dan Grimes a partial answer (or is it a non-answer ?): 
> First of all, bit depth: 8 bit, 10 bit, 12 bit, higher? 

It depends on a lot of things. 

It depends on the noise level. For digitizing a noisy analog signal 
there was a time when 8 bits was enough. For clean computer-generated 
signals you need more than 8 bits, or else you will see contours in 
shallow slopes (like a sky with non-uniform brightness). 

It depends (very much) on the coding, whether it is perceptual or 
not. The most famous perceptual coding applies a power law with a 
gamma of approx. 2.4, and you can use 8 bits or perhaps 9 or 10. 
The extreme opposite is linear-light coding (a gamma of 1.0) and 
you'll need approx. 14 bits to get the same quality near black. 
In-between you have e.g. the signal that goes into a (deliberatly) 
badly corrected LCD panel, and you're supposed to apply further 
correction with a LUT before the panel interface. This is where 
you'll see a 10 bits interface instead of the usual 8 bits, 
because the coding is not nearly perceptually uniform anymore. 

It depends on the spatial and temporal resolution, because a higher 
resolution gives you more opportunity for spatial and temporal 
dithering to move the quantisation noise to higher, invisible, 
frequencies. Dithering is a controlled form of additive noise, 
and like analog noise it allows you to use fewer bits ! 

It depends on the correlation between R,G,B channels. They can 
co-operate to reduce mostly the luminance noise, at the cost of 
increaing the less visible chrominance noise. 

It depends on the number of cascaded quantisation operations. 
8 bits may be enough for a single digital stage, like a D1 recorder, 
but it is not enough for many digital signal processing operations 
in cascade. The quantisation errors will accumulate. 

It depends on any conditions that affect the adaptation of the eyes. 
When the screen and ambient brightness go down, the eye will adapt 
and be able to better see the quantisation steps near the black. 
(This is where backlight dimming comes in very handy !) 

And it depends on the expectation for the money that you pay. 
You expect more from a paid digital cinema experience than from 
a free amateur movie rendered on a cell phone display. 

And I won't even go into (MPEG) compression artefacts, which are 
essentially also quantisation artefacts (in the DCT domain). 

Best regards, 
-- Jeroen

+------------------------------------------+------------------------------------+
| From:    Jeroen H. Stessen               | Phone:  ++31.40.27.40246  |
| Deptmt.: Philips Applied Technologies    | Mobex:  ++31.40.27.99650  |
|          Digital Systems & Technologies  | Mobile: ++31.6.4468.0021  |
| Address: High Tech Campus 5 - room 5.025 | Skype:  Jeroen.Stessen      |
|          5656 AE Eindhoven  - Nederland  | VoIP: 
Jeroen.Stessen.at.Philips  |
| Website: http://www.apptech.philips.com/ | E-mail: 
Jeroen.Stessen@xxxxxxxxxxx |
+------------------------------------------+------------------------------------+

Other related posts: