QTR-Quadtone RIP

Tyler Boley wrote:

>--- In QuadtoneRIP@yahoogroups.com, Ernst Dinkla <E.Dinkla@c...> ...
>snip...
>  
>
>>The targets you mention are the ICC profiling targets that usually
>>    
>>
>build 
>  
>
>>on linearised inkchannels. Linearisation targets exist in many shapes 
>>CMYK-CcMmYK-Ncolor and they have ranges of 11 to 41 patches per 
>>inkchannel. the lowest number can be used for printsystems that
>>    
>>
>actually 
>  
>
>>address the heads separately but on a CMYK driver that is linearised as 
>>CMYK but drives a CcMmYK printer it is better to use the 21 patches so 
>>there are 10 steps measured per head. With the 4800 that has 3Ks you 
>>could go up to 31 steps and that has been used for Colorspan models
>>    
>>
>that 
>  
>
>>have 3 ink dilutions of CMYK each = 12 heads.
>>    
>>
>
>My impression of the process is a bit different, correct me if I'm
>worng. Individual ink components of a color, C and c for example, or
>the 3Ks you mention, are generally not addressed individually in the
>linearization process.  Some control is usually given with respect to
>how the transitions will happen, but then the color (all inclusive
>"C", or"K") is linearized as one. 
>
I thought that I described that in similar words.

>How exactly the total patch  count
>may divide up between those indivicual components may not be very tidy.
>  
>
Whether it exactly divides up to equal linearisation points per channel 
doesn't matter much. But if you check for example the linearisation 
correction curves of an Epson 9000 then I wonder how 3 points per head 
would be sufficient. And that printer isn't using different droplet 
sizes per head. That the linearisation in the end has to set 256 shades 
right in all setups doesn't mean that there will be no differences in 
its task between different B&W inksets/printers. The extreme case of 7 
non linearised head curves bulging upwards  including 2 droplet size 
curve nicks per head + 6 points where the curves have to join, will not 
happen I guess because the partitioning and overlap in the partioning 
point cuts droplet sizes, smooths both droplet size transfers and the 
partitioning point tranfers. The upper part of most of the non 
linearised head curves is cut off anyway as the density isn't increasing 
there. Where the smoothing happens, where some droplet sizes vanish is 
hard to predict, we partion on the max density per channel and where it 
fits the black channel density. A bit of a black box there, which may be 
better controlled by 50 linearisation points than with 20. That the 
partitioning has an important role to play is clear to me. The trend 
towards K7 inks isn't one I see as a solution and I think that using 
more heads + less dilutions of black  ink + effective use of droplet 
sizes and actually using the mix you get from heads weaving (as intended 
in the printer design) will in the end deliver less banding, better 
linearisation, more consistency in time. I have written this before in 
relation to BO printing and on the K7 threads. There are indications 
that Epson did change the weaving for the Adv. B&W driver in the K3 
printers to give better B&W printing. So far we didn't address that 
aspect much (Paul Roark did however in his distribution of the quad 
inks, then more or less directed by the limits of the Epson driver), and 
thought it would be perfect to add black dilutions where the head 
quantity increases. Let's say it is like putting the 4 wheels of a car 
in line, has some advantages but not in all aspects.

>I'm sure this varies from RIP to RIP. I recall Roy asking if we
>thought there would be an advantage to individul ink linearization,
>and wonder the same thing myself.
>  
>
I do not think that's wise either. Adding the separation curves after 
linearisation introduces the imbalance of the curves themselves.  
Profiling to the perceptual curve could smooth that out of course. It 
will also take more time to make curves for new ink/paper combinations. 
I have been on that route and I do not want to go back.

>  
>
>>21 should be good for B&W systems K2-K3-(K4).
>>    
>>
>
>I have on occassion seen an advantage to doing many more here with 4,
>and wound up at 80. I've seen a higher count like this fix problems
>between say 95% and 100%, more than once.
>  
>
It must be quite difficult to get the readings right for 80 steps but 
averaging readings on more targets before computing helps.
Something I did throw out in my first message: the steps in the 
linearisation target/software are fixed on for example 5% increase. A 
target of 30 steps with some step variation where it counts would keep 
the target number down but increase the linearisation quality. But like 
you mention the partitioning differences in quad setups may make it 
difficult to see where it counts to have a higher frequency in steps. 
The shadows are without doubt a rewarding spot to have more steps for 
measurement. Maybe a separate profiling target could get another step 
frequency related to the perceptual values. It will diminish the 
iterative function of the measurements but make the last step more 
suited for the task. It probably would require higher step frequency in 
the linearisation target where it builds on to keep the steps you print 
for the profile making more related to the first readings (and less on 
the interpolated parts of the linearisation curve).

>  
>
>>But for a K7 set it means 
>>3 measure points per ink channel (head) which is too low in quantity.
>>    
>>
>
>The above may apply here again. It seems, under the curcumstances, we
>are somewhat at the mercy of the partitioning scheme. If it's  very
>well behaved and designed for the ink set and media, a good overall
>linearization (with an adequate patch count as you are discussing)
>should work well.
>In fact this seems to be what we are all doing for now anyway. The
>downside is that a less then stellar partition scheme can be bad
>enough that overall linearization will not fix it.
>With what most people are using, that doesn't seem to happen much.
>  
>

>  
>
>>The CMYK linearisation of the Wasatch SoftRip has 14 patches per color, 
>>on the Cc and Mm inkchannels of the 9000 it means 7 patches per C,c,M,m 
>>head and I think that's about the lowest you can go.
>>    
>>
>
>So it does allow you to linearize each color component individually?
>  
>
No, it takes the C+c channel together but limits the linearisation to 14 
patches so 7 patches for C and c individually if they are distributed 
equally to both sides. With (extra) linearisation done with other 
profiler software you could add patches but internally it is not possible.

>  
>
>>The ICC profiling done with the 21 step target on top of the 
>>linearisation done with the same 21 step target may work as an
>>    
>>
>iterative 
>  
>
>>measurement as well and so average the result a bit (it produces
>>    
>>
>another 
>  
>
>>curve of course) but whether that is sufficient for B&W sets above K4 
>>isn't clear. I think there's improvement possible.
>>    
>>
>
>no doubt. But the partitioning step needs to be considered as a major
>part of the overall performance picture. How many patches and how much
>work needs to be done at the end can have a lot to do with how well
>behaved the ramp is to begin with.
>That few have trouble with QTR speaks well of Roy's existing
>partitioning scheme. Setting it up to pull 7 inks in and out without
>major problems must have been fun, or not!
>  
>
I have no 7 ink experiences as I have written before but having done my 
share of experiments I think that there are more ways to get smooth prints.

The linearisation etc of QTR is very user friendly and has been 
sufficient for me so far. The almost custom quad inks + two different 
hue toners distributed in another way on the 9000 that I have were easy 
to get in line. The prints are not the smoothest around given the 
printer and number of k inks and one toner but I have done my best to 
get the linearisation right and customers are satisfied. That printer 
with its limited nozzle count and differences between head output 
learned me something about partioning and linearisation.

I'm trying to get a picture of what is possible in linearisation. Where 
finer step differences still contribute to the result or get in conflict 
with the consistency etc of the spectrometer and the repeatability etc 
of the printer. The partitioning is important, the distribution of the 
inks, the number of heads and nozzles as well. I'm not ultimately 
seeking higher target patch numbers but whether the linearisation and 
consistency of printing can be made better, in repeatability, banding, 
detail etc as well. Setting the target step number against 256 shades is 
one thing, setting the number against the number of heads/nozzles/inks 
involved is another thing. If all 7 heads were used for BO they even out 
their differences and the linearisation curve would be a simple one I 
guess. Not to mention the one droplet size the printer is forced to then.

Ernst