Linearization Problem 9900

Hi guys, so as you know, we’re working with a 9900 with a head that we replaced after you (walker) left. The new head is missing orange, which houses very light warm, so no problem. However, I am having an impossible time getting this thing to be perfectly linear, and sometimes even close to linear.

When I replaced the head, I was unaware that the only zeros and ones are used in the head rank ID. I am not sure if I put them in as the letters “o” and “i” but I imagine that could really mess things up voltage-wise. Once I realized this, I went back and made sure to put it in correctly (even scanned the little QR code to verify the ID was correct).

The initial 129 reads for most curves look like this at best:


After printing another 129 with that curve, I got this:


Even when it reads as very close to linear, I still get visible banding in 16bit gradients, especially in the transitions from RGB 10 - 0 values. Here’s the photo rag curve that looks linear, but bands still. Printed using a 51step linearized curve:


Doesn’t seem to matter if I do single pass 129 lin, or 51 / 129 lin, this still happens. I am using this image to test print. It’s a ~5.5x5.5" 16bit GG2.2 image generated using the gradient tools in PS CC 2017.


I have also tried different computers with different OS, different USB cords, re-entering the head rank / clearing the head counter, all readings are very carefully read via i1pro2 colorport spot reading (I’ve even tried calibrating three times throughout the 129 chart). Some are better than others, but after seeing first-hand the possible smoothness of the gradients achievable with this system, I really want to get our system optimized to that level!


Link to the full size test file I’m using… wish I could capture the banding being printed, but it doesn’t really translate properly.



HPR308_Cool_129.txt (2.13 KB)

EdEtch_C_129_InitialReading.txt (2.11 KB)

What is your dry-down time between printing and linearizing?




Looking at your raw measurement data it looks smooth enough on the initial readings to work just fine. Your Error Corrector could still be modified a little to your specific device to smooth out the readings a bit more (this is partly why we host the Error Corrector on Google because we can modify each one specific to a user’s device if need be). My thinking is that you possibly DO have an error in the print-head voltage. Does the un-linearized K5 Cool .quad that comes with Community Edition band as well? If so, it’s most likely the head.

I suggest building an ICC profile using the new Piezography ICC Tool published with PPE 1.2 and see if this smooths your tonals out. This uses an entirely different set of algorithms (inside of Xrite i1Profiler) as well as more than 256 grey steps.

I am finishing the documentation for this tool this week (while making ink) but the QuickStart PDF included with the beta tool inside of >Piezography>Tools>Piezography ICC Tool> should be enough to get you rolling.

let me know if this fixes it. If so, I will modify your Error Corrector a bit.


Also, L* of 12.14 on Hahnemuhle Photo Rag with UltraHD MK ink. That, I think, is a new world record!

dMax of 1.85. WOW



Hey Walker, thanks for your reply! I fear you might be right about the head… they really should make the damn thing easier to read! Well, guess that’s what the QR code is for… I have tried both overnight dry and hairdry, and even both, all with more or less the same results. I al also re-installed QTR and PCE per your suggestion. The photo rag and ultrasmooth curves I’ve made are almost perfect, but still sliiiight shadow gradient posterizing despite the curve reading as lin. Yes, the unlinearized stock curves band quite severely.


So basically I can make an ICC that will still allow it to print linear by correcting for the banding values more or less? We use ICCs now for some clients, but they’re more for screen matching.


SAY-- we are getting a new head in our 9890 because the LK crapped out. If George (epson tech) is right and the heads are in fact the same on 9900/9890, would it be worth swapping into the piezo and remapping?