Howtek D4000 vertical streaking

[georgl]

I recently bought a Howtek D4000 drum scanner, the first scans were fine. I cleaned the drum and lens and then noticed vertical streaking. I noticed that pads on the right side of the drum were missing and I switched to the other, nearly new drum - now it runs much smoother even at higher speeds. Pads, calibration strip, belt - everything looks fine, but I still get streaks?

What could this be? How can I solve it? Could there be dirt on the encoder? It doesn't seem to be a serious hardware issue - works 18 years and fails the second day I own it...

 

[pfigen]

The first thing I would check is the drive belt on the left side of the drum. Those cause more problems than anything else. Make sure it's not cracked or slipping. I usually put a new one on every six months regardless. Also, make sure that the lead screw is clean and has been lubed with the special lubricant that Aztek sells. If none of this helps, call Evan at Aztek. He's very knowledgeable, friendly and helpful.

And, never scan without three good wear pads on the drum. When I had a 4500, those were always coming off, but I've never had a single issue in over nine years with my 8000.

 

[georgl]

I've already ordered the Klüber lubricant - although the streaks would be showing in horizontal direction, wouldn't they!?

The belt makes sense, it looks fine, it doesn't seem to slip - but the streaks are pretty small (<0.1mm on the drum?) and the belt is 14 years old...

Thank you!

 

[Lars]

There are many possible explanations for streaks.
- Mains power fluctuations can affect brightness of scanner lamp
- dirt in the calibration area can affect whitepoint calibration
- mechanical vibrations, caused by various problems such as pads, belts, worn bearing etc.

The image you posted doesn't tell me much unfortunately. Can you describe the appearance of the streaks, or post a better sample image?
When you say vertical, do you mean in the direction of rotation or parallel with the drum axis?
Are you scanning film or reflective media?

When it comes to pulling detail out of deep blacks, these scanners are not so good unfortunately - you only get so far with 10 linear bits and the technology is old and (comparatibely) noisy (my Howtek 4500 was manufactured in 1991, it has a 386 in it!!!). I took my Howtek to Aztek in L.A. for a complete service before moving back to Europe, definitely worth the $1500 cost but I still can't pull out nearly as much shadow detail as with a Nikon LS-8000 w 16x multisampling or a high-end Imacon or Hasselblad, or a new Aztek Premier for that matter.

Belts tend to crack within months - I remove the belt and put it in a ziploc bag when not using the scanner for more than a few days, to get some more life out of it.

 

[pfigen]

Lars - If your 4500 won't record more shadow detail than any CCD scanner, then you probably need new PMT's. They wear gradually with use and will need replacing. It's a bit of a chore, but it's not that hard. Make sure to get the Hammamatsu (sp?) version of the 931B tube. The last time I checked they were only about $70US.

I don't know about the 4000 per se, but all of the other Howtek drum scanners to use the same scanning electronics, pmt's and overall optics, so there shouldn't be any real difference other than maximum resolution between models. Well, almost. Because the Aztek Premier/Howtek 8000 model spins at a higher RPM, there is a greater smear factor in that model in areas of maximum adjacent contrast - clear drum next to film rebate - an effect that is present on all drum scanner to some extent.

My 4500 was every bit as good than my 8000, and the 8000 had been meticulously gone through by Evan and was for all intents and purposes the same as a Premier, with the improved bearing and custom alignment.

Hammamatsu does make a side window pmt with the same physical dimensions as the 931 B, except that it has simultaneously greater sensitivity AND lower noise. The downsize is that they were about $250 per. I mentioned this once to Evan and he was aware of them but had never tried them. In theory, those tubes could improve the d-max further with lower noise for a relatively small price.

 

[georgl]

I've ordered a new belt and some other spare parts from Aztek, let's see how it works out, thanks!

Aren't the old Howtek-machines 12bit-scanners? Seems hard to push this color depth under real-world-conditions anyway.

I've been trying to decide for a film scanner since 5 years, had some consumer-models, tested Imacons, drum scanners....
The result was always the same:
the "better" consumer scanners (from Minolta to Nikon) were good enough as long as the focus was right (never over the whole frame!) and the density wasn't too high, either.
The Imacon Precision II was sharp but had the same issue with density and noise, the Imacon 848 (and the other "big" models like 949, X1 & X5) were great but out of my financial range.

The drum scanners are large, heavy and slow beasts. But once the slide is properly mounted,, you already got rid of most of the dust & dirt, colors don't seem to be far off from the very beginning and I can see into dense parts of the slide, I can barely see on a light-table! At 4000ppi you're close to the physical resolution limit, also 5400ppi-scans from consumer-scans are less detailed and exhibit more grain at the same time!
Since I also want to upgrade to 4x10"/8x10", there was simply no choice, it had to be one of the "beasts".

@pfigen
You can replace the PMTs? Isn't there some kind of alignment issue or is it achieved by the fibre optics (just like the lamps)?

Here is an old test-scan of mine, 35mm-Velvia with a D4000 vs. Imacon Precision II (which was already better than the Minoltas and Nikons with single-scan):

 

[Lars]

Nope D4000 and 4500 are 10 linear bits, which translates to less than 5 bits in a gamma 2.2 space like AdobeRgb, so it's not a whole lot of dynamic range. The aztek software can load a gamma curve into the A/D converter of the scanner and thus circumvent this limitation, but it's an expensive software (far more than what I paid for my Howtek).

The Nikon LS-8000 with 16x oversampling is truly remarkable WRT dynamic range, you get 14 bits with close to no noise at all. A 6x9cm MF scan takes over an hour and there is the issue with film flatness, so it's not in any way an ideal process. I wetmount on mine so flatness is not an issue anymore. Unlike with drums, flat wetmounting is real easy, just takes 10 secs, no tape needed.

pfigen - there is just no way that the 10 bits of a Howtek can compete in terms of DR with the Nikon's 14 bits. The math is simple, and it shows in the results. I've scanneed the same dense slide on both, and the Howtek is all blocked up and noisy. Mind you, the noise is very far down in the shadows, only visible when intensity pulled up by eight stops or so in photo editor. With the same treatment, the Nikon scans show almost no noise at all, and lots of clean shadow detail.

Re PMTs - mine were replaced at the last service. It's still noisy.

I feel almost passionate about my drum scanner - the 1.8GB scans of my 8x10" slides are gorgeous, with incredible detail. But I trust what I see when I compare DR in scans.

 

[georgl]

Where do you get the 10 bit from? Howtek clearly states 12 bit in every document I saw.

 

[pfigen]

Lars - If what you are saying was in fact true, there would be no way to get even a really crappy scan from any Howtek, when in fact, even pulling 8 bit per channel scans yield files that you can yank and twist without any banding even when outputting to sensitive media like LVT. A five bit scan would be banded from the getgo the same way many 6 bit laptop screens are.

That your scanning results are the opposite of everyone else who has done CCD to PMT comparisons only means that there is something seriously wrong with your scanner. The PMTs used on all drum scanners have very similar abilities to record shadow density. They all clock in at around 3.9 plus or minus. Howtek's 931B's are no different. There is a way to falsely limit the d-max if you make an input profile incorrectly by not using an infinite density patch in the equation. But it's the pmt itself that provides the density range and the bit depth that determines how many steps are available.

It's been a while since John Panazzo and I had the bit depth conversation when I was a beta tester for Trident, but I do remember asking him about the internal bit depth vs. the exported bit depth, and he basically told me that there was bit depth, and then there was bit depth - the quality of the bits from the pmts was that much better than ccd's and even the "limited" 12 bits per channel were far better than, say the 14 bpc coming from an Imacon. And of course, all you have to do is look a how crappy Imacon's are in the shadow's to confirm that.

I haven't used a 4000, but I've owned a 4500 and own currently a 7500 and an 8000, and there is simply no visible difference between the three, even though the 8000 is claimed to be 16 bit throughout. It was my understanding that the 4500 was 16 bit internal and output 12 bits.

 

[georgl]

It took me nearly one month to get my hands on new belts and lubricant - to make a long story short:

It didn't work!

Still these vertical streaks! What can it be? The new belt is noticeably less lengthened, it thought this might be the issue.

 

[timparkin]

It took me nearly one month to get my hands on new belts and lubricant - to make a long story short:

It didn't work!

Still these vertical streaks! What can it be? The new belt is noticeably less lengthened, it thought this might be the issue.

I'm pretty sure this is the encoding ring. The vertical streaking is a lack of absolute positioning in the rotational direction.

Tim

p.s. Howtek is 12bit and combined with DPL produces some great results. Also, I use two custom profiles (one for the shadows and one for highlights) and combine them together to produce a true 16+ bit scan. Alignment hasn't been an issue and in a couple of cases I've created a third custom profile scan to extract exceptional detail (e.g. use all 12 bits for the top or bottom 10% of the image).

Just to be clear - DPL Pro custom profiles create new hardware settings, redistributing the 12 bits anywhere throughout the image channels that you like (hence why it does great neg scanning - you only scan the actual range of data - not lots of empty colour space)

 

[Professional]

Is the Flextight X5 scanning better than say Howtek 4000-8000? better mean the quality of the scan file not the speed of scanning, and i am talking about scanning @2400dpi, X5 can't scan at 4000dpi for larger format, so i was talking about MF and often with 4x5, any recommendations?
About what size i want to print if you may ask then it can be anywhere between A3 up to 40x60", i am not thinking to go over 40x60" which is too much for most times, but this size is there when i want to do [i did have some prints larger than 40" wide, i loved it even the quality of the print because the pic size isn't perfect].

 

[pfigen]

For the original poster, your best bet is to send an email to Evan at Aztek, along with your sample scans and see what he has to say. He would be the best person to help you.

I used to go round and round with Phil Lippincott in regards to how DPL used the hardware. He was adamant that the way DPL used the linear amps was superior to how Trident used the log amps, and would put DPL into log mode to try and demonstrate that. That produced a scan that was about four stops brighter than the "linear" version, and, of course, it looked very noisy in the shadows. He could never understand that I would NEVER ever scan that open and would push the shadows back to where they were supposed to be. Bottom line was that after a solid month of doing side by side comparisons (and having to pay a non-refundable $800 for the pleasure of Haddon's company teaching me the software) it was quite apparent that Trident produced superior scans in almost every way - AND it was properly color managed as well. When you make equivalent scans with DPL and the linear amps or Trident with the log amps and they had roughly the same white and black points, there was no difference in shadow noise, and smooth out of focus tonal and color transitions were smoother with Trident. In addition, the slight blooming or halation you would get around 35mm sprocket holes was actually worse using DPL. The ONLY solution to that is to tape off the sprocket holes with black or silver tape. Every drum scanner exhibits that phenomenon to some degree, but the 8000 is the worst in the Howtek line.

Anyhow, aside from having to run OS9, I have no complaints whatsoever.

 

[georgl]

Short update:

I haven't got response from Aztek from a mail I send several months ago. Still haven't found a solution, it may be a serious hardware problem (encoder)!? Thanks for the advice anyway.

I've used Silverfast multi-scan as a work-around to get rid of the streaks - takes forever...

But I can't scan color negatives, the pre-scan looks fine, the final scan is completely off! Several stops too bright, nearly indistinguishable from the negative itself! Changes before final scan are not applied!?
I have re-isntalled the scanner and Silverfast two times - no success.

Trident doesn't work with XP, my PC doesn't have drivers for 2000...

 

[Lars]

Sorry to hear that.

re Silverfast: All my scanning with Silverfast is for positives. I actually scan HDR mode, which captures the raw bits from the scan, so I can apply a color profile later. I don't know if that would help negative scanning - an advantage of raw scanning would be that you can apply the neg conversion later so you don't have to rescan to experiment? Assuming Silverfast supports neg conversion of a HDR scan of course. I haven't tried this.

BTW my apologies for my silly mistake re bit depth above, of course the 4500 has 12 bit depth. Still not a whole lot but 10 bits would have been unusable.

 

[Lars]

Lars - If what you are saying was in fact true, there would be no way to get even a really crappy scan from any Howtek, when in fact, even pulling 8 bit per channel scans yield files that you can yank and twist without any banding even when outputting to sensitive media like LVT. A five bit scan would be banded from the getgo the same way many 6 bit laptop screens are.

That your scanning results are the opposite of everyone else who has done CCD to PMT comparisons only means that there is something seriously wrong with your scanner. The PMTs used on all drum scanners have very similar abilities to record shadow density. They all clock in at around 3.9 plus or minus. Howtek's 931B's are no different. There is a way to falsely limit the d-max if you make an input profile incorrectly by not using an infinite density patch in the equation. But it's the pmt itself that provides the density range and the bit depth that determines how many steps are available.

It's been a while since John Panazzo and I had the bit depth conversation when I was a beta tester for Trident, but I do remember asking him about the internal bit depth vs. the exported bit depth, and he basically told me that there was bit depth, and then there was bit depth - the quality of the bits from the pmts was that much better than ccd's and even the "limited" 12 bits per channel were far better than, say the 14 bpc coming from an Imacon. And of course, all you have to do is look a how crappy Imacon's are in the shadow's to confirm that.

I haven't used a 4000, but I've owned a 4500 and own currently a 7500 and an 8000, and there is simply no visible difference between the three, even though the 8000 is claimed to be 16 bit throughout. It was my understanding that the 4500 was 16 bit internal and output 12 bits.

Late reply... yes I was mistaken in my comment, I'm not sure where I got the 10 bits from, of course it's 12 bits out.

Whether the internal processing is limited below 16 bits I don't know (the processor is 16/32 bits so it wouldnt make sense to artificially cut bit depth there), possibly the A/D converter has a limit of 12 bits?

My comment on noise floor with my Nikon 8000 scanner was based on (and I emphasize this) 16x multisampling. While this doesn't compensate for obvious shortcomings of a line CCD scanner like stray light, the effect on lowering the noise floor is quite amazing. In essence, the 14 bit depth is fully utilized, yielding close to a true 4.2 DR. Now, multisampling is a software feature rather than a property of the underlying technology so in all fairness if I could do proper multisampling using my drumscanner software then my drum scans would probably benefit most significantly. As georgl mentions, Silverfast supports multipass scans, however last time I checked this means doing multiple complete scans rather than multisampling each scanline before stepping the drum, which leads to alignment issues as the wetmounted film cannot be expected to be completely stable over time.

It's possible that my 4500 is not up to par. When I had it at Aztek for service Evan hooked it up with DPL to show me that nonlinear curves for the A/D converter would improve the noise floor situation. It didn't. We agreed that the logical conclusion of was that the noise floor is produced before the A/D converter - in the PMTs, wiring, or lamp voltage variations. PMTs had just been replaced, and noise floor seems to be the same at Aztek as in my home in Palo Alto and my home in Stockholm, so I can't really blame the voltage feed either. My observation then is that even my fresh PMTs don't produce noise-free 3.9 DR - it's more like 3.3-3.6 when noise floor is taken into account.