ICE TUBE Q3 TEST: MOSFET or PNP TRANSISTOR Moderators: adafruit_support_bill, adafruit

[Russell 27]

BS250 TUBE.JPG (763.74 KiB) Viewed 1245 times

After tests for the chart, I wanted a field effect transistor that had turn on character similar to the PNP. I picked up a few more to test, The BS250 was the best. While just the slightest better than the recommended ZVP2110A, I wanted to see if it delivered as good as on paper. The original chart test was done with 5 volts, clock rail is 4.7. If you subtract the .3 volt difference from the chart, measured voltage in my clock is essentially the same as original chart. As mentioned it's difficult to photograph the tube accurately, I uploaded the image full size so detail could be seen, clock is set at 70 brightness. It's a decent representation.

ICE TUBE circuit.PNG (9.67 KiB) Viewed 1245 times

The circuit Image shows test points, and readings from BS250, as well as Ub voltage @ 70 brightness. The 2907 I originally had in place had measurements of Vt = 4.59v and Vr = 3.24v; subtract the .3 volts from original chart, and it matches those results. There was mention above that the MAXIM chip has minimum working voltage of 3 volts. Max chip is supplied from Vt, and none of these parts supply less than three volts. But again, if you subtract .3 volts from the ZVP3306A (Vt) chart data, Maxim voltage is right at minimum 3 volts logic, and (Vr) filament voltage is 2.25 volts. It's a safe argument that these combined conditions are most notably the cause of display problems with the ZVP3306A. If you like the result of the BS250, cost is 88 cents. I recommend it as another alternative to the ZVP2110A. One more note. Even with AC filament, I have noticed some digits are sometimes a little dimmer, this is visible in Phil's photos (right side). The clock I put together myself doesn't exhibit this as much, not sure why. I have a 30 year old DENON receiver and the VFD display is still beautiful. I imagine the quality of that display is of superior quality to these tubes. The PNP worked well, but there was also times when display was not as evenly lit as my old DENON. I would say this BS250 is very close to the 2907.

[jarchie]

Thank you for sharing your observations on the BS250. I'll have to pick up a few on my next parts order.

Max chip is supplied from Vt, and none of these parts supply less than three volts.

Did you read my previous post in this thread? It described how, on one Adafruit clock, the voltage supplied by the ZVP3306A (Vt) drops to below 3 volts when R3 is replaced by a jumper. Although I didn't mention it, I measured the voltage before and after replacing R3 with a jumper in a second Adafruit clock. I didn't write down the exact numbers, but do remember that the results were similar: the voltage supplied by the ZVP3306A (Vt) was greater than three volts with the 22 ohm resistor installed as R3, but fell to below three volts when R3 was replaced by a jumper. Do you believe my measurements to be wrong?

But again, if you subtract .3 volts from the ZVP3306A (Vt) chart data, Maxim voltage is right at minimum 3 volts logic, and (Vr) filament voltage is 2.25 volts. It's a safe argument that these combined conditions are most notably the cause of display problems with the ZVP3306A.

I agree that the dim digit issue is caused by too little voltage across the filament, but are you making the argument that voltage--and not noise--is the cause of the flaky segment problem?

Even with AC filament, I have noticed some digits are sometimes a little dimmer, this is visible in Phil's photos (right side). The clock I put together myself doesn't exhibit this as much, not sure why.

The handful of xmas rev d clocks that I built were also more even than the photos Phil posted, so I think that your clock is probably more typical.

[Russell 27]

I'll look into it. Chart results were what I found. While I connected filament directly to transistor Drain (Vt), jumper across the resistor should provide same potential of Vt where Vr was, current always follows the least resistance. Resistor would become invisible. As far as noise goes, I'll have to look at that also. I'll try the 128 bitclock setting, wasn't aware that rate was recognized.

[phild13]

Maxim voltage is right at minimum 3 volts logic, and (Vr) filament voltage is 2.25 volts. It's a safe argument that these combined conditions are most notably the cause of display problems with the ZVP3306A.

I agree with that and your original chart does show the ZVP3306A to be marginal. I think it is safe to say that it is due to the failure of the ZVP3306A to fully turn on as has been previously established in a number of posts on the forum.

After I fixed the Transistor version of the clock (damaged crystal) I found very similar numbers for the voltage on the 2907 (Vt) with a 1k base resistor. I have used an 11 ohm R3 in place of the 22 ohm on all versions of the clocks so the voltage at Vr will be different than what Russell has posted in the charts..

On a different clock the ZVP2110A exhibited very similar measurements at Vt to the chart. Again voltage was different at Vr as I use 11 ohm for R3

Without taking them apart again, I think the VCC is very similar on the clocks to the 4.7 volts Russell has measured. That the original chart used a higher voltage would have been my only gripe about the testing for that chart as it skewed the voltages higher by .3 volts. However the chart is still very useful as one can plainly see that the ZVP3306A is marginal even at a higher VCC voltage and one of the commonly suggested options of a transistor or upgraded FET for Q3.

Though the clocks each have their own personality, some of the differences I see in my clocks compared to others may not actually have anything to do with what components are used. Temperature is generally about 58 degrees where the clocks are, clocks are set to turn off between the hours of 2 am and 6 am, and using Johns firmware the display is set to auto dim with low of -5 and high of 20. These settings tend to give me the best range of brightness operation, though the display will become uneven at the lowest brightness levels. The colder than normal temps probably have some affect of emissions from the filament and areas of greater cooling within the tube would translate into greater visual effects. Heck, I have one tube that the minute digits are much more affected at low brightness levels (to point of almost not seeing a full number 11) than either the seconds or hours. The photos I took were in the morning just after the clocks turned on while it was still dark (no lights on/minimal from window) in the office.

All the clocks exhibit very similar digit brightness levels and evenness under normal room lighting conditions.

Though I like the pretty captures of the noise, I am not convinced (so convince me :) ) that noise has much to do with digit brightness unevenness as the architectures between Johns RevD and the Adafruit design are significantly different and so should exhibit different characteristics in any response to noise. Since John drives his filament to spec and uses a different method for dimming than the Adafruit version does, that would seem to further support that noise is not a factor.

Since I work nights and sleep days, I wish to have the room as dark as possible, but still want a display visible on the clock while I sleep and so tend to drive the clock to a dimmer value than others may use..

This does not matter for the subject at hand as it has no bearing on digit brightness unevenness, but in case anyone cares, one clock uses a 1K for R4 and the others use a 5.6K for R4. The difference just biases the response to light a bit differently. All clocks use the same photocell.

My opinions/observations here:
Johns Rev D board exhibits very similar display variation as the other clocks based do that are not of his design when at very low light levels. Since his design drives the tube and the Maxim chip differently, then one needs to look at other reasons for any display unevenness. I think that is just the way the tube is designed. If you go below a certain brightness level, the tubes are going to exhibit digit brightness unevenness.

Using a transistor/base resistor as recommended by Russell instead of a FET would eliminate damage caused by static to Q3 that affects some peoples clocks.

The R3 22 ohm may have been a design decision to limit filament current in the original transistor version of the clock because of the forgotten base resistor and later to compensate for the voltage drop of the ZVP3306A that does not fully turn on. Some evidence of the latter by Russells measurements of insignificant voltage (below 3 volts) at Vt when R3 is replaced with a jumper.

replacing Q3 with a ZVP2110A or a BS250 provides better operation as the voltage at Vt is higher.

changing R3 to 11 ohm provides more current to the filament and when used in conjunction with a recommended Q3 replacement provides for better overall clock operation.

changing R3 to 11 ohm or a jumper without changing Q3 to a recommended replacement will likely result in a flaky clock due to the voltage drop at Vt.

[Russell 27]

I used 5.0 volts for my original chart test, and mentioned that 4.7 actual Vcc would effect things slightly. While most of the transistors behavior does not change much with a .3 volt difference The ZVP3306 does. Others including myself mentioned earlier this transistor didn't have good turn on below 5 volts, I should have used 4.7 volts for earlier test. This test is complete evidence.

The Image below is set up as original chart test, like above ICE TUBE circuit image @ 5.0 volts Vcc. Meter on the left is Vt voltage, meter on the right is Vr. Image to the left is using the 22R resistor and image on the right is same circuit with resistor jumped. Very similar result to the chart.

5.0 VOLT TEST.PNG (784.3 KiB) Viewed 1206 times

Image below is same test with 4.7 volts Vcc, a .3 volt difference.

4.7 VOLT TEST.PNG (717.04 KiB) Viewed 1206 times

As you can see, with no resistor in circuit, voltage drops to 2.6 volts. Current rises about 5 or 6 Milli AMPS, but 3306 can only turn on to supply 2.6 volts in this lower resistance relationship. This is due to part behavior. I originally recommended Jumping the resistor for testing, and while a viable test, drop is much more than I expected, and useless really. Maxim chip is rated at 3-5.5 volts LOGIC Vcc. Does this mean the Maxim VFD driver chip will not function; not necessarily. I've used CMOS chips with a working voltage rating of say 3-18 volts Vcc, and circuit worked fine under 3 volts. The data sheet ratings are recommended, reliable specifications. As far as noise goes, I didn't see that much noise on my scope, I personally don't think it's a variable. I hope this finally clears things up. In all fairness, .3 volts in all but very strict circumstances would amount to nothing. No matter how you look at it, this transistor is a loser at 4.7 volts. At 6 volts I doubt this would have ever amounted to a discussion.

[jarchie]

I used 5.0 volts for my original chart test, and mentioned that 4.7 actual Vcc would effect things slightly. While most of the transistors behavior does not change much with a .3 volt difference The ZVP3306 does. Others including myself mentioned earlier this transistor didn't have good turn on below 5 volts, I should have used 4.7 volts for earlier test. This test is complete evidence.

Thank you, Russell! I appreciate your taking the time to test this.

Without taking them apart again, I think the VCC is very similar on the clocks to the 4.7 volts Russell has measured. That the original chart used a higher voltage would have been my only gripe about the testing for that chart as it skewed the voltages higher by .3 volts.

I believe there's more going on than a simple 0.3v drop. Lowering VCC doesn't just change the transistor input voltage. It also changes the gate-source and drain-source voltages which both affect how much the transistor is turned on. The voltage drop across Q3 also increases with current, so jumpering R3 has a significant effect--especially when the ZVP3306A is only marginally turned on.

Though I like the pretty captures of the noise, I am not convinced (so convince me :) ) that noise has much to do with digit brightness unevenness [...]

I never intended to claim that noise was responsible for display unevenness, but I am making the claim that noise is responsible for the flaky segment problem, which is distinct from issues with evenness (1, 2, 3, 4).

The exact symptoms of the flaky segment problem vary depending on the particular clock, but most commonly, the clock will start normally and segments will quickly stop working until few or no segments are active. Also common is a display that only shows one or two blinking segments. But occasionally the symptoms are more bizarre such as a display with all digits and segments continuously lit or a single digit going blank at a certain time.

My scope image shows voltage fluctuating between ~3.2v (max) to ~0.9v (min) in the logic supply voltage to the MAXIM driver chip. Can you imagine any chip working reliably with that kind of ripple in the supply voltage? Also, clocks exhibiting the flaky segment problem generally have normal voltages at all three pins of Q3--normal, at least, for the ZVP3306A. The way the segments fail on particular digits but not others means the segment failures happen in perfect sync with the multiplexing--which would normally rule out Q3--yet replacing Q3 solves the problem. So Q3 must be causing something else to fail. If Q3 is providing more than 3 volts to the VFD driver chip, how can Q3 cause something else to fail? Power ripple is the only reason I could imagine, and a huge power ripple from Q3 is exactly what I saw. Noise from Q3 was also observed by Frank_tt on a different clock with flaky segments. There's a better discussion later in that thread.

Johns Rev D board exhibits very similar display variation as the other clocks based do that are not of his design when at very low light levels. Since his design drives the tube and the Maxim chip differently, then one needs to look at other reasons for any display unevenness. I think that is just the way the tube is designed. If you go below a certain brightness level, the tubes are going to exhibit digit brightness unevenness.

According to the Noritake guide, tubes should be dimmed using a PWM blanking signal to the VFD driver chip. Reducing the anode/grid voltage causes uneven illumination. In the xmas rev d board, the PWM signal used to blank the display has an extremely short duty cycle at low brightness, and the VFD driver chip only allows the anode/grid voltage to rise to a small fraction of the 50v boost supply. The Adafruit design reduces brightness by reducing the boost voltage directly, so at low brightness settings, I believe that the cause of unevenness is the same in both designs--low anode/grid voltage.

To provide consistent illumination at low brightness with xmas rev d design, I would need to decrease the frequency of the blanking signal by at least sixteen fold, but this cannot be done because the timer controlling the dimming also controls boosting which fixes the frequency.

The R3 22 ohm may have been a design decision to limit filament current in the original transistor version of the clock because of the forgotten base resistor and later to compensate for the voltage drop of the ZVP3306A that does not fully turn on. Some evidence of the latter by Russells measurements of insignificant voltage (below 3 volts) at Vt when R3 is replaced with a jumper.

That explanation makes sense, but my guess is that it has to do with the history of the design. The Ice Tube Clock was based on the inGrid Clock. With a DC filament drive, the IV-18 illuminates more evenly when the filament is driven at lower voltage, so the inGrid design used a 20 ohm resistor to drop the 5v supply to a lower voltage for the filament, deliberating running the tube out-of-spec.

Originally Q3 was a 2907 transistor with very little voltage drop, and R3 served its intended purpose. But without a resistor at the base, Q3 sourced too much current from the microcontroller, which caused some clocks to fail. To fix the issue without redesigning the board, Q3 was replaced by the ZVP3306A. R3 was left unchanged, as the ZVP3306A seemed to work fine as a drop-in replacement... at least until people started reporting the dim digit issue.

[Russell 27]

ADAFRUIT

There has been an endless discussion about the 22 ohm resistor. Would you mind clearing up why you used it. Also, if you could comment on your choice of the 20 Pf load capacitors for the 32.768 crystal. While this has probably been a great source of amusement for you all behind the scenes, I really want to hear it from you.

[jarchie]

Does this mean the Maxim VFD driver chip will not function; not necessarily. I've used CMOS chips with a working voltage rating of say 3-18 volts Vcc, and circuit worked fine under 3 volts.

Agreed. I've seen a handful of clocks where the dim digit problem was resolved by replacing R3 with a jumper... and only one where replacing R3 with a jumper caused the flaky segment problem. In most cases, everything seemed to work fine, so the Maxim driver chip does seem to work on less than three volts.

The data sheet ratings are recommended, reliable specifications. As far as noise goes, I didn't see that much noise on my scope, I personally don't think it's a variable.

I don't believe that the ZVP3306A output oscillates in most circumstances--otherwise the clocks forum would be flooded with display-related issues. So the trick will be to recreate the the flaky segment problem, and then check for significant noise. I was lucky enough to play with a clock where I could induce the flaky segment problem, and observe the huge amount of power ripple on the scope.

[Russell 27]

Are you certain ripple was from the transistor.

[jarchie]

Are you certain ripple was from the transistor.

Yes, but you should check my logic anyway... I might be certain, but that doesn't mean I'm right!

I attached the scope probe to the output of the ZVP3306A, and at that point in the circuit, the ZVP3306A would be the only part supplying current. There was no noticeable ripple on the VCC side of Q3. Unfortunately, I didn't look at the gate. The noise on the output of Q3 disappeared after replacing it with a ZVP2110A. To me, that means the ZVP3306A must be responsible--at least partially--for the ripple.

From what I've read, FET oscillation is not terribly uncommon. Although other factors play a role, part of the cause is the small inductance of the wire going to the gate, which can cause negative resistance and promote oscillation. Some people put a resistor on the gate to ensure the total resistance of the wire stays positive. Sometimes a ferrite bead on the gate is necessary to filter out high frequency noise.


One more bit of evidence: Artificially creating severe Q3 noise reproduces the all-eights variant of the flaky segment problem. To create the noise, I made a small change to the xmas firmware to briefly push the Q3 gate high for a microsecond at semi-regular intervals. My test clock had a ZVP2110A and 11 ohm resistor installed as Q3 and R3 and exhibited behavior identical to the all-eights variant of the flaky segment problem.

[Russell 27]

Why is it then, that sometimes the same digit that is lighter in brightness with a full display, is sometimes the only one illuminated at all.

[jarchie]

Why is it then, that sometimes the same digit that is lighter in brightness with a full display, is sometimes the only one illuminated at all.

Could you provide more details or links to the specific problem(s)? Off the top of my head, I'd say this was probably a coincidence, but I could be way off the mark here. Without more information about the case(s) you mention, I'm really just guessing.


Summary of my position:

There seem to be two major, but distinct classes of display problems.

The first class causes inconsistent illumination or unevenly lit segments or digits. These issues are caused by insufficient voltage on the filament or insufficient voltage on the anodes and grids--not noise. I systematically classified the different types of inconsistent illumination elsewhere, but I think you're specifically referring to the dim digit problem which is caused by insufficient voltage across the filament, and increasing filament voltage evens out the display.

The second class (the flaky segment problem) causes a complete failure of particular segments or entire digits. I am arguing that these issues are caused by transistor oscillation--noise. The symptoms vary considerably from clock-to-clock. Most commonly, the display will light normally but segments quickly fail until few or no segments are active (e.g.). Also common is a display that only shows one or two blinking segments (e.g.). Is that what you mean by "only one illuminated at all"? Occasionally the symptoms are more bizarre such as a display with all digits and segments continuously lit (e.g.) or a single digit failing at a certain time (e.g.).


Summary of the most compelling evidence:

• When identifying the types of inconsistent illumination, I was able to replicate display problems in the first class by changing the filament voltage or the grid/anode voltage. But during that process, I never encountered problems in the second class.
  
  
• Given the way vacuum fluorescent displays work, display problems in the first class would be expected at low filament voltage or low grid/anode voltage. Problems in the second class would not be expected with either low filament voltage or low grid/anode voltage.
  
  
• In clocks exhibiting problems in the second class, forum user Frank_tt and I have both observed noise on the output of Q3.
  
  
• I have been able to replicate a problem in the second class by artificially creating noise on the output of Q3.

[Russell 27]

I'm not so sure this is noise. If you put a frequency on the gate, there will be a frequency on the drain; exactly how the inductor ANODE inverter works, and replicating this does not mean it's origin. I would say this is more the cause of low voltage oscillation. Just like a relay chatters if held right at it's turn on point.

[jarchie]

I don't think I understand what you're arguing.

I'm not so sure this is noise.

Here you seem to suggest that the cause is not noise/oscillation.

I would say this is more the cause of low voltage oscillation. Just like a relay chatters if held right at it's turn on point.

But here you seem to claim that the cause is indeed noise/oscillation, specifically "low voltage oscillation."

Of course, I'm interpreting the pronoun "this" to mean "what is going on"... but perhaps you meant for "this" to refer something else--such as the ZVP3306A being only partially on. If the latter, I do agree that the ZVP3306A being only partially turned on makes it more prone to oscillation. But the oscillation is, in turn, what causes the flaky segment display problem (i.e., problems in the second class).

If you put a frequency on the gate, there will be a frequency on the drain; exactly how the inductor ANODE inverter works, and replicating this does not mean it's origin.

If you're referring to my replication of the all-eights problem, my point was not that I replicated noise on the output. As you correctly point out, noise on the output would be expected because I applied noise to the gate. Instead, my point was that I was able to replicate the display problem by creating noise/oscillation on the output of Q3 where Q3 was a transistor that outputs sufficient voltage.

[Russell 27]

I'm not arguing anything, the evidence is right in the meter shots, earlier in the post. No noise.