Tiny VFD clock

[Mad_Hank]

Hey people I have a few questions for a project that I have in my head, it would be great if anyone can help me a little bit.

I recently got the idea to make a VFD clock with some small IV-6 tubes as a present to some friends/family. Due to the number of clocks and my idea for an enclosure there are a few constraints: it has to be tiny but no SMD and cheap so the MAX6921 is not an option.
Currently the Idea is to use a cheap arduino nano clone as microcontroller, why not a lone atmega? well a friend of mine tried that and had some trouble with it, 3 dollar extra is not worth messing around for hours with an arduino (not much of a software guy). To replace the driver I was thinking of using a 74hc595 for the segments and MPSA42 transistors to switch them. But as I am typing this I realize that the arduino nano has 14 I/O so I can probably just drive 11 transistors straight from my arduino. As a PSU I was thinking of a simple mosfet, 100uh inductor, fast diode and capacitor with a 1:15 voltage divider to measure the output.

My remaining questions are:
I read over here: http://www.noritake-itron.com/SubPages/ApplicNotesE/vfdoperapn.htm that it is best to feed the filaments AC but I was planning on hooking up 4 filaments in series and powering that with a 7805 (and maybe a shottky for extra voltage drop, 48*1.2=4,8v). Is the intensity gradient visible with single tubes?

I also read that it is best to have a slight bias on your filament voltage to avoid ghosting, how can I accomplish that with my simple 7805 idea?

Regarding the transistors, I think a MPSA42 with a 1-2K resistor between base and I/O + a 10k pulldown should be alright, but is there a way to calculate these values besides just guessing. I don't completely grasp how to switch the VFD tube with transistors at the moment.

any suggestions, answers, ideas or whatever you have to say about this is greatly appreciated.

[richard.cs]

I think I follow what you're suggesting but a quick sketch of your proposed circuit would be nice.

I suspect that you won't see the intensity gradient across a single tube but that there will be a noticable difference between tubes. You could compensate for that in another way (e.g. by changing grid or anode voltages to compensate) but it may well be more work than driving it with square wave a.c.

Conventially you lift the filament a few volts above ground so that 0V on the grid corresponds to a small negative bias. If you were to drive the cathodes with a 5V square wave then you'd lift them to an average 2.5V which might be about perfect. You can of course keep the cathode at ground potential but then you need to make the grid drive swing from a few volts negative to a few volts positive (for most tubes) and that makes it harder to use standard logic.

[Harvs]

I couldn't easily find a datasheet for those tubes, so I'm not sure what you need to drive them.

However, if you haven't seen it, this is how I did it with another VFD, I'm sure you could do something similar.
https://www.eevblog.com/forum/projects/'jelly-bean'-vfd-driver/

[Mad_Hank]

Well I am currently on my phone but I will post the data when I get home. The main problem is that those tubes need 60 volt instead of 12 like you used.

I think that I understand how you generated the AC but I don't see how you add some bias.

But I will have a better look when I am back home

[Harvs]

I didn't add bias. There was no need at least for the vfd I was using.

I guess a bunch of transistors are called for if you need 60V

[richard.cs]

One way of driving the anodes is simply to have a resistor to the high voltage supply and pull the anode down to ground with one of the open-collector driver chips to turn the segment off. It's not very efficient but it does allow simple driving with standard logic signals. Of course a couple of extra resistors and a PNP transistor per segment would allow real high-side switching. Or was your question really about generating the +60V?

The bias in Harvs driver comes from the average cathode voltage being 2.25V above ground. The grids can then be driven both above and below the cathode, although driving the grids to +12V would be excessive for most VFDs I think.

[Mad_Hank]

Well my main question at the moment is how to switch the segments on and off. At this point my idea was to do it with MPSA42 is to do it with MPSA92 (PNP version), a 10k pulldown and a resistor between base and I/O which I don't know yet how to calculate. They can easily drive 60 volt since they are rated @300 volts.

The +60 volt supply should not be that much of a problem; MOSFET to switch a coil between V+ and gnd + diode and capacitor. Add a little voltage divider and let the software switch it at ±50khz and adjust the duty cycle according to the voltage seen on the voltage divider.

Harvs already showed a nice idea to create `AC` for the filament; four small MOSFETs and switch the inputs on and off (alternating) although I might use the BS170 due to being very close to their limits with those tubes

attached is a screenshot from a webpage with the specs of the tubes.

[richard.cs]

Interesting, it looks like you have to pull the grid *very* positive on those ones.

With only PNP transistors at the top you'll struggle to drive it directly from a microcontroller - the base of the transistor (and therefore when no current is drawn both ends of the base resistor) will need to be at +60V to turn it off. You need to either use 56V zeners as "pushrods" or (better) use open collector outputs with a >60V rating, perhaps a ULN2003-type array, to drive the bases of your PNP transistors. You'll struggle with zeners on that voltage I think because your 0-5V logic swing is less than 10% of your "pushrod" length.

You will also need pullups from base to emitter on your PNP transistors to turn them off fully against any leakage current from your base drive. Is that what you meant by 10k pulldown?

[edavid]

With only PNP transistors at the top you'll struggle to drive it directly from a microcontroller - the base of the transistor (and therefore when no current is drawn both ends of the base resistor) will need to be at +60V to turn it off. You need to either use 56V zeners as "pushrods" or (better) use open collector outputs with a >60V rating, perhaps a ULN2003-type array, to drive the bases of your PNP transistors. You'll struggle with zeners on that voltage I think because your 0-5V logic swing is less than 10% of your "pushrod" length.

An easier way is to use common base NPN transistors to drive the PNP transistors.  You tie the NPN collector to PNP base, NPN base to microcontroller supply voltage, NPN emitter to resistor to microcontroller output pin.

And OP, do some tests before you decide you need 60V.  30V might be bright enough.

[Mad_Hank]

@ edavid, do you mean like a sziklai pair?
And regarding the grid voltage; it would be easy to adjust in the software or do you mean to make the design easier?

@richard.cd, yeah I still had a NPN transistor in my head so lets flip that around and make it a pullup


I'm going to draw a schematic of that part, maybe that clears some things up

[edavid]

@ edavid, do you mean like a sziklai pair?

No, cascode.

And regarding the grid voltage; it would be easy to adjust in the software or do you mean to make the design easier?

To make the design easier.

[Mad_Hank]

Is this what you mean?

[edavid]

More like this...

[Mad_Hank]

More like this...

what kind of black magic is that Sorry but I have no idea how that circuit works, can you explain it a little bit?

Am I right that you sink the pin with your microcontroller? So with the I/O high no current can flow but when you pull it low the current can flow, triggers the PNP and hopefully the VFD lights up

[Whuffo]

You should run the filament on AC. If you use DC (or pulsing DC) to run it, electromigration will shorten the life of the filament. The frequency of the AC isn't critical, so a simple little switched power supply would be perfect - just use AC right off of the switching transformer. You're going to need a fairly elaborate power supply to supply all the voltages; this is the reason you rarely see VFD displays on cheap products. Actually, if you look closely you might discover that some of those VFD displays are really LCD panels running in "white on black" with a EL panel supplying light from behind; that gets "the look" but is simpler and cheaper to make.

Electromigration is a problem with all vacuum tube (and light bulb) filaments. The metal ions form a cloud around the hot filament, and the electric field pulls them along the filament. If you heat with DC, the filament will gradually get thinner on one end and fatter on the other end - and ultimately burn out at the thin end. On AC, the ions still move, but they just do a little shuffle and the filament stays a consistent diameter for much longer.

Vacuum tubes (and VFD displays) are kind of delicate; run them at their specification and they'll last a long, long time. But they don't last long if you run them out of spec; they won't die instantly like solid state devices do, but they'll die all the same. This is important, especially for something like a clock which will be running constantly.

[Mad_Hank]

Doesn't a switching transformer generate DC?

[richard.cs]

Harvs' design with the H bridge will produce perfectly adequate square wave a.c. for the cathode. Even without worrying about electromigration you'll want a.c. drive for even segment illumination.

Edavid's suggestion should work fine for driving the anodes, I've attached a schematic of what I was suggesting. I don't see much to choose between the two other than a logical inversion although I may well have missed something - Edavid, care to point out what? The one thing I will say in favour of mine is that it can be done in slightly less board area with a transistor array that includes the NPN transistors and the base resistors. The ULN2004, etc are only good for 50V but I'm sure I've come across similar devices with >100V ratings.

[Mad_Hank]

Okay so just build that H bridge (with the BS170 for more current) connect the wires that Harvs connected to U7 to the arduino and flip between 1/0-0/1 at a rate >100hz or something. V+ can then be generated with the 7805 + diode like I was planning and Harvs also does in his design.

The only thing that I am "missing" is how to apply a bit of bias to the filament power supply (in a simple manner) and how that two transistor driver works.

[richard.cs]

From the information you posted it doesn't appear that this display really needs the grid to go negative (achieved by biasing the filament a few volts positive) to turn it off. Harvs design has an inherent +2.2V filament bias anyway and I'd just expect it to work. Are you multiplexing these displays to reduce the number of GPIO pins used? If you are you want to make sure the multiplexing rate is not close to your filament frequency or you may see flickering.

Edavids 2-transistor driver works like this:
Port high, no current flows into the base of Q1, therefore no current flows into its collector from the base of Q2, R2 holds the base of Q2 high, Q2 is off and the VFD output is low.
Port low, current flows from the micro-controller supply into the base of Q1. A larger current flows collector to emitter set by the value of R1 and the micro-controller supply voltage, some flows through R2 and some from the base of Q2. This turns Q2 on and 60V appears at the output.

My circuit works like this:
Port high, current flows into the base of Q1 via R1 and turns it on. This pulls down on the base of Q2 via R2 turning on Q2, 60V appears at the output. Little current flows through R3 as it only has 0.6V across it.
Port low, Q1 is off as it has no base current. The base of Q2 is pulled up to 60V by R3, Q3 is off and the VFD output is low.

There's not much to choose between the two so far as I can see, build whichever you like.

[Mad_Hank]

Yes the plan is to multiplex everything to keep the partcount low. I need everything as small as possible because the idea that I have for the enclosure will be barely bigger than the pcb + components. I'll post pics once I have designed a circuitboard (at that point I can determine dimensions).

But you mention a +2.2V filament bias, I don't really see where that comes from?

But I would like to thank you all for your help thus far, I'm a 2nd year master student in bioelectronics and nanotechnology (biomedical sciences as bachelor). While I love to mess around with electronics my knowledge sometimes lacks a bit. We did have an electronics course but that was more the OPAMP stuff and/or biological electronics and well... projects like these are nice to improve some skills outside of the laboratory

[richard.cs]

But you mention a +2.2V filament bias, I don't really see where that comes from?

In Harvs' design a 4.4V supply is created by dropping a 5V supply with a diode. The mosfets are always in one of two states and the filament always has one end at +4.4V and the other at 0V, but the voltages at the two ends are constantly changing between the two. In the middle the voltage is always +2.2V, at each end the voltage swings from 0V to 4.4V and the average is 2.2V, but at any point in between the voltage is swinging from a bit more than +2.2V to a bit less than +2.2V and the time-average voltage at all points is +2.2V.

This is the square wave equivalent of using a conventional mains transformer of 4.4V output to feed the filament, and connecting the centre tap to +2.2V d.c.

Hope that makes sense now?

[Mad_Hank]

But you mention a +2.2V filament bias, I don't really see where that comes from?

Hope that makes sense now?

*facepalm* I was thinking of a bias above ground, silly me

Okay a LOT of things have been cleared up today. Thank you, I will draw a schematic of everything somewhere this weekend, but I have a lot of things to do so it'll probably be saturday or sunday before I can post it.

One last question, how accurate is an arduino without RTC as a clock? My nixie clock runs off the DCF77 signal (atom time baby O0) but for a little (gift)project like this a few seconds here and there won't matter but if it is minutes per month it'll be annoying.

[richard.cs]

One last question, how accurate is an arduino without RTC as a clock? My nixie clock runs off the DCF77 signal (atom time baby O0) but for a little (gift)project like this a few seconds here and there won't matter but if it is minutes per month it'll be annoying.

I don't normally work with arduinos but if they're crystal clocked and there are no bugs in your timing code it should be fine I would think. Obvously it would lose time if the power is interrupted. The other thing that can be done is counting mains cycles, the short-term variation is pretty bad but the long term stability is good.

[edavid]

Edavid's suggestion should work fine for driving the anodes, I've attached a schematic of what I was suggesting. I don't see much to choose between the two other than a logical inversion although I may well have missed something - Edavid, care to point out what?

It saves a resistor, and it's easy to get the base drive needed for low beta HV transistors, but yes, not much difference.

[Mad_Hank]

Edavid's suggestion should work fine for driving the anodes, I've attached a schematic of what I was suggesting. I don't see much to choose between the two other than a logical inversion although I may well have missed something - Edavid, care to point out what?

It saves a resistor, and it's easy to get the base drive needed for low beta HV transistors, but yes, not much difference.

While we are talking about resistors... How do you calculate the values you need with the transistors?

[Len]

One last question, how accurate is an arduino without RTC as a clock? My nixie clock runs off the DCF77 signal (atom time baby O0) but for a little (gift)project like this a few seconds here and there won't matter but if it is minutes per month it'll be annoying.

Here's an example of a microcontroller-driven clock that uses a 32kHz watch crystal to keep time:
http://learn.adafruit.com/ice-tube-clock-kit/
It's also an example of how to drive an old-timey VFD (IV-18 not IV-6).
I have one and its timekeeping wasn't great until I trimmed it in the firmware. Now it's quite good.

If you want more accuracy, the Maxim DS3231 is excellent but costs around $10 I think.

[Mad_Hank]

Yeah but I wanted to use the 16mhz crystal present on the arduino nano.

But I can always sacrifice my two remaing I/O for something like this http://www.ebay.com/itm/Arduino-I2C-RTC-DS1307-AT24C32-Real-Time-Clock-Module-For-AVR-ARM-PIC-SMD-/170910326110?pt=LH_DefaultDomain_0&hash=item27cb0c9d5e

14 I/O
-11 --> 4 for tube selection + 7 for the segments
-1 for the +60 volt psu
-2 left or for RTC

[Len]

Since the Nano seems to have a real crystal (not a ceramic resonator) it might be OK by itself for timekeeping. I'm not sure. But the RTC chip will keep track of minutes, hours, days, months for you which is convenient.

[Harvs]

One last question, how accurate is an arduino without RTC as a clock? My nixie clock runs off the DCF77 signal (atom time baby O0) but for a little (gift)project like this a few seconds here and there won't matter but if it is minutes per month it'll be annoying.

As long as it's got a real crystal (just look for a shiny tin can), and you can continuously supply it power, then it'll be fine.  The reason for going to an RTC IC is to maintain time when you've got no power.  The watch crystal is no more accurate (in fact most are less accurate than a >1MHz crystal if you look at the specs), but in low power applications they consume a lot less current than running a timer at 16MHz.

Forgive me if I'm wrong but doesn't the Arduino already have a millisecond counter built into the software?  Then it becomes fairly trivial to convert that to hrs, sec, min.

[Mad_Hank]

Yeah they have a real 16mhz crystal an I know about the milis function. Power will be provided by a wall wart. But my idea is a button to increment the hours by one and a button for minutes +1 so setting the time after a power outtage wont be a much of a problem

[carbon dude oxide]

I dd a project using the arduino as a clock an i found that it lat 4 second every hour, it might have been the code but it still lost that much time an hour, i had an ethernet sheild connected so i could sync the time every 30 mins to keep it relativly correct i would recomend using an RTC that way the time does not get affected by your code, and it can also be powered via an external backup battery so that you dont have to re program the time everytime you turn the clock off. Something like a DS1307 RTC works quite well, i use it in my clock and its nice and small

[Harvs]

10ppm is 26sec/month.  The cheapest 16MHz crystal on digikey is 50ppm tolerance, so at worst with no calibration will be just over two minutes per month.  If you do a small amount of calibration (add/sub milliseconds per day at midnight), and the temperature is just normal temp that a room would be, then the performance can be very respectable.

It sounds like your ethernet code may well have been completely hogging the processor (maybe disabling interrupts as part of it?).  It's not hard to just make sure that processor can always service something like a 1 ms interrupt off a hardware counter.

Sure you can put a DS1307, then calibrate the crystal on that, but unless you must have battery backup, the IC is more expensive than an ATmega328p.

[Mad_Hank]

Yeah I'll try to get it going first, cramming a little extra pcb somewhere inside is always an option if it has trouble keeping time. But I have to say, even the cheapest chinese gadgets keep decent time nowadays. I may or may not have some time to draw a final schematic tonight but I already picked up some tiny 100uh coils at a local electronics store.

[edavid]

Yeah I'll try to get it going first, cramming a little extra pcb somewhere inside is always an option if it has trouble keeping time. But I have to say, even the cheapest chinese gadgets keep decent time nowadays.

They are usually trimmed.

I may or may not have some time to draw a final schematic tonight but I already picked up some tiny 100uh coils at a local electronics store.

Tiny coils are not good for DC-DC converters... resistance is too high, saturation current is too low.

[Mad_Hank]

Oh no don't worry, they are not the resistor like coils. They are of the "can" type if you get what I mean, but one size smaller. Most often you see them about a cm thick but these are roughly half a cm. I'll post a pic next to a ruler tomorrow.

[Mad_Hank]

More like this...

I just wanted to start drawing but then I realized that I have no idea of the resistor values. how can I calculate those or what value should they be?

[edavid]

More like this...

I just wanted to start drawing but then I realized that I have no idea of the resistor values. how can I calculate those or what value should they be?

R1 supplies the base current for the PNP... which should be the VFD current / minimum beta * safety factor (2 or 3)

R2 makes sure the PNP is turned off quickly enough... has to supply the leakage current (tiny) and discharge the capacitance (also tiny)... try 100K?

[Mad_Hank]

Okay so for the grid 45 mA because it is multiplexed and the anodes <1.8 mA so I think ±1 mA should be alright.

[Mad_Hank]

It needs to be smaller

I made some sketches and the problem is that I have to fit everything in 18*24 holes, each transistor driver is 5*6  and I need 11 of them, so out of 432 holes I already used 330. SO i'll have to make concessions. I'll swap the arduino nanos for an atmega328 but since that won't give me the extra room necessary I was thinking of replacing the NPN transistor with an ULN2004 to save a bit of space. And adding a small board to connect the tubes to, since having 4 times a dozen of wires takes up 48 holes while I can switch that with 13 wires. 

My only other idea is to find out whether the IV-6 can run multiplexed at 45 volt or something so that I can swap all MPSA92, but I haven't found someone who can confirm this.

[Mad_Hank]

Well I found out that my first design was very inefficient (space wise) 2*7 baby

[Mad_Hank]

I forgot the crystal, is it a bad Idea to add long wires to the crystal and place that above the power supply part?
On another forum someone said that the tubes often are bright enough to run at lower voltages, I might get away with a slightly lower part count but I think I can get away with this circuit. I just need to shift a bit with the  layout.

[richard.cs]

You could put the crystal either on the underside of the board or on short wires on top of IC1

[Mad_Hank]

I think I can alter the layout a bit but I'm continuing this project after my exams.

[Mad_Hank]

Back to the drawing board!

Someone helped me design a great enclosure but now I want to draw a schematic of my plan so I'm trying eagle at the moment.

For the resistors it was mentioned that   "VFD current / minimum beta * safety factor (2 or 3)" should give me the resistor value. So if I am correct that would be 0.0015/25*3 which is a very low number. Am I missing something?

[richard.cs]

For the resistors it was mentioned that   "VFD current / minimum beta * safety factor (2 or 3)" should give me the resistor value. So if I am correct that would be 0.0015/25*3 which is a very low number. Am I missing something?

That very low number is the base current you need, not the resistor value. Divide the supply voltage by that current and get a large number that will be the resistance.

[Mad_Hank]

schematic of drivers at this point


so:
0.0015/25*3 = 0.00018
60/0.00018 = 333333 = 330k Ohm

[edavid]

NPNs aren't connected right... looks like you couldn't decide if you wanted common base or common emitter

[SeanB]

It is right, base current for the high side is slightly dependant on the high state voltage, but it will work very well as a current limited driver. Only issue is if a pin is tristated or set to a mode with a pull up which will turn the driver on. Leakage with open circuit base will tend to turn the drivers on at elevated temperatures, but is not a worry with an active low drive.

[Mad_Hank]

okay so as long as I actively set them to o or 1 in the software it won't be a problem.

0.0015/25*3 = 0.00018
60/0.00018 = 333333 = 330k Ohm for the segments
0.01/25*3 = 0.0012
60/0.0012 = 333333 = 50000 = 50k Ohm for the grid/tube selection.

[Mad_Hank]

Anyone has some experience with buying atmega328 chips from ebay? How big is the chance of these being fake?

[Mad_Hank]

Okay after debating some exotic clock on a chip solutions from the VFD era I decided to go for my original idea.
I will edit this guys source code because I'm not that much of a software guy; http://thecustomgeek.com/2011/06/29/multiplexing-for-a-7-year-old/
I will probably use LM2596 powersupplies off ebay for the V+ because I hear from more and more people that VFD displays at their recommended voltages are way too bright. ***
VFD's have been upgraded to IV-11 tubes because I have more room than expected.

*** just noticed that these are step down only

[Mad_Hank]

NPNs aren't connected right... looks like you couldn't decide if you wanted common base or common emitter

I just realized what you mean!

Where I connected my ground I should have connected my uC pin, and where I connected the uC pin I should have connected the 5 volt supply.

But I posted this on another forum and people over there said that my (mistake) should work.

Well I can order the parts anyway since that is merely a layout thing.

[Mad_Hank]

Yikes I just bumped into another problem:

0.0035/25*3 = 0.00042
60/0.00042 = 140k = 130K or 150K ohm

suppose I lower my voltage to 40 volts because the tubes are very bright:

40/0.00042 = 100k

that is a deviation of 40-50%

I ordered both values but how will this affect my final design.

[edavid]

Yikes I just bumped into another problem:

0.0035/25*3 = 0.00042
60/0.00042 = 140k = 130K or 150K ohm

suppose I lower my voltage to 40 volts because the tubes are very bright:

40/0.00042 = 100k

that is a deviation of 40-50%

I ordered both values but how will this affect my final design.

The resistors supply base current... lower voltage -> lower VFD current -> lower collector current -> lower base current.  So, no problem.

[codeboy2k]

even though people on the other forum said your mistake should work, it's much, much better off as a cascode (common base input) as edavid suggested. If you still have the time, you should correct it and configure it as a common base input cascode, i.e. bases to +5V and control input is via the emitter.

The cascode has much higher isolation from input to output and thus offers much better protection for the microcontroller from the 60V high-voltage supply.  I'm surprised no one has mentioned this yet.  Also, the cascode won't suffer from base leakage that might turn it slightly on, as would be the case with the common emitter (your drawing mistake). i.e. tri-states or pullups on the I/O port won't turn it on.

[Mad_Hank]

Okay thank you, I tried applying for a few samples of the HV5812 anyway, but if those are declined I will go for the big box of transistors

And anything can be changed until my soldering iron is cold