Multiplexing noise

[Joe Croft]

Hi Yall,

I am working on a multiplexed display and am finding that for the lines that have both of their transistors turned off, there are massive swings of voltage that are induced on the lines. The levels are so high I get other digits flickering in the tube and I also have neon bulbs that are attached get lit up.

I've read several things about this being due to timing, but this is not the case (at least for the neon bulbs) I know this because the 'other pin', the pin that should not light up, is the pin that gets lit up.

I had an engineer tell me that ground plains will help isolate these to some degree, are there any other ideas? I am using MPSA42 and MPSA92 transistors for driving the tubes and lamps.

Any ideas will be appreciated.

-joe

[chuck richards]

Joe,

One possible solution (and not a very politically-correct one either)
would be to loose the multiplexing. Run them with straight DC. Keep
them on all the time. No noise that way.

Just a suggestion which I know works because that's how I run
my 16-tube B-7971 text-scrolling array here. Straight DC.

240 tube segments. No flicker. No noise. No ghosting.

Chuck

>>--
>>You received this message because you are subscribed to the Google
>Groups "neonixie-l" group.
>>To unsubscribe from this group and stop receiving emails from it,
>send an email to neonixie-l+...@googlegroups.com.
>>To post to this group, send an email to neoni...@googlegroups.com.
>>To view this discussion on the web, visit
>https://groups.google.com/d/msgid/neonixie-l/CAE%2BVk6P-qH8gAaAFmCbdp
>P9atnTOfBkJ9n%3DQ%2B4TbZkP4iGRVNQ%40mail.gmail.com.
>>For more options, visit https://groups.google.com/d/optout.
>>



$4.95/mo. National Dialup, Anti-Spam, Anti-Virus, 5mb personal web space. 5x faster dialup for only $9.95/mo. No contracts, No fees, No Kidding! See http://www.All2Easy.net for more details!

[John Rehwinkel]

> I am working on a multiplexed display and am finding that for the lines that have both of their transistors turned off, there are massive swings of voltage that are induced on the lines. The levels are so high I get other digits flickering in the tube and I also have neon bulbs that are attached get lit up.

One approach is to slow down your switching. I like to do this by using optoisolators to switch the anodes. They easily let a logic level control a high voltage signal, and they're generally a bit slow to operate, which is an advantage here.

- John

[Terry S]

Actually you need to be careful with planes -- they can actually couple the signal from one switching line to the next. You may have better results minimizing parallelism between traces, lengths that is, and maximizing spacing between those traces.

Slowing down the edges, using termination if possible, even adding capacitance, will result in smaller voltages induced from one line to the next. Slowing the turn on of the drivers will help, at the expense of power dissipation in those transistors.

[gregebert]

If you have an oscilloscope, debug will be easier.

Have you tried biasing the anodes to a lower-voltage during the off state ? If they are at a voltage lower then the sustaining-voltage, the digits will stay dark. Typically this is done with a large-ish resistor tied to 1/2 the HV supply voltage. If that's not convenient, you can use a pulldown resistor. I would experiment with 1Meg-to-gnd and see if that does the trick. If the glow is less, then try a slightly lower value. You want to use as high a resistance as possible to minimize wasted energy.

As far as the cause, my first guess is the anode transistor is a tad leaky. I didn't look-up the datasheet info, but if Vceo (bipolar) of Vdsmax (NMOS/PMOS) is less than the anode supply voltage, you're entering a gray-area (yes, I'm aware that people use lower-rated Vceo  devices because of the voltage-drop across the tube. I never do, but it's my money so I spend a bit more for over-designed devices). Next, check the leakage currents (Icbo, Iebo). Bipolar devices should be well under 1uA. I typically choose devices at 100nA or less because of transistor 'beta'. For NMOS/PMOS, I try to get Idss around 1uA, but my first clock was 25uA and it works fine.

Forget about tube capacitance; I measured less than 10pF on my 5092''s

Someone already mentioned direct-drive, and all my designs use direct-drive (even my wristwatch). Not sure if that's an option for you at this point.

[Terry S]

Joe, what do you mean by "other pin"?

Terry

On Sunday, February 22, 2015 at 10:17:23 AM UTC-6, joenixie wrote:

[Joe Croft]

Hi Yall,

Thanks for the inputs, Sadly direct drive is not an option, I just don't have the I/O pins available nor the space to add the chips needed to 'produce' more.

Terry,

By other pin, I mean the side considered the anode side of the neon bulb. The side that is positive when the bulb should be lit. It will have a dim but visible glow around the tip of it. This goes away when the bulb is lit.

I will try the biasing. would biasing both sides help? This would be sort of like terminating the lines.

-joe

--
You received this message because you are subscribed to the Google Groups "neonixie-l" group.
To unsubscribe from this group and stop receiving emails from it, send an email to neonixie-l+...@googlegroups.com.

To post to this group, send email to neoni...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/neonixie-l/66a56df9-5571-4bfc-a5a8-8f0801a2d978%40googlegroups.com.

[Niek]

Interesting. Can you take some pictures of the problem? And maybe provide the schematic? (at least of the part driving the nixies).

[gregebert]

I will try the biasing. would biasing both sides help? This would be sort of like terminating the lines.

No need to worry about termination at this frequency, as in reflected waves, unless your PCB traces are several feet long...

 I did check the datasheets for the MPSA42/MPSA92 and they are a good choice because they have low leakage and high Vceo.

Therefore driver leakage is not your problem; can you post a schematic of the driver & predriver circuits ? It could be the predriver.

It's possible the device driving your PNP anode driver is leaky, causing the PNP to turn-on slightly. Are you using a NPN predriver --> PNP driver arrangement ?

If so, your NPN device could be leaking 0.1uA and your PNP will conduct "beta" times this. This is easy to fix with a resistor across the PNP's base/emitter to shunt-away

the predriver leakage.

It's also possible there is noise entering the predriver; this is where a scope will help. If you dont have a scope, then proceed with debugging he predriver.

[Joe Croft]

Hi Yall,

I've attached a picture of the display pcb layout and a pdf of the schematic for it. The cathode drivers are just the MPSA42 with a 100K ohm base transistor and the collector is tied directly to the cathode and the emitter is tied to ground.

-joe

--
You received this message because you are subscribed to the Google Groups "neonixie-l" group.
To unsubscribe from this group and stop receiving emails from it, send an email to neonixie-l+...@googlegroups.com.
To post to this group, send email to neoni...@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/neonixie-l/6c5852ba-b890-4d1d-b20f-0a4c2ab78324%40googlegroups.com.

[Niek]

Does the issue happen more to certain tubes/digits than to others? If so, it could have something to do with the layout of the traces: i'd be especially worried about long runs of high voltage lines parallel to the lines switching the transistors. E.g., in your PCB, the trace at the very bottom (originating out of "M" at the header) is a low voltage line that controls the anode switch transistors, and if you follow it up, it runs parallel to some high voltage lines. The high voltage line's fast switching could induce a smaller voltage spike at the base of that transistor, which could then turn it on. If this is the cause, it probably would show up more to certain digits/tubes than others.

If it's the same for all, then there might be some problem with the program controlling it: e.g. you need to make sure to wait a small while after turning off one anode, then switch the cathodes, then turn on the next anode. Are you doing that?

[Terry S]

Joe -- the anode driver I "borrowed" from the open-source clock looks slightly different than yours.

 

It uses the same transistors, but the MPSA92 has 100K directly across the BE junction, and a 470K between the MPSA42 collector and the MPSA92 base.

 

In other words, imagine moving your R22 to the left of R21, and making R21 470K. (Temp 10 tube)

 

Also, mine has a 33K from the driver to the base of the MPSA42, as opposed to your 100K. A little stiffer drive.

 

I'm not saying either one is better -- just pointing out a difference. I haven't built this circuit yet, so it may need tweaking.

 

I do think Niek has an astute observation regarding coupling -- I went to great pains in my layout to keep the low V drive circuit lines short and far away from the high V switching noise. Put my driver pairs right next to the decoder, and the lines from the connector to the decoder are as far from anything else as possible. 

 

Observing what flickers on and what other lines are switching at the time should point you right to the crosstalk source.

 

Terry S.

[gregebert]

I think this is the problem: The base-emitter of the PNP is not shunted with a resistor (see note from Terry). The purpose of the base-emitter resistor is to provide a path for the leakage-current of the NPN predriver.

With the schematic as shown, roughly half of the NPN's leakage current goes thru the base of the PNP. That current is them multiplied by beta, which I think was around 130 from the datasheet. It's not a lot of current, but it could certainly contribute to unwanted glowing.

Before you start hacking-up your PCB, can you try adjusting the timing of your anode drivers so there is some 'dead-time', say 50-100usec, after 1 anode is turned off, and the next one is turned on ? Also, if you can turn off all cathode drivers during the dead-time, that might help. This will ensure the tube is no longer ionized, and that should either reduce the unwanted glow or make it more difficult to re-ionize from leakage. And to be safe, dont turn off the anode & cathodes at exactly the same time (stagger by ~1usec); otherwise you could create a di/dt problem from the stray inductance of the wiring.

Now, if that doesn't work, next thing to try is a true base-emitter resistor, for example, moving the connection of R22 to the other side of R21 (see Terry's note). I dont think the resistor values are critical as long as they are at least 100K and they are 1/4 watt (or larger). You dont want them to overheat P=Vcc^2/R, so with Vcc=160V, R=100K, a 1/4W resistor is at the limit. If your local Radio Shack store hasn't shut the doors yet, grab some resistors. (I just got some stuff last night at a 95% discount)

[David Forbes]

Greg,

I concur that these are fine things to do. I adjsuted my base-emitter resistor
so that it would have about 0.3V across it if the base were disconencted form
the circuit and the driving signal was on.

Also, short wires from the driver to the tubes are helpful to reducing
crosstalk. I made multiplexed clocks with direct PCB traces from tube to tube
under the sockets, and did the timing things you mentioned.

Most importantly, ensure some dead time (~100 usec) between turning off the
anode and switching the cathode driver to the next digit's value. That's where I
ran into trouble.

[gregebert]

I used direct-drive on my first nixie project only because my gut-instinct was to keep it as simple as possible; I've stuck with that ever since.

Too many postings about 'noisy nixies', choosing the correct cathode-current, bleeding, flickering, RFI, etc. I've even seen scary-looking blue arcs between anodes in a panaplex.

Even my wristwatch is direct-drive (lucky me, the SP-151 has separate pins for each cathode segment...)

Since I dont sell anything, the extra cost for direct-drive isn't an issue for me.

That said, I have an unusual 9-segment (not a typo; it has 2 more segments in the middle so you can display characters like T, W, etc) display that requires multiplexing, and I hope to make a clock out of it in the near future. Fingers crossed I wont have any weird problems....just wish I could find a few more of these units for spares.

[Joe Croft]

Hi Yall,

Okay, I scoped the board and I have almost 500us of total dead time between digits. This was measured on the digital outputs from the CPU. My code is written in a way that I turn off all of the segment and digit signals one at a time in a for loop so they are not all triggering at once. They are turned on in a like manner. I do not know the time between these, but the app know's nothing about the actual port and bit that is being turned on or off so I assume the time is in the 1us ball park if not more. I am using a 16MHZ ATMega328.

As for direct drive, I am multiplexing because of a lack of pins and board space. I wanted a nice small board as well as through hole parts because I want to sell this as a kit. I just don't have room for a couple of more DIPS.

I like the idea of the moving the resistors, I will try this. I will also spread my segment lines out going between the 4 digits up top and the 2 digits below. I am still contemplating the ground plains they are easy to do. I will have to ask a couple of people at work about the issue of them possibly increasing the  coupling between the lines.

Biasing the lines is another idea I will try as well.I can do most of these before the next board spin. Depending on how the other things I try go, I will then decide on the plains.

-joe

--

You received this message because you are subscribed to the Google Groups "neonixie-l" group.
To unsubscribe from this group and stop receiving emails from it, send an email to neonixie-l+...@googlegroups.com.

To post to this group, send email to neoni...@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/neonixie-l/1a902863-3f11-44e9-a9fd-f9d507e5b420%40googlegroups.com.

[David Forbes]

On 2/24/15 6:38 PM, Joe Croft wrote:
> Hi Yall,
>
> Okay, I scoped the board and I have almost 500us of total dead time
> between digits. This was measured on the digital outputs from the CPU.
> My code is written in a way that I turn off all of the segment and digit
> signals one at a time in a for loop so they are not all triggering at
> once. They are turned on in a like manner. I do not know the time
> between these, but the app know's nothing about the actual port and bit
> that is being turned on or off so I assume the time is in the 1us ball
> park if not more. I am using a 16MHZ ATMega328.
>

Do you turn off all the cathodes, then wait a hundred microseconds, then
turn off the anode, then turn on the next anode, then turn on the next
cathode?

Because if you don't have that delay between the cathode and the anode
turnoff, then you get ghosting.

--
David Forbes, Tucson AZ

[Terry S]

Greg -- regarding the resistor size and power dissipation, if there are 6 digits (7 in this case with the neons), can't you figure the resistor duty cycle is 1/6th, and therefore smaller wattage resistors can be used? The average dissipation is more like .043 watts, by my back-of-the-napkin math.

I realize the instantaneous dissipation is .25 watts, but resistors do have some thermal mass....

If I'm wrong here please correct me. I know I've worked on a similar problem before and I recall the answer was not so obvious.

Terry

[petehand]

Just move R22 etc to the other end of R21 etc, on the base of the PNPs. Where they are now, those resistors are are doing nothing but wasting power. When the NPNs turn on they're going to draw 1.6mA and drop 160V across the 100k resistors R21 etc. You need no more than 3V to turn on the PNPs, so in theory you could drop the values of R22 etc to 2.2k, but I suggest 10k is more suitable. 100k is way too high and may as well not be there.

[Joe Croft]

Hmmm... interesting observation Pete, are you talking about changing R21 or R22 to 10K? I chose 100K because I have them in my NixieNeon clock.

-joe

On Wed, Feb 25, 2015 at 2:48 AM, petehand <pete...@gmail.com> wrote:

Just move R22 etc to the other end of R21 etc, on the base of the PNPs. Where they are now, those resistors are are doing nothing but wasting power. When the NPNs turn on they're going to draw 1.6mA and drop 160V across the 100k resistors R21 etc. You need no more than 3V to turn on the PNPs, so in theory you could drop the values of R22 etc to 2.2k, but I suggest 10k is more suitable. 100k is way too high and may as well not be there.

--

You received this message because you are subscribed to the Google Groups "neonixie-l" group.
To unsubscribe from this group and stop receiving emails from it, send an email to neonixie-l+...@googlegroups.com.
To post to this group, send email to neoni...@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/neonixie-l/77035e01-fd3f-4085-bc63-83deae194f39%40googlegroups.com.

[Niek]

Perhaps you can post the code that does the multiplexing - it may shed some light on this issue. By the way, there's nothing wrong with multiplexing - and it's not that hard to get it right, just follow a few simple rules like the dead time, and you also haven't mentioned the frequency yet, which should probably be around 200Hz (you may have set it too high). In my opinion it's a complete waste of drivers not to multiplex, especially with a relatively high duty cycle, like 1/3 (e.g. when you have 6 tubes and you switch 2 at a time). Yours is 1/6, so you need to overdrive the tube a bit more, but other than that there's nothing wrong with it.

[petehand]

I did mean to change R22 to 10k, but I can suggest an even better way. This is a circuit I've used to multiplex IN17s, with no dead period and no ghosting. It looks terrifyingly unsafe. Let me explain.

When the processor pin is high, Q1 base-emitter voltage is 0 and the transistor is cut off. The port pin sees no high voltage. When the port pin goes low the transistor turns on as a constant current source, the current set by (5 - 0.6)V/R2 or about 1mA. This drops 170V across Q1 and 10V across R1, which turns on Q2. Q1 is operating in linear mode, not saturated, so it switches in nanoseconds. Resistor R1 is necessary to help Q2 to switch off rapidly.

This configuration of Q1 with the implied transistor inside the MPU is called a cascode.

[petehand]

One more thing about the cascode. Transistor Q1 is dissipating 170mW with the values shown. It may get a little warm - you have to watch that. You can put a "helper" resistor between Q1 collector and Q2 base. The value is completely immaterial since the current is set by the R2 emitter resistor, so something in the order of 100k will do. This will then dissipate 100mW and take the burden off the transistor. But as far as the circuit operation is concerned, it's completely unnecessary.

[Terry S]

Pete, that's a nice application for the cascode circuit... Help me understand how it eliminates concerns about dead time and ghosting.

 

Terry

[Joe Croft]

Hi Pete,

I like it. I will give this a try.

-joe

On Thu, Feb 26, 2015 at 3:12 AM, petehand <pete...@gmail.com> wrote:

One more thing about the cascode. Transistor Q1 is dissipating 170mW with the values shown. It may get a little warm - you have to watch that. You can put a "helper" resistor between Q1 collector and Q2 base. The value is completely immaterial since the current is set by the R2 emitter resistor, so something in the order of 100k will do. This will then dissipate 100mW and take the burden off the transistor. But as far as the circuit operation is concerned, it's completely unnecessary.

--
You received this message because you are subscribed to the Google Groups "neonixie-l" group.
To unsubscribe from this group and stop receiving emails from it, send an email to neonixie-l+...@googlegroups.com.
To post to this group, send email to neoni...@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/neonixie-l/c3c428f4-1aa1-413e-827c-b06f48549db0%40googlegroups.com.

[petehand]

With the conventional circuit, you have two saturated transistor switches, each of which needs to turn off to blank the anode. Each may take a few microseconds, and the second stage doesn't start its time delay until the first stage turn off is completely finished. With most designs the transistors are over-driven, which makes the delays worse. Switch-on time of the next stage is not delayed, so you can easily get more than 10 microseconds overlap with the next anode.

The cascode stage has essentially zero switching time, and with a bit of attention to base current and resistor values you can cut the switching time of the PNP to a minimum, so even if there is a slight overlap it's likely to be less than the ionization time of the next tube. I don't claim that this circuit will always eliminate ghosting entirely, but I do assert that I've made half a dozen multiplexed clocks using a similar circuit to this (I use a 2ms digit period), switching the anodes simultaneously with the digit cathodes, and I've never had any ghosting.

My example 1mA current to turn on Q2 is ridiculously high, by the way. Experiment with the R2 emitter resistor - it should work with 100k. But as you increase R2, make R1 an equal value, otherwise Q2 won't turn on.

[Terry S]

Thanks Pete -- my design uses the SMT versions of the transistors so I'm particularly interested in reducing the power dissipation. Please clarify the placement of the "helper" resistor -- in the vertical leg of the Q1 collector circuit, or in the horizontal leg of the Q2 base connection?

 

If I understand the circuit right, it's the vertical leg.

 

Terry

[petehand]

You put it in Q1 collector to the junction of R1 and Q2 base. Its only purpose is to drop voltage so that the transistor doesn't have to drop so much. Based on the value of the constant current, set by Q1 emitter resistor, you want it to drop some 100 odd volts. The current doesn't need to be as high as 1mA - check the gain of your PNP transistors. The MPSA92 data sheet says it's minimum 25, but the typical is over 100. Pick the value of R1 so that it drops a few volts at the current you choose, then all that current will be drawn from the base of Q2, but if R1 is too small and doesn't drop at least 0.7V the transistor won''t turn on.

Note that Q1 is non inverting in that configuration, so the port pin goes LOW to turn on the anode driver.

[Dekatron42]

Pete, will the cascode circuit work properly as a cathode driver if you use for instance a 74HCT42 or a 74HCT138 to drive the transistor (they both have inverted outputs going low when selected), using the collector of the transistor to drive the cathode of a Nixie?

/Martin

[petehand]

Yes, it should work perfectly in that application with a CMOS gate. I would not try it with a TTL gate though, as it relies on the output going to the 5V rail to turn the transistor off and bipolar can't get up there. To use TTL you would need to add pullup resistors to 5V on the gate outputs.

[Dekatron42]

Thank you for your answer!

I'll try that as that makes it possible to use either the 74HCT42 or a pair of 74HCT138s instead of a 74141 with just a few extra transistors and resistors, unless you don't want to use anything more modern like any of the Supertex HV-VFD drivers. Thanks for pointing out the difference between TTL and CMOS in this case.

Would an extra diode in series with the base or lets say a lower voltage on the base work with TTL? If so a simple voltage divider or a zener and resistor to the base would be a simple solution.

/Martin

[petehand]

A lower voltage on the base would work, or an extra diode in series with the EMITTER to raise the turnon voltage by another 0.6V, or just pullup resistors on the chip outputs to make sure they go to a righteous 5V. Remember you still need those emitter resistors, otherwise the B-E diode will short the driver output to Vcc and cook it. It may be worthwhile to use those resistors to set the cathode current, since they have to be there anyway, and not use an anode resistor. I don't know, I've not tried it myself, so you're a pioneer. Be sure and tell us how it works out.

[Dekatron42]

Thanks, I will try it but I am currently renovating my whole apartment, including my hobby room, so that will be many months into the future but it is on my list - maybe I can't keep from checking it before I have finished renovating, you know how it is when your fingers itch to do something you really want to do instead of what you should be doing.... ;)

/Martin