Constant current boost converter problems

[Artlav]

Hi.

I want to make a constant current driver for a Xenon arc lamp.
It takes 12V as input, should give out 8A at 15 - 20V, and maintain 50V idle voltage.

I used a TL494 to control it, single-ended output driving a boost converter.
100KHz frequency.
Schematics attached below.

It basically works, but there are problems.
The thing surges and whines in audible range, the input is whole 20A at 12V with output at 8A at 15V, and the current is often a few A above the setting.

The symptoms are here sometimes, and not here the other times - depending on the load or voltage, or just nothing specific at all.

What am i doing wrong?

Tried different inductors, from 10uH to 500uH, capacitors on the current feedback input, chokes - the whine changes tone, but never goes away.

 

[Mordred]

Hi.

The thing surges and whines in audible range, the input is whole 20A at 12V with output at 8A at 15V, and the current is often a few A above the setting.

The symptoms are here sometimes, and not here the other times - depending on the load or voltage, or just nothing specific at all.

What am i doing wrong?

Tried different inductors, from 10uH to 500uH, capacitors on the current feedback input, chokes - the whine changes tone, but never goes away.

The intermittant sounds like a intermittant resistance point due to possibly a poor connection, forming an unplanned RC circuit. I've seen circuit boards do this and usually solved the problem by re-flowing the connections. I don't see any problems on the shematics but if you can identify which component is whining, might help in isolating the signal generation

 

[meBigGuy]

It depends on the nature of the irregularity.

For example, one issue I have seen is what I call "clumping". It is cause by poor layout and is characterized by groups (bursts) of switching pulses during which the current goes way up. I've actually fixed it by re-orienting the inductor (it was talking to pcb loops), but generally it takes more. It varies with load, etc as you described.

As for the sound, what oscillator frequency is it actually running at? The chart says about 20K. What frequency are you hearing? "Clumping" can cause lower frequency variable audible effects. Not sure what could make audible sounds other than the inductor.

Have you layed out switching regulators before? I was surprised there was no layout note in the app note or data sheet. Getting the grounds right so there are current loops affecting the feedback signals can be tricky. Google TL494 pcb layout and see what you might want to change. Sometimes you can add some braid to hack the grounds.

Try a cap from pin 15 to ground.

Try disconnecting the current feedback and drawing some current at 50V. How stable is it then?

 

[Artlav]

Clumping sounds about right.

If i put a 50Ω load at full voltage it would whine with a clean high sound, and the gate would get a regular missed pulse (first attach).

If i get it into constant current at 2A, then it starts making a much lower tone with some surging, but similar picture of bursts of pulses at semi-regular intervals (second attachment).

Third attach is the controller board layout.
The power part is basically a fet and a diode on a radiator with capacitors on both sides and shortest possible wires - i'll take pictures this evening.
The controller board was powered by a separate battery, so big ground loops are unlikely.

As for the sound, what oscillator frequency is it actually running at? The chart says about 20K. What frequency are you hearing?

High, but audible whines.
The oscillator runs at 100KHz as it supposed to.

Have you layed out switching regulators before?

Yes, with varied amount of success or lack of it.
It's a hobby of mine, so i don't usually get to know all the problems.

 

[meBigGuy]

I'm not an expert in tracking ground loops in designs. You have them, for sure.

You have to think about where the currents flow when things switch. That flow causes voltages that, if they get into the feedback path, cause problems such as you are seeing.

There are all sorts of examples of good layouts (google switching regulator pcb design, etc), but what you are doing isn't even close.
Here are two
http://www.edn.com/design/component...leties-in-switching-regulator-and-LDO-designs
http://cds.linear.com/docs/en/application-note/an136f.pdf

I would say 50% or more of the design effort in a switching regulator is proper PCB design, with the rest being calculations and component selection.

You could play with LTspice and add trace resistances and inductances to see what happens.
http://www.linear.com/designtools/software/

 

[Baluncore]

You have built a voltage booster with a capacitance across the output.
An arc lamp needs inductance to stabilise the arc current.
As it is current spikes will probably damage your output capacitor driving the non-linear arc.
Insert another inductor, but after the capacitor, in series with the discharge lamp.

 

[jim hardy]

You have built a voltage booster with a capacitance across the output.
An arc lamp needs inductance to stabilise the arc current.
As it is current spikes will probably damage your output capacitor driving the non-linear arc.
Insert another inductor, but after the capacitor, in series with the discharge lamp.

I'm with Baluncore .

Discharge lamps are non-linear.
If I understand , in slide rule days we'd have called this a "Relaxation Oscillator".

 

[Artlav]

Overview of the build attached.
I don't see any obvious current loops to interfere with the sensing.

You have built a voltage booster with a capacitance across the output.
An arc lamp needs inductance to stabilise the arc current.

The lamp refuses to ignite without considerable output capacitance.

Even so, the "clumping" also happen with a resistive load, and even in constant voltage mode.
So whatever is wrong here is likely wrong regardless of the arc lamp.

Insert another inductor, but after the capacitor, in series with the discharge lamp.

I think the series igniter acts as such an inductor.
For resistive load it's presence or absence makes no difference.

 

[Artlav]

Hm, moving big output capacitor negative end to after the sense resistor makes the whine go away.
It still gives out 8A instead of 5A it's set to, but the input is now 15A instead of 21A.

 

[Artlav]

Here is what i see across the current sense resistor (with 10nF capacitor at the input), with the arc lamp and current set to 5A (0.5V on the amp input).

I think it sort of averages to 0.5A, but it does not look good at all, and DC meter shows 8-9A.
So, no whine does not equal no problem...

 

[Artlav]

Ok, I'm making progress even though I'm still moving ahead blindly.

I've rearranged the components as shown below.
There is a 30uF capacitor before the sense resistor and a 100uF one after it, the sense resistor have a 10nF capacitor across it.
This makes it much more stable, but still far from perfection.

The driver is set for 5A, and it gives about 5A, give or take one.
The thing is very sensitive to anything, and seem to settle in one of a few stable states - a bit more current, a bit less current, a bit more whine, a bit less whine.
An extra pulse from the igniter, moving wires around, or the lamp heating up - many things can make it jump from one state to another.

Here is a video of it, you can hear the whines and noises it makes if you turn the volume up:


For reference, there is a high pitch whine at the start, which turns into white noise-ish sound when i move the lamp around, then turns into a series of lower tones when i fire the igniter one more time, jumping an amp up and down for some time, then settle at half an amp above the setting with a very high whine.

Just in case it's some recognizable issue.

All in all, I'm at loss what else to try.

 

[meBigGuy]

It is ground loops and isolation, bypassing, etc. Did you read through the two example layout application notes I posted?

Do you get the same issues when you use a resistive load?

Once it works cleanly for resistive loads you may need to compensate for the lamp load. Others would be more helpful for that.

I think you are essentially trying to add components to cover up design issues. Sometimes you can get close enough to call it working, sometimes not.

 

[Artlav]

Did you read through the two example layout application notes I posted?

Yeah, I'm going through them slowly.
Quite a bit of it is irrelevant, but many things from the second one are.

For example, i found out that i can reliably switch between modes of wrongness by moving the current feedback wires a tiny bit relative to each other.
This is so wrong...

The big question is, however, how to do it right?

Do you get the same issues when you use a resistive load?

Pretty much.
It's a bit more stable with a resistor, but same kind of issues remain.

 

[Baluncore]

I once tried to stabilise magnetron current by using a vacuum tube as a constant current supply. No matter what I did I could not outrun the hard switching characteristics of the magnetron. In the end I reduced supply capacitance and used an RF inductor.

You have two problems here. You have a voltage regulator topology with a capacitor, not an inductor close to the load. You have no-low pass spike filter between your current sense resistor and the regulator.

Once started the arc must continue and be stable, that is the job of a series inductor, probably with an RF ferrite core, not LF iron powder. The current feedback must be low pass filtered to remove spikes faster than the regulator switching rate. That is confusing the slow cycling regulator. It explains why moving your output capacitor across the current sense resistor calms things down a little.

What type and part number is the xenon lamp ?
How does it start ?

 

[AlephZero]

Looking at your photos, it might be worth reviewing the "design rules" for the layout of the circuits and havign a general tidy up. With wires carrying 8A or more at 100 kHz, you don't need much "random" coupling to create measurable feedback currents and voltages. If everything is solidly "nailed down", at least you might have just problem to debug instead of random variations.

But sometimes you can solve this type of problem by NOT "nailing things down". We once had an instability problem in a machine generating about 50MW of power, where we calculated the amount of power that needed to get to the wrong place and trigger the problem was only about 5W, i.e. one millionth of the total power available. We fixed that one by reprogramming the electronic control system to continually "waggle the gas pedal" by a tiny amount, which had no practical effect on the machine except that whatever conditions caused the feedback loop never existed for long enough to let the problem develop. That was a lot quicker and cheaper than trying to find where the 5W was coming from, and stop it.

 

[jim hardy]

Baluncore asked two very good questions..

here are two articles on xenon lamps, maybe they'll help

first one
http://www.excelitas.com/Downloads/Cermax_Eng_Guide.pdf
describes trigger, boost and DC phases around pages 20-26
I found section 4.2.3 interesting, and looking at your current waveform I'd wager that your trouble lies in boost to DC transition.

second one
http://ww1.microchip.com/downloads/en/AppNotes/01372A.pdf
has an interesting description of lamp startup surrounding figure 3 on page 2

now I've never messed with these lamps
but your scheme looks to me like it might need a few more parts.

 

[meBigGuy]

Regardless of the xenon lamp issues, you need to first fix the instability issues with resistive loads. The other instability issues due to load will generally take a different form and should be attacked later (unless you want to do LTspice simulations with a realistic model of the lamp now).

There are many different examples and application notes addressing proper switching converter layout. They all have large low inductance ground areas that properly channel the input and load currents. I agree the 2nd article had more pertinent information. There are better ones out there. You read enough and suddenly you see the current paths and what you need to do. No shortcuts.

I've seen articles on "clumping" or "bunching" also, but I can't find them at the moment. There was one solution that fed back a switching transient into the reference, or something like that (a coworker told me about it). I'll keep looking.

Try adding a snubber also, just for grins.

The only time I had to troubleshoot a problem such as yours it was a circuit someone else designed. I added additional bypassing, some ground straps, and reversed the inductor and was lucky enough to get it to work "good enough".

 

[meBigGuy]

From an ON semi paper:
There are two rules of thumb for PCB layouts:
“short and fat”
for all power carrying traces and
“one point grounding”
for the three different types of grounds within
a switching power supply. Traces that are short and fat
minimize the inductive and resistive aspects of the trace,
thus reducing noise within the circuits and RFI. One point
grounding keeps the noise sources separated from the
sensitive control circuits. The three types of grounds are
the input power return ground, the output power return
ground and the low-level control ground.

 

[jim hardy]

you need to first fix the instability issues with resistive loads.

second that.

Your current feedback voltage looks as if the lamp fires every 60 microseconds
current reaches a 20 amp peak then decays to near zero on ~25-30 μsec time constant

and the 494 doesn't pulse again until current goes below 5 amps

control theory says feedback needs to be faster than the process
but yours is instantaneous

if you can risk blowing up a lamp
or try it with a 2 ohm resistor load

insert a little bit of RC delay in series with pin 16
perhaps 100 microsecond time constant for starters
so that the poor little 494 is responding to time average of its last several cycles, not the instantaneous current

things should get either much better or much worse.

but you'll learn something. And if you post another photo, so will I.



old jim

 

[Artlav]

here are two articles on xenon lamps, maybe they'll help

Helps quite a lot. :)

The first one in particular, since it gave me a few ideas - i haven't realized that you might get away with a buck converter if you use a diode-isolated boost for initial ignition, instead of a 64V battery or something impractical like that.
That might be something to try if all else fails, since a buck sounds like something much more stable in a constant current role.

Also, nice detailed description on how the lamp starts - no wonder the output capacitors were necessary, without them there is nothing to provide the current between ignition and the driver catching up with it.

The current feedback must be low pass filtered to remove spikes faster than the regulator switching rate.

insert a little bit of RC delay in series with pin 16
perhaps 100 microsecond time constant for starters
so that the poor little 494 is responding to time average of its last several cycles, not the instantaneous current

The effect is completely opposite - the current regulation pretty much goes out of the window.
No difference between 100Ω or 1KΩ or 10KΩ, with 10nF capacitor.

Tried with 3Ω and 5Ω loads - 5Ω goes up to voltage limit (and 8A), 3Ω goes over the input limit, essentially unregulated.
Both work almost fine without the RC filter.

The scope shows a flat line on the cc input, with progressively less noise as the R increases.
Without it, there is noisy pulsing.

What type and part number is the xenon lamp ?
How does it start ?

It's a soviet 150W short arc xenon lamp, ДКсШ-150, runs at 7.5A.
Starts with an in-series igniter - an HV capacitor discharges into a primary of the ring transformer with the main wire wound around as secondary.

And if you post another photo, so will I.

Not sure what to make of that...

 

[jim hardy]

Not sure what to make of that...

what I meant was - if you post another photo, I too will learn something.
haste makes bad grammar I guess.

 

[jim hardy]

no wonder the output capacitors were necessary, without them there is nothing to provide the current between ignition and the driver catching up with it.]

I'd been wondering if a bigger cap and an ohm to limit the current might stretch your "boost" interval

feedback delay: 10K and 10 nf = 100 microseconds, should have been enough.
try 1k and 1 μf just to push it to an extreme, a whole millisecond. At some point it should oscillate at ~ that delay period.

also --- 494 can draw [STRIKE]0.3 milliamp [/STRIKE] some current into its feedback terminal , so try RC delay with lower r and more c to avoid that voltage drop across the R of your RC delay... maybe 1 uf and 100 ohms.

http://www.ti.com/lit/ds/symlink/tl494.pdf page 3 "Recommended operating conditions" says design for 0.3 milliamp, but bias current is only 1 μa per Error Amp section on page 4.

If input current to error amp runs higher while inputs are unbalanced, that may explain your loss of current regulation .

 

[Artlav]

I think i am loosing my mind.
Something is working backwards in here.

Let's start back where we left.
The flat line with RC filter was due to the zenner on the current sense input, which draws considerable leakage current, overwhelming the RC filter.

Without it, things started happening.

More filtering - lower note, less stability.
Less filtering - higher note, more noise.
Eventually, i got sick of the thing picking up everything from it's own switching noise to hand waves over the wires to apparently a local radio station (or some other audio source).

So i decided to murder that problem with as much overkill as i had on hand.
MAX4080T current sense amplifier, small 0.01 Ω sense resistor, high side sensing, no sense ground loops, clear low impedance feedback signal.
Here is the appearance now:

Got the signal filtered with 100Ω by 0.1uF RC filter, and it looks quite clean now.
But it does not work at all, and now i can see the anomaly cleanly.

The feedback is backwards, for some time at least.

All of the below is with a 5Ω load.

The current limit is set to 1V, here is what i get:

At first i thought it was due to too much output capacitance, somehow.
So, down to 30uF:

See anything strange?
Let me add the duty cycle on the second channel.
The limit is 0.5V here and below.

The current is growing when DC goes down, for some time.
The current is decreasing as DC increase.

It seems that a change in duty cycle results in a momentary change of current in opposite direction.
Then it starts to follow the duty cycle, but by that time it's already far over or under the setting.
No wonder it's oscillating like crazy...

What is causing this?
Note how rapidly the current starts to climb, and compare with the rate of climb between the cycles when the DC is high.

As far as i can visualize it, it's the inductor getting charged and discharged - with the DC down to 0 it starts to discharge all the current in it into the load, with duty cycle starting to increase it's getting charged while the load runs on capacitors.
Does that sound plausible?

In any case, is that a fundamental flaw of attempting a constant current boost converter, or is there some sort of a solution to this?

I've been looking at automotive HID ballasts, and they boost 12V to something high with a flyback, then regulate down from there with a constant current buck regulator.

 

[Baluncore]

As you have found, all those long wires will make an oscillator rather than a regulator.
Your converter cannot be a switching voltage booster and a fast current regulator in the same time.

What low pass filter do you use to attenuate current spikes before feeding them back to the regulator amplifier?
Those two tasks have very different time specifications.
Your design, if it regulates anything, regulates the peak amplitude of it's current spikes.

I've been looking at automotive HID ballasts, and they boost 12V to something high with a flyback, then regulate down from there with a constant current buck regulator.

Now you are starting to recognise the necessary circuit topology.

 

[jim hardy]

I'm just wondering out loud here.

your 100 ohm / 0.1 uf cap introduces 10 μsec delay into feedback.
Your basic cycle time is a lot longer than that, 400 μsec with 30 uf (DC output filter across load?)
and about 900μsec with previous capacitance.
Both of those are a lot longer than your 10 μsec delay in the sense line. So I don't think it's hurting you, yet, ,,...

So - what region of your closed loop has a response time around a half millisecond?

5 amps DC charges 30 uf at rate of 167 volts per millisecond
but your charging current is delivered in short gulps, not steady DC.

Are above traces with lamp load or resistor load?
What is the yellow trace now - current through load or voltage across output capacitor?

 

[jim hardy]

What's purpose of C1 on pin 2 of TL494 ?
I don't see one here:
http://www.ti.com/lit/an/slva001e/slva001e.pdf fig 35

 

[Mordred]

reference 4.2.3 describes the function of C1 ties into the master clock operation. coupled to Q1,

edit yeesh I thought motorolla manuals were bad, this one has a helter skelter layout. Would have been nice if they had a complete circuit layout instead of having to find each block control individually. From what I can tell though can't be sure c1 is for an external clock tie in, controlled by RT, still trying to figure out the D1, ah kk its for synchronization of 2 or more tl494's and can be disabled by tying RT to the reference supply. Used only for generation of a sawtooth wave form.

 

[jim hardy]

Hmmmm, I read section 4.2.3 as describing a sync input for master oscillator to an external clock via a capacitor they called C1 connected to a timing pin.

Artlav's C1 is tied instead to an error amp input pin.

I'm trying to re-learn this IC so please don't mistake my questions for statements of fact.

I'm looking at his circuit for something that'd make transit time around his loop almost a millisecond to explain his 400 and 900 microsec cyclic behavior. R9C1 looked like a candidate ; if only I could work this thing in my head.
Twenty years ago I could have - ahh the joys of aging !

yeesh

indeed this isn't one of TI's more straightforward appnotes.
Subject matter was so familiar to the author that in his examples he didn't make it clear to a beginner which components are inside vs outside the IC. Fig 20 for example, I think that Q1 is part of an external synch circuit and not the internal Q1 output transistor.. but again don't take that as implying certainty...


old jim

 

[Mordred]

your correct on the dual function on pin 2 its also used as a comparator with pin 1. I started looking at other sites, for some better details.

http://www.bcae1.com/switchingpowersupplydesign/switchingpowersupplytut01.htm#strc

this site may help as it also includes calculators, save us some math to run numbers.

here is a datasheet with the electronic characteristics and some circuit examples.
http://www.hqew.net/product-data/TL494/TL494-DataSheet.html
http://www.hqew.net/files/pdf/PartsDictionary/ProductDatasheet/69f919a0-d947-45d8-88ec-71844f765909.pdf?key=TL494

those links may shed some insight by supplying details missing in the pdf

edit first site is a different board

 

[jim hardy]

It seems that a change in duty cycle results in a momentary change of current in opposite direction.
Then it starts to follow the duty cycle, but by that time it's already far over or under the setting.
No wonder it's oscillating like crazy...

What is causing this?
Note how rapidly the current starts to climb, and compare with the rate of climb between the cycles when the DC is high.

is this lamp load or resistor load ?

arc lamps have a region of negative resistance
and probably some time constant as gas ionizes and warms up around the initial arc
I note your 12V can feed through your 10μh choke and provide lamp current at zero duty cycle
but apparently that's not enough to keep lamp lit(if that's your load)

you know just where you've connected 'scope, and whether your probes are 1X or 10X, and what is the nature of your load,
but we don't

a trace of voltage across and current through your load would remove some doubt for me...

now -- A closed loop that is perfect except for having too much gain for the amount of delay around itself will oscillate wildly, as yours seems to be doing.
But before making that diagnosis I'd want to be sure what is the nature of your load,

and understand your 494 circuit. Is it your own design or from a reference someplace else ?
Are R9, R10, C1 just for power up sequence ?
494 is beginning to come back to me - haven't messed with one since late 90's...
so kindly be patient

could be Baluncore is right -
a current source with a capacitive load will have inherent delay.
there's time constants that have to be kept well apart.

old jim

 

[Mordred]

I recommend we isolate the lamp with a close resistor load, measure the load on initial lamp start up as it will, I would think drop once its fully on. Then try a potentiometer on the load side and check the signal behavior. If the behavior improves then we can look into a means to regulate it. If it doesn't improve then we know were still configured wrong on the 494

 

[Artlav]

Are above traces with lamp load or resistor load?
What is the yellow trace now - current through load or voltage across output capacitor?

5Ω resistive load.
The traces are across current feedback output - 20:1 amp of current across 0.01Ω sense resistor, with RC filter of 100Ω and 0.1μF.

So - what region of your closed loop has a response time around a half millisecond?

The ignitor transformer?
It's 200μH worth of choke in series with the load.


All in all, I'm likely to give up on this design, and try a CV boost followed by CC buck instead.
So, thanks all for help.