Author Topic: [Solved]Type 43 core resonance (Hartley oscillator frequency jump during tuning)  (Read 2822 times)

0 Members and 1 Guest are viewing this topic.

Offline szoftveresTopic starter

  • Regular Contributor
  • *
  • Posts: 108
  • Country: us
Hey All,

Of the few Hartley-oscillators I've built, I've noticed a common symptom: while adjusting the tuning capacitor to the more open setting (least capacitance), the oscillation frequency suddenly jumped - and skipped a small frequency band.

Specifically I've built the attached circuit, (grid leak: 100k and 47pF, Tuning capacitor: 10-330pF with a 5-30pf trimmer, Inductor: 4+17 turns on an FT50-43 core), which is supposed to produce frequencies between ~500kHz and 1700kHz. While inspecting the oscillation on an oscilloscope (very loosely coupled), the frequency goes up and then "hesitates" at around 1500kHz as I keep turning the tuning capacitor, then suddenly jumps up to around 1600kHz and the tuning continues to 1700kHz as I max out the tuning capacitor. It's not producing this behavior when adjusting the frequency the other way.

I've noticed the same thing a while ago with a VHF (transistor) oscillator, and attributed this behavior to the supposedly loose coupling of the lower turns of the air-core coil to the upper turns, but this shouldn't be the case here with a type 43 core material (u=850).

I'm wondering what plays role here, i.e. what could cause this circuit to prefer staying at 1500kHz instead of following the LC tank resonance. One wild guess is that the tube internal (grid to cathode and screen) capacitance forms a series (fixed) resonance point. Any suggestions, especially regarding how to mitigate this thing would be greatly appreciated.
« Last Edit: November 02, 2022, 05:30:58 am by szoftveres »
 

Offline TimFox

  • Super Contributor
  • ***
  • Posts: 7934
  • Country: us
  • Retired, now restoring antique test equipment
Re: Hartley oscillator frequency jump during tuning
« Reply #1 on: November 01, 2022, 11:42:49 pm »
You are seeing strange behavior at the minimum capacitance (maximum frequency) end of the tuning capacitor range.
Is there any possibility that you are not at precisely the minimum where you see the jump.
Also, your circuit should not have a problem with "squegging" (q.v.) unless the time-constant of the grid-leak circuit is too long.
Maybe try reducing the grid resistor by a factor of 5 and see what happens.
Are you monitoring the output with an oscilloscope when the strange jump occurs?
 

Offline szoftveresTopic starter

  • Regular Contributor
  • *
  • Posts: 108
  • Country: us
Re: Hartley oscillator frequency jump during tuning
« Reply #2 on: November 01, 2022, 11:49:33 pm »
I still have ~100kHz of tuning range -up to 1700kHz- after this jump happens. No squegging.
Yes, I'm on the cathode with a 1:10 probe (also seen this phenomenon in an electron-coupled mode, i.e. on the anode after the screen grid, which is pretty much 100% isolation)
 

Offline TimFox

  • Super Contributor
  • ***
  • Posts: 7934
  • Country: us
  • Retired, now restoring antique test equipment
Re: Hartley oscillator frequency jump during tuning
« Reply #3 on: November 02, 2022, 03:27:50 am »
Is it possible you have a local problem with a bent plate in your multi-plate tuning capacitor?
That could give a non-monotonic capacitance vs. angle.
 

Offline szoftveresTopic starter

  • Regular Contributor
  • *
  • Posts: 108
  • Country: us
Re: Hartley oscillator frequency jump during tuning
« Reply #4 on: November 02, 2022, 03:39:01 am »
Tried with a 10k grid-leak resistor, didn't help.
After the oscillation reached ~1500kHz, the frequency won't change any more, but the amplitude shrinks a tiny bit as I slowly keep tuning upwards - then it abruptly jumps to the new frequency (I don't see this happening when I tune the other way). This makes me think that it's some kind of secondary resonance point.
Checked the capacitor for any mechanical issues, nothing I could point at.
 

Offline szoftveresTopic starter

  • Regular Contributor
  • *
  • Posts: 108
  • Country: us
Re: Hartley oscillator frequency jump during tuning
« Reply #5 on: November 02, 2022, 04:27:53 am »
Did some more investigation:
It appears that there's a resonance of a very specific frequency around 1400kHz and the jump happens when tuning the other way as well (it's just more subtle and I didn't notice it).
This resonance frequency is very specific and doesn't change (stays within 1kHz), even if I adjust the trimmer, attach random cables to the grid, change the R or C values of the grid leak, or change out the tube.
Total mystery.
« Last Edit: November 02, 2022, 05:03:34 am by szoftveres »
 

Offline szoftveresTopic starter

  • Regular Contributor
  • *
  • Posts: 108
  • Country: us
Re: Hartley oscillator frequency jump during tuning
« Reply #6 on: November 02, 2022, 05:06:31 am »
Got the coil out, it seems like it has a resonant peak exactly where I see the weirdness. No change in the frequency regardless of the wiring or capacitive loading of either end.
Wound another coil on an identical ferrite core from the same stash and it doesn't produce a resonant peak.
 

Offline szoftveresTopic starter

  • Regular Contributor
  • *
  • Posts: 108
  • Country: us
Re: Type 43 core resonance (Hartley oscillator frequency jump during tuning)
« Reply #7 on: November 02, 2022, 05:30:31 am »
The solution: https://www.pa3fwm.nl/technotes/tn11a.html
Besides 190kHz, there are peaks higher as well (odd harmonics or other modes of the core).
« Last Edit: November 02, 2022, 06:12:56 am by szoftveres »
 
The following users thanked this post: mag_therm

Offline JohnG

  • Frequent Contributor
  • **
  • Posts: 567
  • Country: us
FWIW, ferrites used for inductors and transformers fall in two main groups, manganese-zinc (MnZn) and nickel-zinc (NiZn). Type 43 (from Fair-Rite Corp.) is a NiZn material. NiZn tend to be used at higher frequencies, and also have much higher resistivity, than MnZn. However, they are susceptible to having their magnetic properties damaged when exposed to a strong enough DC field, either due to too much DC current in a coil, even momentarily, or if you bring them too near a magnet. I don't know the actual mechanism, though one is proposed in the article linked. That's why you don't see NiZn used much for inductors in power converters.

If I recall correctly, you can actually get the core properties back by raising the temperature of the core above the Curie temperature and letting them cool back down. The low-perm NiZn cores tend to have quite high Curie temps (300C to 500C), but the high-perm (Fair-Rite 43's mu_r = 850 would fall into that class) tend to have Curie temps below 200C. So, you might try baking the bad core at 200C for a bit, and see if it gets better.

MnZn ferrites are much better behaved in this regard, but they don't have good performance beyond a few MHz.

I once heard a story about a company that manufactured cores who had a customer complain about starting to get bad cores for a design they'd been shipping for years. The cores had passed testing, so finally they asked the customer to send them cores, and sure enough, they tested bad, even though they had tested fine before shipping. No one could figure out how they went bad, as the customer had never had this problem previously. Finally, they went to the customer's assembly line, and saw the assemblers using a stick with a magnet on the end to pick up the cores out of the shipping tray, thereby ruining the core. This practice had just been started when they began to have problems.

John
« Last Edit: November 02, 2022, 02:09:21 pm by JohnG »
"Reality is that which, when you quit believing in it, doesn't go away." Philip K. Dick (RIP).
 
The following users thanked this post: antenna

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 21609
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Agree, could be acoustic effects.

Degaussing will do it, though you may need an oddly shaped / unusually powerful tool to do it, given the lumpy shape of the core.

Alternately, annealing above Tc will also do.

I was once testing a ferrite ring, by pulse excitation (square voltage pulse, measure the nominally-triangular current pulse), and had it suddenly snap into about 12 equal sized pieces!  Seems it caught exactly that kind of magnetization (trivial as I was doing unipolar pulses up to saturation) and I swept the pulse rep. frequency past a resonance, and, off it went.

That was at some 30kHz or so, I think, which seems suspiciously low given the figures on the link, and the number of fragments it went into, though? *shrug*

Forget if that was MnZn, NiZn or what (or if I had any way to know at the time -- it was a salvaged core in any case).


That's why you don't see NiZn used much for inductors in power converters.

Ehh.... more likely because of adequate performance and cost.  Power converters aren't much into the MHz just yet, and the air gap makes up for a lot of core losses -- even if the tan delta is near 1 for the raw material at the driven frequency, the Q factor goes up proportionally(?) to mu_r/mu_eff.  Basically, you're storing most of the energy in the air gap, and only the fraction of energy stored in the actual core material (which is small) is lost.

Maybe we'll see more NiZn use as GaN becomes more widespread; or maybe not, as powdered iron materials serve these applications well, too.  Your good old fashioned #2 mix is still quite attractive for these frequencies (low MHz); a molded composite, of similar blend, would do fine for SMT chip inductors and whatnot.

Nickel is rather expensive, as base metals go, so it will make up a significant fraction of component cost here.

Well, the same is true of tantalum... but despite that, it stuck around quite a long time.  It's largely displaced nowadays, by affordable large ceramics, and the introduction of polymers, I think?

On a similar note, there are "bleeding edge" sort of mixes of MnZn, with mu_r and losses straddling the line of NiZns like #43.  They're not as common as ye olde 3C90, 3F3, N49, etc. though.

---

Speaking of which, the same solution applies to the present problem.  The fundamental problem is trying to use a transformer core (high mu) as an inductor.  Expect problems!  The biggest problem being the massive tempco, poor manufacturing consistency (what's the tolerance on that stuff, 20%?), and high losses.  Evidently, resonances are also a possibility!

So, simple enough, use a gapped style core. :-+

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline mag_therm

  • Frequent Contributor
  • **
  • Posts: 633
  • Country: us
Hi sz, Thanks for the link to pa3fwm note.
pa3fwm  had problems with Amidon FT50-43 which is NiZn ferrite.
Amidon does not show magnetostriction on type 43 data sheet.
My reference book shows that  NiZn materials B1, B5 and B10 have magnetostiction ranging from
-15 , -38, and -65  [ * 10^-7 mm/mm] respectively at H = 3200 A/m

These values are ~ 10 times higher than typical MnZn cores for example  3H1 is -5  and 3E2 is about +2.

MnZn cores like 3E1, 3E2, and 3E3 are OK for use in HF radio

What type of cores are you using?

Comments about the circuit :
Grid Leak Bias:
Grid leak resistor is shown as as returning to ground through the gapless toroid.
The  DC will be multiplied by the turns  to give an offset on the core's BH curve,
which will cause its relative permeability and hence inductance to change with the DC current.
Further, if the grid goes positive during the half cycles of the oscillator,
 the resulting grid current pulses will swing the inductance cyclically at half the oscillator frequency.

For these reasons it may be better to return the grid leak resistor directly to the cathode, as is usually done.
 The blocking capacitor then eliminates DC from the inductor.

Grid resistor value:
Usually 1 MOhm or higher to get a useable AC swing on the grid before the grid goes positive.
It is difficult to measure the actual grid voltage due to the high source resistance, measuring the leak resistor  current is more accurate.

The plate characteristics of the tube with Vg as parameter, can be compared with  measured grid amplitude of the oscillator.
For example if the grid bias is at -1.0 Vdc , the Hartley tank voltage should be not be more tha abouit 0.75 V peak-to-peak.
 

Offline jonpaul

  • Super Contributor
  • ***
  • Posts: 3322
  • Country: fr
try  Micrometals powdered iron cores.

Consider self resonance of the high self capacitance toroïdal wdg.

I would use a different core shape and winding techniques

Jon

Jean-Paul  the Internet Dinosaur
 

Offline JohnG

  • Frequent Contributor
  • **
  • Posts: 567
  • Country: us
It's not just material cost for NiZn inductors, at least in a buck converter, though that is an important factor. If you are running near the edge of saturation (pretty common in a hard-switched converter), you might have to start thinking about protecting the inductor. That could be a pretty big headache, best avoided.

Also, if you run a core near saturation, how does a reduce the energy stored in the core? I thought that just depends on the flux in the core. At least in a buck converter, you want to push as much average current through the inductor as you can, meaning you will design so that it is near saturation at full load. Core loss is a different story, of course.

I'm interested to hear about some of the newer MnZn ferrites. Do you know the names of some? I have not used raw cores for a few years, but I may be soon. I'm interested in all power applications in the low MHz range (PWM inductors, resonant inductors, and transformers).

John
"Reality is that which, when you quit believing in it, doesn't go away." Philip K. Dick (RIP).
 

Offline szoftveresTopic starter

  • Regular Contributor
  • *
  • Posts: 108
  • Country: us
Quote
Comments about the circuit :
Grid Leak Bias:
Grid leak resistor is shown as as returning to ground through the gapless toroid.
The  DC will be multiplied by the turns  to give an offset on the core's BH curve,
which will cause its relative permeability and hence inductance to change with the DC current.
Further, if the grid goes positive during the half cycles of the oscillator,
 the resulting grid current pulses will swing the inductance cyclically at half the oscillator frequency.

For these reasons it may be better to return the grid leak resistor directly to the cathode, as is usually done.
 The blocking capacitor then eliminates DC from the inductor.

The grid current has to go through the 100k resistor either way, which in my case is going through the coil as well. The tube cuts off at no more than 5V, the average grid current (~50uA) via 21 turns (~1mA*turn) will probably not magnetize this core. Also, mating the tube input capacitance (5pf) with an adequate grid-leak capacitor sets an upper limit on the resistance of the grid-leak resistor if one wants to avoid squegging.
I suspect the magnetization came from the cathode current somehow - either way, looks like ferrite ring isn't the way to go in this circuit. Thanks for the comment anyways!
« Last Edit: November 03, 2022, 05:05:19 pm by szoftveres »
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 21609
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
It's not just material cost for NiZn inductors, at least in a buck converter, though that is an important factor. If you are running near the edge of saturation (pretty common in a hard-switched converter), you might have to start thinking about protecting the inductor. That could be a pretty big headache, best avoided.

(Protecting?)

Saturation isn't as important as you might expect, especially at high frequencies where core losses restrict flux density further.

Or if ripple is very low, you're probably using cheaper powdered iron materials instead.  Which can have quite high Bsat, but much lower B(p-p) especially at frequency.  With high-ripple controls (like peak current mode) being most common, and when prioritizing compact size, these aren't so practical though.


Quote
Also, if you run a core near saturation, how does a reduce the energy stored in the core? I thought that just depends on the flux in the core. At least in a buck converter, you want to push as much average current through the inductor as you can, meaning you will design so that it is near saturation at full load. Core loss is a different story, of course.

"how does a" a = air [gap]?

The energy stored in the core is the same; it's given by e = B^2 / (2 mu).  (e has units of pressure, which is also to say, energy density.)  The difference is, you're storing much more energy in the air gap.  So, as a fraction of total -- yeah.

And again, yeah, can't always push so much current, because size, control method, or topology, may dictate that for you.  Like, I recently-ish did a BCM (quasi-resonant) PFC, which ended up with quite generously gapped ferrites (EE36 or so, 5mm spacing, so, effectively 8mm+ total gap length).  Maybe they could be smaller, but the litz wire was pretty well filling up available winding area.  Peak current is around 13A but I measured saturation at 40A.  Which was unexpectedly high due to the huge air gap, and will tend to be high anyway because of the lower flux density (something like 200mT at 200kHz -- BCM, so that's 0 to 200mT every cycle!).


Quote
I'm interested to hear about some of the newer MnZn ferrites. Do you know the names of some? I have not used raw cores for a few years, but I may be soon. I'm interested in all power applications in the low MHz range (PWM inductors, resonant inductors, and transformers).

Mostly very incremental improvement, AFAIK.  Like Ferroxcube's 3F3, 3F4, 3F45, etc. series.  I haven't perused material listings in a little while but they're easy enough to find on the usual suspects (EPCOS/TDK, Ferroxcube, etc.).

Oh hey, sweet compilation: https://people.zeelandnet.nl/wgeeraert/ferriet.htm

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline JohnG

  • Frequent Contributor
  • **
  • Posts: 567
  • Country: us
Saturation is important if you have a core whose properties can be degraded upon saturation, which includes a large subset of NiZn cores, the main topic of this thread. If you are using MnZn cores, then generally it is not a big deal, though it can be. I have seen a number of designs (mostly buck converters) where total losses went through the roof because the MnZn core was pushed too hard, the inductance dropped at the flux peaks, and ripple and RMS current went up a lot. The additional core losses did not limit this to any significant degree.

I did not say these were good designs, BTW, but they were existing commercial designs. You might be surprised (or perhaps not) at the number of existing shipping designs where inductor current has never been measured.

Thanks for the link. It's the most comprehensive list of core manufacturers I have ever seen.

John
« Last Edit: November 02, 2022, 08:50:43 pm by JohnG »
"Reality is that which, when you quit believing in it, doesn't go away." Philip K. Dick (RIP).
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 21609
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Reminds me, once upon a time we (PPoE) were tasked with a custom induction heater, lower frequency rating than usual (a kHz or so).  So, did the design work, needs a transformer of such-and-such, called around a few shops, eventually got a reasonable quote (a toroid about a foot across, wound on fine, I think 3 thou, GOSS), got the parts in and starting testing protos.  One transformer, after being hit with saturation (either because it was direct drive from an H-bridge, or through a coupling cap but without damping so the startup transient heavily magnetized it), never ran quiet again, always with a noticeable buzz (which was especially noticeable at this frequency!).

I suppose the same thing was probably in play, something like, residual magnetism, probably in a nonuniform manner (are there usually sites pinned so hard they only shift up at saturation?), or because of inhomogeneity (could be a lumpy formulation I suppose -- patches in the steel, or pockets in ferrite?), or because of geometry (of note, the start/end of the core spiral won't be magnetized equally with the rest; not sure how well glued down they were, I never saw the interior of those things), and so the magnetostriction was permanently increased despite the presumable degaussing effect of running at AC for millions of cycles.

Which, I think, is a non sequitur, the AC exposure part?  It seems tempting to think about continuous wobbling as tending to drive a system to equilibrium, like shaking a pile of sand, or straightening out a piece of wire: even if you don't deform it much on each step, you're still deforming it back towards that base condition.  But magnetic materials aren't in equilibrium like that, there may be a lower energy state with spontaneous magnetization (indeed there is, hence domains even at zero net magnetization) than at zero.  And so, magnetization can persist more or less forever, until the coercive force is finally exceeded.  (Of course, we use an oscillating decaying field in the hope that, at the initial peaks, all important sites are being pushed around, and so as time goes on, they're pushed around a little less, and a little less, averaging zero as the amplitude falls below the coercive threshold.)

Hm, if that's the case [high-coercive-force defects], I suppose degaussing -- well, it should still do the job, but it may take much more external field than you would otherwise expect; and maybe it's not feasible at all at that point to degauss it (could such a defect exist, which, once it takes a set in one direction, takes significantly more force to reverse it..??), but maybe it's just more practical to anneal above Tc at that point, assuming you can (heat ferrite to ~soldering temperature, good to go; obviously, a bit harder to do that for steel!).

Hadn't thought of it that way before, but that seems reasonable... domain pinning could be a population thing, where most take low force (giving the Barkhausen effect in aggregate), but very few have extraordinarily high coercivity, and then in combination with NiZn's higher magnetostriction, you get the observed effect.  I would imagine the distribution is a power law of some sort.  Maybe the population statistics vary with material as well, and it happens to be worse for NiZn?

I wonder how you would quantify that... Well, if it's a magnetization thing, and if you set up to measure Barkhausen noise, I suppose that would be a possible route.  Basically count the steps and see at what points the largest steps fall at; then reverse it and hope to count the same again.  Well, there's not really any reason for domains to be consistent from cycle to cycle, I suppose, is there... it's not a perfectly fixed array of discrete dipoles, huh.  But also maybe you can still identify, oh maybe even with local field probes around the core, such localized sites, that only change at extremely high field strength?

The ultimate implication being, if you knew exactly where those sites were, you could, like, literally drill them out of a core, ending up with a quieter / lower remenance part as a result.  Like defects in semiconductors causing yield reduction instead of absolute failure, it's a statistical process, not a homogeneous one.  Which, I mean, it's no accident, ferrites are semiconductors too, just ferromagnetic ones.

Tim
« Last Edit: November 03, 2022, 10:31:36 am by T3sl4co1l »
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline JohnG

  • Frequent Contributor
  • **
  • Posts: 567
  • Country: us
Those are great questions for a materials scientist >:D.

More seriously, magnetic materials have been poorly modeled for a century now. Maybe things are starting to improve, but I am not so sure that a full physics model, if ever developed, will be all that useful to an engineer. This is especially true for bulk materials. There are umpteen variations on the Steinmetz equation, and they are still all a fancy curve fit in the end.

On top of that, materials science is an expensive endeavor, and ferrite cores are cheap. I would guess that the margins are too low to justify serious material development, at least for most of the companies that make cores. No wonder progress is so slow.

Capacitors seem to make slightly faster progress, but not by much.

John
« Last Edit: November 03, 2022, 01:09:31 pm by JohnG »
"Reality is that which, when you quit believing in it, doesn't go away." Philip K. Dick (RIP).
 

Offline WPXS472

  • Regular Contributor
  • *
  • Posts: 67
  • Country: us
AT my previous employer, we had a serious problem caused by magnetostrictive resonance in a ferrite core. Seems a tech was using a magnetic screwdriver too close to the core. This was a major CATV manufacturer, and they got in a panic thinking that they were going to have to replace a bunch of product. It took a couple of days to identify the issue. The core was made by a major foreign brand. I found that heating beyond the curie point cured the problem. My suggestion that they change vendors to an American core that didn't exhibit this property fell on deaf ears. Since the problem had only actually been identified with 2 units, they decided to put a label on them saying not to use magnetic tools, and let the sleeping dog lie.
 

Offline szoftveresTopic starter

  • Regular Contributor
  • *
  • Posts: 108
  • Country: us
Annealing/de-magnetizing in real time (just for the sake of science):
https://youtu.be/Z_t8PBLibXA
Interesting to see how the mu disappears (S1,1 shoots up at the end of the video) with temperature.
« Last Edit: November 30, 2022, 11:02:59 pm by szoftveres »
 
The following users thanked this post: T3sl4co1l


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf