Speakers, Output Transformers and overshoots - some data

[nickb]

There's been some discussion of late on this topic. I was a bit dubious about some of the things being said and what I thought seemed to be borne out by simulation. Anyway, you can't beat an actual experiment for real data so I grabbed a 4 ohm speaker from the spares and started looking.

Experiment 1 - Tap cone - No load
Shows cone resonance at 65Hz with a surprising 2Vpk signal. The speaker is acting as a generator. A very sensitive microphone!



Experiment 2 - Apply a 5V pulse to see how big an overshoot we get
The cone motion is identified by the damped 65Hz oscillation. The energy stored in the suspension is dissipated as sound and heat causing the signal to decay. But wait! Look carefully at the bottom of the screen, just between the "Ch2" and "2.00V" labels. There is a short and very big negative spike.



Experiment 3- Let's check out the spike.
2V pulse this time. The spike is hard to see as it so fast so look carefully. At over 50V or 25x the applied signal! When reflected through the output transformer x30= 1500V. This is the kind of spike that can start an arc. And that was only a 2V stimulus. The stray capacitance of of an actual OPT will reduce that quite a bit but I didn't have one to test with and in any case I wouldn't want to damage one. This spike is from the voice coil inductance, not the cone motion.

The speaker was face down on the bench so that the air is damping the cone motion so you don't see the 65Hz resonance but rather an exponential decay. This was a surprisingly big 10V due to cone motion. I'm guessing the air pressure causes the cone to return to the center quickly and the greater speed causes greater rate of change of flux and a bigger voltage out. I don't think you would ever see this much in practice. I just mention it out of interest. Still, it makes you appreciate how the cabinet can affect the results.




Experiment 4 - add a source impedance to simulate the driving amplifier

The cone resonance is now heavily damped by a 16 ohms resistor across its terminals. This is a damping factor of 0.25. Why, 16 ohms? I just had that handy.

[pdf64]

How interesting and well presented! Thanks Nick.
I think that when the speaker is driven by a heavily overdriven amp, eg bias shift may cause the output to flip between one tube and the other with a transition time during which neither side is conducting.
During those transitions, the output source impedance may be very high, allowing the big inductive spike of Ex3 to form and stress the OT primary circuit insulation.

[nickb]

How interesting and well presented! Thanks Nick..

Thanks, Pete

I think that when the speaker is driven by a heavily overdriven amp, eg bias shift may cause the output to flip between one tube and the other with a transition time during which neither side is conducting.
During those transitions, the output source impedance may be very high, allowing the big inductive spike of Ex3 to form and stress the OT primary circuit insulation.

Indeed. That is why I did most of the tests into an open circuit for the worst case.

[pdf64]

Further to that, my experience / hunch is that for the same speaker, power and degree of signal overdrive, cone cry is much more of a problem when using a smallish overdriven power amp as the main clipping stage, cf an overdriven pre-amp and beefier power amp working within its linear range, due to the beefy power amp source impedance damping the speaker movement at all times.

[J M Fahey]

Excellent and very useful test.

And yes, the narrow high voltage pulse comes from the VC inductance which is purely electrical, while the mass and flexibility are *equivalent* to a large *mechanical* inductor and capacitor which account for speaker resonance but very little to electrical phenomenons such as flyback pulses.

[nickb]

Excellent and very useful test.

And yes, the narrow high voltage pulse comes from the VC inductance which is purely electrical, while the mass and flexibility are *equivalent* to a large *mechanical* inductor and capacitor which account for speaker resonance but very little to electrical phenomenons such as flyback pulses.

Nice to see you back JMF - you (and your wisdom) have been missed!

[Mike Sulzer]

Experiment 3- Let's check out the spike.
2V pulse this time. The spike is hard to see as it so fast so look carefully. At over 50V or 25x the applied signal! When reflected through the output transformer x30= 1500V. This is the kind of spike that can start an arc. And that was only a 2V stimulus. The stray capacitance of of an actual OPT will reduce that quite a bit but I didn't have one to test with and in any case I wouldn't want to damage one. This spike is from the voice coil inductance, not the cone motion.

The speaker was face down on the bench so that the air is damping the cone motion so you don't see the 65Hz resonance but rather an exponential decay. This was a surprisingly big 10V due to cone motion. I'm guessing the air pressure causes the cone to return to the center quickly and the greater speed causes greater rate of change of flux and a bigger voltage out. I don't think you would ever see this much in practice. I just mention it out of interest. Still, it makes you appreciate how the cabinet can affect the results.




Experiment 4 - add a source impedance to simulate the driving amplifier

The cone resonance is now heavily damped by a 16 ohms resistor across its terminals. This is a damping factor of 0.25. Why, 16 ohms? I just had that handy.

Very interesting test, and relevant to what I am working on now (making a solid state output stage look like a tube output stage to the speaker). As JMF said, the spike is electrical in nature. It comes from trying to kill the current through an inductor instantly when the driving amplifier is high impedance, that is, approaching a current source. (A guitar amp using pentodes and little or no global feedback looks a lot higher in impedance, that is, looking back into it, than the load impedance.)

If you used a voltage source (such as nearly all SS amps are closer to), when you apply the voltage pulse, the inductor charges up smoothly to its final current. When the voltage drops, the current is continuous (no significant transient if the amp is very low impedance), and the current is absorbed without much impact on the voltage. (Careful, though. If the current goes too high, the current limiting cuts in and the voltage might go way up blowing your transistors.)

Guitars do not have transients such as that, but the difference between tube and SS impedance still has a significant effect on the frequency response. It turns out that it is easy to make a high output impedance SS amp with modern devices, but it just is not what is normally done.

(Got a house full of family now; will have some results sometime after Christmas.)

Merry Christmas, and/or other holiday of your choice!

[nickb]

Further to that, my experience / hunch is that for the same speaker, power and degree of signal overdrive, cone cry is much more of a problem when using a smallish overdriven power amp as the main clipping stage, cf an overdriven pre-amp and beefier power amp working within its linear range, due to the beefy power amp source impedance damping the speaker movement at all times.

Well, things are never what you expect...

Since I was now happy that the speaker model was close enough to reality. I took up your suggestion and modelled these situations. I'll start off with the simplest case but it gets messy really quickly so you'll have to keep your wits about you.

Situation 1: Overdriven sine wave into resistive load
I used a perfect zero leakage inductance transformer to show only the effects of the load. As expected, you see some severe blocking distortion but no nasty overshoots. The supply voltage is 450v and the peak input signal is 2x the bias voltage 100Vpk for some serious overdrive.



Situation 2: Same as 1 but with a speaker for a load
Now it gets a little messy since I've got three plots to show what it going on. Stick with it and it will become clearer...

Start with the RED trace. This is the load current. This looks very much like the plate voltage from above.

Next, the GREEN plate voltage plot. This is very similar to the resistive case except we now have a kink in the middle.
Note the current lags the voltage due to the reactive load.

Last, the BLUE trace. This is the load current differentiated with respect to time. Notice how it has the same shape as the plate voltage? This is because the load looks like an inductor and since V=L.d(i)/d(t) the plate voltage follows the voltage across the voice coil. Note this inductance is the almost entirely due to the voice coil - the cone motion has almost no effect.




OK. Now it's time for the low power overdrive vs high power with preamp overdrive tests ...

Situation 3: Low Power amp over driven with 1kHz sine wave at 2X bias.

This amp is driving about 11W out. Notice the max plate V is only 200V over the supply. If you look at the effect of the speaker load, then it's only causing only +140V over the supply. No worries here.




Situation 4: High Power Amp under driven by a square wave (coming from an overdriven preamp).

Again about 11W out. Now trouble's a-brewing. The overshoot is now 850V over supply to almost 1.3kV. The width of the overshoot comes from the voice coil inductance (not motion). The oscillitary fuzzies come from the transformer leakage inductance resonating with the tube plate capacitance. Why is this voltage so high? It because it is rate of change of current that matters and the rise time of the square wave is much higher that in the low power amp slow sine wave. Notice I limited the rise time to 50uS to keep this realistic. Watch what happens if we have a faster rise time next...




Situation 5: High power underdriven amp with 10uS rise time square wave input

Ouch! We are up to almost 4kV! Proving once again it's d(i)/d(t) that matters.




Situation 6: Low Power Amp with 5kHz over driven sine wave.

So, since it d(i)/d(t) that matters, lets look at the low power amp with a 5kHz input. No surprises, I hope. Since the rise time is higher the voltages are higher too.



Situation 7: Adding catch diodes to the amp in (5)

The diodes chop off the low going undershoot but, because of leakage inductance, they don't help the overshoot much. This is why MOV suppressors are such a good idea.




It has to be said that the real word is more complicated that what I have here ( for example I have ignored the OPT capacitance) but the important points are these:

1) It's high frequencies in conjunction with voice coil and leakage inductance that cause the problems. Real waveforms are a a lot more complex that I have above but if you can suppress the higher frequencies, then you will have less overshoots. Guitar speakers don't do much over a few kHz anyway so you might as will lose them before they cause breakdown problems.

2) The cone motion doesn't come into it. Really. For these fast things the cone is hardly even moving.

3) MOV protection is cheap and very effective

VERY IMPORTANT: NO AMPLIFIERS WERE HARMED IN THE MAKING OF THESE PLOTS...
...yes, there are very real benefits to simulation

[trobbins]

Top stuff Nick.

For the measured results, did you use a constant voltage supply with sufficient current to suit the voice coil resistance? It looks like when the DC source is disconnected, then the 'motor' voltage rises to a peak as the voice coil moves from its initial extended position, and through the central (resting) position (maximum velocity of coil).

As I understand your simulation set up, the dI/dt of the conducting valve at the time it is driven in to cut-off (the other valve being in cut-off) is the principal arbiter, along with that primary half-winding's leakage inductance, that is generating the spike?

Did the simulation use a signal source to drive the output stage, or did you use a practical driver stage to drive the output stage?

Ciao, Tim

[octal]

At the risk of mild thread jacking, this thread touches on a question I have. As I understand it, the max voltage quoted in tube datasheets takes in to account these back EMF transients. (i.e. an EL84 is rated for "300v max" but in actuality, it's probably going to see 600V+ peaks on its plates in use.) Given that the output transformer works like a step up transformer for the back EMF, is it safer to "bend" the the max voltage rating when you're running a quad of tubes versus a duet? For our EL84 example, a duet might use an 8K:8ohm transformer whereas a quad might use a 4K:8ohm transformer. Driving the same speaker, I would expect to see much higher voltage spikes in the 2XEL84 amp vs. the 4XEL84 amp due to the difference in transformer ratios. Am I conceptualizing this correctly?

[nickb]

Top stuff Nick.

For the measured results, did you use a constant voltage supply with sufficient current to suit the voice coil resistance? It looks like when the DC source is disconnected, then the 'motor' voltage rises to a peak as the voice coil moves from its initial extended position, and through the central (resting) position (maximum velocity of coil).

I used a 9v battery for the first test and a 3A lab PSU set to 2V for the rest.

As I understand your simulation set up, the dI/dt of the conducting valve at the time it is driven in to cut-off (the other valve being in cut-off) is the principal arbiter, along with that primary half-winding's leakage inductance, that is generating the spike?

It's the d(i)/d(t) in the inductance under consideration that matters.

Did the simulation use a signal source to drive the output stage, or did you use a practical driver stage to drive the output stage?

Ciao, Tim

I used voltage sources to keep it fast and simple.

[nickb]

At the risk of mild thread jacking, this thread touches on a question I have. As I understand it, the max voltage quoted in tube datasheets takes in to account these back EMF transients. (i.e. an EL84 is rated for "300v max" but in actuality, it's probably going to see 600V+ peaks on its plates in use.) Given that the output transformer works like a step up transformer for the back EMF, is it safer to "bend" the the max voltage rating when you're running a quad of tubes versus a duet? For our EL84 example, a duet might use an 8K:8ohm transformer whereas a quad might use a 4K:8ohm transformer. Driving the same speaker, I would expect to see much higher voltage spikes in the 2XEL84 amp vs. the 4XEL84 amp due to the difference in transformer ratios. Am I conceptualizing this correctly?

You're thinking voltages. It's current that matters.

To a first approximation, your quad amp will delivery twice the power of a dual. Therefore the current in the speaker is sqrt(2) higher and so the voltage due to the d(i)/d(t) is also sqrt(2) higher. However your OPT turns ratio is also SQRT(2) less than the dual so the voltage due to the voice coil is about the same at the plates. BUT, the energy is doubled due to higher current. That means twice the hazard.

Also, the lower primary impedance (half) means the primary inductance will double all things being equal. The current in the primary is doubled since we have the same supply voltage. Thus the voltage due to the leakage inductance is x 2 x 2 = x 4 since v= L x d(i)/d(t). The energy stored in the leakage inductance is octupuled=0.5 x L x i^2, ouch!

Use the biggest MOV's appropriately voltage rated you can get and sleep easy trobbins has done a bit of work with MOV's and I seem to recall he had a sort of white paper on them. He's your go-to guy on that.

EDIT: Sorry, the inductance halves so the voltage is x1.4 and the energy is x4 see post #17

[octal]

You're thinking voltages. It's current that matters.

To a first approximation, your quad amp will delivery twice the power of a dual. Therefore the current in the speaker is sqrt(2) higher and so the voltage due to the d(i)/d(t) is also sqrt(2) higher. However your OPT turns ratio is also SQRT(2) less than the dual so the voltage due to the voice coil is about the same at the plates. BUT, the energy is doubled due to higher current. That means twice the hazard.

Also, the lower primary impedance (half) means the primary inductance will double all things being equal. The current in the primary is doubled since we have the same supply voltage. Thus the voltage due to the leakage inductance is x 2 x 2 = x 4 since v= L x d(i)/d(t). The energy stored in the leakage inductance is octupuled=0.5 x L x i^2, ouch!

Use the biggest MOV's appropriately voltage rated you can get and sleep easy trobbins has done a bit of work with MOV's and I seem to recall he had a sort of white paper on them. He's your go-to guy on that.

Thanks for setting me straight!

[Mike Sulzer]

Also, the lower primary impedance (half) means the primary inductance will double all things being equal.

I believe that the primary inductance goes down a factor of two. The required impedance of the magnetizing inductance at some reference low frequency is some multiple of the primary impedance, and therefor scales with that impedance.

[J M Fahey]

Thanks

Consider it a (very short ) Sabbatical Year .... or a very long Sabbath !!!!!!!!!!!!!!