Output transformer protection, flyback diodes, MOVs, 2KV capacitors,

[mhuss]

Good luck finding any Thyrectors (or anything containing Selenium) these days. I note the date stamped on the datasheet says 1968(!). These are clearly from the "olden days" when GE actually made commercial parts.

The diodes are probably a good second choice, certainly better than nothing.

[nsubulysses]

R3000 FTW

[nevetslab]

At present, the feedback is connected. When I make the component upgrades, it will be disconnected, which is how they are presently building the amp.

I haven't seen Loudthud's scope photos of the anode voltages under speaker load, guitar signal applied. Do you know what the Thread # is?

[trobbins]

I am away from my PC with exact details, but I think it was just in:
http://music-electronics-forum.com/t28096/

I used his plots in:
http://dalmura.com.au/projects/Outpu...protection.pdf

I tried that technique on an amp last month. My digital scope allows xy and persistence, but I think this is one time when a good old analog oscilloscope is best. I also just used a sine oscillator overdriving a fixed bias EL34 amp (Eminar). The OT I was testing didn't seem to have the same leakage inductance, as I couldn't discern excessive voltage disturbances (just distorted ellipses), but I also didn't use a guitar input.

[robrob]

It's very cool to have R.G. Keen commenting in this thread. In case you don't know he is THE amp protection guru.

[tedmich]

among his many titles! RG is a true mensch

[trobbins]

MOVs slowly degrade with each over voltage conduction, so they gradually drop in breakover voltage until they conduct at normal voltage, then they die FAST.

RG, the reliance on a MOV as an integral part in almost any commercial or industrial device over the last few decades has meant a lot of performance improvement and characteristion since the earliest days of MOVs. MOV's are certainly seen in an exploded form by service people - pretty much due to lighning and swell conditions on the mains side, where transient and continuous energy levels can be huge.

Protection of output transformer windings from over-voltage is a much more benign environment for an appropriately sized MOV (or MOVs in series) than for mains connection.

Datasheets show derating curves for pulses, which includes an indefinite curve, along with an average power dissipation rating. From what I can deduce, even the small 7mm MOV disks I use do not exceed the indefinite derated curve for the worstcase one off inductive event I could define in an amp, or is the average power dissipation rating exceeded from continuous overloading with leakage inductance induced events such as the type that loudthud exposed in his x-y testing in threads #28096 and 28786.

[trobbins]

As I understand it, loudthud's test amp was a Fender 5F6A using 6L6, with feedback to the LTP PI and without any output stage grid-stoppers. The lack of grid-stoppers, along with 0.1uF coupling caps and feedback, probably cause strong blocking distortion, and consequentially cross-over distortion. The cross-over distortion would leave the OT primary windings without any loading for substantial periods, especially with a sharp attack strum.

I can see two possible causes of the plate voltage excursions observed:
- the speaker may be forcing the unloaded primary winding voltages
- dI/dt at the start of cut-off may be causing plate voltage excursions depending on the reflected loading of the speaker at that point in time.

My test amp used 56k grid stoppers, had no feedback, and I didn't use a guitar input, so my test rig may have been a bit more immune to cross-over distortion.

[loudthud]

I heard a ringing in my ears, but I thought it was just tinnitus. If you look at these threads, I think you'll find what trobbins was speaking of.

http://music-electronics-forum.com/t37469-2/

http://music-electronics-forum.com/t39829-2/

The 5F6A re-issue does have 0.1uF coupling caps and has, get this, 47 Ohm grid stoppers on the PCB from the factory. Gerald Weber worked on the amp before I got it, it has 5881s. He didn't pull the PCB and crudely tacked a resistor across the "slope" resistor on the top side.

Edit: The attachments in threads #28096 and 28786 were corrupted in the big virus attack.

[nickb]

RG, the reliance on a MOV as an integral part in almost any commercial or industrial device over the last few decades has meant a lot of performance improvement and characteristion since the earliest days of MOVs. MOV's are certainly seen in an exploded form by service people - pretty much due to lighning and swell conditions on the mains side, where transient and continuous energy levels can be huge.

Protection of output transformer windings from over-voltage is a much more benign environment for an appropriately sized MOV (or MOVs in series) than for mains connection.

Datasheets show derating curves for pulses, which includes an indefinite curve, along with an average power dissipation rating. From what I can deduce, even the small 7mm MOV disks I use do not exceed the indefinite derated curve for the worstcase one off inductive event I could define in an amp, or is the average power dissipation rating exceeded from continuous overloading with leakage inductance induced events such as the type that loudthud exposed in his x-y testing in threads #28096 and 28786.

Here's how I see it. I think the speaker and the leakage inductance both contribute significantly to the overshoot problem, but at different times.

There is plenty of room for debate on the actual values I have chosen, but I think they still sufficiently in the ballpark to lead to a valid conclusion.

Consider a nominal 50W amp with a damping factor of 1 driving an 8 ohm speaker with an inductance of 55mH. The largest current of 2.6A in the inductor occurs at resonance of 24Hz in the model I used. This translates to a power of 4.2 watts that is going to be dissipated somewhere and I'll assume it all ends up in the MOVs.

For the same amp the peak plate current is going to be around 200mA and the leakage inductance is around 20H * (1-0.995) = 100mH using 20H as the primary inductance. Each tube sees 1/4 of this so 25mH. The energy is 0.5 * .025 * 0.2^2 = 0.5mJ i.e. 0.5 Watts at 1KHz.

So the MOV must be able to handle a worst case of 4.2W continuously. We don't know how often we're going to hit this worst case, perhaps 10% of the time as WAG? Looking at the EPCOS S20 (23mm) series they are rated for 1W continuous and so at 10% duty cycle we are quite safe with 420mW at one MOV per plate.

What about no speaker at all? Taking 50Hz as the frequency of interest and the unloaded inductance seen by each tube as 5H. V=L.d(i)/d(t) so in our 10mS time and with 450V on the plates the current should be 900mA, in practice limited by saturation to ~300mA. The energy is 0.5*5*0.3^2 =0.225J or 11 watts at 50Hz. Again we won't reach this figure as it depends on how hard we are banging away (duty cycle) and the frequency is likely be higher than 50Hz and that will reduce the dissipation further. Our 1W MOV will probably not last long, but hopefully long enough for the user to realise there is a problem and fix it before it's too late.

I think that's the best I can come up with as my crystal ball is also a bit murky, as RG said...

[trobbins]

nickb, I agree that proposing and assessing scenarios is a good method to work through a better awareness and understanding of protection schemes - be it MOV based, or some other format.

Your scenarios relate to continuous operation, and I think the 'speaker-connected' scenarios are also for overloaded operation in an amp that is experiencing noticeable cross-over distortion. The 'speaker connected' plate current waveforms would then have portions of time when both PP valves are in cut-off, at which time there would be no loading on the primary half-windings, and hence the plate voltages are somewhat free to be driven by other forces. The duty-cycle for both valves in cut-off is likely to be somewhat indicated by loudthud's plots that show the relative brightness of traces experiencing over-voltage conditions compared to the time spent with one valve conducting and the other in cut-off - perhaps a lowish duty-cycle value.

Leo Gnardo described the speaker as a generator eloquently in t37469, where it's not just the speaker coil inductive energy wanting current to continue, but also kinetic energy from the moving mass driving an emf. With both those contributions, the plate voltage is likely to respond to the reflected OT secondary situation. One comment is then that probably all MOVs on primary windings would be acting as loads, and that loading would not be a 100% duty cycle, as the next valve to be driven in to conduction would take over the loading.

As I see it, primary leakage inductance could be an influence when again both PP valves are in cut-off, and dI/dt in the primary windings becomes substantial. dI/dt is probably highest at the time when a valve is driven in to cut-off (the other PP valve already being in cut-off). As you indicate, the energy in the leakage inductance could be consumed in a MOV across each half-primary. The power in to a MOV would then be frequency/2 and peak current squared dependant. One comment is that although plate voltage could be an over-driven square type waveform, the peak plate current reached will be frequency dependant.

For the scenario with no speaker loading, then worstcase plate current and MOV power consumption occurs with a continuous dominant bass note where the plate current is allowed sufficient time to increase along the saturation V-I curve to the max current region. A comment is that multiple MOVs may be deployed on the OT primary (eg. across each half-winding, and perhaps in series). Another comment is that the OT primary inductance may be larger than normally measured (typical inductance measurement is at 5-10Vrms, whereas inductance will typically increase up to at least the design full power voltage swing). There may be a residual level of energy consumed by the plate when it is turning off (ie. a switching loss) due to finite switching time, which would depend on the achievable slew-rate of the output valves, but that energy is likely to be a low %. So I guess its best to 'test' that the audio chain is working properly by strumming strings that include higher frequency strings, and not strumming continuously!

[nickb]

nickb, I agree that proposing and assessing scenarios is a good method to work through a better awareness and understanding of protection schemes - be it MOV based, or some other format.


Your scenarios relate to continuous operation,

It's neither a transient nor a small signal AC analysis or even continuous. I took the full output currents and then worked out the energies and then calculated the power for a few cases.

..,and I think the 'speaker-connected' scenarios are also for overloaded operation in an amp that is experiencing noticeable cross-over distortion. The 'speaker connected' plate current waveforms would then have portions of time when both PP valves are in cut-off, at which time there would be no loading on the primary half-windings, and hence the plate voltages are somewhat free to be driven by other forces. The duty-cycle for both valves in cut-off is likely to be somewhat indicated by loudthud's plots that show the relative brightness of traces experiencing over-voltage conditions compared to the time spent with one valve conducting and the other in cut-off - perhaps a lowish duty-cycle value.

Yes indeed. We are talking about serious hell-raising working conditions here. I assumed the mode of operation was such that all the energy ended up the MOV for the worst case. Yes, the intensity of the traces is a clue, but it's not much to go on.

Leo Gnardo described the speaker as a generator eloquently in t37469, where it's not just the speaker coil inductive energy wanting current to continue, but also kinetic energy from the moving mass driving an emf. With both those contributions, the plate voltage is likely to respond to the reflected OT secondary situation.

The model I used had the speaker motional energy i.e. Les in the Thiele-Small infinite baffle model. I ignored the voice coil inductance as that is so small by comparison.

Having thought a little overnight, I'm not at all sure that the speaker's motion generates any significant back-emf at all. It could only be significant if the cone were suddenly to move quickly and that would require an outside force. At full excursion the energy is stored in the suspension and the cone is stationary. The force from the suspension accelerates the cone so inducing a voltage in the coil. In Thiele-Small model terms the energy in the motional inductance/capacitance is radiated acoustically, dissipated in the source and stored again. In the model I used the capacitance was 700uF so if ALL of the inductive energy ended up in the capacitor the voltage would be V= sqrt( 2 * energy/ Capacitance) = sqrt ( 2* 0.175/ 700uF) = 22.3V. It has a natural resonance that determines how fast it moves. If you apply a voltage to force to the cone the back-emf only opposes that driving voltage but does not exceed it. If the driver is suddenly removed as is the case where the output valves become non-conducting for whatever reason, the cone moves at the natural resonant frequency as there is no longer anything to force it so the voltage is not larger enough to make it to the MOV limits.

On the other hand, there is no similar mechanism for the coil inductance so it will try it's best to maintain the current so producing very high voltages, but not much energy, about 5mJ. However, there is no reason this can't happen at a pretty rapid rate e.g 1KHz will give you 5 watts which will end up in the MOV, again taking the worst case view, for as long as that the situation persists.

One comment is then that probably all MOVs on primary windings would be acting as loads, and that loading would not be a 100% duty cycle, as the next valve to be driven in to conduction would take over the loading.

I have no doubt that the output valves will dissipate some of that energy in real life. But that wasn't the objective, it was to determine a worst possible case.

As I see it, primary leakage inductance could be an influence when again both PP valves are in cut-off, and dI/dt in the primary windings becomes substantial. dI/dt is probably highest at the time when a valve is driven in to cut-off (the other PP valve already being in cut-off). As you indicate, the energy in the leakage inductance could be consumed in a MOV across each half-primary. The power in to a MOV would then be frequency/2 and peak current squared dependant. One comment is that although plate voltage could be an over-driven square type waveform, the peak plate current reached will be frequency dependant.

The 1/frequency dependence you are thinking of is only a factor when there is no (significant) load as the primary inductance is large. At normal loads the plates see the only leakage inductance in series with the magnetizing inductance and that is shunted by the load. Of course a real speaker's impedance rises with frequency so... it's complicated

For the scenario with no speaker loading, then worstcase plate current and MOV power consumption occurs with a continuous dominant bass note where the plate current is allowed sufficient time to increase along the saturation V-I curve to the max current region. A comment is that multiple MOVs may be deployed on the OT primary (eg. across each half-winding, and perhaps in series). Another comment is that the OT primary inductance may be larger than normally measured (typical inductance measurement is at 5-10Vrms, whereas inductance will typically increase up to at least the design full power voltage swing). There may be a residual level of energy consumed by the plate when it is turning off (ie. a switching loss) due to finite switching time, which would depend on the achievable slew-rate of the output valves, but that energy is likely to be a low %. So I guess its best to 'test' that the audio chain is working properly by strumming strings that include higher frequency strings, and not strumming continuously!

Agreed, but the voice coil comes into to play at higher frequencies too. Certainly the actual inductance is another uncertain quantity, but as I said, "there's plenty of room to debate on the values I've chosen"

I have a feeling the only way we're going to know for sure is shove in the MOV's you think will work and test, maybe have some parallel back-up ones just for the purposes of the test.

I think the best solution might be a combination of MOVs and muting the drive after a preset time of current detected in the MOVs.

[trobbins]

My comment about 'continuous operation' was to separate the scenario under discussion from scenarios based on a 'one-off' transient, such as a fuse blowing. One aspect of MOV's is that although they may have a substantial single event energy spec, they typically have quite lower power dissipation specs, and so a continuous operation type scenario is more likely to need a protective device based on a power dissipation spec.

I think there needs to be a practical awareness that a duty-cycle factor would be part of the assessment for a crossover distortion type scenario where both primary half-windings become effectively unloaded for repetitive portions of time. Even for a worst-case study the duty cycle can't be one, it can only be a limit approaching one, as the valves need to conduct for some time duration in order for OT related energy to cause a problem. It also points to the use of circuit design as a means to alleviate OT stress, rather than just rely on insulation or over-voltage protection, whereby lessening the opportunity for blocking distortion to occur would in effect lower the duty cycle that both PP valves were in cut-off together.

Bringing in T-S parameters is an excellent addition to the assessment. I guess the max signal power and the signal frequency applied to the speaker, prior to both valves entering cut-off, is going to equate to the max potential energy stored in the mass/spring assembly, which is then the max available driving force to accelerate the coil back through the magnetic assembly (as you say that transient damped oscillation would end up exhibiting the speaker resonant frequency if the cut-off duration was long enough). That may mean that the worst-case potential energy would be related to a low frequency signal, or the attack start of a strum - rather than for a higher frequency signal.

The topic of detecting an over-voltage situation, and for example muting them signal level input, is interesting. MOV's across primary half-windings have a 'hot' and 'cold' end, so it may well be practical to insert an opto device in the cold end, and use the opto coupler output as a compression or limiter device. Some shunting diodes would be needed around an opto LED.

I have a low power 6M5 PP amp with relatively low B+ and a cheap modern OT that I may be able to pull out for some testing. The last tests were done on a higher powered vintage amp and I chickened out on driving the cathode current up too far along the saturation curve with no speaker loading, as the aim was to capture an anode voltage waveform with/without over-voltage protection. My other thought was to use a Philips 6CM5/EL36 PP amp, as it also runs a low B+, and valve base flashover won't be a concern