square wave testing and slew rate

[vladn]

While recovering from a minor surgery I've been toying with various amp configurations in LT Spice (note - I make my living as an EE but analog amp design is not really my specialty, more of a curiosity). I have some trouble understanding the limits/envelope of square wave testing the amp especially when using difficult capacitive loads.

Let's say you have an amp with fairly deep NFB limited to a healthy 100V/us at the VAS stage and no output coil.

1. Does it make sense to test it with an input signal demanding a higher slew rate (bypassing front end HF filter)? Obviously the the input signal will generate huge error signal in the front end. Even with anti-saturation circuitry everywhere the output will not look pretty.

2. With a perfect square wave input a 100V/us slew rate amp will immediately trip over-current protection at very low output voltages (few volts or less) when presented with a capacitive load in uF range. Overriding the SOA protection in Spice is safe but not very informative since beta of the output stage is seriously drooped at this point and it is not clear how to interpret the results.

It is obvious that a good design should recover reasonably fast and not enter continuous oscillation under these extreme conditions (since it may happen during clipping and such). However do you really expect a clean output when test fixture exceeds slew rate and/or current limit of an amp ?

 

[tedr]

I know more about mixers than amps but I can give some hints about the situation. In many ways audio is a nasty application for electronics. Power amps WILL be abused, input overload, wrong line voltage, wrong load impedance, thermal overload, mechanical damage, output shorts to ground, and at the end of the day the amp that is still working despite the abuse has a lot going for it. It is also preferred that things degrade gracefully and recover spontaneously.

My answer to question 1 is No.

My answer to question 2 is maybe. Loudspeaker loads may include significant capacitance, but it is a well known torture test to add a few microfarads to see what happens, which should be non-catastrophic self protection.

Input overload in voltage terms is much more common than exceeding slew rate, especially since everything off CD has nothing above 22kHz anyway. Live mics, synths, Cd playback, it is all slow stuff, nothing much to worry about in slew rate terms. It is normal for an amp to be band limited, this means the output square wave will show the HF roll off due to all the stages combined. Applying the "degrade gracefully" concept the band limiting may be applied where it does the most good, which probably means before or in the input stage. You will get much better detailed technical advice from some very knowledgable folk here, I just wanted to get some basics over from the practical side.
Degrade gracefully, recover spontaneously, the load is connected all the time and someone (maybe a lot of people) is listening to it. Anything nasty that triggers protection should not only protect the amp from itself but more importantly protect the speaker (costing $$$$) from the amp.


Ted

 

[Bonsai]

Try reading this to get a handle on th e slew rate thing:-

http://waltjung.org/Classic_Articles.html

It was written in th e late '70's so a lot of the stuff we take for granted now (e.g. LTP emitter degeneration/JFET input, importance of high slew rate, open loop linearity, etc) as applied to power amp design were still being discovered or developed.

 

[gootee]

No. It makes no sense to use too-high slew rates at the input. I would definitely keep the RF filter in place, for that kind of testing.

As has been said, by some famous amp designer, you can make the performance of an amplifier driving squarewaves into a capacitive load look as good as you want, by just adjusting the input's slew rate. So, conversely, I guess, if it's too high then any amp would look bad.

The magnitude of the maximum rate of change of a voltage sine wave is:

slew rate max of sine (in volts per microsecond) =

[(2 x Pi) x (freq in Hz) x (amplitude in volts)] / 1,000,000

So, for example, for a 10V 0-to-Peak 20 kHz sine, the maximum slew rate is about 1.257 V/us. For 20V 0-P it would be about 2.513 V/us.

So, for the rise and fall times of squarewaves used for amplifier testing or simulation, for example, IF we assume that we don't need to have slew rates at an amplifier's output that are outside of some maxfreq range, then the maximum slew rate for the input would be 2 x Pi x peak output voltage x maxfreq / gain / 1000000, in V/usec.

That would mean that for a gain of 20, and a 40v p-p squarewave output, with maxfreq = 22kHz, the input would be a 2v p-p squarewave that would only need rise and fall slew rates of <= 0.138 V/us, i.e. about 14.47 usec risetime and falltime, minimum, giving a maximum slew rate of about 2.764 V/us for a -20v to +20v squarewave transition at the output.

However, I was told, here, by Bob Cordell (Post # 1985, here: http://www.diyaudio.com/forums/showthread.php?s=&postid=1230123&highlight=#post1230123 ), that it is probably a good idea to shoot for at least 50 V/us, at the output (for a discrete-component amp). On the other hand, some chipamps will only slew at a maximum of about 10 V/us, and some even less than that. So, obviously, you wouldn't want to try to exceed that, with them, except maybe for pathological-type tests.

P.S. HINT: For driving large capacitive loads with squarewaves, you can get down to basically zero overshoot and ringing, but still with very high slew rates for almost the whole edge (of the square wave), if you have a large excess current-dumping capacity (and just the right feedback and output networks, etc, of course). When using the TI OPA541 (E version) model with LTspice, I got the excess current capacity by simply paralleling several of them, for example. (But I only got really good performance after I enclosed them all in an fast opamp's feedback loop, and came up with some well-tuned compensation schemes.)

 

[Bonsai]

Tom, I agree with your comments about input slew rates. In my post I was refering to general slew rate requirements of an amp.

 

[AndrewT]

Hi,
following on from Tom's excellent summary.

Take a hypothetical power amp with a slew rate of 30V/uS allowing a 100Vpp (150Winto 8r0) signal @ 100kHz, and the high frequency response set by the input filter were 3db down @ 200kHz (RF=0.8uS).

Would increasing the slew rate higher than 30V/uS be audible?

Or asked another way, should the slew rate match the input filter?
What factors determine this requirement if any?

 

[roender]

An high SR amplifier always mean a very short GNFB path ?

 

[AndrewT]

An high SR amplifier always mean a very short GNFB path ?

physically short or very few components in the string?
Is there a rule for slew rate to GNFB path length?

 

[jan.didden]

An high SR amplifier always mean a very short GNFB path ?

It's the amp that has to prevent the signal from slew limiting. The fb is just (most times) a passive divider. I can't see how the feedback physical arrangement can make a slew limiting amp un-limit. Nor how the physical fb arrangement can make a fast amp start to slew limit.

We built these amps to reproduce audio. Audio signals go up to 20kHz in frequency.

There may be stuff in the input signal that is above 20k, like DAC switching artifacts, EMI and noise, but we don't want that to got to the speaker so hopefully the input filter will attenuate it.
The only reason to test the amp with higher freq components would be to make sure it doesn't upset the circuits to harm the actual audio.

So, for instance, if you have a non-oversampling DAC with no or little output filtering, you'll have a lot of problems in your power amp unless it is supersonic

Jan Didden

 

[roender]

Sorry, the question was wrong ... I mean with few stages. As you know, an FC amp has a very high SR

Mihai

 

[AndrewT]

The fb is just (most times) a passive divider.

Hi,
I'll accept you said most times but the feedback is a time delayed signal that is out of phase with the input signal. It is this delay that causes the stability/slew/overshoot problems.
Leach explains well.
The added delay of nS due to physically long paths may have an effect on faster (audio) amplifiers.
I would not be prepared to write off size as inconsequential, at least not yet.

 

[cabbagerat]

Hi,
I'll accept you said most times but the feedback is a time delayed signal that is out of phase with the input signal. It is this delay that causes the stability/slew/overshoot problems.
Leach explains well.
The added delay of nS due to physically long paths may have an effect on faster (audio) amplifiers.
I would not be prepared to write off size as inconsequential, at least not yet.

Nanoseconds? Unless your feedback path is well over 20cm long, it'll be sub nanosecond. And anyways, 1ns is 0.00013 radians (0.0075 degrees) shift at 20kHz, so I don't think it's going to make much difference at all

Of course, I have never understood why so many audio people build amps with passbands out into the MHz range.

 

[AndrewT]

..... feedback path is well over 20cm long, it'll be sub nanosecond. And anyways, 1ns is 0.00013 radians (0.0075 degrees) shift at 20kHz,....

Hi,
That's the sort of evidence that might convince me.
If we take a 300mm GNFB path length and use that in a 200kHz amplifier, then your phase angle delay is just about 0.1degrees.
But do we need to ask the question:- what does the amp do with a fast transient while it is waiting for the feedback signal to arrive?
Does it go open loop during that delay? Is that what overshoot is all about?
Does that bring back the question: does slew rate and RFfilter @ input need to be inter-related?

should the slew rate match the input filter?

 

[jan.didden]

[snip]I would not be prepared to write off size as inconsequential, at least not yet.

I would.

Jan Didden

 

[jan.didden]

Hi,
That's the sort of evidence that might convince me.
If we take a 300mm GNFB path length and use that in a 200kHz amplifier, then your phase angle delay is just about 0.1degrees.
But do we need to ask the question:- what does the amp do with a fast transient while it is waiting for the feedback signal to arrive?
Does it go open loop during that delay? Is that what overshoot is all about?
Does that bring back the question: does slew rate and RFfilter @ input need to be inter-related?

Andrew,

The amp doesn't wait for anything. If you look at the spectral components in that fast transients, you will see there are all sorts of freq components. They all are going back through the feedback. There is NO time that the fb is 'waiting'. It's just the phase shift. I really don't know how to eplain that, it's so basic.

Like the idea that a capacitor somehow delays a signal. NO! If you start putting charge in a cap, for instance you connect the cap to a signal, the cap voltage starts to change at the *exact* instant that you put the signal in. There is no 'delay' as such.

What does happen is phase shift between the charging current and the cap voltage. If you use a sinusoidal signal to charge a cap, the voltage across the cap will lag 90 degrees with respect to the signal that caused the charging current.

So, if you look at the input sine and the output sine you see that the latter peaks 90 degrees after the input peaks, that can be called 'delay', but the voltage at the cap DOES start to rise exactly when you put in the signal, there's no 'dead-time' or delay in this sense.

Jan Didden

 

[jan.didden]

Hi,
That's the sort of evidence that might convince me.
If we take a 300mm GNFB path length and use that in a 200kHz amplifier, then your phase angle delay is just about 0.1degrees.
But do we need to ask the question:- what does the amp do with a fast transient while it is waiting for the feedback signal to arrive?
Does it go open loop during that delay? Is that what overshoot is all about?
Does that bring back the question: does slew rate and RFfilter @ input need to be inter-related?

As long as the amp is not clipping (and slew rate limiting is also a form of clipping) the amp will not go open loop at all. What happens with high signal frequencies, the feedback signal is not exactly in (opposite) phase anymore to the input signal. The subtraction at the input stage of Vin-Vfb will no longer be exact, so the feedback works less and less with higher frequencies. That is why you see higher and higher distortion with higher frequencies.

The overshoot or instability is because at still higher frequencies, with still higher phase shifts, the nfb turns into positive fb. The subtraction of Vin - Vfb now actually starts to turn into an addition (because the fb signal starts to become of opposite polarity, going to 180 degr phase shift). And positive fb means you start to increase Vout, which starts to increase pos fb, which starts to increase Vout etc. Unless there is a physical limit like power supply voltage, Vout would continue to grow. We have now an oscillator. If there are enough losses in the circuit, or if we are not yet at 180 degr phaseshift, the oscillations die out and we have 'only' overshoot.

Jan Didden

 

[AndrewT]

Hi Janneman,
you have not convinced me.
Your explanation is completely at odds with a few papers I have understood, see Leach as an example.

We will have to differ.

 

[mlloyd1]

i haven't seen any (or don't remember if i did), but maybe someone who has could comment on the schematics/designs of some of the monster slew rate amps from the 70s'? I am thinking Kenwood and Sansui.

I think Sansui used class AB VAS to push output for faster slewing?

mlloyd1

 

[vladn]

The slew rate impact on sonic qualities is an interesting topic but somewhat distinct from the original question of how to interpret results from an out of the envelope testing.

In my opinion there is approximate envelope boundary in any amp design. If you operate the amp within the boundary it can be mathematically treated as a linear control system to a reasonable degree of accuracy. Once you exceed the boundary the amp can no longer be treated as a linear control system. I am not talking about signal distortion only, the phase response and even monotonicity assumptions become invalid beyond this boundary.

For example a simple calculation shows that an ideal amp (Zout = 0) driven from infinite slew rate source and limited to 100v/uS by the VAS stage driving a 1uF load will generate a current of I = C * dV/dt = 100A regardless of the output voltage swing !!! The only thing that changes with the swing is the duration of the pulse. Short pusle duration extends the output device SOA somewhat but does not help linear system approximation accuracy at all.

Now the real thing will have finite Zo and that would determine the output current limit. Testing this amp with a capacitive load that can not be driven reasonably linearly to 100v/uS by the output stage is exceeding the envelope in the same way as testing an amp with infinite input slew rate.

So the conclusion I am coming to is this - there should be a fairly clear distinction between testing an amp inside and outside the envelope defined by linear system approximation. The criteria for this are (i) input signal should not cause clipping, (ii) the input signal should not exceed amp slew rate divided by amp gain, (iii) the output capacitive load should not exceed Vopp / (Zolim * slew rate) where Zolim is an amp output impedance consistent with *both* linear system approximation and SOA (the highest of the two).

My thinking on testing the amp outside the envelope boundary is to make sure it (i) does not self distruct, (ii) does not damage speakers, (iii) does not trigger sustained oscillations, (iv) recovers reasonably fast (this is rather vague term).

On top of that there should be means that (i) prevent an amp from going outside the boundary from an input side such as input limiter and RF filter, (ii) protection circuitry that shuts the amp if the load exceeds amp drive capacity in the sense defined above.

BTW thanks for everyone who responded.

 

[Samuel Jayaraj]

Read Giovano Stochino's articles on Ultra-fast amplifiers in EW. He had one article on a Voltage Feedback amp for audio, which had a SR of > 300v/us. Build the amp to know its superior sonics and great dynamic impact. I have built quite a few and have not yet come across a commercial offering that even comes close - not that I have exhausted 'all' commercial offerings.