(Suggestion on) Testing power amplifiers according to IEC 60268-3 standard

[DonH56]

Back when I was more involved in the audio world a number of manufacturers performed very extensive internal testing and produced large characterization reports. Those almost never went "outside"; they were considered proprietary, and most consumers would not understand them. The consumer/marketing datasheet was a small part of what was actually tested. I suspect that is true now as well, though a lot of manufacturers probably do little to no testing if they are using modules and boards supplied from their OEM/ODM companies. Emotiva for one showed an internal report far more extensive than what they publish in the manual or marketing material, and you used to be able to request the internal test report for your amplifier from companies like Bryston (again. do not know if still true today).

Bottom line is some of the CEC tests would be interesting to me as a private party, engineer, and ASR member, but I do not expect manufacturers will in general provide such data. Given Amir's overload and time constraints it may be worthwhile discussing which tests are most relevant and worth adding to the test process, but of course every one adds time and effort on his part as well. Maybe someday we'll have a team making up our own ASR Test Laboratory.

 

[MaxwellsEq]

I have done some reading and will investigate more what makes sense to pull in.

I certainly think that the Noise Output Voltage would be a useful thing to sample over several different amplifiers and architectures.

In most domestic environments SINAD greater than 110dB is probably adequate, but high noise voltages are potentially significant. Designs for mixing desks where there are sequences of gain stages followed by attenuation stages followed by gain stages etc. minimising noise is always very important. Personally, I prefer systems which minimise noise over minimising distortion (whose audibility is questionable).

 

[pma]

Output power frequency response at several power levels into 4ohm

.... and corresponding THD vs. frequency plots

Rated full power 250W/4ohm test passed through whole audio band 20Hz-20kHz.

 

[pma]

Short-term maximum output voltage and power
Characteristic to be specified
The maximum voltage or the corresponding power which the amplifier is capable of producing across, or dissipating in the rated load resistance (regardless of non-linearity) 1 s after the application of a specified short tone-burst signal. Each channel is operated independently.
It is recommended that the manufacturer state the rated value in the specification.
IMPORTANT
It is fundamental to the concept of this characteristic that, for all samples of a given amplifier, the measured value of short-term maximum output voltage or power shall not exceed the value stated by the manufacturer.

Method of measurement

The amplifier is brought under rated conditions with a true r.m.s.-responding level recorder connected to the output terminals.
a) The source e.m.f. is then applied to the amplifier under test in the form of a sinusoidal tone burst of 1 s duration and the output voltage U2 of the amplifier at the time 1 s from the start of the pulse is measured from the level recorder chart. The frequency of the tone burst shall be 1 kHz unless otherwise stated.
b) The source e.m.f. is then increased until the measured output voltage U2 reaches the maximum.
c) The value U2 is then the short-term maximum output voltage and (U2)^2/R2 is the short-term
maximum output power where R2 is the rated load impedance.
d) If the test is repeated, the repetition period of the signal pulses shall be 60 s or greater.

1s sinusoidal burst of 1kHz, record of initial part from the amplifier

1s sinusoidal burst of 1kHz, complete shape from amplifier

1s sinusoidal burst of 1kHz, record of final part (from the amplifier) and true rms measurement
Load is 4 ohm, short-term maximum output power is 290W into 4ohm.

 

[restorer-john]

The 1 second on 60 seconds off is a very low duty cycle, but the 1 sec is 50 times longer than the toneburst 20ms tests.

 

[pkane]

I don't see a lot of DIM 30 or DIM 100 tests being posted. These are all various tests that I'll be adding to next release of Multitone:

Reducing the measurement bandwidth to 20kHz:

And EIAJ burst @ 72W (8 cycles on, 32 cycles off):

 

[restorer-john]

I don't see a lot of DIM 30 or DIM 100 tests being posted. These are all various tests that I'll be adding to next release of Multitone:

View attachment 270259

Reducing the measurement bandwidth to 20kHz:
View attachment 270256


And EIAJ burst @ 72W (8 cycles on, 32 cycles off):
View attachment 270263

Can you display the toneburst on a linear scale? That's 16 cycles, not 8 BTW. And oscilloscope view (linear) is useful on the bench for determining what an amp is capable of in EIA/IHF or EIAJ style tonebursts. The beauty of the EIA/IHF is there is a -20dB constant 1kHz which the 'scope can sync to, interspersed with the 0dB bursts. (20cycles on 480cycles off - 500mS)

 

[pkane]

Can you display the toneburst on a linear scale? That's 16 cycles, not 8 BTW. And oscilloscope view (linear) is useful on the bench for determining what an amp is capable of in EIA/IHF or EIAJ style tonebursts. The beauty of the EIA/IHF is there is a -20dB constant 1kHz which the 'scope can sync to, interspersed with the 0dB bursts. (20cycles on 480cycles off - 500mS)

Is it 16 cycles? MT has no problems syncing to this signal, this is actually 20 times average in the time domain:

CEA2006 burst:

CEA2010 100Hz burst:

 

[restorer-john]

When you test an amplifer for its toneburst capabilities, there is no averaging. That is the complete opposite of what you want.

We test each burst and look at how the amplifier clips on the group of cycles from one individual cycle peak to the next. The goal is to take the amplifier to the very edge of clipping. Sometimes that will be on the 1st cycle of a set, other times it could be the fourth. Your 20 average is showing supply collapse/limiting after the 3rd cycle with some recovery, but you can't be sure with so much blurred detail.

Multitone sure is becoming an impressive piece of software!

 

[pkane]

DIM100 vs Power sweep measured @ 96kHz bandwidth:

 

[pma]

Measurements of Hypex NC252MP

PMA NC252MP power amplifier

IEC 60268-3 related tests start from article #4 in the link above. No LPF was used, because cutting frequencies above 25kHz invalidates harmonic distortion measurements. Significant component of harmonic distortion is at least the 5th, in case of NC252MP. With BW limited to 22kHz, 4kHz is the highest frequency where we can declare the value of harmonic distortion. At higher frequencies the results are invalid. With BW 45kHz, the highest measurable frequency is 9kHz. Noise floor of the spectrum gets rapidly upstairs above 22kHz, in case of Hypex modules, and other class D amplifiers.

Quote (from the standard):
Some amplifiers produce a spectrum of harmonics including small but measurable harmonics of high orders. The highest frequency component of this spectrum, the amplitude of which is significant, may in general be taken as the highest harmonic whose r.m.s. value (see 15.12.5) exceeds one-third of the total harmonic distortion at the same fundamental frequency.

The plots clearly show why such amplifiers are mostly measured at 1kHz and BW limited to 20kHz.

 

[pma]

Is it 16 cycles? MT has no problems syncing to this signal, this is actually 20 times average in the time domain:



CEA2006 burst:


CEA2010 100Hz burst:
View attachment 270327

Level in the burst tests should be shown in Volts. Averaging should not be used. Soundcards are not good to catch burst overload behaviour, they are too slow. CEA-2010 burst is intended rather to test speakers than amplifiers.

Below the same burst test on my A250W4R amplifier.

There should be no oscillations during clipping.

 

[pkane]

Level in the burst tests should be shown in Volts. Averaging should not be used. Soundcards are not good to catch burst overload behaviour, they are too slow. CEA-2010 burst is intended rather to test speakers than amplifiers.

Below the same burst test on my A250W4R amplifier.

View attachment 270424

View attachment 270425

There should be no oscillations during clipping.

I misstated the use of averaging: it's not done in the time domain unless coherent averaging is selected. It wasn't for any of the plots I posted.

The main reason I came to implementing bursts was to measure transient/dynamic power of an amplifier rather than seeing how or when it clips, so W is probably a better unit for the Y-axis for that purpose. Any of the standard units are selectable in Multitone, including Volts.

I'm curious about the ringing behavior at clipping of PA5. If the ADC is too slow (it was running at 192k) then I'd expect Gibbs-style ringing at the start and end of a clip. What I'm seeing is larger swings in the middle of the clipped region than at the start/end. What's more, the frequency of this ringing is at 10kHz (tested with a 100Hz burst). Somehow this doesn't seem related to the speed of ADC and may just be the strange way that PA5 clips, although I can certainly try this with a scope to compare:

Edit: here's what Gibb's looks like on a 1kHz square wave using the same set up:

 

[restorer-john]

I misstated the use of averaging: it's not done in the time domain unless coherent averaging is selected. It wasn't for any of the plots I posted.

The main reason I came to implementing bursts was to measure transient/dynamic power of an amplifier rather than seeing how or when it clips, so W is probably a better unit for the Y-axis for that purpose. Any of the standard units are selectable in Multitone, including Volts.

I'm curious about the ringing behavior at clipping of PA5. If the ADC is too slow (it was running at 192k) then I'd expect Gibbs-style ringing at the start and end of a clip. What I'm seeing is larger swings in the middle of the clipped region than at the start/end. What's more, the frequency of this ringing is at 10kHz (tested with a 100Hz burst). Somehow this doesn't seem related to the speed of ADC and may just be the strange way that PA5 clips, although I can certainly try this with a scope to compare:

View attachment 270433



Edit: here's what Gibb's looks like on a 1kHz square wave using the same set up:

View attachment 270439

But that shows nothing of note.

One is showing the amplifier heavily overloading, the other is a low level square (~5V) with interpolation.

If your section of the waveform on the PA5 is correct, we see around 20.4V when it starts to go seriously non-linear, which (assuming a 4R load) is ~104W.

About here:

At around 23V, it's completely out of control (on your plot) and that corresponds quite nicely to the maximum power Amir measured.

Whatever way you look at it, capturing when an amplifier is that far into overload with a slow capture rate (192kHz) isn't going to show anything useful without serious mental interpolation.

You need to pull out a proper scope. 20MHz bandwidth minimum for audio.

 

[pkane]

But that shows nothing of note.

One is showing the amplifier heavily overloading, the other is a low level square (~5V) with interpolation.

If your section of the waveform on the PA5 is correct, we see around 20.4V when it starts to go seriously non-linear, which (assuming a 4R load) is ~104W.

About here:
View attachment 270560

At around 23V, it's completely out of control (on your plot) and that corresponds quite nicely to the maximum power Amir measured.

View attachment 270561

Whatever way you look at it, capturing when an amplifier is that far into overload with a slow capture rate (192kHz) isn't going to show anything useful without serious mental interpolation.

You need to pull out a proper scope. 20MHz bandwidth minimum for audio.

This was into 8 ohms, not 4, but yes it's still overloaded in the burst test. What I was responding to was Pavel's mention of the fact that there shouldn't be any oscillation during clipping and that the oscillations were due to the ADC being too slow. The ADC is not overloaded, and the oscillations don't look anything like an ADC that is bandwidth limited as was demonstrated by a square wave. That's the reason for those plots.

 

[pma]

Measuring overload restoring time

15.6.6 Overload restoring time
15.6.6.1 Characteristic to be specified
The time interval (which occurs after the amplifier working under standard measuring conditions is overloaded by a certain amount for a specified period) between the moment when the input voltage is restored to its original value and the moment when the output voltage has again reached its original value within specified limits (see Figure 5).
The manufacturer may optionally state the rated value in the specification.
15.6.6.2 Method of measurement
a) The amplifier is brought under standard measuring conditions.
b) The source e.m.f. is increased by 20 dB during a time interval of less than one-quarter
period of the input signal and kept at this value for approximately 1 s.
c) The source e.m.f. is then reduced to its initial value during a time interval of less than onequarter
period of the input signal.
d) The time that passes before both the positive and the negative output peak voltages have
reached their final value, within 1 dB unless otherwise specified, is measured by means of
a suitable calibrated oscilloscope.

------------------------------------------------------------------

Measuring process:
a) amplifier brought to standard measuring conditions, i.e. -10dB below rated power, i.e. 1/10 of rated power,
b) source voltage increased of 20dB (10x) for 1s,
c) then reduced to 1/10 of rated power,
d) the recovery time as defined above is evaluated from the scope plot

Measurements on my A250W4R amplifier with 4ohm load (rated power 250W, standard measuring conditions power 25W):

Full record

Recovery area zoomed

Note: this test makes +10dB amplifier overload for 1s, not every amplifier will be happy with this.

 

[solderdude]

What does it look like when remaining just under overload conditions ?
I mean... the amp is clipping substantially here.

Studying clipping behavior is very important though. Most certainly with lower powered amps.

 

[restorer-john]

We've discussed this and advocated for similar tests before.

does it make sense to test how amps behave when they clip?

Yes just in case that the probe 1:10 was not calibrated and trimmed with 1kHz calibrator square. I think it was the case and you probably hit the nail. Before ANY measurement with 1:10 probe switch position, calibration is necessary. We may have more than one scope ....... with the same probe...

www.audiosciencereview.com

 

[pma]

What does it look like when remaining just under overload conditions ?
I mean... the amp is clipping substantially here.

Studying clipping behavior is very important though. Most certainly with lower powered amps.

... and the signal frequency here is 20Hz, so great interaction with overloaded power supply and its output ripple.

 

[MaxwellsEq]

Measuring overload restoring time

15.6.6 Overload restoring time
15.6.6.1 Characteristic to be specified
The time interval (which occurs after the amplifier working under standard measuring conditions is overloaded by a certain amount for a specified period) between the moment when the input voltage is restored to its original value and the moment when the output voltage has again reached its original value within specified limits (see Figure 5).
The manufacturer may optionally state the rated value in the specification.
15.6.6.2 Method of measurement
a) The amplifier is brought under standard measuring conditions.
b) The source e.m.f. is increased by 20 dB during a time interval of less than one-quarter
period of the input signal and kept at this value for approximately 1 s.
c) The source e.m.f. is then reduced to its initial value during a time interval of less than onequarter
period of the input signal.
d) The time that passes before both the positive and the negative output peak voltages have
reached their final value, within 1 dB unless otherwise specified, is measured by means of
a suitable calibrated oscilloscope.
View attachment 270863
------------------------------------------------------------------

Measuring process:
a) amplifier brought to standard measuring conditions, i.e. -10dB below rated power, i.e. 1/10 of rated power,
b) source voltage increased of 20dB (10x) for 1s,
c) then reduced to 1/10 of rated power,
d) the recovery time as defined above is evaluated from the scope plot

Measurements on my A250W4R amplifier with 4ohm load (rated power 250W, standard measuring conditions power 25W):

View attachment 270865
Full record

View attachment 270866
Recovery area zoomed

Note: this test makes +10dB amplifier overload for 1s, not every amplifier will be happy with this.

Fascinating, thanks for working on this. Along with noise voltage, this is a very valuable measurement that should be in more tests. It would be particularly interesting to compare classic linear vs SMPS powered amps.