• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required as is 20 years of participation in forums (not all true). Come here to have fun, be ready to be teased and not take online life too seriously. We now measure and review equipment for free! Click here for details.

AB(x) testing of Power Amplifiers

pma

Major Contributor
Forum Donor
Joined
Feb 23, 2019
Messages
1,630
Likes
2,633
Location
Prague
#1
AB(x) testing of Power Amplifiers

I had to do it. Once again I was convinced that I hear a difference in a sighted test, so I had to prepare a properly level matched A/B test. Two amplifiers, both of 2 x 50W/4ohm rated continuous sine power.
  • conservatively designed class AB, with acceptable distortion up to 0.02%/1kHz/4ohm/30W
  • class D TPA3255, AIYIMA A07
The task was simplified as the AIYIMA has its own volume control, so there was no problem with level matching. I draw the simplest test circuit that would connect/disconnect both wires feeding the speaker, not only to prevent any ground loops, but also for the reason that AIYIMA has BTL output with both binding posts “Live” - none of them is grounded.

test_AB_sch.png


The relay used was just what I had in stock – a 6 amps TYPE6013 with 3 switching contacts. Hopefully this type can be disassembled and the first think I had to do was to brush and clean the contacts, as their contact resistance was creating non-linear distortion, which must be avoided.

IMG_0514_relay.jpg


So this is the photo of the DIYed A/B test circuit.

IMG_0513_testAB.jpg


I made a test with 2 different music samples, Beethoven's No.7 Symphony (Wiener Philharmoniker, Simon Rattle) and Rolling Stones, “Anybody Seen My Baby”. Before the test, the level was matched with a 1kHz sine and with the pink noise, measured at speaker terminals. Matching at about 0.07dB.

So what? Though I did not expect it, there was a difference. The class D was punchier, with better impact on bass and better distinguishing of instruments, and also more dry. The class AB was more on the mellow side, but tending to blend, melt. Average power was about 4W on the Stones sample. The symphony has much more difference between loud and silent parts so speaking about power is difficult. No clipping, of course.
 

Tangband

Active Member
Joined
Sep 3, 2019
Messages
152
Likes
156
#2
AB(x) testing of Power Amplifiers

I had to do it. Once again I was convinced that I hear a difference in a sighted test, so I had to prepare a properly level matched A/B test. Two amplifiers, both of 2 x 50W/4ohm rated continuous sine power.
  • conservatively designed class AB, with acceptable distortion up to 0.02%/1kHz/4ohm/30W
  • class D TPA3255, AIYIMA A07
The task was simplified as the AIYIMA has its own volume control, so there was no problem with level matching. I draw the simplest test circuit that would connect/disconnect both wires feeding the speaker, not only to prevent any ground loops, but also for the reason that AIYIMA has BTL output with both binding posts “Live” - none of them is grounded.

View attachment 113160

The relay used was just what I had in stock – a 6 amps TYPE6013 with 3 switching contacts. Hopefully this type can be disassembled and the first think I had to do was to brush and clean the contacts, as their contact resistance was creating non-linear distortion, which must be avoided.

View attachment 113161

So this is the photo of the DIYed A/B test circuit.

View attachment 113162

I made a test with 2 different music samples, Beethoven's No.7 Symphony (Wiener Philharmoniker, Simon Rattle) and Rolling Stones, “Anybody Seen My Baby”. Before the test, the level was matched with a 1kHz sine and with the pink noise, measured at speaker terminals. Matching at about 0.07dB.

So what? Though I did not expect it, there was a difference. The class D was punchier, with better impact on bass and better distinguishing of instruments, and also more dry. The class AB was more on the mellow side, but tending to blend, melt. Average power was about 4W on the Stones sample. The symphony has much more difference between loud and silent parts so speaking about power is difficult. No clipping, of course.
Very interesting !
Is the more ”punchy” sound from the Aiyima a07 a coloration of the sound, or is it actual sounding better, more natural ?
I have the Aiyima a04 (tpa3251) and I have the same sound experience as you have, when compairing a04 with different class A/B amplifiers.
 
OP
pma

pma

Major Contributor
Forum Donor
Joined
Feb 23, 2019
Messages
1,630
Likes
2,633
Location
Prague
Thread Starter #3
Very interesting !
Is the more ”punchy” sound from the Aiyima a07 a coloration of the sound, or is it actual sounding better, more natural ?
I have the Aiyima a04 (tpa3251) and I have the same sound experience as you have, when compairing a04 with different class A/B amplifiers.
This is difficult to say. As a next step, I want to measure distortion of current that flows into the speaker, in other words speaker/amp interaction. At the moment, I only have a comparison of THD(+N) vs. frequency into 4ohm resistive load at 1kHz. As you can see, the class D under test has lower distortion at 1kHz than the class AB under test. I am not saying this is the main reason, however it is a measurable difference.

classABxD_THD.png
 
Last edited:

Murrayp

Member
Forum Donor
Joined
Oct 30, 2016
Messages
40
Likes
45
Location
Napier, New Zealand
#4
Was the test blind? Likely there is such big differences between the amps that this doesn't matter. An easy way to make the test blind (to most intents and purposes) is to use a rotary 12 way switch and connect every second pole to close the relay. Spin the switch with you eyes closed to a random position and try to ignore any hints (relay sound etc) - then toggle back and forth between two positions (without knowing which is which). Once you have made your mind up - check to see whether your assumption matches reality. Rinse and repeat (many times).

My guess is with this pair of amps you'll get 100% accuracy - so no need for many tests. But with more equivalent amps it may be more interesting (eg quality high power AB vs class D). I recently tested speaker leads this way and found more to it than I expected (could identify blind the effects of very small resistance differences in the leads - gave tonal variations of just 0.14dB). Most amps likely have bigger output impedance than the speaker leads did so I'd expect them to sound different in my case (ie with fairly low impedance speakers).
 

dominikz

Active Member
Forum Donor
Joined
Oct 10, 2020
Messages
143
Likes
549
#5
Most amps likely have bigger output impedance than the speaker leads did so I'd expect them to sound different in my case (ie with fairly low impedance speakers).
This is in line with my thinking too - if the amps that are being compared have significantly different output impedances, the difference in FR deviation with a complex load may be audible - however they could also fairly easily be linearized with EQ.

@pma - Very interesting test - thanks for taking the effort and sharing the results! Have you thought about doing FR measurements of both amps with the same speakers connected, to see if there's significant differences in response due to output impedance?
 

direstraitsfan98

Addicted to Fun and Learning
Forum Donor
Joined
Oct 1, 2018
Messages
828
Likes
1,149
#6
The results would be much different if you ran the test again on a speaker with a benign impedance load. Also you should not know which speaker is being used. What speaker was used for the listening test?
 

restorer-john

Major Contributor
Joined
Mar 1, 2018
Messages
6,746
Likes
16,975
Location
Gold Coast, Queensland, Australia
#7
The class D was punchier, with better impact on bass and better distinguishing of instruments, and also more dry. The class AB was more on the mellow side, but tending to blend, melt.
Getting worried about you Pavel.

Lots of subjective words. Class D...mmm. ;)
 

SIY

Major Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
6,048
Likes
12,786
Location
Phoenix, AZ
#8
I assume you'll make recording of the outputs for ABX or other analysis? Frequency response into the load?
 
OP
pma

pma

Major Contributor
Forum Donor
Joined
Feb 23, 2019
Messages
1,630
Likes
2,633
Location
Prague
Thread Starter #10
I think I have enough data to explain audible differences.

1) FR at speaker terminals
AB_2V_FR_speaker_lines.png

Interestingly enough the FR elevation of the class D amp (LC output filter) compensates for decay of highest frequencies due to inductance of speaker cable.

2) distortion at speaker terminals (behind the speaker cable)

AB_2V_THD_speaker.png

class AB distortion at speaker terminals, 2V. THD = 0.0062% is at 400Hz.


D_2V_FR_speaker.png

class D distortion at speaker teminals, 2V. THD = 0.0034% is at 400Hz.

Conclusion

As always, we MUST measure the whole audio chain and not only a single component of the chain.
 

dominikz

Active Member
Forum Donor
Joined
Oct 10, 2020
Messages
143
Likes
549
#13
I think I have enough data to explain audible differences.

1) FR at speaker terminals
View attachment 113524
Interestingly enough the FR elevation of the class D amp (LC output filter) compensates for decay of highest frequencies due to inductance of speaker cable.

2) distortion at speaker terminals (behind the speaker cable)

View attachment 113525
class AB distortion at speaker terminals, 2V. THD = 0.0062% is at 400Hz.


View attachment 113526
class D distortion at speaker teminals, 2V. THD = 0.0034% is at 400Hz.

Conclusion

As always, we MUST measure the whole audio chain and not only a single component of the chain.
Very nice! The FR difference definitely seems like it could be audible - around 0,2-0,3dB at LF, ~0,15 at 1kHz and more than 0,6 dB above 10kHz. So I assume one probably sounds slightly brighter than the other.
Makes me wonder though if the audible difference would disappear with some EQ to align the FR of the two amps. One could use REW to divide the two magnitude plots and then it should be relatively easy to flatten the difference curve with EQ.
 
OP
pma

pma

Major Contributor
Forum Donor
Joined
Feb 23, 2019
Messages
1,630
Likes
2,633
Location
Prague
Thread Starter #14
Very nice! The FR difference definitely seems like it could be audible - around 0,2-0,3dB at LF, ~0,15 at 1kHz and more than 0,6 dB above 10kHz. So I assume one probably sounds slightly brighter than the other.
Makes me wonder though if the audible difference would disappear with some EQ to align the FR of the two amps. One could use REW to divide the two magnitude plots and then it should be relatively easy to flatten the difference curve with EQ.
Thanks for your feedback, there is a shift of 0.2dB dB added in today's measurement of FR (for better visibility), so the green plot (AB) to be shifted 0.2dB up. The remaining difference is mostly above 2kHz. I am posting D/AB plot and it is really the effect of the LC filter of the class D, as I had shown in another threads. Re sound, as described in the post #1 of this thread.

DdividedAB.png
 

solderdude

Major Contributor
Joined
Jul 21, 2018
Messages
8,779
Likes
18,547
Location
The Neverlands
#15
Now that you know the FR deviations you can shelve the offending one so those differences are minimized.
In that case... can they still be told apart ?
Is it an FR thing only ?
 

boXem | audio

Major Contributor
Manufacturer
Forum Donor
Joined
Jun 19, 2019
Messages
1,046
Likes
1,406
Location
Europe
#16
...
Interestingly enough the FR elevation of the class D amp (LC output filter) compensates for decay of highest frequencies due to inductance of speaker cable.
...
Inductance of speaker cable or inductance of speaker itself?
Are there "official" plots of the speaker that can be taken as comparison point?
Now that you know the FR deviations you can shelve the offending one so those differences are minimized.
In that case... can they still be told apart ?
Is it an FR thing only ?
That would be another interesting experiment.
 
OP
pma

pma

Major Contributor
Forum Donor
Joined
Feb 23, 2019
Messages
1,630
Likes
2,633
Location
Prague
Thread Starter #19
I promised in another thread to measure distortion of current flowing into the speaker, for both amps. So here it is, nothing much new revealed. The measurement was not straightforward, because both output terminals of the class D amp are floating and the shunt (0R22) voltage of small value is thus floating with high CMV. I will probably try transformer isolation later, this was a "standard" balanced input, however the resolution was limited due to input divider scaling with respect to CMV.

This is the current vs. frequency plot and it reflects inverse impedance plot
AB_D_speaker_current.png


class AB x class D current distortion (removed because of low measurement resolution)
 
Last edited:
Top Bottom