Amplifier Distortion Measurements at Low Power

[paulg1]

I have to agree that those measurements at low power are certainly good. This is a Topping B100, we should expect that the B200 and the others at the top of the list perform similarly at low power levels.

 

[Panelhead]

I do not think the tests went all the way down to 100uW on the other three. The tech from Topping who responded to me said that for my speakers the B100 was the best match. High efficiency and 8 ohm.
He followed up that for driving speakers that required more power the LA90 Discrete would address both scenarios. Almost as low noise, low distortion, and high linearity. Higher output.
No mention of the B200. He may have smelled that I am a Cheap Bastardo.
I almost purchased an AHB2 several times. Also a Lyngdorf TDA-1120. A dealer near Austin carried them. But never pulled the plastic.
Was using a MP Evolution and then a Micromega integrated. Both more than 16 bit resolution. Was worried I would not be able to hear any difference with a better measuring amplifier.

 

[John_Siau]

Hello,

A recent thread pointed out the relationship between output power requirements, speaker sensitivity, and the range of power preferred for different listening.
https://www.audiosciencereview.com/...s-decibels-and-sensitivity.62146/post-2278881. This suggests that 17-20bB is desired above the average listening level (50X to 100X times the average power) in a typical application for music/movies.

This got me thinking about typical listening and softer sounds. For me, in a domestic environment, typical average listening level is 60-85 dB. During listening, sometimes the level drops lower, and seldom, only for brief moments it will rise above 90 dB. So let's say we have a speaker with a sensitivity rating of 88 dB like in the example. So at the typical listening level of 78 dB the speakers are using 0.1W. In my case, with typical orchestral music, a lot of the musical details, are at a level considerably lower than this, 20-30 dB down from this level, maybe even lower in rare instances, such as harmonics, decays, hall reverberations, etc. That's 48 dB, so the amplifier power in those instances is 1mW-0.1mW and even perhaps lower in rare instances.

Here is a completely different way of viewing the data contained in a THD+N (%) vs Power plot. I believe this plotting technique will answer your questions.

The above curves show the THD+N vs Output level for the AHB2. The blue curve is 8-Ohms stereo. The Orange curve is 4-Ohms stereo. The green curve is 8-Ohms mono.

The horizontal and vertical units are expressed in dB SPL at the listening position. This assumes 90 dB [email protected], 1m. It also assumes a 2m listening distance in a "typical" listening room. You can easily adjust the values of the X and Y axis by a few dB if your speakers have a higher or lower sensitivity relative to 90 dB.

From X axis above, we can see that with this configuration, we achieve 110 dB peak SPL in stereo and 116 dB mono. From the Y axis we can see that the noise floor is -117 dB SPL in stereo and -115 in mono. This 3dB difference is expected because the two channels are connected in series in bridged mono and the amplifier noise is uncorrelated white noise. Notice that the output is 6 dB higher in bridged mono, so the SNR is actually 3 dB better in mono than it is in stereo, even though the noise is 3 dB higher.

More importantly, if you look at the Y axis, you will see that all three curves are entirely below the 0 dB SPL threshold of normal hearing. This means that the noise and THD never exceed the threshold of hearing, even when the amplifier is delivering full output (100W, 190W, or 380W). The green curve appears to reach about +4 dBSPL, but this rise is largely caused by the limitations of APx555b analyzer, but that is another topic.

What this graph shows is that the AHB2 does not show an increase in THD when driving 4-Ohms instead of 8-Ohms. This is very unusual. Most amplifiers produce more THD into lower impedances. The graph also shows that there is absolutely no audible distortion. The distortion is below 0 dB SPL which means that it would be inaudible if the distortion could be played while the music was muted. With the AHB2, we are not using the music to mask the THD produced by the amplifier.

Just to provide a bit of context, here is the AHB2 response with that of 6 other top-rated amplifiers. Notice the significant difference in the noise floors and notice the significant rise in THD that occurs with most amplifiers as the output level increases. All of these curves are only showing the 8-Ohms response.

And, to provide a bit more context, here is what happens with many amplifiers when driving 4-Ohms vs. 8-Ohms:

Notice how the THD is 5 to 10 dB louder when this amplifier is driving a 4-Ohm load. Also notice that the THD is 40 to 50 dB above the 0 dB SPL threshold of hearing. With this amplifier, you would need to hope that the music masks the THD, because it is louder than normal conversation.

I have just written an application note on the topic of plotting THD+N in terms of dB SPL. This plotting method makes it easy to see the noise floor, the SPL of the noise, the SPL of the THD, the peak SPL of the system, and the differences that are caused by changes in impedance loading. None of these can easily be extracted from a traditional THD+N % vs Power graph, but all of the data is there and it can be extracted and replotted in THD+N SPL vs Output SPL.

Here is the link to the paper:

How Loud is the Distortion from Your Power Amplifier? - John Siau, Benchmark Media Systems, Inc.

 

[Panelhead]

Thank you John, many of us have focused on amplifier performance. Just at a common output of 5 watts. When we listen at levels of 500mW or 100mW. That is what allows the constant noise level to become audible.
Plugged my system in the app note above. Seemed to have several inputs that were capped. I plugged in distortion as percent and app displayed as -110dB. Also speaker output seemed limited to 110dB.
The noise was zero at listening, not positive or negative.
Used 0.00007%, 50 watts@8 ohm, 105 dB, 11 meters, and open space as inputs. One of these may be hanging up the calc.

 

[pma]

@John_Siau
It would be nice to add human ear masking curves, to see which of these distortions might be audible. I agree that if you remove the stimulus, then the residual might be audible, in some of the cases shown .

I remember that some time ago you recommended to use dBW at Y axis? Is not it all the same distortion, depending only on scale interpretation = reader's knowledge?

 

[John_Siau]

@John_Siau
It would be nice to add human ear masking curves, to see which of these distortions might be audible. I agree that if you remove the stimulus, then the residual might be audible, in some of the cases shown .

I remember that some time ago you recommended to use dBW at Y axis? Is not it all the same distortion, depending only on scale interpretation = reader's knowledge?

View attachment 469544

I believe dB SPL vs dB SPL is much easier to read and much easier to adjust for changes in speaker sensitivity. It also allows direct comparison of the amplifier output at different loadings. When you plot power on the X axis, the 8-Ohm, 4-Ohm and 2-Ohm curves are all progressively shifted more to the right. Voltage drives the speakers, and so we really need to align the voltages not the estimated power. If the speaker sensitivities are also expressed relative to 2.83 Vrms instead of 1W, we can read everything directly from the SPL vs. SPL plot.

All of the data for my SPL vs SPL plots was taken directly from the THD+N (%) vs Power (W) plots published by Stereophile.

As for including human masking curves, this could be done as a separate plot, but this could not be rescaled to account for changes in speaker sensitivity or listening position. Curves at threshold are not the same as curves above threshold. You would need to create separate plots for different speaker sensitivities. You would also have to determine the spectral content of the THD. This all gets very complicated very quickly.

Step back and take a simple approach:

If we can show that the residual is < 0 dB SPL, at all output levels (as it is with the AHB2), then we can conclude that the residual is inaudible and masking curves become irrelevant. If the levels are higher than 0 dB SPL, then the masking curves become relevant.

From my point of view, it is much easier to design and build electronics that will keep the THD+N below 0 dB SPL than it is to determine how much THD we can produce before it will be audible. An entire Benchmark audio chain will achieve that < 0 dB SPL end-to-end goal. Electronics should not be producing THD at SPLs that are 50 to 60 dB above 0 dB SPL. Nevertheless, most still do. Even "top-rated" audiophile products!

 

[John_Siau]

Thank you John, many of us have focused on amplifier performance. Just at a common output of 5 watts. When we listen at levels of 500mW or 100mW. That is what allows the constant noise level to become audible.
Plugged my system in the app note above. Seemed to have several inputs that were capped. I plugged in distortion as percent and app displayed as -110dB. Also speaker output seemed limited to 110dB.
The noise was zero at listening, not positive or negative.
Used 0.00007%, 50 watts@8 ohm, 105 dB, 11 meters, and open space as inputs. One of these may be hanging up the calc.

We are correcting the calculator now. Our Shopify web hosting periodically "sanitizes" our HTML code. This has now happened several times and we do not yet know how to prevent this. It will be fixed shortly, thanks for the notification.
In the mean time, there is a working version here:

THD+N Audibility Calculator

 

[Panelhead]

Thank you, calc appears to be working correctly now.

 

[paulg1]

Wow. Thank you so much for this post. It is really appreciated. This is a really good way to evaluate amplifier components. I'm just getting into a careful read of your paper now, Sorry for my late reply,

 

[paulg1]

John Siau, your paper is excellent. https://benchmarkmedia.com/blogs/ap...d-is-the-distortion-from-your-power-amplifier.

I believe that the dB SPL vs. dB SPL graphical analysis should become a standard for ASR, at the very least, an addition to the existing graphical formats. It puts the measurements in perspective, makes the data immediately accessible when compared to a THD+N % vs Power graph. Those traditional graphs stink by comparison. I plan on making dB SPL vs. dB SPL graphs for any component of interest to me from now on.

This post: https://www.audiosciencereview.com/forum/index.php?threads/dynamic-range-how-quiet-is-quiet.14 shows further how frequency dependent this all is. It would be most revealing if a frequency dependent composite overlay could be developed for component evaluation, similar to the room noise level graph (last graph).