An alternative Amplifier THD plot

[Omid]

John Siau (@John_Siau Benchmark media) recently posted this post suggesting a more useful way of representing the performance of an amp. A graph of THD+N vs in room loudness is offers a practical way to judge an amp compared to the standard THD+N vs power plot (using the same data). He explains it very well in his post here, on benchmarkmedia's website. Amir (@amirm) if you read this, I'm wondering if you might still have the data used to generate your THD+N vs Measured W plots. If so it'd be possible to create a series of charts plotting 'THD+N vs in room loudness' for all previously measured amps, making comparisons across their whole power range easier.
To be clear I'm not suggesting changing any past or future reviews. If the data is available, I'd be happy to try create these new graphs, and post them here as another useful way to review and compare amps.

 

[NTK]

The Benchmark paper by Siau is total BS. It completely deny the existence of auditory masking. For example, this is one of the graphs he showed. At below 94 dBSPL of sound output from the loudspeaker, the noise + distortions of all the eight seven amplifiers were no more than 20 dBSPL. (Note: The horizontal axis label in his chart was wrong. It should be the loudspeaker output in dBSPL.)

20 dBSPL is about as loud as a ticking watch or leaves rustling, well below the typical background noise level of a quiet room. Do humans have a problem with room noise even in a quiet room (~30 dBSPL) when listening to loud (94 dB SPL) music? If we do, then we'll have to conclude that nobody will be able enjoy music listening in a normal residential house or apartment, which is flatly false.

Here is some more info from hearing studies. If a normal pair of ears are exposed to more than about 17 minutes of noise at 94 dBSPL, the ears' ability to detect to a 4 kHz tone (close to our most sensitive frequency) is degraded from below 0 dBSPL to 20 dBSPL, 2 minutes after the 94 dBSPL noise has stopped. It is simply ludicrous to claim some distortion products at 20 dBSPL will matter when the loudspeakers are blasting away at 94 dBSPL (i.e. SINAD of 74 dB).

Source: https://www.sfu.ca/sonic-studio-webdav/cmns/Handbook Tutorial/Audiology.html

 

[Omid]

The Benchmark paper by Siau is total BS. It completely deny the existence of auditory masking. For example, this is one of the graphs he showed. At below 94 dBSPL of sound output from the loudspeaker, the noise + distortions of all the eight seven amplifiers were no more than 20 dBSPL. (Note: The horizontal axis label in his chart was wrong. It should be the loudspeaker output in dBSPL.)

20 dBSPL is about as loud as a ticking watch or leaves rustling, well below the typical background noise level of a quiet room. Do humans have a problem with room noise even in a quiet room (~30 dBSPL) when listening to loud (94 dB SPL) music? If we do, then we'll have to conclude that nobody will be able enjoy music listening in a normal residential house or apartment, which is flatly false.
View attachment 470270

Here is some more info from hearing studies. If a normal pair of ears are exposed to more than about 17 minutes of noise at 94 dBSPL, the ears' ability to detect to a 4 kHz tone (close to our most sensitive frequency) is degraded from below 0 dBSPL to 20 dBSPL, 2 minutes after the 94 dBSPL noise has stopped. It is simply ludicrous to claim some distortion products at 20 dBSPL will matter when the loudspeakers are blasting away at 94 dBSPL (i.e. SINAD of 74 dB).
View attachment 470271
Source: https://www.sfu.ca/sonic-studio-webdav/cmns/Handbook Tutorial/Audiology.html

You make a very good point. Most of us will have a hard time hearing less than 20 dBSPL in ideal conditions, plus in real life the room noise and music will mask faint sounds. So the threshold that an amp must reach may not need to be as low as some claim.

Amir’s contention (from what I understand it, I don’t want to put words in his mouth) is that if we’re looking at a SINAD of 115 dB, no one could possibly argue that there would be any audible distortion.

Also, the devil is in the details. When you listen to music at 94 dB, you‘re presumably not listening to a continuous white noise. Music is dynamic. There are loud passages and quiet ones. So the masking effect it’s not a full proof argument. The same goes for the room noise: is the room noise a white noise, or is it just a certain spectrum. Is the noise of the amp in a certain spectrum that’s not covered by the room’s noise.

So in brief, I agree that the threshold that is required is debateable. Each person can decide what suits their needs, but that doesn’t take away from the merit of this graph. I personally think it’s a really useful way to look at the results. Each person can then decide if they feel a particular amp is quiet enough for them or not.

 

[NTK]

The paper by Benchmark focuses on the SINAD of the various amplifiers at the (relatively) higher power levels that will drive the speakers to generate high SPL (> 90 dB SPL).

The Benchmark AHB2 is very well technically engineered and has exceptionally low distortions up till the output approaches clipping power. Benchmark is trying to vastly exaggerate its value as a selling point. The other 6 amplifiers showed higher rising distortions with rising output power. At 110 dBSPL loudspeaker output (clipping thresholds of the AHB2 at 8 ohms in stereo mode), the worse performing amplifier of the six non-Benchmark ones has its noise + distortion products at 60 dBSPL, which was still 50 dB below the output signal. If one is in a room with a sound pressure level of 110 dBSPL, some distortions products at 50 dB below should be the least of her/his concern.

When the music goes soft and the power output is reduced, amplifier distortions will also goes lower, and idle noise (from the amplifier and from the other upstream components amplified by the amplifier) becomes the dominating concern. In that case the worst of the other 6 amplifiers produces an idle noise of ~17 dBSPL, and the rest are 10 dBSPL or below. If the amplifier noise is broadband and the room noise is also broadband, they are likely well masked by the room noise. IMHO a differential of 20 dB (i.e. room noise exceeds the amplifier idle noise by 20 dB) is more than sufficient for all intents and purposes.

Below shows the masking levels of white noise. For example, with a background white noise of 43 dBSPL (see the 43 dB curve), a normal person will not be able tell the presence of a 19 dBSPL single tone at 500 Hz (24 dB below the noise SPL).

Source: Figure 2 - Level of tone just masked by the white noise (https://community.sw.siemens.com/s/article/masking)

 

[Omid]

The paper by Benchmark focuses on the SINAD of the various amplifiers at the (relatively) higher power levels that will drive the speakers to generate high SPL (> 90 dB SPL).

The Benchmark AHB2 is very well technically engineered and has exceptionally low distortions up till the output approaches clipping power. Benchmark is trying to vastly exaggerate its value as a selling point. The other 6 amplifiers showed higher rising distortions with rising output power. At 110 dBSPL loudspeaker output (clipping thresholds of the AHB2 at 8 ohms in stereo mode), the worse performing amplifier of the six non-Benchmark ones has its noise + distortion products at 60 dBSPL, which was still 50 dB below the output signal. If one is in a room with a sound pressure level of 110 dBSPL, some distortions products at 50 dB below should be the least of her/his concern.

When the music goes soft and the power output is reduced, amplifier distortions will also goes lower, and idle noise (from the amplifier and from the other upstream components amplified by the amplifier) becomes the dominating concern. In that case the worst of the other 6 amplifiers produces an idle noise of ~17 dBSPL, and the rest are 10 dBSPL or below. If the amplifier noise is broadband and the room noise is also broadband, they are likely well masked by the room noise. IMHO a differential of 20 dB (i.e. room noise exceeds the amplifier idle noise by 20 dB) is more than sufficient for all intents and purposes.

Below shows the masking levels of white noise. For example, with a background white noise of 43 dBSPL (see the 43 dB curve), a normal person will not be able tell the presence of a 19 dBSPL single tone at 500 Hz (24 dB below the noise SPL).

Source: Figure 2 - Level of tone just masked by the white noise (https://community.sw.siemens.com/s/article/masking)

You’re totally right: when the music goes quiet, the noise level comes down, so part of my argument was incorrect.

Do you not agree, though that this graphic representation is useful? Granted that a lot of amplifiers exceed what we need, and the amplifiers in the white paper might all be perfectly adequate, but that’s not necessarily the case for all amplifiers that are reviewed here. I think this graph can demonstrate that.

 

[NTK]

The Benchmark graph assumes a certain speaker sensitivity and listening distance to estimate the distortions + noise levels. The SPL numbers are only applicable for that specific use case. It will be more useful if it is a interactive calculator that you can enter the pertinent info and it will calculate the noise level at the listening position for you.

 

[Omid]

The Benchmark graph assumes a certain speaker sensitivity and listening distance to estimate the distortions + noise levels. The SPL numbers are only applicable for that specific use case. It will be more useful if it is a interactive calculator that you can enter the pertinent info and it will calculate the noise level at the listening position for you.

True, there are assumptions as you said. One may need to add or take a few dB, depending on their setup. On the positive side, the graph allows you to compare amps, and gives a more useful context than the traditional thd+n vs Watt graph.

As a point of interest, I created a quick html file that allows you to play a white noise, and then add a pulsing tone at various loudness levels. The white noise represents music. The tone, a noise generated by the amp. This is a worse case scenario, as an amp's noise would not be as easy to detect as a pulsing single frequency tone. If you listen with headphones, you can indeed confirm that it's not easy to hear the pulsing tone when it falls 30dB below the level of white noise. This confirms your comment that all of the amps in the white paper would far exceed what is audible.

PS I had to zip the html file, because I could not upload an html.

 

[Omid]

True, there are assumptions as you said. One may need to add or take a few dB, depending on their setup. On the positive side, the graph allows you to compare amps, and gives a more useful context than the traditional thd+n vs Watt graph.

As a point of interest, I created a quick html file that allows you to play a white noise, and then add a pulsing tone at various loudness levels. The white noise represents music. The tone, a noise generated by the amp. This is a worse case scenario, as an amp's noise would not be as easy to detect as a pulsing single frequency tone. If you listen with headphones, you can indeed confirm that it's not easy to hear the pulsing tone when it falls 30dB below the level of white noise. This confirms your comment that all of the amps in the white paper would far exceed what is audible.

PS I had to zip the html file, because I could not upload an html.

And for anyone interested I changed the html file a bit. I flipped things around. The sine wave is the signal (or music) and the white noise, the noise floor of the amp. With headphones, I can hear the noise when it's 68dB lower than the signal, but can't pick it up lower than that. So I suppose a SINAD of 68dB might be good enough...

 

[dlaloum]

The Benchmark graph assumes a certain speaker sensitivity and listening distance to estimate the distortions + noise levels. The SPL numbers are only applicable for that specific use case. It will be more useful if it is a interactive calculator that you can enter the pertinent info and it will calculate the noise level at the listening position for you.

Benchmark do in fact have such a calculator embedded in their website.... you can go there and plug in your use case parameters...

I consider it a useful alternate perspective...

It provides another different way of conceptualising noise and distortion levels - although as pointed out in this thread, room background noise and masking tend to confound "absolutist" approaches to this!

There is always a tension between good and best, and the old saying of best is the enemy of good.... You can spend a lot of fruitless effort choosing the ultimate SOTA solution, and achieving inaudible improvements in a specific part of your system chain, when the same time and effort expended in parts of the chain that have audible imperfections, will achieve greater audible improvements.

The most imperfect components in our hobby remain the mechanical transducers, not the electronics, with imperfections (distortion) being several orders of magnitude higher than what even mid-market, value electronics produce....

The same attitude / approach to conceptualising the audibility of distortion should be applied to speakers.... most people don't realise that speakers tend to be in the 0.5% (state of the art) to 10% (most subs) distortion area !!!

So you get the people who point to their AHB2 amps "see how low distortion my setup is" - while happily including subwoofers specified at 10% THD... and running with speakers that measure at 1% to 2% THD.

How audible will the AHB2's exemplary performance be, when running a pair of speakers that overlay an additional 2% of distortion over the 0.0015% (or less) the amp produces?

Which brings us back to the old addage - SPEAKERS FIRST! (why - because they are the most imperfect of our components - and therefore each improvement in the speaker area, provides a relatively outsize impact to the system outcome) - and once you have the speakers, that then "drives" the amp selection... (some speakers will mate well with almost any amp, others not so much! - requiring higher power, and/or current, and/or stability with reactive loads etc...)

Here is an example measured by Erin on Erin's corner - the Arendal 1723 bookshelves:

Arendal 1723 BOOKSHELF S Bookshelf Speaker Review

Arendal 1723 BOOKSHELF S Review

www.erinsaudiocorner.com

It is worth noting that this is by no means a bottom of the market, flawed "cheapie", but a well regarded mid market speaker.

With actual distortion at around -60db at moderate listening levels, is there any point getting into a "pissing match" with regards to amps achieving -80db SINAD vs those achieving -100db SINAD???

Which is not to say, that given a good fit with budget, one should not consider the AHB2's - they are clearly a superb amplifier, and a SOTA design.

But if budget is tighter - there are certainly alternate more economical options that will do the job without being audibly different!

 

[Omid]

Benchmark do in fact have such a calculator embedded in their website.... you can go there and plug in your use case parameters...

I consider it a useful alternate perspective...

It provides another different way of conceptualising noise and distortion levels - although as pointed out in this thread, room background noise and masking tend to confound "absolutist" approaches to this!

There is always a tension between good and best, and the old saying of best is the enemy of good.... You can spend a lot of fruitless effort choosing the ultimate SOTA solution, and achieving inaudible improvements in a specific part of your system chain, when the same time and effort expended in parts of the chain that have audible imperfections, will achieve greater audible improvements.

The most imperfect components in our hobby remain the mechanical transducers, not the electronics, with imperfections (distortion) being several orders of magnitude higher than what even mid-market, value electronics produce....

The same attitude / approach to conceptualising the audibility of distortion should be applied to speakers.... most people don't realise that speakers tend to be in the 0.5% (state of the art) to 10% (most subs) distortion area !!!

So you get the people who point to their AHB2 amps "see how low distortion my setup is" - while happily including subwoofers specified at 10% THD... and running with speakers that measure at 1% to 2% THD.

How audible will the AHB2's exemplary performance be, when running a pair of speakers that overlay an additional 2% of distortion over the 0.0015% (or less) the amp produces?

Which brings us back to the old addage - SPEAKERS FIRST! (why - because they are the most imperfect of our components - and therefore each improvement in the speaker area, provides a relatively outsize impact to the system outcome) - and once you have the speakers, that then "drives" the amp selection... (some speakers will mate well with almost any amp, others not so much! - requiring higher power, and/or current, and/or stability with reactive loads etc...)

Here is an example measured by Erin on Erin's corner - the Arendal 1723 bookshelves:

View attachment 470380

Arendal 1723 BOOKSHELF S Bookshelf Speaker Review

Arendal 1723 BOOKSHELF S Review

www.erinsaudiocorner.com

It is worth noting that this is by no means a bottom of the market, flawed "cheapie", but a well regarded mid market speaker.

With actual distortion at around -60db at moderate listening levels, is there any point getting into a "pissing match" with regards to amps achieving -80db SINAD vs those achieving -100db SINAD???

Which is not to say, that given a good fit with budget, one should not consider the AHB2's - they are clearly a superb amplifier, and a SOTA design.

But if budget is tighter - there are certainly alternate more economical options that will do the job without being audibly different!

Thank you. I 100% agree. The speakers and the room distortion are orders of magnitude higher than the amps’. That’s why my speakers are five times more expensive than my amps
Just a small addition: harmonic distortion, specially low order harmonics (second and third) are less audible/objectionable than inter-modulation distortion or noise. So a speaker that produces second and third harmonics in the low frequencies might not be too bad. As opposed to a woofer that shows break up modes.
I’m guessing, if we pick a DAC, a preamp, and an amplifier that are each on the edge of being audible, the sum of these three levels of noise, upstream of the speaker might add up and be audible. So if one can afford to keep SINAD well below threshold, it takes this concern out of the equation.
Anyway, circling back to the initial topic, from my perspective this type of graphic could be useful.

 

[valerianf]

Apart from the ambient noise of the room, there is something important about the noise floor of the amp: the lower it is, more dynamic room is available for the music itself.
And many amps have a higher noise floor at low volume.

 

[dlaloum]

Apart from the ambient noise of the room, there is something important about the noise floor of the amp: the lower it is, more dynamic room is available for the music itself.
And many amps have a higher noise floor at low volume.

Only from an electronic perspective... your actual dynamic range, starts from your room noise floor - and goes up to your system capabilities.

Having more extended range at the low (level, not frequency) end gains you nothing, once it is below the noise floor of your listening space.

So with my listening space at circa 30db, and my normal listening level at 72db SPL, and a max value of 92db SPL (allowing for 20db peaks) - I have a true dynamic range of circa 62db (max peak of 92db - noise floor of 30db) - any additional range below 30db is academic.... as is to a lesser degree, capability beyond 92db SPL.

It would require major changes to heating / cooling / environmental isolation of my home to improve the situation, and an order of magnitude greater investment than my entire system.

(not saying it ain't desirable... but it ain't practical/realistic)

 

[Omid]

Only from an electronic perspective... your actual dynamic range, starts from your room noise floor - and goes up to your system capabilities.

Having more extended range at the low (level, not frequency) end gains you nothing, once it is below the noise floor of your listening space.

So with my listening space at circa 30db, and my normal listening level at 72db SPL, and a max value of 92db SPL (allowing for 20db peaks) - I have a true dynamic range of circa 62db (max peak of 92db - noise floor of 30db) - any additional range below 30db is academic.... as is to a lesser degree, capability beyond 92db SPL.

It would require major changes to heating / cooling / environmental isolation of my home to improve the situation, and an order of magnitude greater investment than my entire system.

(not saying it ain't desirable... but it ain't practical/realistic)

Try the html file I uploaded. Play a white noise at x dB and a sine wave at x-20dB. You’ll be surprised that you can still hear the tone in spite of the noise floor being 20dB louder. In other words you have more dynamic range than you think.

 

[Omid]

Just for giggles here is the plot for the Carver 350 amp (https://www.audiosciencereview.com/forum/index.php?threads/carver-raven-350-review-tube-amp.36601/)

72	14
85	37
92	48
98	56
105	64
109	67

 

[valerianf]

@dlaloum I fully understand the mathematical aspect of your point but my real life experience contradict it.
I live in LA with a lot of noise coming from outside.
I switch from an anemic Yamaha RX-A700 to a Onkyo TX-RZ30.
And surprise, the sound is night and day at low level, with a larger dynamic for the Onkyo.
I looked at the DAC datasheet, the announced dynamic for the Onkyo is far better.
Then the noise floor is lower.
But it also could be the amp section that is performing better with low level signals.
My understanding is that we need another measurement than the Sinad at 5W to detect the amps or AVR that are performing better at low level.

 

[dlaloum]

Try the html file I uploaded. Play a white noise at x dB and a sine wave at x-20dB. You’ll be surprised that you can still hear the tone in spite of the noise floor being 20dB louder. In other words you have more dynamic range than you think.

Yes - dynamic range extends beyond noise floor... it's old news - those of us that grew up with Vinyl, were well aware that there was signal beyond the noise floor of the groove.... and similarly, there is signal beyond the noise floor of many components...

But is it valuable?

(Digital has shifted the paradigm, as below it's bottom... there is nothing - unlike the analogue world)

But even if you add more peak and more low level signal range - you still are never going to get anywhere near the sort of dynamic range the AHB2 has - hence also, some savings can be achieved, by targeting amps that have a lesser dynamic range - but sufficient!
Sure you can take the old world Rolls Royce / Bentley / Aston Martin approach - they used to advertise their motors as having ample power under ANY conditions... and Aston Martin, "more than ample power under ANY conditions".

I sort of feel that the AHB2 is in the Aston Martin category of MORE than ample performance under ANY conditions.... - which is great, but many of us can achieve the same results with something not quite as "ample". (Some may argue the Mark Levinson and Krell provide the Aston Martin equivalent, and Benchmark the Rolls Royce/Bentley.... I would not argue that characterisation)

 

[Omid]

Yes - dynamic range extends beyond noise floor... it's old news - those of us that grew up with Vinyl, were well aware that there was signal beyond the noise floor of the groove.... and similarly, there is signal beyond the noise floor of many components...

But is it valuable?

(Digital has shifted the paradigm, as below it's bottom... there is nothing - unlike the analogue world)

But even if you add more peak and more low level signal range - you still are never going to get anywhere near the sort of dynamic range the AHB2 has - hence also, some savings can be achieved, by targeting amps that have a lesser dynamic range - but sufficient!
Sure you can take the old world Rolls Royce / Bentley / Aston Martin approach - they used to advertise their motors as having ample power under ANY conditions... and Aston Martin, "more than ample power under ANY conditions".

I sort of feel that the AHB2 is in the Aston Martin category of MORE than ample performance under ANY conditions.... - which is great, but many of us can achieve the same results with something not quite as "ample". (Some may argue the Mark Levinson and Krell provide the Aston Martin equivalent, and Benchmark the Rolls Royce/Bentley.... I would not argue

Yes the AHB2 may be overkill, but just saying you'd probably benefit from an amp with more than 62dB dynamic range. Thankfully a lot of well designed amps can be had for a good price nowadays.

 

[Omid]

@dlaloum I fully understand the mathematical aspect of your point but my real life experience contradict it.
I live in LA with a lot of noise coming from outside.
I switch from an anemic Yamaha RX-A700 to a Onkyo TX-RZ30.
And surprise, the sound is night and day at low level, with a larger dynamic for the Onkyo.
I looked at the DAC datasheet, the announced dynamic for the Onkyo is far better.
Then the noise floor is lower.
But it also could be the amp section that is performing better with low level signals.
My understanding is that we need another measurement than the Sinad at 5W to detect the amps or AVR that are performing better at low level.

This graph (as well as the std THD+n vs w graph) plot SINAD throughout the whole power range of the amp (not just 5W).

 

[amirm]

Having more extended range at the low (level, not frequency) end gains you nothing, once it is below the noise floor of your listening space.

So with my listening space at circa 30db, and my normal listening level at 72db SPL, and a max value of 92db SPL (allowing for 20db peaks) - I have a true dynamic range of circa 62db (max peak of 92db - noise floor of 30db) - any additional range below 30db is academic.... as is to a lesser degree, capability beyond 92db SPL.

None of this is correct. Please watch this video (I address the above point at 11:00):

 

[amirm]

Amir’s contention (from what I understand it, I don’t want to put words in his mouth) is that if we’re looking at a SINAD of 115 dB, no one could possibly argue that there would be any audible distortion.

That's correct but you need to note that SINADs above 100 dB routinely are dominated by noise, not distortion. Noise you can hear in absence of music. Distortion is not the same. Noise also combines when you have more speakers.

As to your OP, multiple people have reached out to me about it and my answer is the same as @NTK. It is not an argument that can be made with straight face. It is unlike John to have come up with this as his other papers make plausible cases for high fidelity.