Flat Frequency Response/Different Sound?

[Cars-N-Cans]

... and load impedance, and input impedance, and temperature, and supply voltage, and output voltage and output current...

Now as far as the measurements done at ASR, I would say they are reasonably comprehensive and should tell you if there will be any audible issues. As you say there will be variations, but designers know how to handle these, and the bandwidth of audio is low enough that feedback within the design of the amplifier can cope with changes in impedance of the speakers or headphones. A good example of this was me deciding to foolishly send a 100 Hz square wave thru my amplifier with speakers attached to see what happened while measuring with an oscilloscope. At the output of the amplifier, it was nice and square, with very short rise and fall times and sounded just as it should: extremely annoying. For all intents and purposes, it was electrically perfect. At the speaker terminals it was a much different story. There was substantial distortion of the waveform. Was it because the speaker wire is not transparent? An audiophile like Danny Richie would certainly say so. However reality was much different. The speaker and its cross-over were becoming reactive at some frequencies, and the distortion in the square wave was due to the large reactive currents circulating. This combined with the resistance of the actual speaker wire is what caused the visual distortion in the waveform. Obviously this is not a good situation, but it did show that the amplifier was doing what it should, namely ensure that the output is a larger faithful replica of the input irrespective of what the speakers were doing. The only potential exception will be at very high frequencies where the gain margin starts decreasing, however most of the amplifiers where this is a concern, such as linear amplifiers, are also powered by unregulated linear supplies. The 100/120 Hz ripple (edit: And rectified audio as well) has to be rejected by the amplifier's feedback, and if its low enough to cause issues with speakers, then it will cause rising distortion with frequency and show up in Amir's measurements. This will be captured in the power vs. distortion sweeps done at varying frequencies. The recently reviewed Hypex DIY amp shows variations in distortion with frequency. I don't know the sources of the actual distortions, but it does provide a good example of what variations can look like:

Here we can say the amplifier does show some load dependence with frequency. Almost all amplifiers will show some dependence, but its still well below what we can hear, and further the distortion products at high frequencies are multiples of the fundamental, so often they fall outside of our hearing. This is why they are truncated on speaker distortion plots. No point in worrying about 3rd harmonic distortion at 10 kHz when the actual distortion will be 10 kHz x 3 or 30 kHz. We can no longer hear it. As far as IMD goes, Amir also provides those in the multitone tests, and the "grass" as he refers to it, should not be higher than about -50 dB if we want it to be inaudible.

 

[IAtaman]

I hear you, but I don't think I am not making my point very well. Let me try and be more specific. Let's take this THD+N as an example.

In the graph you also shared, THD+N is around 130db at 1V output. But that is at Rsource = 0. If you scroll down in the datasheet, there is a graph that shows (if I am reading it correctly) that THD+N goes up by ca 10db if Rsource is 600 ohm. Then of course 130db is for 1Khz. Another graph shows at 10Khz this is 5db higher. This is also at unity gain and 2K load. Make load 600 ohm that's 5db more. 130db is, as Amir puts it in his reviews, best case scenario, probably one which very smart people who are designing these amps are optimizing for as we speak But what is clearer is that amps are not linear devices, there are a lot of parameters that effect their output performance, I am not sure just because THD+N at one output voltage for one source and load impedance at one frequency is -120db, that qualifies that amp as "certified transparent" across all variety of circumstances that it can find itself outside the lab. I think it would be great if @amirm or someone made a video or wrote an article explaining all sources of non-linearity in amplifiers, what measurements are taken to get a better understanding of these parameters and most importantly, what are the limitations of measurements so that we can all sleep well knowing we understand the topic, instead of assuming that a few proxies that has been chosen as performance indicators for a complex, non linear device can tell us everything about it

 

[Cars-N-Cans]

I hear you, but I don't think I am not making my point very well. Let me try and be more specific. Let's take this THD+N as an example.

In the graph you also shared, THD+N is around 130db at 1V output. But that is at Rsource = 0. If you scroll down in the datasheet, there is a graph that shows (if I am reading it correctly) that THD+N goes up by ca 10db if Rsource is 600 ohm. Then of course 130db is for 1Khz. Another graph shows at 10Khz this is 5db higher. This is also at unity gain and 2K load. Make load 600 ohm that's 5db more. 130db is, as Amir puts it in his reviews, best case scenario, probably one which very smart people who are designing these amps are optimizing for as we speak But what is clearer is that amps are not linear devices, there are a lot of parameters that effect their output performance, I am not sure just because THD+N at one output voltage for one source and load impedance at one frequency is -120db, that qualifies that amp as "certified transparent" across all variety of circumstances that it can find itself outside the lab. I think it would be great if @amirm or someone made a video or wrote an article explaining all sources of non-linearity in amplifiers, what measurements are taken to get a better understanding of these parameters and most importantly, what are the limitations of measurements so that we can all sleep well knowing we understand the topic, instead of assuming that a few proxies that has been chosen as performance indicators for a complex, non linear device can tell us everything about it

I think what is being lost sight of here is the relative sizes of the distortion relative to the signal. Best case for distortion that we can hear is something like -40 to -50 dB. Distortion at -50 dB is around 300x smaller than the audio signal itself. Distortion at -120 dB is over a million times smaller that the audio signal. Here the distortion and noise will be down in the microvolt region. So just to get this out of the way we need to be clear when one sees figures like -120 dB, -130 dB, etc. none of this is even remotely audible. If you want a quick assurance from the reviews to sleep better anything with a SINAD of about 60 dB or higher is, for all intents and purposes, guaranteed to be transparent (edit: provided it’s distortion dominated). As such, these are primarily academic discussions. Also when you hear Amir saying that “-110 dB is the absolute limit” for audibility, he is referring to the 0 dB hearing threshold with 110 dB as the reference for the signal. If you are listening to music at an average of 110 dBA, you won’t have to worry about distortion for long! Your next set of IEMs will be hearing aids! He says this to quell even the most pedantic audiophiles. For us, though, that limit can be taken as -50 dB. If you have REW, it allows you to add distortion to tones, and it’s instructive to listen since it gives a feel for what harmonic distortion sounds like. Note that here, too, the details come back to bite us as distortion for pure tones is much easier to hear than in music due to things like masking and such.

Now as for the nonlinearities, that is a very complex subject unfortunately. This is why Amir provides things like the SINAD measurement. Without detailed knowledge it’s hard to get a feel, otherwise. If you want to know more, getting Audio Power Amplifier Design by Douglass Self would be a great place to start. While it’s information-dense, he is very knowledgeable, and the book quite comprehensive. There are lots of good videos on YouTube as well. But, the subject is far too involved for one video alone since there are so many aspects to amplifier design and the various mechanisms that cause distortion, IMD, etc. As a quick example, a potential source for the changing THD versus load impedance would be something like variation in transistor beta with respect to output current in the transistors that comprise the emitter-follower output stage of the amplifier. There are more, and listing them all is definitely beyond my pay grade

I think a better place to start would be audio examples with distortion and other defects deliberately applied so you can physically hear them for yourself. There have been threads here that do have such exercises. If you know what they sound like you are better informed to make a decision on their relative impact. For electronics, however, unless it’s a tube amp this is just a numbers game and doesn’t have any bearing on real life. For us, in terms of sound quality we can regard electronics as being perfect.

 

[IAtaman]

I think a better place to start would be audio examples with distortion and other defects deliberately applied so you can physically hear them for yourself. There have been threads here that do have such exercises. If you know what they sound like you are better informed to make a decision on their relative impact. For electronics, however, unless it’s a tube amp this is just a numbers game and doesn’t have any bearing on real life. For us, in terms of sound quality we can regard electronics as being perfect.

Are you recommending I focus more on listening tests and less on measurements?

If you want a quick assurance from the reviews to sleep better anything with a SINAD of about 60 dB or higher is, for all intents and purposes, guaranteed to be transparent (edit: provided it’s distortion dominated).

I don't agree with that and I think it depends very much on what are the measurement parameters. 60db SINAD measured at 4V for 1Khz for might very well be distortion at audible levels say at 5Khz and 500mV output measured at high source and load impedance. Whether one can hear it, that is a different story of course.

Now as for the nonlinearities, that is a very complex subject unfortunately.

Yes, exactly my point. Amir says in A90D review that SINAD of 119 dB is superb and assures transparency in all conditions. Does it really assure transparency in all conditions? If it does, it is not immediately apparent to me, and I suspect to many people as well , and think there is value to have a conversation around that.

 

[HarmonicTHD]

Are you recommending I focus more on listening tests and less on measurements?



I don't agree with that and I think depends on very much on what are the measurement parameters. 60db SINAD measured at 4V for 1Khz for might very well be distortion at audible levels say at 5Khz and 500mV output measured at high source and load impedance. Whether one can hear it, that is a different story of course.


Yes, exactly my point. Amir says in A90D review that SINAD of 119 dB is superb and assures transparency in all conditions. Does it really assure transparency in all conditions? If it does, it is not immediately apparent to me, and I suspect to many people as well as, and think there is value to have a conversation around that.

You are casting doubts. But can you show us some actual evidence? Especially of their audibility?

 

[IAtaman]

You are casting doubts. But can you show us some actual evidence? Especially of their audibility?

I am not trying to cast doubt on measurements in general or on ASR in particular. As I said earlier, maybe with different words, I found myself assuming the measurements demonstrate transparency, 120db SINAD sounds impressive indeed, but the more I thought about it, more I realized I actually don't know if it is true and would like understand it better.

 

[Cars-N-Cans]

Are you recommending I focus more on listening tests and less on measurements?



I don't agree with that and I think it depends very much on what are the measurement parameters. 60db SINAD measured at 4V for 1Khz for might very well be distortion at audible levels say at 5Khz and 500mV output measured at high source and load impedance. Whether one can hear it, that is a different story of course.


Yes, exactly my point. Amir says in A90D review that SINAD of 119 dB is superb and assures transparency in all conditions. Does it really assure transparency in all conditions? If it does, it is not immediately apparent to me, and I suspect to many people as well , and think there is value to have a conversation around that.

This is why I say it’s instructive to actually hear these types of distortions at varying levels. People throw these numbers around with little concept of what they actually are. There is indeed a conversation to be had, but to properly have it requires going to get a degree in electrical engineering These topics are quite complex. Better would be to actually discuss the distortion itself since you won’t need a prerequisite of small signal design.

For practical purposes since you just want to use this equipment, actually hearing what distortion 20, 30, 40 dB below the tone or music will show why the topping is several if not many orders of magnitude cleaner than any distortion we could ever hear under any circumstances. Then even with respect to noise even the most sensitive IEM would come back empty handed most likely.

 

[Cars-N-Cans]

Are you recommending I focus more on listening tests and less on measurements?



I don't agree with that and I think it depends very much on what are the measurement parameters. 60db SINAD measured at 4V for 1Khz for might very well be distortion at audible levels say at 5Khz and 500mV output measured at high source and load impedance. Whether one can hear it, that is a different story of course.


Yes, exactly my point. Amir says in A90D review that SINAD of 119 dB is superb and assures transparency in all conditions. Does it really assure transparency in all conditions? If it does, it is not immediately apparent to me, and I suspect to many people as well , and think there is value to have a conversation around that.

Also please note that the decibel is a logarithmic unit. The actual ratios it represents can be enormous. Millions or even billions of times difference between the fundamental that we want to hear and the distortion and noise we don’t.

 

[JJFUSA]

Always be wary of statements like that - whether they come from others, or are your own conclusion. The cognitive biases talked about upthread can - and often do - result in perception of differences that seem like "night and day" or "hearable pretty well" even though those differences don't actually exist in the sound reaching the listeners ears. Everyone is subject to this. It doesn't mean you are less skilled, or have faulty hearing - it is just the way people are built.

The simple fact is we can't trust sighted listening as giving valid results - even when it is us listening.

If it is just a case of "do I like what I hear" then it is fine. We can just enjoy it if we like it. If, however we are going to make statements we expect others to use in their own decision making "onboard audio is always bad" then we should really have something other than our own sighted listening to back that up.

It's a perception of the sound of similar level devices, and I totally agree with your statement, listening pretty much is very subjective and highly dependable on the listening conditions, actually it's very well related to my original question here about "different" soundings of the identically measured DACs/AMPs.

In this particular situation I should correct myself, I was rather talking about the sound quality, not the sound perception, you would agree that an old radio would sound "worse" than a modern digital player in terms of the audio quality, and you definitely wouldn't need any measurements to hear the difference between them, same goes to the most on-board PC sound chips like Realtek for example, which I compared to my LG G7 ThinQ (ES9218P) when I said "no needs for any measurements the difference is drastic and hearable pretty well".

 

[IAtaman]

There is indeed a conversation to be had, but to properly have it requires going to get a degree in electrical engineering

Got mine 20 years ago, ready when you are

 

[antcollinet]

It's a perception of the sound of similar level devices, and I totally agree with your statement, listening pretty much is very subjective and highly dependable on the listening conditions, actually it's very well related to my original question here about "different" soundings of the identically measured DACs/AMPs.

In this particular situation I should correct myself, I was rather talking about the sound quality, not the sound perception, you would agree that an old radio would sound "worse" than a modern digital player in terms of the audio quality, and you definitely wouldn't need any measurements to hear the difference between them, same goes to the most on-board PC sound chips like Realtek for example, which I compared to my LG G7 ThinQ (ES9218P) when I said "no needs for any measurements the difference is drastic and hearable pretty well".

I would be able to easily measure the differences on an old radio, and describe the mechanisms that make them sound bad.

I'd actually be surprised if you were able to hear serious problems with on board PC audio if you were listing blind - unless your audio setup was bad causing noise via (eg) ground loops. Modern DACS (even on computer motherboards) are so good that any problems tend to be below the level of audibility.

 

[antcollinet]

I am not trying to cast doubt on measurements in general or on ASR in particular. As I said earlier, maybe with different words, I found myself assuming the measurements demonstrate transparency, 120db SINAD sounds impressive indeed, but the more I thought about it, more I realized I actually don't know if it is true and would like understand it better.

Here is the current hypothesis of what are audible levels for various types of noise and distortion.

Audibility thresholds of amp and DAC measurements

Introduction A recurring theme on ASR is whether or not the various measured qualities of the devices are audible. In this post, I'll present some clear and visual thresholds for when those imperfections can be considered a potentially audible concern. I will not explain the basics of amp/DAC...

www.audiosciencereview.com

No one has yet demonstrated that they can hear any effects below them (even the lenient limits) in normal listening conditions - if you want to understand it better, then by far the best way would be to try listening distortionn at various levels much higher than these suggested limits and see if you can hear them.

I know (at least on one test I tried) my ability to detect distortion ran out at about -45 dB. Some other people here reported similar results. It is quite an eye opener when you realise the levels of distortion produced by modern electronics is up to 10,000 times lower than you personally can actually hear.

 

[Cars-N-Cans]

Got mine 20 years ago, ready when you are

I think then a detour into psychoacoustics would be instructive In that regard if you have experience in things like analog signal design I can see why you may have reasons to doubt. One thing to keep in mind is that the auditory center’s function is to allow you to hear. In that regard it’s a very powerful signal processor. But, it also is geared towards information recovery, and as such, it won’t go out of its way like electronics do to be perturbed by noise, and will even deliberately ignore it under certain circumstances. There is also the resolution limits of the ear and things like its equivalent rectangular bandwidth and masking that make distortion harder to hear in music. It’s why LPs can actually be listenable and even exhibit good subjective sound quality despite their abysmal objective performance.

Amir has done videos on the topic:

Understanding Audio Frequency Response & Psychoacoustics (Video)

For some of you this is old news but for many, I thought it would be good to do a tutorial on frequency response. No, not the trivial aspects of it but digging into psychoacoustics so that you can properly analyze headphone, speaker and room measurements and determine audibility of deviations...

www.audiosciencereview.com

 

[Cars-N-Cans]

Here is the current hypothesis of what are audible levels for various types of noise and distortion.

Audibility thresholds of amp and DAC measurements

Introduction A recurring theme on ASR is whether or not the various measured qualities of the devices are audible. In this post, I'll present some clear and visual thresholds for when those imperfections can be considered a potentially audible concern. I will not explain the basics of amp/DAC...

www.audiosciencereview.com

No one has yet demonstrated that they can hear any effects below them - if you want to understand it better, then by far the best way would be to try listening distortionn at various levels much higher than these suggested limits and see if you can hear them.

I know (at least on one test I tried) my ability to detect distortion ran out at about -45 dB. Some other people here reported similar results. It is quite an eye opener when you realise the levels of distortion produced by modern electronics is up to 10,000 times lower than you personally can actually hear.

We need to be cautious with that as the first post in the thread indirectly references the somewhat infamous “110 dB limit“ of audibility. This is being referenced to the 0 dB threshold of hearing with an average 110 dB signal or music level. You won’t even be able to hear distortion at -20 dB at those levels, and likely will experience pain and hearing damage, as you are no doubt aware. At a more reasonable listening level of 75 dBA, a more realistic limit will be ~25 dB absolute as the audible threshold for distortion. I think it’s obvious when one reads on in the thread, but def. a good idea to read further in than the first post for those not familiar with the topic.

As you say, -45 dB is where I run out as well Just can’t hear anything below that.

 

[antcollinet]

We need to be cautious with that as the first post in the thread indirectly references the somewhat infamous “110 dB limit“ of audibility. This is being referenced to the 0 dB threshold of hearing with an average 110 dB signal or music level. You won’t even be able to hear distortion at -20 dB at those levels, and likely will experience pain and hearing damage, as you are no doubt aware. At a more reasonable listening level of 75 dBA, a more realistic limit will be ~25 dB absolute as the audible threshold for distortion. I think it’s obvious when one reads on in the thread, but def. a good idea to read further in than the first post for those not familiar with the topic.

As you say, -45 dB is where I run out as well Just can’t hear anything below that.

I think it is less problematic when taking the green "lenient" limits - which as stated "are also "strict" from an audible point of view", I"ve edited my post slightly to reflect that.

 

[IAtaman]

Here is the current hypothesis of what are audible levels for various types of noise and distortion.

Amir has done videos on the topic:

Really appreciate the links, I read the article on audibility thresholds and am reading Amir's article on dynamic range now. Very informative.

In the meantime please allow me to share a graph to try and illustrate my point a bit better. This is from an article written by an application engineer in TI discussing the effects of headphone impedance characteristics, especially phase angles on amp design. Details are less relevant but what is more informative is the graph below.

These are two different amps with best case THD ratios of 95db and 112db respectively roughly (estimating from the graph). Can we claim that because the best case THD of 95db is below limits of threshold of audibility, Traditional Amplifier will be transparent and sound the same as Amplifier with Rx/Cx network?

If someone said the Traditional Amplifier sounded warmer while the Amplifier with Rx/Cx sounded more analytical compared to Traditional Amp, would not there be a way to interpret the measurements maybe to see where they are coming from?

 

[antcollinet]

FIrst question.

Based on my statement above about my ability to detect distortion falling off at about -45dB I would expect both amps to be transparent to me. Somone with hearing more sensitive to distortion may be able to hear the distortion in the "traditional" amp above, but given that sensitivity to distortion is typically lower at lower frequencies (I think - not sure where I read that), then perhaps not.

Second question - it is diffiult or impossible to relate "Audiophile poetry" to particular measurements, since the terms are simply undefined. What does analytical mean? Ask 10 different audiophiles, get 10 different answers. Secondly most of the descriptions "clean" "analytical" "warm" "cold" etc are more likely to relate to frequency response - which is generally pretty flat for amps, so are unlikely to be vailid for electronics in any case.

Audible distortion (such as my be created by overdriven tubes) might create a warmth, or fullness to sound, at severe cost to accuracy.

A well measuring amp won't have a sound - it will be the proverbial "wire with gain".


Again - I strongly reccomend you try an online distortion ABX test to get a first hand idea of how audible distorion is for you. I'd link you to the one I used, but can't find it.

Here it is:
https://www.klippel.de/listeningtest/

I need to be listening at the top end of my normal volume range to even get close to -45dB


Go on, try it and post your results here.

EDIT some more : I just tried, and only got -27dB this time. Room setup is significantly different now, and I've not yet re-run room correction. I'll try again when I have.

 

[Cars-N-Cans]

I have used that test from Klippel before. Oddly enough it got stuck on "B" for some reason, which meant I could hear distortion down to -96 dB or whatever it went to. I must have amazing ears!

Actually, this time around it appears to be working:

The audio starting out sounds like its equivalent to about a 64 kbps mp3. Don't know if they did that on purpose or not, or if it changes as you go. Last time I got stuck so it didn't change at all, so I think I will tinker with it a bit more later on that it seems to be working properly. Also minor correction, but the address included above also has the user progress as part of the HTML link. The general link is: https://www.klippel.de/listeningtest/

Edit: Taking a quick run thru now that it works the quality of the sound makes more sense since its really for comparative statistics between listeners I would assume:

Not surprisingly, I'm pretty much smack in the middle just casually taking the test using my 560s'. That song was pretty much the only thing I could stand hearing for more than a few times in a row, which makes it a bit of a handicap since intentional distortion was used artistically in the song itself.

 

[antcollinet]

I have used that test from Klippel before. Oddly enough it got stuck on "B" for some reason, which meant I could hear distortion down to -96 dB or whatever it went to. I must have amazing ears!

Actually, this time around it appears to be working:
View attachment 264250

The audio starting out sounds like its equivalent to about a 64 kbps mp3. Don't know if they did that on purpose or not, or if it changes as you go. Last time I got stuck so it didn't change at all, so I think I will tinker with it a bit more later on that it seems to be working properly. Also minor correction, but the address included above also has the user progress as part of the HTML link. The general link is: https://www.klippel.de/listeningtest/

Thanks - I've corrected the link in my post...

 

[Cars-N-Cans]

Really appreciate the links, I read the article on audibility thresholds and am reading Amir's article on dynamic range now. Very informative.

In the meantime please allow me to share a graph to try and illustrate my point a bit better. This is from an article written by an application engineer in TI discussing the effects of headphone impedance characteristics, especially phase angles on amp design. Details are less relevant but what is more informative is the graph below.

View attachment 264231

These are two different amps with best case THD ratios of 95db and 112db respectively roughly (estimating from the graph). Can we claim that because the best case THD of 95db is below limits of threshold of audibility, Traditional Amplifier will be transparent and sound the same as Amplifier with Rx/Cx network?

If someone said the Traditional Amplifier sounded warmer while the Amplifier with Rx/Cx sounded more analytical compared to Traditional Amp, would not there be a way to interpret the measurements maybe to see where they are coming from?

Its hard to say as the op-amp is being used to drive the headphones directly with no sort of buffer, so its not really representative of what you would encounter in a final commercial product. Id imagine that sort of thing is usually only done in mobile devices (edit: More likely to use a dedicated IC intended for driving the HP's to ensure good efficiency and drive capability, fwiw), which really do have that much distortion. I know my Samsung S9 has an output impedance of about 2.2 ohms or so, and it starts to get somewhat crummy into low impedance loads and doesn't look too much different. Typically in something we would use for driving headphones for more critical listening like my Schiit Magni 3 there are unity gain followers after the input op-amp to buffer it so the distortion doesn't get out of hand (not to mention having useable power as well).

I'm sure you are aware of this, but still worth noting that the above example you provide is not really an actual usage case for most of us. Would it be audible? From using my Truthear Zero IEM with my S9, no, but only just since its right on the edge of having audible distortion, and is only saved by the rising impedance of the LF driver it has. If it was 10 ohms all the way across, it would go straight into current limiting at low frequencies and have severe distortion.

Edit: One thing that does change when using the Truthear with my S9 is the frequency response. The low end gets boosted by a dB or two due to the changing impedance. It can be corrected for using some simple math and EQ, but it does change the tonality with respect to a low impedance dedicated headphone amp without it.