• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Why SINAD is not important (article)

mdsimon2

Major Contributor
Forum Donor
Joined
Oct 20, 2020
Messages
2,478
Likes
3,315
Location
Detroit, MI
Missed this post. With speakers, @amirm does measure speaker's resolution in terms of frequency response, distortion, just to name a few metrics. It tells a lot about the performance of the speakers. Do you really think @amirm would measure only internals of the active speakers (in terms of SINAD) to make meaningful conclusions about speakers performance? It just doesn't make sense and you're taking things out of context. SINAD in active speakers amps is only part of the big picture, but of course important.

Ok, enough for now, hopefully we will put this debate to an definitive end...

No, I am not recommending that Amir measure internal electronics of active speakers.

@JJB70 post #118 sums up my stance on this nicely. There is a disconnect between worrying about SINAD for discrete electronics and completely ignoring it for internal electronics of an active speaker.

I am not taking anything out of context, if anything I am asking to take things in context. If an active speaker can sound good with such poor measuring internal electronics it says that good electronic measurements are not as important as other aspects (speakers, room, DSP, etc). I am in 100% agreement that you need to look at the whole picture.

Michael
 
Last edited:

confucius_zero

Addicted to Fun and Learning
Joined
Nov 22, 2018
Messages
541
Likes
345
so... test bench update soon?
 

dshreter

Addicted to Fun and Learning
Joined
Dec 31, 2019
Messages
794
Likes
1,226
There has often been something of a double standard in chasing SINAD in amplifiers and DACs (and deriding equipment with less than stellar SINAD) at the same time as ignoring audible hiss in active speakers or rationalising excuses for it. I actually agree that in almost all cases the hiss doesn't affect listening enjoyment and is outweighed by the benefits of a single well integrated package and I don't go around putting my ears against speakers looking for hiss anyway. But I think people should be honest about what it says if they suddenly see no issue with audible hiss.

I see no disconnect. The person hooking up a 305P to an $800 DAC has a disconnect. Evaluating a piece of equipment for excellence amongst its class and its peers though just makes sense. That DACs have already achieved a very high standard for what they need to do is besides the point if one still wants to compare them.

This is a growing hobby especially with headphone listeners, so many are still making their first or second purchases. There’s no reason they shouldn’t understand the full playing field in determining how to apply their expected outlay.
 

Jim Matthews

Major Contributor
Forum Donor
Joined
Mar 25, 2021
Messages
1,051
Likes
1,286
Location
Taxachusetts
If you want transparency of your current active speakers, I would definitely start with the proper source...

I concur that this is a reasonable approach. The "gate" through which the data must travel is the stage where noise (measurably) distorts the signal.

It's like losses in any power transfer system - the wasted power is inefficient use.
 

tmtomh

Major Contributor
Forum Donor
Joined
Aug 14, 2018
Messages
2,636
Likes
7,491
This is a fascinating thread - thanks all! I appreciate the detailed, in-depth discussion and hope it continues. At the same time, I also think there's value in keeping a simple point in focus: SINAD is far from useless. As I see it, the article seems to be making a good point that SINAD could perhaps be more useful if the "SIN" component were given more weight and the "D" component less weight, instead of merely being added together. I have no idea what a proper factor would be, but if the limit of human hearing as far as noise is about -115dB and the auditory masking threshold of low-order harmonics for trained listeners is, say, 80dB, then perhaps multiplying the S/N by 1.1 or 1.2 and the distortion by 0.8 and then adding them might help? Again, I have no expertise in this, so this could be completely dumb - just a thought.

On an even simpler level, as a layperson and potential consumer, I have to say I'm with @amirm here when it comes to the bottom line: given the performance capability of the best, affordable DACs, amps, and other electronic components today, you can basically just pick a SINAD threshold and know you're good to go at that figure no matter what the noise and distortion components are. True, if you do that you will miss some components that have lower SINAD but a "goldilocks" level of noise and THD within that SINAD where they will still be audibly transparent to most people in most situations. But given that such a wealth of affordable options are available that easily clear the threshold, I'm not entirely sure what the reason would be to bother looking into those details, except perhaps with AVRs, where there are unique features some folks need, and the performance levels are on the whole lower than standalone DACs and stereo amps.
 
Last edited:

HiFidFan

Addicted to Fun and Learning
Forum Donor
Joined
Feb 22, 2021
Messages
723
Likes
906
Location
U.S.A
We will do an article that recommends aggregating noise and distortion over my dead body, but I'm definitely hoping to do a piece on some of the more-subjective-results-correlated alternatives for distortion and...I mean, there really isn't anything wrong with SNR.

I truly hope you do. And explain what "subjective-results-correlated alternatives for distortion" means.
 

Wes

Major Contributor
Forum Donor
Joined
Dec 5, 2019
Messages
3,843
Likes
3,788
This is a fascinating thread - thanks all! I appreciate the detailed, in-depth discussion and hope it continues. At the same time, I also think there's value in keeping a simple point in focus: SINAD is far from useless. As I see it, the article seems to be making a good point that SINAD could perhaps be more useful if the "SIN" component were given more weight and the "D" component less weight, instead of merely being added together. I have no idea what a proper factor would be, but if the limit of human hearing as far as noise is about -115dB and the auditory masking threshold of low-order harmonics for trained listeners is, say, 80dB, then perhaps multiplying the S/N by 1.1 or 1.2 and the distortion by 0.8 and then adding them might help? Again, I have no expertise in this, so this could be completely dumb - just a thought.

On an even simpler level, as a layperson and potential consumer, I have to say I'm with @amirm here when it comes to the bottom line: given the performance capability of the best, affordable DACs, amps, and other electronic components today, you can basically just pick a SINAD threshold and know you're good to go at that figure no matter what the noise and distortion components are. True, if you do that you will miss some components that have lower SINAD but a "goldilocks" level of noise and THD within that SINAD where they will still be audibly transparent to most people in most situations. But given that such a wealth of affordable options are available that easily clear the threshold, I'm not entirely sure the reason to bother looking into those details, except perhaps with AVRs, where there are unique features some folks need, and the performance levels are on the whole lower than standalone DACs and stereo amps.

The suggested weighting would still produce a univariate figure of merit. I dunno if that will be correlated or better correlated with SQ*

It may be necessary to decompose things and present a response surface.

* and what we want is something that is easy to measure that is so correlated, whether a single measurement or a suite of them
 

tmtomh

Major Contributor
Forum Donor
Joined
Aug 14, 2018
Messages
2,636
Likes
7,491
The suggested weighting would still produce a univariate figure of merit. I dunno if that will be correlated or better correlated with SQ*

It may be necessary to decompose things and present a response surface.

* and what we want is something that is easy to measure that is so correlated, whether a single measurement or a suite of them

Actually, the entire critique of SINAD is that it's a singular figure, but precisely not a univariate one - the author's point is that THD and S/N behave differently and have different thresholds of audibility, so it's a multivariate figure. And these two variables that compose SINAD - S/N ratio and THD - are already easy to measure and widely acknowledged as useful. So we could say SINAD isn't terribly important pro or con, and what we really need to know about are the noise and THD measurements that go into that SINAD figure. Except we don't need to say that, since Amir already breaks those out in some detail in his reviews, as does pretty much every other reviewer I'm aware of who posts SINAD measurements.

But part of my point - and I apologize if I was implicit when I should have been explicit - is that the article, in and of itself, does not persuade me that SINAD is useless of should be abandoned. The article's topline claim makes it sound like there's something actively bad about SINAD. But from what I can tell, the only issue with it is that if you set a SINAD benchmark sufficient to ensure transparency, there will be some equipment that doesn't meet that benchmark but is still audibly transparent because its SINAD is dominated by THD. Fair enough - but there are also other measurements that folks around here are quite interested in beyond the point of audible transparency too. So doing away with SINAD wouldn't solve the core issue, which is that most folks who believe in measurements want some "buffer" and/or "feeling of engineering excellence" in their measurements, beyond what they might actually need.
 
Last edited:

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,591
Likes
10,727
Location
Prague
If there is a good 1kHz SINAD, it tells that some of the design steps were done right. Some. We have no information about higher frequency linearity, no information about higher frequency intermodulation, no information about bandwidth, no information about output impedance and no information on stability into complex load impedance. Thus, as a single number score, it really does not tell much. So why this, just this specific number should be used as a distinctive parameter in the comparison chart? I think that just for the reason it is easy to measure and we have something to wave with. Nothing more.
 

tmtomh

Major Contributor
Forum Donor
Joined
Aug 14, 2018
Messages
2,636
Likes
7,491
If there is a good 1kHz SINAD, it tells that some of the design steps were done right. Some. We have no information about higher frequency linearity, no information about higher frequency intermodulation, no information about bandwidth, no information about output impedance and no information on stability into complex load impedance. Thus, as a single number score, it really does not tell much. So why this, just this specific number should be used as a distinctive parameter in the comparison chart? I think that just for the reason it is easy to measure and we have something to wave with. Nothing more.

Sure - but by that logic, why should any one measurement be used? The answer is there's no reason for any one measurement. So we have multiple ones. In any review, one still has to be first. The only "special" status of SINAD here at ASR is the ranking charts. But Amir color-codes them into large quintiles for a reason - to de-emphasize chasing 1 or 2dB of difference here and there between individual models and units.

Put more simply, the idea that any one measurement is a necessary but not sufficient condition to make an informed judgment is an idea that I doubt anyone here would disagree with. But let's not forget that this is not what the article in question says - or at least not what its headline says. It goes beyond that and says SINAD is not important. I disagree.
 

KeithPhantom

Addicted to Fun and Learning
Forum Donor
Joined
May 8, 2020
Messages
641
Likes
658
Well, I'll start by saying that this article has very good points, especially explaining masking windows of distortion vs. the fundamental. Not only that, it breaks down both the components of Signal-to-noise-and-distortion ratio very well, I commend this aspect.

But there is a problem with the approach of the article and the video, it assumes that the measurement only matters for audibility. The importance of these measurements is vital in the design and testing phase of audio products, not just to assure audible transparency, also technical competency (or any other objective like the addition of the desired distortion in case of some amplifiers designed for it). In terms of audibility, the sensitivity of the ear is pretty limited, being sounds masked if they are from -30 (or -40-50 depending on the research you read) dBr to -infinite (threshold of hearing masking) from reference, this area where the fundamental can mask close-by sounds is called the masking window (being the line that defines the function the masking threshold). It is level and frequency-dependent due to our ears and how they are more sensitive to some frequencies than to others. On top of that, our hearing is not linear but follows the equal-loudness contour that states our insensitivity at low frequencies and our increased sensitivity at high frequencies.

Before discussing audibility, I want to state the use of objective metrics such as SINAD/THD+N/ENOB is completely acceptable in its current form if used to rank the technical performance of a device, regardless of audibility, also being general enough of including single-point variables. Why? These describe the addition of mathematically-unwanted signals or components to the desired output of the transfer function (or its real-world approximation). These objective measurements allow us to measure the interaction of noise and distortion (both unwanted) and add them into a single number. Now, the single number is just at best an overall representation, like any other single measurement. If a more in-depth examination is warranted, reporting noise and distortion measurements separately can reveal more about the characteristics of a device. In this case, ears do not need to intervene since their insensitivity is more of a liability than an asset, because they are pretty bad at picking up nonlinearities unless they are high in amplitude and in the right frequency region.

In terms of audibility, I am in agreement with the article, especially in the distortion part.

Now, the approach of the article suggests that SINAD is the only metric given in measurements (except in the last part). We do rank by SINAD, but the reviews of every place I see that provides measurements have many more, in-depth measurements exploring the components of this aggregate number. SINAD is a good 'welcome' measurement that invites you to explore the other measurements that compose it.

Another point that I am somewhat in disagreement with is the title and the conclusion. Why should we settle for less performance (assuming subjective qualities and features are the same) when we can get more? For the ones looking for fidelity, if there are offerings of DACs and amplifiers in the range of 120 dB SINAD (as I said, it is just a general measurement, go to the review for specifics), why buy the ones that score worse when we can build both devices at the same price. If they had more features, that is a valid point; if you like how they look or feel, that is another valid point; but if you want fidelity, even if you are not really going to hear it, why settle for less? Even if you won't hear it, let your ears do their filtering, put their shortcomings on the faithful sound that hits them. It also gives you peace of mind in terms that at least the limiting factor by a huge lot is going to be your ears.

TL;DR: SINAD matters for technical aspects or preference of fidelity by the user. Do not buy by SINAD only, your ears won't hear it.
 

Raindog123

Major Contributor
Joined
Oct 23, 2020
Messages
1,599
Likes
3,554
Location
Melbourne, FL, USA
In digital RF communications, there is an aggregate parameter - EVM (error vector magnitude) that attempts to characterize the degree of "imperfection" of the channel, "the amount of [signal reconstruction] errors" regardless of the exact origin of these errors. In essence, its a "max SINAD across operational spectrum".

There are many effects contributing to the EVM: spectral nonlinearities, oscillator phase noise, signal clipping, intermodulation, channel effects (eg multipath reflection and frequency-selective fading), etc. Now, while EVM is widely used for system debugging, the "design requirements / performance specifications" are usually given by other metrics: SNR, phase noise, sensitivity, frequency budget (filtering, aliasing), ...


EVM is the "smearing" of each information state ('constellation point') related to the signal strength for this point:
zzzz.png
 
Last edited:

Wes

Major Contributor
Forum Donor
Joined
Dec 5, 2019
Messages
3,843
Likes
3,788
Where are the listening tests of a top quintile box vs. a lowest quintile one?
 

tmtomh

Major Contributor
Forum Donor
Joined
Aug 14, 2018
Messages
2,636
Likes
7,491
Where are the listening tests of a top quintile box vs. a lowest quintile one?

Not the point - because you could ask the same question about any measurement, including frequency response and including the individual S/N and THD measurements that comprise the SINAD number.

The issue is setting an appropriate threshold or benchmark for any measurement. There’s nothing special about SINAD in this regard, and so your rhetorical question does not provide any support for the position you’ve staked out here RE SINAD.
 

KeithPhantom

Addicted to Fun and Learning
Forum Donor
Joined
May 8, 2020
Messages
641
Likes
658
Where are the listening tests of a top quintile box vs. a lowest quintile one?
You can do that by creating files that have the noise + distortion that is equivalent to the SINAD you want. If you want to test a distortion-dominated SINAD of 60 dB at 1kHz, you can create a file with a fundamental of 1 kHz and a harmonic series where the second distortion is -60 dBFS from the fundamental. That will net you the same as testing with equipment. This is what the article did to explain the insensitivity of our ears to nonlinearities.
 

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,835
Likes
16,497
Location
Monument, CO
Many years ago I created a set of files with a single tone and two tones for IMD, with added distortion of 0%, 0.1%, 1%, 5%, and 10% IIRC and posted them (on a different website) for people to try. The results fit the classical "high-fidelity" assumptions: hardly anybody picked up on 0.1%, most got 1% after a few trials, and everybody could tell 10%. I think I added 0.5% for IMD as people could pick it up more readily than HD. I don't remember the spread and doubt I still have the data (I lost a lot of files in a HDD crash about ten years ago; the drive was going bad and I did not know it, so my backup was also corrupt when the drive finally died). Note 1% is -40 dB, 0.1% is -60 dB, in voltage (the basis for my files).
 

Wes

Major Contributor
Forum Donor
Joined
Dec 5, 2019
Messages
3,843
Likes
3,788
Not the point - because you could ask the same question about any measurement, including frequency response and including the individual S/N and THD measurements that comprise the SINAD number.

The issue is setting an appropriate threshold or benchmark for any measurement. There’s nothing special about SINAD in this regard, and so your rhetorical question does not provide any support for the position you’ve staked out here RE SINAD.

my position is to care if one can hear a difference or not - you seem confused
 

tmtomh

Major Contributor
Forum Donor
Joined
Aug 14, 2018
Messages
2,636
Likes
7,491
my position is to care if one can hear a difference or not - you seem confused

There are scientifically established thresholds for what people can and cannot hear. For example, the noise threshold is about -115dB. But what volume are you listening at? How loud or quiet is your listening space? Are you listening to speakers, and if so near-field or far-field? Or are you using headphones, and if so what is their impedance and efficiency? Perhaps even more fundamentally, what kind of equipment is being measured - where is it in the chain? Is it a source, or is it processing or amplifying a source? All of these make a difference in what people can or cannot hear when comparing gear with differing measurements.

This is why some objective measurement standard is needed - use cases vary. If you listen to far-field speakers in a room with 35dB ambient noise, you might be able to go farther down Amir’s quintile scale before you hear a difference than if you listen with closed-back, over-ear headphones in a quiet space.

But you can’t figure any of that out unless there’s some kind of objective benchmarking system to use.

So it does indeed seem that one of us is confused.
 

Wes

Major Contributor
Forum Donor
Joined
Dec 5, 2019
Messages
3,843
Likes
3,788
The "scientifically established thresholds" for what people can and cannot hear are not always accurate, and are often based on old, and highly simplified methodology. They may need to be updated. If you do research in this area, then I am all ears, so post your citations.

You are confused indeed if you think I claimed that no "objective measurement standard is needed" as you put it. Please refer up thread to what I actually said. I did mention that it may not be possible to collapse all objective measurements into a single figure of merit and, if so, then a response surface or some other means of presentation would be needed. I hope this clarifies things.
 
Last edited:

stevenswall

Major Contributor
Forum Donor
Joined
Dec 10, 2019
Messages
1,366
Likes
1,075
Location
Orem, UT
Makes sense. But if you want a single score to rank devices, what do you use?

Something like a pass fail grade for different things:

Audibility Test: 0/5
Impedance: Does this change the frequency response in an audible way for more than 1% of speakers? No? Fail.
Power: Can this drive 99% of consumer speakers to their limits without audible amp distortion? No? Fail.
SINAD: Is this transparent to 99% of users? No? Fail.
Hiss: Can anyone hear any hiss in a non-anechoic environment? Fail.
Other Issues: Does this have an audible fan or pop or click or is a turntable? Yes? Fail.

Engineering Test: 0/5
Distortion: Is this in the top quin-tile for it's price category? No? Fail.
Power Output: Is this in the top quintile for its price category? No? Fail.
Connections: Does this output a balanced or digital signal instead of just having archaic RCA and speaker wire terminals? No? Fail.
Output filters: Does this follow best practices to avoid forwarding ultrasonic distortion? No? Fail.
Build: Is the device well made without flex and with a volume control that will last for a century and costs 10 cents more? No? Fail.

Then have customers make a 5/5 in both tests the absolute minimum for anything to ever spend a penny on it, and we can categorically dismiss the rampant amounts of trash. Trash=Something that isn't best in category, best for the price, best all around, etc. EX: Many models of cars don't have the best suspension, best acceleration, best reliability, or best of anything else. They should never be recommended to anyone unless they are looking for a visual piece to add to their collection or tell you that they are a fashion slave/marketing addict.
 
Top Bottom