How much SINAD is enough

[Blumlein 88]

Designers or testers? You know more about it than me, but my hope is that designers at the top level focus on everything. That testers use only 1kHz ain't their fault. E.g., Benchmark's 1kHz happens to be great, but so is the whole 20 - 20k. Are other manufacturers gaming the tests, by knowing the testers focus only on 1kHz? Is it a VW-style thing?

A lot of your early Icepower amps quoted only 1 khz specs for many parameters. They suffered quite a bit from distortion rising with frequency. I don't know if you'd call it gaming. The distortion even at 20 khz was still plenty low, but not nearly so low as the 1 khz number. They did the same thing with SNR and dynamic range.

 

[JSmith]

It's not all about SINAD... has anyone said it is?



JSmith

 

[restorer-john]

Designers or testers? You know more about it than me, but my hope is that designers at the top level focus on everything. That testers use only 1kHz ain't their fault. E.g., Benchmark's 1kHz happens to be great, but so is the whole 20 - 20k. Are other manufacturers gaming the tests, by knowing the testers focus only on 1kHz? Is it a VW-style thing?

No, not to a great extent, but 1kHz is a best case number for a lot of designs, be they D/A converters, preamplifiers or particularly power amplifiers. THD at low frequencies and high frequencies is just as important or more so.

It goes back to the full disclosure specifications where any frequency across the audible spectrum is treated the same. (so they should be).

Here's an example.

The 1kHz plot (blue) is basically a best case number. As the frequencies go up, the THD becomes worse of course. What is also something to consider, is the raised THD is not representative of the actual total THD as harmonics start to fall off due to the 45kHz bandwidth. So, a 10kHz gets H2-H4. A 15kHz gets only H2 and H3 is right on the bandwidth stop- with nothing after. And a 20kHz, gets ony H2 with nothing afterwards. So the light blue, green and red plots are also domewhat deceptive as their real trajectory is going up rapidly, not falling as the plot shows.

We can see the HF distortion is actually very poor in this design.

The same happens at low frequencies where power supply issues (especially linear) can result in THD rocketing up below a few hundred Hz.

 

[restorer-john]

I don't know if you'd call it gaming.

I would call it selective disclosure.

Let's face it, if we had THD specified at any power level from 20Hz-20kHz, many of the so-called SOTA amplifiers would fail the grade.

 

[Blumlein 88]

I would call it selective disclosure.

Let's face it, if we had THD specified at any power level from 20Hz-20kHz, many of the so-called SOTA amplifiers would fail the grade.

Oh, I agree. I'd be more upset if distortion became clearly audible at other frequencies which doesn't seem to be the case. But 1% at 20 khz and .005% at 1 khz is a big difference at the rated power output.

 

[Ron Texas]

SINAD of 120 db is fine. 110 db is almost surely fine. Other than that hard to say because as others point out you need to separate out noise and distortion at a minimum. You also need to weight the noise on top of that. A-weighting goes a ways in that direction. Other weightings are more sophisticated and would be better, but aren't seen as commonly.

THD of -60 db with music is fine. Noise values need to be less than this. I'd say you probably want A-wtd noise levels of at least -90 db, and -100 db is more like it.

Very informative. There are many amplifiers with a s/n better than 100 db but distortion pulls down the SINAD down into the mid 70's.

Around here members get obsessed with SINAD as the only thing. Likewise with speaker preference ratings to the exclusion of dynamic range.

 

[rdenney]

And residual noise in uV, A-WTD.

Amen. Now, if I could just figure out how loud 40 uV of hiss is going to be in my 90-dB speakers.

Rick "come to think of it, I need to measure quiescent voltage of my amps to find out what the level is of this hiss, now that I have a voltmeter of the required sensitivity" Denney

 

[Tangband]

So been getting upgrade envy lately with some of the latest products returning amazing results - but it had me wondering is it worth it - is there a baseline acceptable performance a product should meet and anything else is just chasing numbers ?

Two things :
1) the sound from my average scoring SINAD desktop dac/amp is much better than my higher rated portable dac/amp for both IEMS and on ears.
2) aside from sound signature, is there any audible difference between last years 90s SINAD products to this years 100s SINAD products ?

Seems to me that the measurement reports should be compared to baselines rather than other products . And @amirm recommendations are really saying " there is nothing technically wrong with this product" after that consider its design , features and warranty, read the subjective reviews or listen to it to decide if its right for you.

I would say 80 dB SINAD is good enough in power-amplifiers. Very few loudspeakers have a distortion lower than about -60 dB.

Genelecs investigations and AES papers tells us that the amplifier should have less distortion than the loudspeaker. If the loudspeakers driveunits have 0,1 percent distortion, the amplifier should have less than 0,01 percent distortion.

In a dac/preamp/streamer I would like to have at least 100 dB SINAD, because of the digital volume control.
You have to have 20 dB headroom for that. Thats for practical use, and good enough. CD material is 96 db though, so theoretical with digital volume control included you would need a SINAD of maybe 116 dB DAC resolution.

 

[Qbd]

This post by @ctrl probably reflects reality the most. As with noise if you hear it you hear it if you don't you don't. no need for audibility thresholds.

This is the only sensible approach to noise IMHO.

To provide a couple of examples:

Crown XLS 1502 Specs say SNR > 103 dB, and amirs measurement of its smaller sibling puts it at >100 dB (at full power).

I can say both from personal experience (and a quick google search will show it’s a common issue) that this amp is noisy, such that it is clearly audible at MLP in my room. So obviously, one cannot say that 100 dB SNR is some «good enough» threshold in general.

Another example is the JBL M2 with its recommended amp Crown I-Tech 5000 HD. Specs say >112 dB SNR. Owners of this package have also noted audible hiss (can’t find the relevant thread right now).

So 112 dB doesn’t cut it as a general «good enough» cutoff either.

Now of course, if used with inefficient speakers in a noisy room, both of those Crown amps would be fine from a noise perspective. But trying to put a one-number-fits-all «good enough» number on SNR just isn’t feasible. If you hear it you hear it.

 

[abdo123]

This is the only sensible approach to noise IMHO.

To provide a couple of examples:

Crown XLS 1502 Specs say SNR > 103 dB, and amirs measurement of its smaller sibling puts it at >100 dB (at full power).

I can say both from personal experience (and a quick google search will show it’s a common issue) that this amp is noisy, such that it is clearly audible at MLP in my room. So obviously, one cannot say that 100 dB SNR is some «good enough» threshold in general.

Another example is the JBL M2 with its recommended amp Crown I-Tech 5000 HD. Specs say >112 dB SNR. Owners of this package have also noted audible hiss (can’t find the relevant thread right now).

So 112 dB doesn’t cut it as a general «good enough» cutoff either.

Now of course, if used with inefficient speakers in a noisy room, both of those Crown amps would be fine from a noise perspective. But trying to put a one-number-fits-all «good enough» number on SNR just isn’t feasible. If you hear it you hear it.

From my experience the noise of the DAC is the most limiting factor by far since it gets amplified repeatedly downstream, if an amplifier has a 112 dB S/N at full output and the hiss still bothered me i would probably look at my DAC first for sure before jumping into a Hypex / Purifi solution with 120 dB S/N.

 

[Qbd]

From my experience the noise of the DAC is the most limiting factor by far since it gets amplified repeatedly downstream

Oh for sure. Though I can happily note that the Topping EX5 together with Hypex NC502MP is completely silent to my ears. They are of course well matched with both DAC and amp having roughly 120 dB SNR.

 

[oivavoi]

This is the only sensible approach to noise IMHO.

To provide a couple of examples:

Crown XLS 1502 Specs say SNR > 103 dB, and amirs measurement of its smaller sibling puts it at >100 dB (at full power).

I can say both from personal experience (and a quick google search will show it’s a common issue) that this amp is noisy, such that it is clearly audible at MLP in my room. So obviously, one cannot say that 100 dB SNR is some «good enough» threshold in general.

Another example is the JBL M2 with its recommended amp Crown I-Tech 5000 HD. Specs say >112 dB SNR. Owners of this package have also noted audible hiss (can’t find the relevant thread right now).

So 112 dB doesn’t cut it as a general «good enough» cutoff either.

Now of course, if used with inefficient speakers in a noisy room, both of those Crown amps would be fine from a noise perspective. But trying to put a one-number-fits-all «good enough» number on SNR just isn’t feasible. If you hear it you hear it.

Very much this. IME noise is the most noticeable and annoying type of degradation of the signal from amps and dacs. Otherwise all modern amps and dacs which are not deliberately made to be coloured are in practice (as opposed to in theory) transparent to the human ear with music playing. Can be mitigated with gain matching though, so it's good to have amps (or active speakers) where it's possible to adjust the gain.

 

[Absolute]

Even with L-pad and the Crown CDi 4x1200 set to lowest gain (26 dBu) there's very audible hiss from 4 meters away on my compression drivers. SINAD doesn't tell me anything I need to know when it comes to one of the two things that matter, namely noise.
Without the L-pad to quiet the noise and signal down by about 13 dB the noise is horrendous.

Death and destruction upon all noisy electronics, and death and destruction upon all these measurements that don't tell me diddly about what matters most.

 

[restorer-john]

Even with L-pad and the Crown CDi 4x1200 set to lowest gain (26 dBu) there's very audible hiss from 4 meters away on my compression drivers. SINAD doesn't tell me anything I need to know when it comes to one of the two things that matter, namely noise.
Without the L-pad to quiet the noise and signal down by about 13 dB the noise is horrendous.

Death and destruction upon all noisy electronics, and death and destruction upon all these measurements that don't tell me diddly about what matters most.

Their published S/N numbers will tell you how much noise you are dealing with. They are 20-30dB away from SOTA in amplifier residual noise.

 

[Absolute]

Their published S/N numbers will tell you how much noise you are dealing with. They are 20-30dB away from SOTA in amplifier residual noise.

Do you have any tips on how to make sense of such ratings for laymen like me? When is S/N ratio very good and what about uV?

I think the Purifi stuff comes in around 30-32 uV residual noise, but what on earth does that mean?

 

[restorer-john]

Do you have any tips on how to make sense of such ratings for laymen like me? When is S/N ratio very good and what about uV?

I think the Purifi stuff comes in around 30-32 uV residual noise, but what on earth does that mean?

Signal to noise ratio is a somewhat deceptive number. For a given amount of noise (hiss, whatever you want to call it), the signal is the level above said noise, and measured with respect to full rated output into a given load.

It's a voltage relationship, specified in dB (decibels). So, manufacturers pick the highest voltage they can and the lowest residual in V (uV normally) to come up with a number. They will also "weight" the noise and specify a bandwidth. The bigger the number the better, as it sells more amplifiers.

Trouble is, the more powerful your amplifier is, the more voltage it develops, and, for a given residual noise level, the S/N number will be bigger by default. So, free ride!

Let's take an example:

A 100 watt per channel amplifier at 8 ohms will develop 28.28V RMS over that 8R load. (take the sqr root of 100x8)
Let's say it has 50uV of residual noise (unweighted).
Take 28.28V, divide it by 50uV and you get 565,685.42.
Log that, you get 5.7526.
Multiply by 20 (as we are talking voltage, not power for dB) and you get 115dB.

So, all things being equal, your 100 wpc amplifier with 50uV residual has a 115dB S/N ratio.

Let's do the same with the Purifi you mentioned (off the published specs)

It has 11.5uV of residual, A WTD. (a very low residual noise- amazing in fact, due to its very low gain in part)
It is rated at 425 watts at 4 ohms.

Let's see how it works out:
425 watts at 4 ohms is 41.23V (take the sqr root of 425x4)
The residual is specified at 11.5uV
Divide 41.23V by 11.5uV and you get 3,585,309.24
Log that, you get 6.5545
Multiply by 20 (as we are talking voltage, not power for dB) and you get 131dB.

What is Purifi's dynamic range spec? (in real terms, it's the same as S/N in this case)

So, we know the math works out.

So, the Purifi is incredibly quiet, but it also has very low gain of only 12.8dB (x4.3) without a buffer stage. For some rare individuals, that may be enough. For most, it is not remotely enough. A normal amp is around 26-29dB (x28). So, you put a buffer up front to increase the gain, but with that gain comes some noise. Noise that the amplifier dutifully amplifies. You can see the issue?

 

[rdenney]

John, you didn’t answer what I think is a good question: what should we expect as hiss output from a given quiescent hiss output voltage? If my amp puts out 40 microvolts of hiss just idling, will I hear that in 90-dB-sensitivity speakers? Connect those dots for us, and then people can know the practical meaning of the spec.

Quiescent output voltage is an absolute value, not a ratio.

Rick “is that good?” Denney

 

[Absolute]

@restorer-john Thank you, it makes perfect sense now. So this is how Crown can point to good S/N numbers even though it's noisy as hell. All very powerful amplifiers will have "artificially" high S/N ratios. One question, is the uV (unweighted) a constant regardless of signal strength so that we can turn the math around to find the uV based on rated power at specified S/N number?

 

[restorer-john]

John, you didn’t answer what I think is a good question: what should we expect as hiss output from a given quiescent hiss output voltage? If my amp puts out 40 microvolts of hiss just idling, will I hear that in 90-dB-sensitivity speakers? Connect those dots for us, and then people can know the practical meaning of the spec.

Quiescent output voltage is an absolute value, not a ratio.

Rick “is that good?” Denney

It's tough to answer. I'm super sensitive to residual noise (hiss) and, as such, I'm unforgiving of it anywhere in the reproduction chain. I guess it was all those years trying to get rid of it prior to the introduction of CD. But then, we became even more fussy, preamplifiers added noise, high gain amps added noise we didn't need, and processors totally ruined the "quiet backgrounds" we wanted.

I don't know how close you sit to your tweeters, how quiet your room is, and how good your hearing is. I also don't know what your normal playback gain is. Hiss (residual) is dependent on all those factors.

You have a millivoltmeter- use it. Make up an A-WTD filter and compensate for it if the bandwidth of yours when wide open is too much, especially if you do Class Ds.

One question, is the uV (unweighted) a constant regardless of signal strength so that we can turn the math around to find the uV based on rated power at specified S/N number?

Pretty much, yes, you can mostly assume that. But beware. Manufacturers will pick the highest output voltage they can, even if it is at extreme level of distortion in some cases. Pro amps will often reference the BTL (bridged) output power as it has twice the voltage. And, at the other end, they will specify an A-WTD curve or a curve that kills off the mains related contributions (50/60Hz and harmonics) like a 400Hz HPF (aka hum and noise filter).

 

[restorer-john]

Funny, just listening to headphones and could hear ultra low level hiss. Then it started lightly raining outside. No hope of hearing hiss now. Off to bed.