How to interpret SINAD numbers of separate DAC + Amp combos

[restorer-john]

SINAD is lazy, pointless, and doesn't really tell us anything useful at all.

I want absolute A-weighted residual noise in uV in the absence of signal, and I want THD at arbitrary levels separately. This has been the standard for many decades. How hard is it?

Residual noise in uV vs maximum output level in volts tells you what the DR of your gear can actually achieve.

For modern consumer digital line levels use 2.0 V, after all, that is the standard.

As for any for amplifier levels, use 1 watt at 8ohms. (2.83V) You know, the value that has been used for 50 years or so. Not some arbitrary number dreamed up. Not 5W into 4 ohms or some other stupid number.

 

[Krunok]

Residual noise in uV vs maximum output level in volts tells you what the DR of your gear can actually achieve.

And how would you compare that value among different gear when manufacturers are not specifying it?

 

[DonH56]

Noise colors: https://en.wikipedia.org/wiki/Colors_of_noise
Pink noise: https://en.wikipedia.org/wiki/Pink_noise

Pink is equal energy per octave and is often used for audio measurements. Not as annoying (and tweeter-intensive) as white noise.

Fun factoid: I used colored (bandlimited) noise to decorrelate spurs in a radar data converter back in the 1980's. Dither (noise decorrelation) has been around a long time.

I find SINAD useful as a relative comparison but from there it is nice to know what part is noise, what part is distortion, and the frequency spectra of each. If SINAD is 120 dB for one audio device and 124 dB for another chances are I won't care enough to dig deeper. If one is 80 dB I'd like to know why the poor performance. Note it is difficult to do proper noise separation so most THD curves in reviews are actually SINAD, sometimes labeled THD+N though noise can include a lot of things... I'd like to know if it is broadband noise, discrete tones like clock coupling or power supply spectra, etc. So not quite pointless to me but usually just the starting point, a quick and dirty assessment of performance.

I suspect the noise floor of many components is near the noise floor of the instrumentation, and measuring low levels is very hard -- difficult to keep other spurs out of the signal path (fluorescent lights are a royal PITA, for example). Without a lot of effort chances are the no-signal noise floor might not look that great. At in some cases it is the opposite -- without a signal, some components mute outputs (e.g. when receiving digital 0) or have static outputs from data converters (DACs) that does not include quantization noise so may be unrealistically good. Always something...

 

[Krunok]

SINAD isn't a function of frequency in that way, it is a ratio of powers within a measurement bandwidth as stated above.

But are SINAD components measured at single frequency or they are average within the measured frequency range?

Both SINAD components, THD and noise, change value with frequency so how do you exactly calculate a single value?

 

[DonH56]

SINAD = signal to noise and distortion. It is usually a single tone (signal) measured relative to everything else except that tone. For audio it is common to specify the number with a 1 kHz tone but also modern analyzers make it easy to do a sweep so you get a number valid over the sweep range (typically 20 Hz to 20 kHz) and power level (stated, often max rated output power for an amplifier). The datasheet should clarify if it is specified over a range (of frequency and/or power/voltage level) or a single-point measurement.

SINAD is very general and in that I agree (though less colorfully) with John. The "NAD" part includes THD, IMD (if multiple tones are used), power supply noise, clock spurs and noise, RFI/EMI, the components noise floor from thermal, shot, flicker, and other noise sources, etc. etc. etc. If data conversion is included then quantization noise is also present, which may or may not be shaped and filtered. And so forth and so on.

 

[Krunok]

For audio it is common to specify the number with a 1 kHz tone but also modern analyzers make it easy to do a sweep so you get a number valid over the sweep range (typically 20 Hz to 20 kHz) and power level (stated, often max rated output power for an amplifier). The datasheet should clarify if it is specified over a range (of frequency and/or power/voltage level) or a single-point measurement.

Ok, I understand this, but when measured over a frequency range and at say max rated power, does the value represent average or minimum value measured?

 

[DonH56]

Ok, I understand this, but when measured over a frequency range and at say max rated power, does the value represent average or minimum value measured?

It should be the minimum value. That was an IHF/FTC requirement but I have not kept up with those specs in years and they are woefully lacking for multichannel systems. I have seen a number of datasheets that include both single-tone and swept results, but IME most of them use power instead of SINAD as the variable. That is, an amplifier may spec e.g. 0.01% THD+N from 20 Hz to 20 kHz at 100 W, and max power of 120 W at 0.01% THD+N for one channel at 1 kHz. By mixing up the variables they (Marketing) can highlight what they want (and feel will best sell the product).

Normally a single tone is swept in frequency and at each frequency the tone is compared to everything else so the SINAD slides along with frequency. The minimum value is reported over the frequency range that is swept. One frequency at a time.

Frankly, THD specs are so good these days, I rarely do more than glance at them. Speakers usually dominate overall system performance, not the electronics. And THD specs are more about power output (since that's what sells) -- a manufacturer may choose to rate an amplifier at 0.1% instead of 0.05% or whatever just to publish a power rating a little higher than their competition. Even when in the real world they are essentially identical, consumers go for power. Hmph.

 

[Krunok]

Speakers usually dominate overall system performance, not the electronics.

What a simple truth is that and yet so few folks seems to be aware of it.

 

[amirm]

So the value you get is average SINAD vaue within that range, or..?

As John explained, that is not how the measurement is done. Think of removing the main tone and then using a voltmeter to measure everything that is there. In other words, the measurement is without taking apart the waveform in frequency domain.

 

[bennetng]

@amirm
Could you show AP's SINAD and AES17 DR reading of dithered and undithered 16/44 data without going through a DA-AD loop? Thanks!

 

[amirm]

@amirm
Could you show AP's SINAD and AES17 DR reading of dithered and undithered 16/44 data without going through a DA-AD loop? Thanks!

Undithered 16 bits? Why? Regardless, if I use the analog generator in the APx555, there is no concept of dither anyway.

Here is the SINAD for unbalanced:

And Balanced:

Is this what you were asking for?

 

[amirm]

And here is the dynamic range in unbalanced (2 volt output):

And balanced:

 

[Krunok]

Undithered 16 bits? Why? Regardless, if I use the analog generator in the APx555, there is no concept of dither anyway.

Here is the SINAD for unbalanced:

View attachment 18077

And Balanced:
View attachment 18078

Is this what you were asking for?

I can see a small spike at 60Hz which is obviously coming from power supply and not to mention those 2 spikes at 2k and 3k. - clear sign of not so good engineering!

 

[restorer-john]

Those AP loopback numbers are amazing. What a wonderful piece of test gear.

 

[bennetng]

Is this what you were asking for?

Sorry, maybe my English is bad. I meant something like this:
https://www.audiosciencereview.com/...ts-of-schiit-yggdrasil-v2-dac.3607/post-88290
https://www.audiosciencereview.com/...ts-of-schiit-yggdrasil-v2-dac.3607/post-88328

The SINAD and DR of the synthesized 16/44 digital data itself, dithered and undithered, without any analog conversion, like SPDIF. Should be somewhere between 90-100dB.

However, the AP self-loop result is also a useful reference, thanks.

 

[amirm]

Sorry, maybe my English is bad. I meant something like this:
https://www.audiosciencereview.com/...ts-of-schiit-yggdrasil-v2-dac.3607/post-88290
https://www.audiosciencereview.com/...ts-of-schiit-yggdrasil-v2-dac.3607/post-88328

The SINAD and DR of the synthesized 16/44 digital data itself, dithered and undithered, without any analog conversion, like SPDIF. Should be somewhere between 90-100dB.

Those measurements require use of the DAC in AP which I don't normally use. The DAC generator is only used for special signals that can't be generated using the high-performance sine generator. Or if you have a lower end version of the AP that doesn't have the high performance generator.

What is the motivation for wanting to see that given the above?

 

[bennetng]

What is the motivation for wanting to see that given the above?

To have some reference. Some reviews (e.g. Modi3, Asus STX II) have compatibility issues with ASIO and unable to show 24-bit performance. I would like to simulate this situation and see what number will show up on AP, also the pattern of the FFT analysis.

 

[signalpath]

SINAD is lazy, pointless, and doesn't really tell us anything useful at all. I want absolute A-weighted residual noise in uV in the absence of signal, and I want THD at arbitrary levels separately. This has been the standard for many decades. How hard is it?

Maybe not "pointless" -- but the sentiment is valid.

I would go further and ask for broadband, unweighted noise reading in uVrms (or dBu) in the absence of signal. This is the gold standard in our lab. Separate THD readings, sure, but below a certain threshold (0.05%?) it really doesn't tell us anything of perceptual value.