What to Measure to Evaluate DAC Sound Quality

[Wes]

Besides the obvious things like Frequency Response, Noise, and Total Harmonic Distortion there may be other factors in SQ. In other words, SINEAD & FR may not completely characterize the performance of a DAC.

I found the following list on a Subjectophilicly Censored Forum (you can guess which one), but it was posted before that site converted to a single party system via fatwah.

Here is the list:

1. shape of the oversampling filter

2. analog filter roll-offs

3. correlated and non-correlated in- and out-of-band distortion and noise components

4. phase response

5. TIM

6. step overshoot

7. modulator behavior

I am curious as to what thoughts (or even data) folks here may have on this list or on the general question of whether the current measurements provide a complete* characterization of DAC SQ.


* Kurt Godel sends his thanks

 

[Wes]

PS - I'm not an EE, and thought a new thread would be a good place to vet this issue before posting any or all the above in the 'what measurements I'd like to see added' thread...

 

[majingotan]

I just add RCA output impedance to an active monitor input impedance. There are DACs out there (for example PS Audio Directstream) that doesn't even output the standard 75 ohms in RCA or 110 ohms in XLR out. Some active monitor amp have very low input impedance (600 ohms) which can cause frequency response deviation from flat

 

[RayDunzl]

There are DACs out there (for example PS Audio Directstream) that doesn't even output the standard 75 ohms in RCA or 110 ohms in XLR out.

Aren't those impedance standards for coaxial (S/PDIF) digital and AES/EBU inputs, not for the DAC analog outputs?

 

[majingotan]

Aren't those impedance standards for coaxial (S/PDIF) digital and AES/EBU inputs, not for the DAC analog outputs?

IIRC they're both for inputs and outputs. Then again if there's no standard OI, < 1 ohm output impedance would be ideal

 

[beefkabob]

Definitely gotta measure number 5 on your list.

 

[ZolaIII]

Yes there definitely is more to it like; noise, noise & more noise. All do it sounds vulgar plainly said as that there are a lot of sources for it (from process & materials used to voltage regulation, operating temperature, shielding and cetera). Factors you cannot influence is how IC whose made but you can pretty much eliminate lot of other things by applying shielding and using clean more stable power supply and satisfactory operating conditions. All measurements are heavily influenced by mentioned conditions and they rarely meet the real use one's.

 

[mansr]

1. shape of the oversampling filter

In most cases it's impossible to access the digital oversampling filter output. The best we can do is capture the analogue impulse response. A lot of DAC reviews include this. Amir usually doesn't for whatever reason.

2. analog filter roll-offs

Isn't the overall frequency response what ultimately matters? If you want to know the characteristics of the analogue portion alone, that might be tricky to measure.

3. correlated and non-correlated in- and out-of-band distortion and noise components

That's quite a catch-all description.

4. phase response

The impulse response gives some indication of this. In most cases, it's not interesting beyond noting whether it is linear phase, minimum phase, or something in between.

5. TIM

Isn't that just a special case of intermodulation distortion?

6. step overshoot

The step response and impulse response both fully characterise a linear system. It is enough to measure one of them.

7. modulator behavior

That's a can of worms. The less secretive chip vendors (i.e. not ESS) usually reveal the number of levels and modulator order in the datasheets. The noise spectrum is easy enough to measure, typically presenting in the output as a hump rising above the noise floor somewhere in the 30–60 kHz range and falling off as the analogue filter kicks in. The spectrum usually scales with the master clock frequency, so it can vary between devices based on the same chip. Teasing out any other characteristics is tricky at best and unlikely to reveal anything interesting anyway.

 

[mansr]

Aren't those impedance standards for coaxial (S/PDIF) digital and AES/EBU inputs, not for the DAC analog outputs?

Yes, the 75 Ω and 110 Ω specs apply to digital connections only. Analogue audio signals are low enough in frequency that impedance matching is irrelevant.

 

[miero]

I'd extend the list with a recording of this song: https://youtu.be/dQw4w9WgXcQ

 

[Wes]

mansr, do you think that the measurements on ASR completely account for any SQ for a DAC?

or could any of the above (Miska's post mainly) have an effect outside of that?

 

[mansr]

mansr, do you think that the measurements on ASR completely account for any SQ for a DAC?

It is probably possible to construct a DAC that exhibits some anomaly not easily seen in the usual set of measurements. Doing so while performing well in those measurements is probably rather tricky, though one could of course take a stellar device and add a DSP that detects music-like signals and makes it go bonkers only then. Let me know if Volkswagen starts selling DACs.

 

[majingotan]

Yes, the 75 Ω and 110 Ω specs apply to digital connections only. Analogue audio signals are low enough in frequency that impedance matching is irrelevant.

Would this imply that the impedance matching matters only to transducers/amp matching (IEMs in particular) as some of them have really low impedance (12.8 ohms for my IEM for example)? What about those that use separate passive preamp (ignore the tube mode measurements) with impedance changing as a function of attenuation? See this example here: https://www.audiosciencereview.com/...easurements-of-schiit-saga-tube-pre-amp.6520/ . If you use this passive preamp to an professional active monitor with only 600 ohm input, surely you will encounter damping issues in the low frequency, altering the sound from the original. Unless someone want to use digital attenuation on a DAC with fixed line output to avoid that impedance issue but one needs to be careful when implementing this because it could lead to max output straight to speakers if not being careful to check the levels first

 

[mansr]

Would this imply that the impedance matching matters only to transducers/amp matching (IEMs in particular) as some of them have really low impedance (12.8 ohms for my IEM for example)?

At audio frequencies, the source impedance should be much lower than that of the load. If they are equal and purely resistive, the effective signal level is halved. This is wasteful but otherwise harmless. If, as is often the case, either end has a capacitive or inductive component, a high source impedance results in uneven frequency response. When driving a speaker, a low output impedance additionally helps control mechanical oscillations (see damping factor).

If you use this passive preamp to an professional active monitor with only 600 ohm input, surely you will encounter damping issues in the low frequency, altering the sound from the original.

The active monitor input is likely somewhat capacitive, so a high source impedance will give some high-frequency roll-off. Damping issues would only arise if the built-in power amp has too high output impedance for the drivers, something the designers hopefully thought of.

 

[A800]

Listen to it.

 

[LTig]

If you use this passive preamp to an professional active monitor with only 600 ohm input, surely you will encounter damping issues in the low frequency, altering the sound from the original.

Do audio amplifiers (standalone or integrated into a speaker) with 600 ohm input impedance still exist? I thought impedance matching was used in telephone networks mainly, and I'm quite sure that in audio voltage matching is standard since at least 50 years. This means as long as the source impedance is 10 times lower than the load losses are small.

 

[sejarzo]

Do audio amplifiers (standalone or integrated into a speaker) with 600 ohm input impedance still exist?

I don't think so. A quick review of specs suggest that they are primarily in the range of power amps (as well they should be!) with typical 20k ohm balanced/10k ohm unbalanced input impedance, or thereabouts.