AudioQuest White Paper: Evaluation of Digital Devices and Proper Warm-Up for Ideal Listening and Measurements

A market that at the end of the last century seemed all but extinct is once again at the conversational forefront. Thanks in part to audiophiles’ rapid acceptance of digital music, and in part to the average consumer’s increased interest in desktop and computer-based audio, the digital-to-analog converter is once again a significant part of the audio ecology. DACs are definitely back. Along with this welcome return comes the likelihood (among consumers) to purchase, (among dealers) to evaluate, and (among journalists) to review just about every DAC that comes down the pike. As we reestablish our interest in this unique component category, we must also reestablish certain basics of auditioning. One of the most critical aspects in the auditioning of any digital device is proper pre-evaluation warm-up.

The purpose of this paper is to call attention to the changes in audible and measurable performance that occurs to any DAC over a 24-hour period. Listening to a cold digital device (any DAC that has been unplugged for one hour or more) will significantly shortchange the experience and negatively impact the desired outcome. However, a 24-hour warm-up period can – and most often will – significantly improve DAC performance. Moreover, comparisons between a suitably warm DAC and one that is cold can result in invalid findings.

Phase Noise
In an ideal world, a perfect digital oscillator would generate a pure square wave at its intended frequency, and the signal would have power at only that single frequency. However, in reality, all real-world oscillators are imperfect and have phase-modulated noise which spreads the power into and across various sidebands. This noise can spread across the entire frequency spectrum; it is a significant cause of unwanted jitter. Here’s a simple formula, illustrating how phase noise correlates with jitter: Jitter = Phase error (degrees) / 360 x Frequency (hertz) (Jitter is a numerical output represented in seconds or, in our case, picoseconds (pS), or xxx E-12.)

Interestingly, this phase noise is not necessarily constant; phase noise may constitute differing levels of distortion at various frequencies. But the real message of this paper is that phase noise tends to be significantly higher when a DAC is cold; a 24-hour warm-up period significantly reduces phase noise. Additionally, you will see in the following measurements that there can be a period of settling, during which measurable phase noise oscillates up and down before it finally settles to its lowest and most stable level.

click for larger image

For the purposes of providing valid measurements, a DragonFly DAC was used on a Symetricon Phase Noise Test Device. The Symetricon is one of, if not the most, accurate pieces of test equipment. Tests were run at the following increments:

  • Cold/“Zero Minute” (as soon as the DAC was plugged in)
  • 15 minutes after plug-in
  • 1 hour after plug-in
  • 24 hours after plug-in
Cold — At startup, we see at 1Hz, a dBc (decibels relative to the carrier signal) of -73. However, notice the spike from 6Hz to 11Hz. There is a significant rise in phase noise within this range. Also notice how ragged the response is from 1Hz to 100Hz. This volatility will have effects on the DAC chip’s ability to recreate an accurate waveform. Jitter is measured at 3.7pSeconds.

15 Minutes — Here we begin to see a slight drop in noise at 1Hz (-.8dBc), as well as a reduction of the cold-related spike (6Hz to 11Hz). Things appear as though they’re about to begin tracking nicely. Jitter is measured at 1.4pSeconds.

1 Hour — We begin to see the settling process occur. The measured performance takes a significant turn for the worse. At 1Hz, phase noise has risen by 11dBc. However, the good news is that the initial spike from 6 to 11Hz has not returned. But overall, we are seeing a much noisier signal. Additionally, notice that the response is more ragged than it was when tested in the 15-minute range. This will have a very negative impact on the DAC’s ability to output an accurate (and pleasant) musical signal. Jitter has risen significantly to 5.7pSeconds.

24 Hours — At this point and beyond, the DAC appears to have reached a stasis. Notice that the response is now very stable, that there is a smooth line from 1Hz all the way out to 1kHz. At its greatest point, the delta in phase noise has improved by 17.2dBc over the initial cold test. We can now begin to critically listen to this DAC, confident that our findings will be accurate and illuminating. Jitter is dropped and settled at an astonishingly low level of 0.85pSeconds.

Audio-based jitter is most accurately measured from 1Hz to 100Hz range. This is where averaging cannot correct or hide errors as easily as it does at say 1KHz. Ironically, most published jitter measurements are given at this frequency range.

How does a high level of phase noise affect our music? While it’s hard to draw a simple conclusion that will apply to all DACs, it’s absolutely clear that higher levels of phase noise create higher levels of jitter. And higher levels of jitter correlate to poor sound; the more jitter, the worse the sound.

If you’re auditioning a new DAC – whether for personal purchase, store merchandise, or critical review – be patient and give that contestant a day to warm up, so that it can truly strut its stuff.

COMMENTS
junker's picture

Nice to have some jitter data vs. warm-up time. I'd also love to see jitter data vs. length to see one should go as short as possible, or longer than 1m to avoid reflections. I've never seen anyone quantify USB cable performance vs. length.

Vincent Kars's picture

Looking at the graph the differences are below 20 Hz only, well below our threshold of hearing.
Makes me wonder if this is audible at all.

Archimago's picture

Indeed. What's the problem?

So warming up the Dragonfly DAC (not exactly the most accurate DAC of course) made a little difference in noise and jitter. Regarding noise level, we're looking at maybe at most 5dB difference between runs down at -105dB (20Hz) and by the time we hit 50Hz, there's barely any difference down at -125dB or so! Seriously, does anyone have ears with an acuity anywhere close to this resolution, or recorded music for that matter where this would make an iota of difference?

As for jitter, it's literally "scraping the bottom of the barrel" to post up a chart of picosecond jitter at *1Hz*.

If anything, this data suggests we should be relieved that there's nothing to worry about! Especially since they're using a Dragonfly DAC to get this level of excellent performance!

CG's picture

I believe you might find some insight here: http://www.davidgriesinger.com

David Griesinger is a highly credentialed and regarded researcher in the field.

Keep in mind that phase noise is attached to every single tone output by a DAC as sidebands to that tone. That's just how the conversion process works. So, although you might not be able to hear 1 Hz energy by itself, if it's modulating the spectral output of, say, a trumpet, you sure as heck can hear it. It's really a question of what level phase noise is audible (which partly depends on the audio content) and what its spectrum is (also content dependent).

Archimago's picture

And where on that site is there talk about phase noise of the magnitude being graphed here?

Why in the world is Audioquest not demonstrating these sidebands at a reasonable frequency that is audible and showing the difference in sidebands at the different times?

CG's picture

If you are asking where on Dr, Greisinger's site you can find information related to this, you'll have to read through the papers yourself. I did 99% of the leg work by recalling his web page URL.

If you are asking why Audioquest did not include that information, well, you'd have to ask them.

I don't want to cone across as snarky or impolite - that's not my intention at all - but aren't you the person who authors authoritative articles about audio reproduction on your own web site? Shouldn't I, a bumpkin amateur, be asking you these questions?

My reply was really directed toward Vincent Kars, but of course I hoped it might be helpful to others. Sorry if you felt antagonized.

fmak's picture

It is clear that you do not understand the effects of phase noise on sound quality. This is also clear from your 'mesurements' on your website.

5dB is nearly half I voltage terms and is significant.

Archimago's picture

5dB difference at -100dB is exactly how many volts? Why don't you simulate this level of phase noise into some music for us to have a listen?

As usual. Long on hyperbole, brief on demonstrable tangibles.

fmak's picture

It is not SPL or SN or THD.

Read up and stop complaining in ignorance

Archimago's picture

Yes. I know that. It's phase noise. Very significant in MHz and GHz wireless designs. Engineers I know at Intel scratch their heads when I mention 20Hz - 20kHz.

Look... There's no point arguing about this unless someone can demonstrate that this is a problem that's audible. I have no problems with the idea that noise and jitter fluctuates with warm-up time for a DAC. You say it's significant, and I have doubts that what is shown on the graph and in that table has audible relevance. It's a "white paper" designed to convince businesses/consumers of something the company wants to convey - not exactly held to high academic standards.

I'm certainly happy to change my stance if there is a way to demonstrate that this is significant. Maybe someone can create an audio sample with these levels of simulated jitter/side bands in place for us to have a listen.

Archimago's picture

Actually...

Should have said 44kHz - 192kHz above to correspond to the sampling rates. Still, quite low frequency in the context of technology these days. Then we also have to take account of the frequency limitations of human hearing on top of this of course.

Michael Lavorgna's picture
When we talk about listening, you want measurements and/or ABX tests. When we talk about measurements, you want to listen. It seems to me you have all of your bases covered so that you never have to do anything (or learn anything new ;-).
Wavelength's picture

All,

Sorry for the delay in getting to this.

First the results of this paper are not limited to the DragonFly. I thought of this when I saw a reviewer writing about the differential of one dac to another. If the first dac was plugged in for days then the other dac was plugged in for minutes what kind of effect this would have on the differential of sound between them.

The oscillators we use are very specific in nature. The reason we look at the jitter below 10Hz is because above that the error becomes averaged. Most modern day oscilloscopes can do jitter measurements at like 1K, 100K, 1M whatever. Well this really doesn't help us in audio. Jitter/Phase Noise errors created by these oscillators only happens below 10Hz. Above that these errors get averaged out and don't play any significance to the audio output. Errors below 10Hz will cause problems in the audio output over the entire range.

This also has nothing to do with voltage so just because it is dB, doesn't mean you can apply this to a voltage or current. It is a reference like signal to noise ratio. The higher it is the better the sound your going to have.

These oscillators also have nothing to do with USB and cables. We typically put these audio oscillators as close to the DAC chips as possible. Like a couple mm away and then buffer these back to the USB controller which creates the audio feed to the DAC chip via I2S.

Thanks,
Gordon

Wavelength's picture

All,

By saying higher the number I guess needs clarification. These are all negative numbers so by that I mean -100dB is better than -80dB and so forth. Which I guess is the opposite of what I said. So really the lower the number the better.

Thanks,
Gordon

junker's picture

I do agree that proper component warm-up is a valid testing methodology consideration.

But, why is >1KHz not meaningful to audio? While close-in phase noise is interesting why omit the regime from your graph? The noise floor of the oscillator is still interesting data to have. And if you are truly concerned with using a moving average, then just reduce the averaging with you FFT. Surely, this is selectable with the most advanced test equipment in world. It doesn't seem that you have enough averaging to suppress the triplet @ 600-850Hz.

Wavelength's picture

Junker,

All oscillator companies will present Jitter & Phase Noise at the standard rate of 1KHZ. Remember they are not designing specifically for audio. Most of the stuff is computer oriented or some RF usage.

Audio related Jitter/Phase Noise errors only occur in the really low frequencies. This is why the specialized oscillators for audio are tested down there and not at 1KHz. If you look at the CCHD-957 from Crystek the data show 10Hz and above for this very reason. Sure they also show in the data sheet at 100Khz and 10Mhz. These are there to show companies outside of Audio that they the capabilities to make custom frequencies for say military or whatever.

Thanks,
Gordon

Archimago's picture

Another thread looking at phase noise and discussions of audibility. I find Phaeton's comments particularly helpful in terms of the context:

http://www.computeraudiophile.com/f6-dac-digital-analog-conversion/lowes...

CG's picture

Perhaps the most interesting aspect of this is that it's a direct measurement of the effects of burn-in, warm-up or whatever you choose to call it.

I just did a quick web search on burn-in for electronic equipment and found that most hits were to the effect of:

"crap!"

"Imaginary"

"Impossible"

"All in your head"

"%#{+^!!"

"Show me some concrete proof!"

Well, here is some concrete proof of change over time. Where are all the guys who spat at the "idiots" who made the burn-in observation in the first place? Off doing a new edition of The Boy Scout Manual?

gefski's picture

...if we're using our dacs daily, we should leave them on 24/7? Guess it won't hurt; lots of people do.

highstream's picture

Useful info, though harder to compare dacs if listening is limited to <1 hr each.

Unless I'm misreading the graph and table, the reason for the conclusion of 24 hrs, as opposed to somewhere between 1 and 24, is not apparent. Was there something more than convenience involved?