Interesting tidbits about "jitter" in digital audio reproduction

It is perfectly possible to build a DAC without a internal clock. This is called NOS (Non oversampling)
It is driven by the rate of the incoming bit stream.
The moment you want oversampling you need a clock to generate this higher sampling rate.

Inherent to the DA is that higher-frequency multiples of the audible range are created, the so called aliases. In case of CD audio, the sampling rate is 44.1 kHz, the audible range is the half, so the first alias will start at 22.050 kHz.
To avoid having these aliases ruining your tweeter, in general a brick wall filter is applied. As it is very steep, all kind of artefacts are created like pre-ringing etc.
To solve this problem almost all DAC's today use oversampling.
Again aliases are generated but due to the oversampling (8 times) now the first one starts at 330 kHz
This allows for very smooth low pass filters or if you don't use one, it is probably outside the range of what your amp or tweeter can reproduce.

If you have a NOS but don't use a anti alias filter you won't hear the aliases as they are outside the audible range.
But you run the risk that you smell them one day as this high frequency content can ruin your tweeters.
http://www.thewelltemperedcomputer.com/KB/Sampling.htm

One of the things I have learned about my journey through audio is that just because there may be something that sounds better out there, that does not mean that I have necessarily sounds bad.

NOS does not mean without a clock; it means a standard Nyquist DAC instead of an oversampling (delta-sigma or some other flavor of PCM) DAC. I believe this is essentially what Vincent Kars said, yes? I agree with the word "internal" added; there is virtually always a clock from someplace. Not always, but without a clock time skew and jitter quickly becomes a killer if you have more than very low resolution and/or output frequency.

There are all sorts of advantages to oversampling, and some disadvantages (natch); a discussion is probably beyond the scope of this forum (though certainly not to many folk on it). You can oversample using a conventional DAC, avoiding the modulator loop, with arguable pros and cons.

Timely, sort of, as I just got pushed onto a high-rate (GHz+) delta-sigma converter proposal for a week or so, after working on a high-rate Nyquist ADC. - Don

Vincent, NOS DACs are used almost without exception in SET systems. Audio Note builds speakers with dome tweeters. There is no problem. My tweeters are gossamer light and are twenty feet long per speaker. They will tell all concerning a system. Oversampler players have failed on my system.

http://www.sakurasystems.com/articles/Kusunoki.html

Hmmm... The stuff I have from past presentations is for undergrad and grad EEs in college, not a general audio forum, so I'd have to work on it. My expertise is mostly Nyquist converters, not delta-sigma, though I have a working knowledge of the latter (as in, I have worked on them, not that they always worked ). I'm willing, but it will take a little time (I have put in 39 hours so far this week, with Monday off and one day to go, ouch!) Also, it wouldn't fit in a single post very well unless I really cut it down, and I love pictures... I could probably hit the high points. Let me think on it a bit. I may have some time this weekend. Are you willing to proofread for me? I think it would be helpful to have somebody with one foot in both worlds (engineering and consumer) review it first.

Also a little nervous after getting singed last time I tried to explain data converters to an audio group (you remember when, right? ) It ain't always pretty when marketing meets reality (cue Dilbert cartoon now). As bad as those first few CD players were, few realized how awful the early ADCs were! - Don

Thanks Amir. Let me see what I can put together, maybe a series of posts (or threads) to break it down into readable chunks.

Reminds me of doing an invited paper at a conference -- the guy who invited me set the topic (a new device technology) but failed to tell me the audience wasn't all that technical... They thought Fermi levels related to some sort of creature from the Goonies... (Not quite that bad, but closer than I expected!)

Thinking - Don

Got tied up today (morning job ended up taking all day, go figure). I did get some writing done, starting with a description of basic terms, like volts and amps and watts, and the difference between resistance and impedance, and got four pages in Word. Guess there's a reason it took all those years in school to become an EE. Maybe I should just write a book... Need to review it then perhaps I'll post it here, but I'm afraid it's way too long. After that I'll work on sampling theory and bits and stuff. - Don

Dancing with Jitter Bugs

As mentioned in earlier posts, a treasure-trove of jitter related papers all stems from the work of the late Julian Dunn at his colleagues in the UK.

Juergen Reis, chief engineer and designer at MBL, kindly pointed these out to me a few months ago, for which I am most grateful. While the papers are dry and technical, they do a good job of explaining the essentials. If you go through them carefully, make notes and diagram things for yourself, a great deal can be learned.

This post is an attempt to simplify and summarize a lot of details into some general observations...

One of the most important basic points to understand is that "digital" signals are really just very strange-looking analog waveforms. All the discussions about "How can there be errors in digital systems?" confuses the idealized concept (repeat, concept) of a binary digit with the reality of its electronic representation (notice I didn't say equivalent). The concept is logical, not physical: It's either a zero or one. That's great, and looks well on the chalkboard, but try making a perfect zero or one using real circuits.

Those burdened with advanced education in the field will recal lthat a square wave in fact comprises a large number of component sine and cosine waves, of various frequencies and phases, all of which sum up to something that's got sharp corners and flat tops. When you start passing this "package" through real circuitry, the fun begins...

Where people often get off track is on the subject of how can one hear 100 picosecond timing errors? Well, of course you can't, directly. Human hearing can detect differences in arrival time of clicks to milliseconds and microseconds, but not picoseconds. However, the consequences of such timing errors are readily audible in many cases, that that they distort ordinary audio signals in various obvious ways.

One of the most insidious realities of jitter, in all its lovely forms, is that the distortions are seldom related to the natural harmonic structure of music. You could infer this from some of the comments, but I thought it was a good idea to mention it explicitly. A-harmonic distortions are very hard to hide, so to speak.

Another common-sense reality of jitter is that it affects higher frequencies more than lower, thus contributing to screechy violins, edgy transients and all that.

Basically the amplitude of the jitter sidebands relates to the magnitude of the timing slop, and the frequency offset of the sidebands relates to the periodicity of the timing error. More or less

There can be jitter in USB, in S/PDIF, in IEEE 1394/FireWire, or basically any form of interface. There are jitter artifacts caused by the bit patterns of the digital data, according to Julian. And of course during A/D or D/A sampling conversion, once again, sample clock timing errors distort the waveform.

A particularly bad situation is when the sampling clock is derived from information coming across the interface. More advanced designs use phase-locked loop devices to essentially smooth out the inevitable jitter and decouple various subsystems in the A/D or D/A.

Jitter is often pretty obvious in lower-end USB DACs, to pick on one type of component Sure, they can handle 96 and 192 kHz signals, the audio gets progressively worse. High levels of jitter have a lot to do with such issues.

With higher-quality gear, where extensive measures are taken to control and isolate various types of jitter, the reverse occurs: the sound keeps getting better and better with higher sampling rates, as one would expect.

Hi Nicholas

Why does jitter affect higher frequencies more than lower frequencies?

Nicholas can answer better I am sure, but the short answer is that the impact of jitter is related to frequency, and as frequency goes up it takes less jitter to do the same damage as at lower frequencies. The faster the signal moves, the more sensitive it is to smaller deviations from things like jitter. Equations are coming...

I thought I would be a smart ass. how about these

http://www.maxim-ic.com/app-notes/index.mvp/id/3631

Don's essentialy right, higher frequencies of input signal are more sensitive to the effects of jitter distortion. And higher frequencies of jitter are also more potent than lower frequencies. So there's two things to consider here... and the technical paper referenced by Steve is excellent. The drawings help a lot. Dan Lavry also has some good whitepapers with illustrations... see http://lavryengineering.com/index_html.html

1. Regarding the input signal, without resorting to the horror of equations just think of it visually... as in the top or the bottom of a waveform. The slope is zero. If you slip it left or right slightly (earlier or later in time), so what? It's hard to notice. If you temporally "stagger" the part of the waveform where the slope is maximum, the high frequency part, there's an obvious artifact.

2. Regarding the jitter itself and its periodicity of the jitter, "slow" jitter creates artifacts that are very close to the center frequency, and are often masked. Higher-frequency jitter creates sidebands that are farther and farther away from the center frequency, and thus show up much more readily.

These are general observations; there are myriad details...

Nice job, Nicholas. Pictures, I have to figure out how to embed a few pictures. Explaining this stuff without them is hard, and harder to follow.

The Maxim paper is a very good overview of random jitter. Unfortunately for listeners and designers, deterministic (signal-related) jitter is harder to deal with and causes the most problems (for all of us). I wish the paper had used RJ_total or something; the total random jitter is not what we think of as the actual total jitter.

IMO! - Don

Useful papers on jitter and dithering...

Here they are... some of the concepts and tricks involved in measuring and reducing dither are also related to managing quantization noise. The IEEE paper was recommended by Pflash Pflaumer of Berkeley Audio Design.

Downloaded, saved for later reading, thanks Nicholas!!! - Don

This material is good stuff and well worth the effort of slogging through the explanations. A good part of these papers is devoted to how to measure jitter in the first place,which isn't easy. You can see the entire topic is fraught with complexity. Trumpeting of "low jitter" by vendors often means that they really don't know how to operate the test gear very well.

These papers basically illustrate why digital has been so difficult to get "correctly": The nature of the recording and playback process is completely different from analogue. Now, I'm sure that our colleagues who are expert in the analogue domain have their own set of gripes with the process, but it does seem to me that digital is even worse

Gentlemen, thank you. While much of this is over my head, it's good to hear that jitter manifests itself most noticeably in high frequencies, and to read the descriptions of the results. I think I may have actually heard it once or twice. I had a handful of DACs and headphone amps in once for listening, and while harshness in the high frequencies might be hard to distinguish sometimes from passive crossover distortions, it should reveal itself well in headphones. One piece of kit, a tube headphone amp with a usb DAC, sounded awful, and awful in the opposite way one would expect tubes to sound -- thin, brittle, harsh. Bi-passing its DAC completely changed the character of the thing. An amazing transformation. Do you suppose that was jitter I heard? I'd like to know it in case we meet again.

None of the other DACs I had in-house then, or since, have had this problem. In fact they have been pretty hard to tell apart, when playing 16/44.1 files.

P

The problem is you eliminated 2 distinct components, the DAC and the USB receiver.
Hard to tell "Who dunnit".

I'm inclined to suspect the USB receiver (adaptive mode probably) and/or the direct electrical connection between the PC and the DAC over the power and/or data lines of the USB.
Nice bit of guesswork anyway.

It would be interesting to try a modern USB implementation (asynchronous) preferably combined with galvanic isolation using the same source.