[主观翻译]A Matter Of Time——The Audibility Of Clock Jitter

得鱼买醉

交易区新帖推荐

A Matter Of Time

The Audibility Of Clock Jitter

首次在紫坛发帖，如有不妥还请指正包涵，谢谢大家

声音的回放过程相必大家都很清楚——经过音响器材间相互协调工作，环环相扣，声音最终得以呈现，期间虽会经过诸多环节，可如果我们换一种角度审视这一流程，不难发现整个过程就像一条流水线，在源源不断的“生产”着声音。音响系统如同一件精密仪器，而这台仪器可以广义的类比成一台DAC，正在以Digital数字信号为原料，锻造出精美的Analog模拟信号。当然，如果你是模拟发烧友，就不能比喻为生产线，信号搬运工相比之下会更为恰当。模拟信号搬运就谈不上Clock Jitter了，所以咱们这回主要聊聊前者。

高精密生产。其中“高精密”说的好听，其实都是比出来的，间或还是会产出一两件不合格品。这就是为什么有些产品，包括硬件产品以及软件算法，会产生属于“自己”的声音，因为有些“残次品”只有特定的产品才引进的来。在设计一款DAC时就会遇到这样的问题。

唯一能影响Clock Jitter产生的环节就在产品的时序电路。模拟电路是无法改变Jitter量级的。Clock Jitter经常以皮秒为单位。1皮秒等于一万亿分之一秒，为了直观的表明这一变化，以光举例，在1000皮秒内大约能传播30.5厘米，在1皮秒内，光也只能前进0.3048毫米。

从追求还原的角度看，DAC的工作必须准确无误。事实上超低jitter会在初听时会不符合心灵上的预期——少了韵味，一定是这样。在初听的时候，上了年纪的产品会更加“讨喜”。在短暂聆听之后，细听对比，刚刚的结论会有所变化。老电路的声音不是暖，而是浑浊，相比之下少了一些细节，结像也不理想。这就引出了大量问题，有些可以靠仪器解决有些则必须靠听音训练来解答。

• 阀值点。jitter低到什么程度便不会再带来对听感上的提高？

• 高于阀值点的Jitter听起来会是什么样子？

• jitter在不同频域的影响又会如何？

• 我们是否会被欺骗，认为一台低性能的dac的表现要优于高精度DAC？如果是这样，那么一套完整的听音训练将会是必须的。

• 科技性能的极限在哪里？音频硬件制作工艺的极限在哪里？

为此Dave Hill工程师设计了一套实验，它允许生成带jitter的音频文件，并且是可控、可重复的。这个实验的重点在于在保持时基对齐的情况下增减jitter，这样才能准确的控制变量，使得文件之间我们听到的区别仅仅在于jitter的区别。模拟信号经过Ad形成数字文件，在ad的过程中靠改变时钟信号来施加抖动。这个实验花费了大量时间和努力最终证明此实验是有效的。

时钟电路中的许多环节都会引起抖动。简化描述，跳过锁相环和其他电路不谈，只看时钟的源头——晶振。关于晶振的类型在网上有大量的介绍。最早在无线电微波通信领域人们便发现了时钟jitter对通信质量的影响，当年在无线电和计算电路的优秀工程师已经深入探讨过抖动的来源和影响，所以目前时钟jitter早已不是一个新的问题。时钟源材质基本上都是石英晶振，给石英施加交变电压便会产生规律的正弦波。少数顶级电路会采用铷钟，铷钟的精度可以保持10天不差10皮秒，这个精度足以使任何人听不到变化。

数字信号中的变化也就是抖动，如果在瞬时达到纳秒级别，将会明显略化听感。其实引起jitter最大的原因并不是来自于晶振本身，而是来自于周围电路的电噪声干扰，无论采用何种晶振它都会显著劣化时钟电路指标。噪声程度普遍伴随着频率的降低而加强，低频抖动是最难以抑制的。一个常见的例子便是电源噪声。

克服噪声后，接下来问题便是如何使得时钟电路生成系统可识别的时钟波形。这个问题的精髓在于，把缓慢的的正弦信号转化为高速，锐利的方波。一旦我们生成了方波，下一步会便是如何克服逻辑电路引入的jitter。逻辑电路的jitter，根据逻辑电路的类型，Jitter 可以是非常大的。如果你想实现亚皮秒级别的低抖动，至少你会面临逻辑电路可选性降低、高能耗电源以及高元件成本这三个问题。

如果Jitter足够低，它将不会被听到，但这涉及到jitter水平和频率之间的关系。

最后，如果说通过这次实验得出了什么结论，那便是我们都需要专门的听音训练。教育在人的一生中是不可或缺的。

Education will help everyone.

得鱼买醉

以下为原贴
A Matter Of Time
The Audibility Of Clock Jitter
By Dave Hill
In the course of developing gear, one sometimes does experiments that do not work as
planned. The resulting issues include why are things working the way they are and why
the device or algorithm sounds the way it does. In the process of developing a new
Digital to Analog converter(D/A), such questions arose.
The only changes being made at the time that questions were raised about were
in the clocking circuits; the analog circuits were not changed.The clock jitter decreased
from around 13 ps to about 2 ps.To put picoseconds in prospective, light travels
approximately 12 inches in 1000 ps; in 1pS of time, light travels 0.012 inches.
The measurements indicated that the D/A was working correctly. The first
listening test did not go as one expected. On the first listen, the old circuit sounded
better. After listening for a short time, with one’s ears learning what the sounds are, the
observation changed. The old circuit was not warmer; it was muddier, less clear and did
not image as well. This raised a large number of questions (some are technical and
some have to do with ear training):
• What is the threshold point where lower jitter no longer offers performance
improvement?
• What does jitter sound like, can mixes be influenced by a D/A with higher than
threshold jitter and how do different frequency ranges of jitter influence performance?
• Can we be fooled into thinking that a device which has poorer performance is
the better converter? If so, there is an ear education problem that needs to be
addressed.
• What are the limits of technology and the level of audible artifacts?
An experiment was designed that allowed making recordings with jitter that was
controlled and repeatable. An important part of the experiment was to be able to record
the files so they would, when time aligned and subtracted, result in an audio file in
which we can hear only the jittered component. The experiment produced a set of five
files (the reference and the files with induced jitter) that can be listened to with the set
of four difference-only files.
The test was set up with an Avid Pro Tools system using digital I/O. The D/A used
an ASRC to remove any jitter in the audio before feeding it to an A/D and back into Pro
Tools.
The A/D was built so that lab equipment could be used as the clock source and
this equipment could be frequency modulated to produce jitter. There were a great
many problems in making this work, and once the first audio files were recorded and
listened to, there was a new problem: If one did not hear what was expected, was the
experiment valid? It took a fair amount of time and effort to be able to state that the
experiment was valid.
Jitter can be caused by many things in clock circuits. To keep this relatively
simple, we are not looking at PLLs used in word clocks and elsewhere—just oscillator
circuits, the source of the clock. There are many types of oscillators and a large amount
of information on the net, in books and papers that a circuit designer can make use of
when developing a clock circuit. Some of the best sources of information come from
amateur radio work, and in microwave and space communication. Jitter in clocking is
not a new problem; technical information exists showing that since the earliest days of
radio and computing circuits, many very good engineers have worked to investigate the
sources and effects of jitter.
Looking at all of the data, you learn many things that are surprising. A pendulum
clock is more accurate than a quartz watch. It is possible to make a quartz crystal
oscillator that will outperform most other types of oscillators, including rubidium and
other atomic clocks, for short-term accuracy (and the problem with clocking for audio is
short-term accuracy). Rubidium is great if you need to be remain within 10 ps in 10
days—a change that no one could hear.
Digital audio clock variations (jitter) on the order of a few picoseconds or greater
in a short time window, like 10 ms or 100 ms, will be audible. The largest cause of jitter
is noise, which can affect any clock, no matter the type of clock base (crystal, atomic
and so on). Noise increases as frequency goes down, thus LF jitter is the hardest to
work with. A common example is power supply noise; in clock generation, power supply
noise is more critical than for mic-pres or tape-head playback electronics.
The next problem is turning the oscillator output to a clock waveform; all
approaches to this have issues. The basic problem is the need to turn a slow-moving
sine wave into a very fast, sharp, vertical-edged square wave. Once we have a square
wave clock, the next thing to overcome is jitter induced by logic circuits. Jitter in logic
circuits, depending on the type of logic, can be very large. If you are trying to achieve
sub-picosecond jitter, there are very few logic choices and the better parts are very
power hungry and costly.
If the jitter is low-enough level, it will not be audible, but it is a level versus
frequency relationship.
The five jittered audio files have been online since June. The phase-canceled
(subtractive) files are now available. Some were played at Audio Days in Paris and the
AES Convention in Rome. Listen to the source files to form an opinion first, then listen
to the phase-canceled files. Think of what you might feel compelled to do with an EQ to
change the sound of the jittered source. What is the result of playback on a very low
jitter playback system? The effect of jitter does not sound like what most people think. If
one can draw any conclusion from these experiments, it is that we all need ear training
when it comes to recognizing new problems; education will help everyone.

wqq

得鱼买醉 发表于 2016-5-8 10:03
以下为原贴
A Matter Of Time
The Audibility Of Clock Jitter

好帖

nightelf_zl

这篇文章说的light travels approximately 12 inches in 1000 ps;是指空气中的传播速度，而不是PCB板上，在相对介电常数4的FR4的PCB板上大约是6 inch/1000ps；Maxim公司介绍说最低要求是100ps。

darkfrank

把原网站贴出来吧
http://cranesong.com/jitter_1.html

这里有文章里提及的人为增加jitter的音频文件试听，可以很好的听出jitter对声音的影响

micsunnyfish

大赞！