Sunday, 3 July 2011

Low Jitter Clocks

Over the last few months I have been developing, building and testing a low noise clock.  I've mentioned it in previous posts, and now it is finally finished.  Before I go into the details of the clock, here's some information on jitter.

Jitter and its effects

When digital audio is played back its speed is controlled by a clock.  In a perfect world the period of each clock cycle would be the same, but in the real world they always vary.  This variance from an ideal clock is called jitter.  The variance causes distortion in the reproduced waveform, as shown below:

The example above isn't particularly dramatic, but it does show how clock errors translate into distortion.  Note that jitter is differences between the length of clock periods, and is quite separate to errors in overall clock frequency.

Sources of jitter

Jitter has two sources:
  • A non-ideal clock source.  A CD player, or any other digital audio device, plays back audio according to a master clock.  All real work clocks are non-ideal, but some are closer to ideal than others.  The best way to deal with this source of jitter is to use a clock with as low jitter as possible.
  •  Downstream manipulation.  The clock signal is manipulated many times over after leaving the master clock.  It will be divided and buffered many times over as it goes between the different ICs of the digital source.  S/P DIF transmission can add a particularly large amount of jitter due to its complexity.  The best way to avoid this source of jitter is good board layout and a simple clock chain.  Unfortunately, unlike changing a master clock, altering the board layout and clock chain is a very difficult task.
Jitter can also be found as an intrinsic part of audio recoding, this is caused by the same issues, but during the analog to digital conversion.  I won't go any further into this source as nothing can be done about it in a player.

Jitter in master clocks

 The jitter produced by different master clocks varies immensely.  It's hard to tell how much jitter a given clock has, as many won't list any specifications, and those that do will often list specifications that are irrelevant or even misleading.  The list below details most varieties of clock and how they perform (in order of highest to lowest jitter):
  • A gate oscillator integrated into one of the CD player's ICs, usually the decoder or digital filter.  Gate oscillators aren't very high performance to begin with, and integrating them into another IC, forcing it to share a noisy supply with that IC, really doesn't help.  These are the most common form of master clock found in CD players.  An example is pictured below:
  •  A gate oscillator using a separate IC, usually a 74HC04 or some variety of that IC.  These are slightly higher performance than the former due to separation of power supplies.  An example is pictured below:
  •  A single package crystal oscillator, abbreviated to XO (or TCXO and OCXO with added features).  These contain a crystal and an oscillator circuit in a single package.  These are becoming more popular in modern high end CD players.  An example is pictured below:
  •  A discrete crystal oscillator circuit.  These can be higher performance that the other types, which all have limitations like size and available voltages.  There is quite a range of circuits in this category.  I couldn't find an example CD player schematic where this class of master clock is used, but I know of a few that do.
This list is just a broad overview.  The best in some classes will be better than the worst in the next class.  A Tent Clock (a single package crystal oscillator) will probably be better than many discrete clocks.  The power supply a clock uses is also very important.  Separate regulation, or even better, a separate transformer winding or transformer will do a clock a world of good.

Chosing a master clock

Master clocks are very hard to chose.  Jitter is specified in a variety of ways, the best of all being a phase noise plot.  The most important aspect for audio is the noise level near the fundamental frequency (within 100Hz).  Many clocks specify noise at 1kHz or 10kHz, this measurement isn't very useful.  Even less useful is accuracy specifications, which don't relate to jitter at all.  Be wary of clocks whose sole specification is something like "Ultra low jitter ±1PPM", which refers to frequency accuracy, which just isn't important for audio.

Since most clock won't provide jitter specifications (even the good ones), I'd recommend choosing a good, well known discrete clock.  These are made by companies such as LC Audio, New Class D, Hagerman and Sercal.  Tentlabs and Burson also make XO based clocks which are worth looking at.  Prices range widely, starting at about USD120 and going up to USD600 or more.

A cheaper alternative is a DIY clock, that's the option I chose.  There are schematics such as Elso Kwak's 'Kwak Klock' that are fairly simple and easily followed.

My own clock, the LJC

Finding the prices for ready made master clocks a bit too high, and not really liking most available DIY circuits, I designed my own.  After a lot of research I decided to use a differential Colpitt's oscillator.  This type of discrete clock has low jitter and good power supply rejection, and is popular in the telecommunications industry.

This is the second clock I designed.  The first was overly large, and though it performed well it was too bulky to fit into most CD player.  This time I made size a focus.  The clock has two versions, a low voltage DC powered one (pictured at the front) which is 32 x 75mm and a mains voltage AC powered one (pictured at the back) which is 32 x 100mm.

The clock has performed well in testing, and provides a very stable output with a good waveshape.  I've tested the clock at most CD player frequencies; 4.2338MHz, 11.2896MHz, 16.9344MHz and 33.8688MHz.  The only frequency I've yet to test is the 1024fs, 45.1584MHz that some newer Sony players use.

I'm about to fit the clock to a Marantz CD72, I'll post listening results along with other modifications to that player soon.