Mutec REF 10 Masterclock

[JohnSwenson]

Hi All,

I'm sorry I haven't responded in this thread earlier, but I have been very busy (mostly with moving stuff), I have some general comments on what has been brought up here.

 

But first I want to thank Julian for his posts, he has done an awesome job trying to explain some of the complexities and confusion on this subject.

 

First a little more on the subject of clock types, word, master, reference etc.

 

Word clocks are low frequency, they are the sample rate, 44.1KHz 96 KHz etc. Thus a 10MHz clock is not a word clock.

 

The term "master clock" can mean several different things which leads to a lot of confusion. I recommend not using the term at all to avoid the confusion.

 

As discussed in this thread a high frequency clock can have at least three different "uses".

1) integer multiple of a word clock, DACs frequently have two of these, one for each of the sample rate families.

2) clock used for "processing" not related directly to sample frequencies

3) reference for a frequency synthesizer (which may be producing #2 frequencies)

 

Very few DACs actually use word clocks any more. Most of them use some form of #2 clock.

 

The Ref10 is designed to be a #3 source. For most audio equipment it cannot be used directly since not very many boxes have frequency synthesizers built in.

 

#2 are clocks that are fed into things like processors, USB receivers, PCIe boards etc. They are not related to audio sample rate frequencies at all, but observations seem to indicated that they can effect ultimate sound quality. The mechanisms for this are not known. I do have a theory I have been working on that explains this but it is very preliminary at this point and I'm not willing to make it public yet.

 

Clock synthesizers come in in many forms, some good, some bad, some cheap, expensive etc. I'm not going to go into all these types now. Until fairly recently they have not been particularly good and a decent crystal oscillator at the exact needed frequency would outdo a synthesizer.

 

But that has changed. Modern frequency synthesizers can be quite good. Thus IF your system is using one of these very good synthesizers, a device like the Ref10 can be very effective, just not universally effective, it depends very much on the system, what synthesizer it is feeding and what the out of THAT is feeding. The MC3+ series and the SOTM board seem to be good synthesizers so they will probably benefit from a Ref10.

 

The the MC3+ and SOTM do different things so it is important to understand the differences.

 

The SOTM board uses an off the shelf synthesizer chip with 4 outputs, each output can be programmed to be a different frequency, but they need to programmed at the factory. It has its own reference oscillator for the synthesizer, but it also has an external port. The board has to be programmed at the factory for the frequency of the external reference.

 

The MC3+ is designed to work with external audio sources. Julian can correct me if I get this wrong, but I think I understand its functioning. There is one frequency synthesizer, but it can have its frequency adjusted on the fly in large or very tiny amounts.

 

Both USB and S/PDIF inputs have their own clocks which run their respective chips. The outputs from these fill a FIFO in the FPGA, clocked by their own clocks. The samples are read from the FIFO using the output of the synthesizer.  The FPGA reads information from the receiver and sets the synthesizer to the correct frequency for the sample rate being received. The FPGA keeps track of the FIFO, if it is getting too full or too empty it changes the output frequency of the synthesizer by a tiny amount to keep the FIFO around half full. This means that the overall average sample rate is controlled by the input source, the quality of the clock generating the output stream is determined by the reference to the synthesizer. Thus a better reference (such as the Ref10) will lower the phase noise of the output data.

 

But what about asynch USB, isn't the DAC in control? Overall yes, the DAC has its OWN FIFO and also checks it, but instead of changing a clock frequency it sends a command back to the computer which tells it to speed up or slow down the average sample rate. So even though the local DAC clock is in ultimate control of the sample rate, as far as the MC3+/USB is concerned the USB data stream is in control, it just passes it on down to the DAC.

 

Now on to cable length. For a properly implemented system the only thing that should matter is the attenuation of the signal with increase in length. Note the "properly" in there, it is very important. Two things primarily affect this: attenuation at different frequencies, and reflections caused by impedance mismatch.

 

Because the Ref10 has a square wave output, the frequency characteristics of the cable are very important. Most coax cables have quite low attenuation at 10MHz, BUT a square wave consists of LOTS of higher frequency components, most coax has MUCH higher attenuation at those higher frequencies. The result is that even for fairly short lengths the high frequency attenuation will cause the rise time of the signal to decrease significantly, which means an effective increase in jitter at the receiver. This can be mitigated somewhat by using very high frequency low loss cable.  (which is much more expensive, but might be worth it if you need a longer distance)

 

The affects of impedance mismatch are much more complex. The mismatch causes a "reflection" on the line which causes a change in amplitude on the signal which sits on top of the original signal. The length of the cable determines where that change is going to occur relative to the original signal. Some cable lengths will result in a reflection that can cause a problem for the receiver, but the same reflection may not cause a problem for a different receiver circuit.

 

If the source board design and layout, cable, receiver board design and layout are all at the same impedance this will not be a problem. BUT the practice of unsoldering an oscillator from a board and soldering some coax to the pads is almost going to guarantee that you have a massive impedance mismatch, meaning you WILL have cable length issues, but it is almost impossible to figure out in advance what they might be. Again it may cause a problem, it may not.

 

I guess that is enough for now.

 

John S.

 

[JohnSwenson]

2 hours ago, rickca said:

Terrific post, very helpful.

 

John, when you have some more time ... SOtM claims there's an advantage to having the 4 clocks on an sCLK-EX board share an internal reference oscillator.  IOW these clocks are in some sense synchronized even though the 4 outputs can be programmed to different frequencies.

 

Does this make sense to you?

 

@romaz please advise if I have misinterpreted what SOtM has said about this.

Yes they definitely are synchronized, there is one PLL that syncs to the reference clock then four fractional/N dividers, since the dividers all run off the same PLL they definitely are all synchronized, whether this is a good thing is debatable. It is good in the sense that you will not have any low frequency beat frequencies between clock outputs due to outputs wandering relative to each other, but noise from each output will tend to directly add since they are all in the same point in the phase noise from the PLL. Independent sources do not directly add since they are not at the same point in the phase noise. I don't really want to go into more detail on this, it is a 20 page dissertation and I really don't have the time now to do that.  

 

Neither of these things is probably that important.

 

John S.

[JohnSwenson]

9 hours ago, Confused said:

I was a little confused when I first read this.  in point 1, 'integer multiple of a word clock' you state 'DACs frequently have two of these', then you state 'very few DACs actually use word clocks any more'.  Am I correct in interpreting this as that point 1 is more generally referring to older DAC designs, with more modern DAC designs being typically per point 2?

In this point I'm talking about what is internal to the DAC box, not an external connection. Most modern DAC chip circuitry don't actually have a word clock. Remember a word clock is a clock at the actual sample rate. Instead they use a clock that is some multiple of the sample rate, 22.5792 MHz and 24.576 MHz are quite popular these days. These clocks are integer multiples of all the standard audio sample rates. If you take a scope and probe around the traces in most DACs today you will find these, but not the actual sample rates (44.1KHz 96KHz etc)

 

Older DACs (and a few modern ones) DO you use real word clocks. The problem today with an external word clock signal is that is that since the DAC doesn't use it, the word clock has to go through a complex circuit to generate the clock frequencies above which are actually used.

 

#1 and #2 are used for completely different things. Look carefully at what I wrote, ALL the numbered points are high frequency clocks, none of them refer to word clocks, I covered that in the paragraph before. #1 are the high frequency clocks used by the DAC chips etc, #2 is used for things like the CPU on the motherboard, USB receiver chips etc, #3 is a reference used by a clock synthesizer to generate #1 or #2 clocks.

 

John S.

[JohnSwenson]

9 hours ago, Confused said:

 

OK - This is the bit that really interests me!  I am in the position that adding any kind of external clock or clock reference to my DAC is not going to be a practical prospect.  In my case I am running dual mono Devialet amplifiers, so modifying these to accept some kind of external clock would require some serious modification work and a lot of specific expertise.  Maybe it could be done, and I would be fascinated if anyone has any insight into this, but for now I am considering this as not a practical option. 

 

So, this effectively leaves me, and many others, as someone with a DAC that will not accept a word clock or external clock reference.  However, my understanding is that if I feed the DAC by it's preferred AES input, the clock signal is imbedded the AES feed and the DAC has the job of 'recovering' this clock.  It is this interface between the clock in the feed to the DAC and the DAC itself that interests me.  I am presuming that it should generally be a case of the better and more accurate the clock in the feed is, the better the DAC will ultimately perform.  Pure speculation now, but I am assuming this could be very DAC / system dependant?  From my own subjective experience I believe this to be true with the Devialet DAC / ADH.  As an example I have listened to a Devialet D800 with a dCS Vivaldi Upsampler, both with and without the optional dCS 'Master Clock', I thought it sounded better with the clock.  Plus, I run my own Devialet's with a Mutec MC3+USB via AES.  This sounds a lot better than anything I have fed the Devialet via USB.  So subjectively I have some experience this stuff works, and I can see there is some hard science behind this.  What I do not know is how big an improvement might be achieved by adding the REF10 to the Mutec.  This will clearly improve the clock accuracy in the feed to the DAC, but does this definitively result in an improvement in the DAC performance?

The circuitry used to generate clocking in a DAC from a S/PDIF (AES3 etc) feed vary radically from DAC to DAC. There are many different ways to do it and they all have very different "clock transfer functions", how the quality of the input signal affects the clock feeding the DAC chip. I have no idea how the Devialet does it so I can't come up with even a speculation as to what that transfer function might be.

 

The only thing I can say is that the phase noise of the signal coming out of the source usually DOES have an impact on the phase noise of the clock fed to the DAC chip, exactly what the mechanism is and by how much will vary dramatically from DAC to DAC. There is no way to make generalizations here.

 

One note, you use the term "accuracy" several times, to me the accuracy of a clock refers to how close the actual frequency of the clock compares to what it SHOULD be. Generally in digital audio this has very little to do with anything. A little bit off is not going to make any audible difference. The phase noise of the clock (another way to talk about jitter) is what matters, not the accuracy.

 

John S.

[JohnSwenson]

1 hour ago, austinpop said:

 

Thanks once again for your insight, John. 

 

This was an aspect of asynchronous USB that I had not fully understood. Let's consider the scenario:

• sMS-200 (the USB source) > ISO-Regen (USB reclocker) > DAC

To repeat back what I understood from your writeup:

• The USB source still generates the data clock, and embeds it in the data stream
• The reclocker reads in and regenerates the data stream with its own (better) clock
• The DAC is the "master" of this communication, not in terms of being the clock generator, but rather in terms of controlling the average sample rate. As far as reading the data itself, the DAC still needs to lock on to the embedded clock in the data stream.

Perhaps this explains why even asynchronous USB DACs are so profoundly affected by the quality of the clocks used by the USB source and any regenrator(s) in the path?

 

Question for you:

• When the DAC, as master controller, sends a command back - to "slow down" the ave. sample rate, say - what are the magnitudes we are talking about, in % terms of the nominal sample rate? 1%, 5%, 10%?
• When this command is sent back, presumably both the USB source and the reclocker register the command, and honor the new rate? Or does that only fall to the USB source, with the reclocker merely passing through what it receives?

I apologize if this is going off-topic for the Ref 10, in which case we can take this elsewhere.

In USB data comes in packets, for digital audio using high speed mode there is a LOT of idle time in between the packets, this means that the actual rate at which bits are transmitted is actually irrelevant. For any given system the bit rate stays constant, the packet rate stays constant. The data rate is determined by how many samples are put in each packet. The source can vary this from packet to packet to achieve a very fine resolution in the average sample rate. This is important to understand, at no point is any clock frequency actually changed, it is all done by number of samples per packet.

 

In the actual transmitted signal each packet starts out with a number of "sync bits", this is a square wave at the bit frequency, the receiver uses this to lock its internal PLL to the bit rate in that packet. This is done for every packet, so even if there is some drift in clocks in either transmitter or receiver the receiver can lock onto the rate the data is coming over the wire.

 

None of these clocks are used to generate the "audio clock". The above clocks are used to put data into a FIFO, the local audio clock (a type #1 in my list) is used to grab the data from the FIFO.

 

Exactly how phase noise from the source gets into the audio clock is the subject of my current research which I am not ready to discuss yet so that is as far as I am going to go right now.

 

As far as how big is the change when the DAC tells the computer to change rates here are some actual numbers I have seen. These are from linux systems which usually let you know what the actual sample rate is. For example if the computer is sending a 44.1KHz signal, if no feedback has occurred it will think the actual rate is 44100. But if the computer's clock is a little off from the clock in the DAC, the DAC tells the computer to change it. I have frequently seen this go to 441001, or 44099, up to 441004 and 44096. 

 

John S.

[JohnSwenson]

13 hours ago, julian.david said:

Between the 50 and 75 ohms outputs, one is not per se better than the other. Use whichever one matches the impedance of the receiving device (DAC, re-clocker, etc.) and make sure you use the same impedance cable in between!

There is a bit of a gotcha with 75 ohm BNC cables. MANY of the cables available "out there" use 75 ohm cable but use 50 ohm BNC plugs!

 

Several years ago I ordered 12 such cables from different companies and only 4 of them had actual 75 ohm plugs. They all had 75 ohm cable, but only 4 got the right connector.

 

Several of these were well known cable assemblers and they even got it wrong.

 

John S.

[JohnSwenson]

As noted above Pasternack is the place to go for custom RF cables. I recently had some 2 ft 50 ohm cables made using LMR-200-UF (thats the UltraFlex version) I wouldn't call it limp but it is quite a bit more flexible than other cables with similar specs.

 

They cost $50 each for the custom cables. You can get them in any length you want and over 200 different coax types and MANY different connector combinations.

 

I had to call them up to get what I wanted, for some reason the web page wouldn't get me what I wanted. Their people on the phone are very knowledgeable and got me what I wanted right away.

 

If you want good RF cables, just go to Pasternack.

 

John S.

[JohnSwenson]

1 hour ago, barrows said:

Remember, relatively thin aluminum (under an inch thick) has virtually no magnetic shielding ability, so the box itself does very little. 

This actually is not true, I did a research project on this many years ago and found out that thick aluminum (greater than 1/4 inch) actually IS a fairly decent magnetic shield. AC magnetic fields generate eddy currents in the aluminum, which produce their own magnetic fields that oppose the original field. How effective this is depends on the frequency of the original field and the resistance of the metal. For a 60Hz field and 1/4 inch aluminum you get about 6db attenuation, at 1/2 inch it is 12db, and at 1 inch you get about 24db. I would say that is better than virtually nothing, its not huge, but it can make a significant audible effect.

 

The normal much thinner aluminum plates used for electronic enclosures, are pretty useless for magnetic shielding. For some reason it is common to use the thick aluminum for the front panel, but the top and bottom have very thin plates. If the devices used 1/4 inch for top and bottom plates you would have 12 db attenuation when stacking them. But VERY few companies do that.

 

John S.

[JohnSwenson]

I'm NOT from Mutec, but I have a little bit of knowledge about crystal oscillators so I hopefully can offer some insight as to what aging is.

 

First off we need to understand that there is not just one aspect about crystal oscillators that have numbers, people here tend to like to latch onto numbers as figures of merit, but this can be fraught with danger since there are at least there different aspect of crystal oscillators that have numbers, before you start comparing numbers you ABSOLUTELY HAVE to understand which aspects those numbers refer to otherwise you are comparing surface tension to the color of the peel of an orange.

 

There are two primary aspects of a crystal oscillator:

 

1) phase noise (I have written exhaustively about this early in this thread so I will not duplicate it all here) This is not a single number, it is a graph.  This graph is the phase noise as an off set from the "carrier", which is the frequency of the signal coming from the oscillator. In a nutshell no oscillator produces a perfectly "pure" frequency. They all vary a little bit over time. Phase noise looks at the rapidly varying frequency changes. It is plotted in regards to frequency. If the output frequency varies a little higher, then a little lower, then a little higher and does this at a regular rate, this will show up as a spike in the graph (refered to as a "spur"). Real oscillators rarely do this, they kind of randomly fluctuate in frequency, such that this plot looks like a jagged continuous line. USUALLY much higher in value at the lower frequencies than the higher frequencies. From listener reports it seems that the lower offsets, (around 10Hz), seem to be the most import for audio. Unfortunately these are usually the most difficult to improve.

 

2) Actual frequency of the output. Due to above there is no such thing as AN actual frequency, it is wandering around. So the term "frequency of the output" is some form of averaging over time. That process can vary all over the place and is very rarely specified. Which of course makes comparing numbers rather difficult unless the same test equipment is used in exactly the same way. For example I have a frequency counter which has at least 30 different ways of measuring frequency, which will all give slightly different numbers.

 

#2 has several different subcategories:

 

#2.1) Accuracy. This is just the frequency out of the box. A high accuracy oscillator might be within 10 Hz of the number specified on the can and a lower accuracy one might be within 200Hz of the number on the can. Usually specified in Parts Per Million (PPM), thus a 1 PPM 10MHz oscillator can be up to 10Hz off the specified 10MHz. Some are so good they are specified in parts per billion (PPB). Unless it is pretty grossly off, this is pretty much unimportant for audio.

 

#2.2)  Temperature coefficient. All oscillators will change their frequency with a change in temperature, the Temperature coefficient (Tempco) specifies how much. It is usually measured in PPM per degree C. Unfortunately it is not a single number. Take an oscillator at 25C, raise the temp 1C and you will have a certain change in frequency, Start with the oscillator at 50C and change it one degree and you will get a VERY different  change in frequency. All crystal oscillators have some temperature where a small change in temperature makes almost no difference in frequency, if you are significantly away from this temperature the change can be VERY large for even a fairly small change in frequency.  Because this is measured in PPM/C a lot of people confuse it and accuracy since they both have PPM in the units but they are VERY different things, You can have high accuracy and lousy Tempco, or lousy accuracy and low Tempco. This has SOME affect on audio, but not a lot. The primary effect is at warm up, when a device is is turned on and the temperature inside the box is increasing. During this time the changing frequency can make a small audio difference. After reaching thermal equilibrium the Tempco has almost no effect on audio.

 

#2.3) Aging. This is the long term change in frequency over long time periods (measured in years). Most crystal oscillators have a fairly large change in frequency from year to year. During the first few years this is fairly large, then slowly goes down to almost no change after about say 15 years or so. Aging has essentially zero impact on audio.

 

#1 is the only one that has any significant impact on audio. Of the #2 categories Tempco is the only one which will have some impact on audo, but only during warmup. After the temperature settles down, almost no impact.

 

So in summary, spend money on low close in phase noise, money spent on high accuracy, low Tempco or low aging, is usually just throwing away your money. The OCXO is an exception to this, see below.

 

In particular a TCXO (temperature compensated crystal oscillator) is almost never a good thing for audio. A TCXO, has a normal crystal oscillator and a temperature sensor of some sort. The voltage from the temperature sensor is fed into a port on the crystal oscillator which causes its frequency to change with a varying voltage. This setup so it does some degree of cancellation of the crystal Tempco. So now we have a temperature sensor with almost always some degree of noise on the voltage output, feeding an input port which changes the frequency, thus rapidly varying the frequency, what is this called? Phase noise. Thus TCXOs ALWAYS have higher phase noise than a regular crystal oscillator using the same crystal and circuit minus the compensation. Yes it might have a smaller impact during warm up, but sound worse otherwise. Not usually a good use of money.

 

The type of crystal used in common crystal oscillators is what is called an AT cut. Its primary claim to fame is that the temperature where the zero Tempco appears ( sometime called the Tempco threshold or "knee" of the Temcp curve) happens near normal room temperature. This gives pretty good temperature behavior without doing anything else. But they do not have the best performance in other parameters. In particular for audio the phase noise of a different cut, called the SC cut, is much lower. BUT the knee in the Tempco curve is way up in the 90C range, at room temperature the Tempco is so bad that even a small temperature change drastically changes the frequency, so even for audio it is useless. This is where the OCXO comes in, the primary purpose is to raise the temperature of the SC cut crystal so it is sitting right at the knee of the Tempco. This gives an oscillator with a very low Tempco, very low aging and very low phase noise.

 

Not all OCXOs are created equal, in particular the less expensive OCXOs (say less than $100) do not use a crystal and circuit with particularly low phase noise, but they DO have very low Tempco and low aging, but the phase noise is no better than a $10 regular crystal oscillator. Again this is just a waste of money, you are spending money on something that doesn't make sound better. (note this is "new" price, not what you can get on ebay for a used one). BUT if you spend the money on a very special SC cut crystal and very special circuitry you can get the lowest phase noise of any oscillator known. It is not cheap, but this type of OCXO IS the way to get the lowest phase noise. OCXOs at this level also give you very low aging and very low Tempco, but these are not primarily the main reason for getting one of these OCXOs. Unfortunately for audio, most applications (other than audio) want very good specs for ALL the parameters, it should be possible for the manufacturers to optimize for phase noise only, thus giving us lower cost oscillators since they are not trying get say extremely low aging.

 

One other VERY important aspect about phase noise: comparing charts can ONLY be done if the frequencies are the same. The phase noise for an oscillator increacess by 6dBc/root Hz per octave of the oscillator frequency. Thus of you have plot for a 10 MHz oscillator and one for the same model oscillator at 20 MHz, the numbers will be 6 dBc/ root Hz higher. If you take that 20 MHz output and run it through a good flip flop, dividing the frequency by two, you will get the same phase noise plot as the 10MHz version.

 

So be VERY careful when comparing phase noise from different oscillators , they either need to be at the same frequency or you apply the 6 dBc/ root Hz rule. (explaining that rule is a little complicated so just take my word for it)

 

Sooo as far as aging is concerned, spec sheet aging has nothing to do with audio.

 

John S.