• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Review and Measurements of Empirical Audio Synchro-Mesh

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,376
Likes
234,515
Location
Seattle Area
This is a review and detailed measurements of the Empirical Audio S/PDIF and Toslink "reclocker." It is on kind loan from the manufacturer and costs USD $699 direct. Also on load is a linear power supply which from what I can figure out is another USD $799. So the combination tested for this review is USD $1,500.

Let me provide some background on why such devices exist and problem they try to solve.

S/PDIF interface was designed decades back by Sony and Philips. It is a way to get digital audio stream out of a device for external connection to a DAC. A poor design decision was made at the time to make the source the master and embed the sample timing within the data stream. Put simply, if you hook a DAC to a S/PDIF (or Toslink) interface, it needs to a) extract the clock to know at what rate it should play and b) stay in sync with the source. You may think that once the DAC has detect the sample rate is say, 48 kHz, it can play at that rate but it cannot. The spec allows the source to go slightly slower or faster or even change from time to time as it pleases. The DAC then, must attempt to follow this change and stay in lock step with it.

Complicating matters is that we have many sample rates so the DAC and its S/PDIF receiver need to capture a wide range from 32 kHz to 768 kHz. Such circuits are not trivial to design well. Fortunately S/PDIF has been around a long time so the knowledge of how to extract the data and clock is rather mature so even the worse implementations still work and produce high fidelity music.

There is a concept in digital transmission called jitter. This is momentary changes in clock rate. This can happen due to noise, digital circuits interfering, ground loops, and tones from the rest of the circuit. Usually you have a combination of all of this. We use a spectrum analysis to find these in the so called j-test I have been running.

There are a few solutions to jitter and clock extraction. The most common one is to build a very good clock tracking system called a PLL or phased locked loop. Think of a PLL as the platter in a turntable. The heavier you make the platter, the less variation it will have in speed due to motor changing its speed. The same flywheel concept applies to a PLL which attempts to ride out variations that it considers inconsequential. On a turntable, you can change speeds and you want the platter to adapt to that new speed in reasonable amount of time. Users won't want to wait five minutes for speed to change. So you don't want to make the platter 100 pounds. Likewise in PLL designs, you have to allow sample rates to change and do so quickly enough. In a playlist of music with different sample rates, the change is ideally less than a second. You can make a PLL very good at rejecting variations and hence jitter but then it may be slow to sample rate changes. There are tricks to have your cake and eat it too by having two PLLs cascaded and such which goes beyond scope of this review. Suffice it to say, if you are buying a high-performance DAC, it better come with sufficient mechanism to eliminate jitter as a concern.

The second solution to clock jitter is to use what is called an asynchronous sample rate converted and is what Synchro-Mesh implements. Idea here is that you use a very high precision clock on one side that is always fixed. In this case, it is set to 96 kHz (they can customize this to other rates). All input rates are resampled to this fixed rate. Note that even if you play 96 kHz samples, your data will change as it has to. Remember, the 96 kHz is just a nominal value. The actual number of samples may be lower and higher at any time and hence, the resample has to either reduce the number of samples or increase them. How does it do this? Simple: interpolation. Using math, you can determine what has to have to change in samples to represent the same waveform but with fewer or more digital audio samples. If I have 20, 30 and 40 as audio samples, I can convert this to 20, 25, 30, 35 and 40 to represent the same signal but with double the number of samples. I say the "same" but it really is not the same. We would have to be smart about what algorithm we use to determine the intermediate samples and not just this linear interpolation I just used.

Additional compromise occurs here in that you are performing mathematical computation and you have to decide how much resolution is good enough. Too little resolution creates new errors that manifest themselves as distortion even though jitter may have been rejected.

Note that regardless of scheme, jitter may be created inside the DAC, or outbound from a reclocker like Synchro-mesh.

These concepts should become more clear as we get into the measurements with real examples.

Measurements
As it happens, I have the perfect platform for this testing. My Audio Precision APx555 analyzer can not only examine jitter in the output of a DAC as I have shown in many reviews, but can also create jitter over S/PDIF. The second part was courtesy of a new, high performance DAC from Gustard: the DAC-X26. The Gustard has a programmable PLL that can be bypassed or set to different modes of operation. We will see how that becomes handy later.

Starting from the top, concern has been raised as to whether our instrumentation is good enough to find out what is going on with a device like Synchro-Mesh. Let's address that simply by testing the Audio Precision APx555 analyzer purely in digital mode. This means it will generate a S/PDIF signal and then capture/analyze the same waveform over a real (green) coax cable (supplied by Empirical Audio).

For test signal, I will be using the same "j-test" signal. J-test signal was specifically designed to find jitter caused primarily over S/PDIF links. It has a pure "tone" that is actually a square wave with its primary frequency at 12 kHz in 48 kHz sampling used here. Embedded in that square wave is a second square wave. That one toggles the least significant bit in a 24-bit audio sample. No DAC I have tested has dynamic range that good so you have never seen this square wave directly. What you do see is the side effect of it causing activity in the DAC. Here, you finally see it in its full glory since there is nothing analog in this test:

Empirical Synchro-Mesh Jitter and Noise measurement.png


Focus on the graph in blue first. This is the Audio Precision generating and capturing its own S/PDIF signal. The main tone is at 12 kHz. But on each side of it, we have a spray of tones at odd harmonics of 250 Hz (frequency of the one bit square wave). The reason we see it so clearly is that our noise floor is an incredibly low -330 dB!!! This is the noise floor that you would have if you had 64 bit audio samples. Here, we have perfect capture of audio data and so the noise floor is simply indicative of the math error in conversion of time domain waveform as a spectrum.

Note that there is NO jitter here. Everything is 100% captured as data and presented as such. What is shown in blue is simply what the j-test signal looks like in frequency domain.

The red graph is what we get when we route the signal through Empirical Audio's Synchro-mesh. Our "noise" floor is now around -180 dB, representing an increase of 150 db! Again, this has nothing to do with jitter or analog noise as there is no analog capture here. The cause of the much increased noise floor is the coarser math resolution in the resampler chip used in Synchro-Mesh. If it had used 64 bit math, then it would be the same as Audio Precision in blue. But it is not so we lose ultimate accuracy even in a perfect scenario of digital to digital. In other words, your digital audio samples have been changed to embed in them, this much error.

Advocates of asynchronous resamplers state that they implement enough mathematical resolution as to have a noise (error) rate that is below anything the DAC can produce. We will put this to test soon.

For the next test, let's instruct the Audio Precision analyzer to generate a jitter tone that is in the form of sine wave at 5 kHz which has a "magnitude" of 1 nanosecond. That is one billionth of a second. We feed that to the Gustard DAC-X26 while either bypassing its external "GPLL" or setting it to default "High" mode: (signal is J-test)

Gustard DAC-X26 SPDIF Jitter Measurement from Audio Precision APx555.png


Focus on blue curve with PLL turned off. We see to new distortion tones, one at 7 kHz and another at 17 kHz. Why those frequencies? Sinusoidal jitter at these levels generates what we call AM modulation (yes, just like how an AM radio station works). The math tells us that if you modulate a 12 kHz tone with a 5 kHz one, you get two tones: one at 12-5=7 and another at 12+5=17. And that is exactly what we have.

Note that despite our timing variation being just one billionth of a second, we have very tall distortion spikes, reducing SINAD to 88 dB or so from its peak at 115! So small timing errors matter greatly in audio. Why is that? It is because of very high resolution in each audio sample (up to 24 bits). It does not take much to obliterate fidelity of many bits with tugging at the waveform left and right a tiny bit.

Note that despite public assumption, you never hear such jitter as echo, smearing in time domain, change in soundstage, ets. How on earth could the ear detect such a timing error in the order of one billionth of a second? It can't. What it can hear are the non-linearities created that result in those sidebands. Make those loud enough and they can become audible at some point. The effect would be as if you played all three tones at once. Depending on the frequency of the jitter, the variation in sound will vary. So please don't substitute your own intuition for audible effects of jitter. It will be wrong.

Let's route our S/PDIF signal through Synchro-mesh with the GPLL set to OFF (bypass) in DAC-X26:

Gustard DAC-X26 SPDIF Jitter Measurement from Audio Precision APx555 through Empirical Synchro...png


Hopefully this is not too hard to see/understand. The graph in pink is the Synchro-Mesh loop. If we compare it to no PLL in light blue, it is much cleaner. Most but not all jitter/noise spikes have been removed. That is the good news.

Not so good news is that there was a price to pay: our noise floor is higher than the graph in light red which was with the internal GPLL. This indicates to us that the mathematical precision of the Synchro-Mesh resampler is not good enough for an excellent DAC like Gustard DAC-X26. Simply turning on its GPLL, not only eliminates all of those jitter components but also gives us lower noise floor. See inset for that comparison.

Making this short, addition of Synchro-Mesh is a negative in this test. Gustard DAC does better without any external help.

Let's dip into the toolbox of APx555 analyzer for some cool measurements I have not shown before. The analyzer is not only able to generate static jitter, but can also vary its frequency while measuring the DAC performance. In this test, I programmed the APx555 to generate a 10 kHz tone at 96 kHz sampling. This represents one tone in your music. I then told the analyzer to induce jitter with a frequency varying between 20 Hz and 200 kHz with an amplitude of 10 nanoseconds (10x more than previous measurement):

Gustard DAC-X26 SPDIF Jitter Frequency Sweep Measurement from Audio Precision APx555.png


Focus on the blue curve: this is with the internal fancy GPLL turned off again. We notice that even this large amount of jitter has no influence over THD+N of the DAC until we get to 1 kHz or so. Above that, we start to induce good bit of distortion, lowering our performance by a whopping 30 dB! Again, I vary the jitter frequency. Our main signal tone remained at 10 kHz.

Turning on the GPLL nails jitter to the wall, completely eliminating this effect in red. We not only don't have that rise in distortion but also have lower noise floor. That is because random jitter creates random output on the DAC. Getting rid of random jitter using a good PLL then, lowers our analog noise floor.

Running the jittered signal through Synchro-Mesh performs similarly to turning on the GPLL but as we noticed before, with a higher noise floor. So once again, we are better of letting a good DAC eliminate jitter on its own than using this external device.

Alternative to the above test is to keep the jitter frequency fixed, but vary its amplitude and plot that. Here is that performance for the four modes of PLL in DAC-X26:
Gustard DAC-X26 SPDIF Jitter Level Sweep Measurement from Audio Precision APx555.png


Here, you nicely see the trade offs I mentioned at the start of this article. With the GPLL off (blue), increasing level of jitter degrades THD+N performance until we can no longer extract the digital audio at about 900 nanoseconds. Note that is is extremely high level of jitter which you are not going to have in short lengths of S/PDIF cable we use at home.

Switching on the GPLL to "High" (default), gives us perfect performance in red until about 50 or 60 nanoseconds. Distortion then shoots way up briefly before we lose lock. Per above though, this limitation is not an issue because we never have jitter that high.

If we did have more jitter, we could use the "CD Opt" (whatever that means). It has similarly good performance and never loses lock until the maximum amount of jitter we can simulate. This would be a very good choice with ultra long cables.

The "normal" mode is the most picky (in green) in that it lost lock even faster than High mode and provided no other benefit.

Now let's throw Synchro-Mesh into the mix, this time using GPLL off so that it gets to do something:

Gustard DAC-X26 SPDIF Jitter Level Sweep Measurement from Audio Precision APx555 and Empirical...png


We basically see the same thing as before. When Synchro-Mesh is working (below about 550 nanoseconds of jitter), it does a nice job of eliminating jitter but costs us in noise floor by a few dB. Conclusions are the same as before: internal PLL wins in the case of DAC-X26.

Now, you may argue that these results are unique to DAC-X26 so let's go to the extreme and test a low performing DAC: the Schiit Modi 2 Uber. We connect everything the same and run the jitter frequency sweep again, with and without Synchro-Mesh:

Schiit Modi 2 Uber SPDIF Jitter Frequency Sweep Measurement from Audio Precision APx555.png


Unassisted, the Modi 2 Uber (in blue) shows a very similar jitter rejection to Gustard DAC with its PLL shut off. Addition of Synchro-Mesh in red nicely gets rid of that jitter sensitivity. There is no penalty here because our DAC already has such a high noise floor that what Synchro-Mesh does is literally lost in the noise.

Conclusions
Using the right set of measurements, we have no issue whatsoever characterizing the performance of Empirical Audio Synchro-Mesh. It works as designed: it resamples everything to 96 kHz and in the process eliminates jitter. But limited math resolution (or possibly random jitter) means that it raises the noise floor of a high performance DAC like Gustard DAC-X26. The internal PLL (GPLL) in the Gustard brings the best of both worlds, eliminating jitter and maintaining excellent noise floor. In other words, there was nothing broken to fix.

With a very low performing DAC like Schiit Modi 2 Uber the Synchro-Mesh does nicely get rid of excessive jitter with no penalty because the DAC itself is not that good. But keep in mind that I simulated very, very high levels of jitter. Your sources are not likely to have 50 nanoseconds of jitter. The actual improvement will likely be far lower.

Let's say the benefits are real with low-end DACs. How does one justify spending $700 or $1,500 on a reclocker when the DAC costs $150? Why not put all the money toward an excellent DAC and be done with it? Indeed the Gustard DAC-X26 costs less than the Synchro-Mesh!

In summary, the Synchro-Mesh does it what it is supposed to do. It seems to be well implemented for that functionality. Problem is, I can't find a use for it even using highly sensitive instrumentation testing. Any audible effect is bound to be imagined especially with any decent DAC. Heck, if folks can't hear the increased noise floor, they should not talk big about what is improved. :)

------------
As always, questions, comments, recommendations, etc. are welcome.

The sun is out and temps are starting to rise so I need to buy seeds and planting soil to start our garden. We need the food in the winter as otherwise, we will go hungry. So please be generous with your donations using:
Patreon: https://www.patreon.com/audiosciencereview), or
upgrading your membership here though Paypal (https://audiosciencereview.com/foru...eview-and-measurements.2164/page-3#post-59054).
 
Last edited:

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,685
Likes
10,700
Location
Hampshire
I'd like to see a digital capture of a J-test with added jitter. It would also be interesting to see the behaviour (frequency response and such) of the resampler with inputs at some other sample rates, both lower and higher than its native 96 kHz.
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,376
Likes
234,515
Location
Seattle Area
I'd like to see a digital capture of a J-test with added jitter. It would also be interesting to see the behaviour (frequency response and such) of the resampler with inputs at some other sample rates, both lower and higher than its native 96 kHz.
The second test is all of that. The signal is j-test plus added jitter (added text to make this clear) and sample rate is 48 kHz.
 

M00ndancer

Addicted to Fun and Learning
Forum Donor
Joined
Feb 4, 2019
Messages
719
Likes
728
Location
Sweden
That was the expected result. Thank you @amirm
Once again we can see that the product do work, but the benefits are debatable, especially if you own a decent DAC, regardless of price.
 

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,685
Likes
10,700
Location
Hampshire
The second test is all of that. The signal is j-test plus added jitter (added text to make this clear) and sample rate is 48 kHz.
Seems to me that's the analogue output of the Gustard DAC. I'd like to see what the S-M does to the digital data. The residual jitter from its PLL ought to be visible there.
 

Hugo9000

Addicted to Fun and Learning
Joined
Jul 21, 2018
Messages
574
Likes
1,749
Location
U.S.A. | Слава Україні
Thank you very much, @amirm , for this review and most especially for the fantastic clarity with which you explained the concepts involved, your process, and how to interpret the results. Like the best professors, I always feel that you elevate those you are teaching or sharing your expertise with, with no arrogance or talking down to your audience! Much appreciated!
 

FrantzM

Major Contributor
Forum Donor
Joined
Mar 12, 2016
Messages
4,337
Likes
7,737
We got our blood pressure raised, in the end ... It does re-clock but ...

Can I paraphrase Shakespeare ?
Nah!.. That would be blasphemy.
 

FrantzM

Major Contributor
Forum Donor
Joined
Mar 12, 2016
Messages
4,337
Likes
7,737
Thank you very much, @amirm , for this review and most especially for the fantastic clarity with which you explained the concepts involved, your process, and how to interpret the results. Like the best professors, I always feel that you elevate those you are teaching or sharing your expertise with, with no arrogance or talking down to your audience! Much appreciated!
+1
 

garbulky

Major Contributor
Joined
Feb 14, 2018
Messages
1,510
Likes
827
The noise floor doesn't really matter to me. No way I could hear it.
I have more of a problem that this resamples everything to 96 khz. Most music I listen to is at 44.1 khz. What impact does this have on timing and a bit perfect signal when it is resampled to 96 khz? Also what if it is fed a 192 khz signal or a 172 khz signal? Wouldn't you lose information? Also can we see a 1khz signal with the modi?

My DC-1 already has an ASRC and it performs it on every input. The older XDA-2 also had one but it did what this unit did - reclock it to 96 khz. The DC_1 actually does no reclocking and processes it at its regular clock rate. It would be interesting to see any effects differences when comparing the two. Because the DC-1's asrc can be turned on and off. And honestly I heard no real difference either way,
 

MZKM

Major Contributor
Forum Donor
Joined
Dec 1, 2018
Messages
4,244
Likes
11,479
Location
Land O’ Lakes, FL
Indeed the Gustard DAC-X26 costs less than the Synchro-Mesh!
Including the price of the LPS; how does it perform with a normal wall-wart power supply?

Oh, and I know you fried the D50, but do you have a DAC to test that’s better than the Modi 2 but not as expensive as the Gustard?
 
Last edited:

MZKM

Major Contributor
Forum Donor
Joined
Dec 1, 2018
Messages
4,244
Likes
11,479
Location
Land O’ Lakes, FL
The noise floor doesn't really matter to me. No way I could hear it.
I have more of a problem that this resamples everything to 96 khz. Most music I listen to is at 44.1 khz. What impact does this have on timing and a bit perfect signal when it is resampled to 96 khz? Also what if it is fed a 192 khz signal or a 172 khz signal? Wouldn't you lose information? Also can we see a 1khz signal with the modi?

My DC-1 already has an ASRC and it performs it on every input. The older XDA-2 also had one but it did what this unit did - reclock it to 96 khz. The DC_1 actually does no reclocking and processes it at its regular clock rate. It would be interesting to see any effects differences when comparing the two. Because the DC-1's asrc can be turned on and off. And honestly I heard no real difference either way,

Most all DACs resample (the Benchmark resamples to ~211kHz), so I doubt its an issue to be concerned over.
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,376
Likes
234,515
Location
Seattle Area
Including the price of the LPS; how does it perform with a normal wall-wart power supply?
It was only sent with this supply. I don't have the standard supply they sell with it.
 

MZKM

Major Contributor
Forum Donor
Joined
Dec 1, 2018
Messages
4,244
Likes
11,479
Location
Land O’ Lakes, FL
It was only sent with this supply. I don't have the standard supply they sell with it.
1475485-empirical-audio-synchromesh-reclocker.jpg


Looks like a random one from the 90’s, so I’m sure any you have laying around would work :).

I was only mildly interested in that (would be interesting if it had mains leakage, like many of the relockers and other similar products you measured).

Besides the fried D50, do you have any well performing DAC at or below the $700 base price of the Synchro-Mesh? Because, the jump from the Modi 2 to the Gustard is pretty high.
 

Newk Yuler

Active Member
Joined
Mar 5, 2019
Messages
155
Likes
252
Thank you, Amir. This test and thorough, easily understood profile has exhibited almost exactly what I expected, so I get to feel good about myself as an amateur hifi tech dork and longtime Empirical Audio user/disciple. It largely answers my question about the need for external digital refining products against a modern, well designed, sophisticated DAC product (Gustard X26 at +/- $1200 USD). It remains to be seen what new DACs will measure well against the Gustard and prices around this point. (The RME, etc.)

Here's hoping and anticipating a profile of the Dynamo LPS alone.

Now the Off Ramp 6 deserves to be measured against the Gustard or similar well performing product. That will be very interesting and better suited for a razor sharp challenge.
 

Newk Yuler

Active Member
Joined
Mar 5, 2019
Messages
155
Likes
252
Looks like a random one from the 90’s, so I’m sure any you have laying around would work :).

It isn't random. Steve looked pretty hard to pick that one. He ships it with both the SM and OR5. It probably measures well for an inexpensive SPS.
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,376
Likes
234,515
Location
Seattle Area
Besides the fried D50, do you have any well performing DAC at or below the $700 base price of the Synchro-Mesh? Because, the jump from the Modi 2 to the Gustard is pretty high.
Sure. I have a bunch more I can test with. Will post later...
 
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,376
Likes
234,515
Location
Seattle Area
Now the Off Ramp 6 deserves to be measured against the Gustard or similar well performing product. That will be very interesting and better suited for a razor sharp challenge.
I am not sure Steve will be in a mood to send me more gear after this review. :) Maybe one of you can send in yours....
 

bunkbail

Addicted to Fun and Learning
Forum Donor
Joined
Feb 27, 2018
Messages
520
Likes
662
Sure. I have a bunch more I can test with. Will post later...
Please do it with a Khadas. I bet it does nothing.
 

HammerSandwich

Major Contributor
Forum Donor
Joined
Nov 22, 2018
Messages
1,137
Likes
1,497
Great review, Amir. Best jitter article I've seen since the original by Bob Adams in The Audio Critic.

Sure sounds like Empirical needs to audition an X-26.

The reason we see it so clearly is that our noise floor is an incredibly low -330 dB!!!
Yikes - this completely breaks the Shoutometer. @RayDunzl, you need to recalibrate out to Alpha Centauri.
 
Top Bottom