For the past few weeks / months, your author has locked himself in a measurement room with lots of switches and loud measuring equipment. This, in order to do some research on the influence of network switches on audio clocks. Will we succeed in finding “proof”?
As an enthusiast, you are always looking for improvements in the audio chain. The biggest steps can be made with acoustics, a suitable speaker and a good speaker/amplifier match. A solid source and and appropriate cabling also belongs within a solid balanced system. Then we get into murky waters, because what about power? Power is power right? Or what about speaker decoupling? And special anti resonance pucks under our audio equipment?
Some things are very hard to explain. We hear something, but don’t really know where the effects exactly come from. Consider cables. Anyone taking simple measurements of cables will not be able to explain why one cable sounds great and another does not. Very accurate measuring equipment is needed to gain some insight. But even those tell only a small part of the story.
Regular readers of Alpha Audio know that we spent a long time testing speaker cables. To give just one example. When we then look at power conditioning, basically the same applies: we hear differences, but it is extremely difficult to explain or measure. We are still working on a solid set-up to measure power conditioners.
If we then look at network audio and then specifically switches, no serious research has actually been done on this at all. Some tests with fiber optics, as we did a few years back, but no real measurements or elaborate theories. At least we have not been able to find it. That prompted your author to dive deeper and start measuring. You have already been able to read the first story. In this part we go further and look at clock jitter in a streamer. How is it affected when we use various switches?
Do NOT look at the data!
“You don’t know what you’re talking about. Learn how a protocol works!” “Another audiophool who has no idea how TCP/IP works. If he would educate himself he wouldn’t waste his time like this!” “Data is data. A one is a one and a zero is a zero. What nonsense. It’s all between the ears. What idiots” A few quotes from dear viewers and readers when talking about switches in an audio network.
We would love to climb on a high roof top, switch on a big PA system and start shouting: there is NO difference in data transmission between switches.
Data just arrives. Bits and bytes don’t get “more corrupt” with a cheap switch. The very robust protocols take care of that as well. Think TCP that has error correction built into it. If those protocols didn’t exist, the Internet wouldn’t work. It’s as simple as that. In addition, a streamer always buffers part of the data stream to prevent problems. That’s not the problem at all.
Huh, what are you saying! So there is no difference in switches?
Yes… correct: in terms of data. But there are definately differences… in generated noise. Yes… here we go again. Power supply noise and other types of noise created by high-frequency switching. That’s where we measure huge differences. There are very “quiet” switches and enormously noisy models. And that’s where we continued to investigate. So forget data… it’s not there.
The measurement setup
For this study, several more members were added to the measurement family. First of all, the Wavecrest SIA 3000. This is a Signal Integrity Analyzer that can literally measure jitter down to the femtoseconds. The internal resolution is 200fs (femtoseconds) and the internal reference clock has less than 1ps of jitter. Our calibrations on the machine itself confirm this (it had to recalibrate itself after traveling from South Korea). This is really very impressive for a machine from 2002. And also for a machine released in this decade. With reason, these machines were practically priceless at the time of release. It really is an extremely accurate – and sensitive! – device with which we can thus measure clock jitter with precision. And phase noise, low-frequency modulations, etc. We don’t have all the licenses available, unfortunately, but enough for what we do. And additional licenses… are somewhat pricey… understatement.
In addition, we purchased a new Rohde & Schwarz probe. This is to achieve a decent bandwidth for the measurements; the R&S probe has a bandwidth of dc to 300 MHz. Enough for this project.
And finally, we purchased an impedance buffer amp to convert from 1 MOhm to 50 Ohm to connect to the Wavecrest with the proper impedance. This machine has an input impedance of 50 Ohms. We chose a model from Matthews Engineering because of the price/quality (dc – 400 MHz and very quiet) and the fact that it works on batteries as well as usb-power. We ended up powering it via an adapter from USB to the lab power supply.
The setup for the measurements was relatively simple. It had to be, because every change in set-up results in slightly different results. By the way, a clock also drifts with temperature and other external factors. So a measurement on day one can be slightly different from one on day two. Very annoying, because it means we really have to measure all switches one after the other to be able to compare things. However, the absolute differences between the switches remain the same. If that were not the case, we would have really gone crazy the last few weeks.
We took a Metrum Acoustics Ambre as a streamer to check jitter. This is a nice midrange streamer with two neat vcxo clocks from Tentlabs. We were able to get to the clock fairly easily, which allowed us to properly connect the measurement probe. We then connected the switch to our LISN (Line impedance stabalizer Network) to both decouple and measure power supply noise. The switch is also connected to the CDN T8 for the same reason: the network is decoupled AND we can measure noise from the CDN. We also measure – one after the other – the power adapter as well as the network cable through an RF current probe to measure common mode. We measure noise from the network in both high and low frequency (low frequency goes through the Prism dScope).
Below is an overview. Thanks to our ‘news guy’ Ronald for transferring the undefinable scribble into a clear picture. By the way, the analyzers (Prism and Spectrum Analyzer) for the LISN and CDN are not interconnected. That line is a bit unfortunate.
The test candidates
Now the other day we did a big test of switches. We did blind listening AND your author did the first series of measurements, which consisted mostly of measurements of all kinds of noise. That laid the groundwork for this in-depth investigation. Because, what soon became apparent: the poorly measuring switches, didn’t sound pleasant either. That must be explainable, right?
We grabbed several candidates for this test to analyze further. These included the Netgear 108E, Dlink 108, Dlink 1210, Pura Ammonite V2 and Elite and, of course, fiber. We also swapped with power supplies to determine the impact of a different power supply. We deliberately don’t say “better” power supply here.
It is important to realize that we had to measure over several days. One series of measurements on a switch takes about 45 minutes to get right. Since we made multiple combinations, we took final measurements, spread over three days, to gather enough data for this article and some conclusions.
As mentioned earlier, everything affects the measurements. We do not have a completely shielded lab, so there are slight variations between day one and day two. However, we also took control measurements each series, so we know what these deviations are. We have factored that into the conclusions.
In the table in the measurements section, we have incorporated everything from one series, so you do have a concrete overview with comparable results.
When you compared switches, does noise peaks appeared the same frequency points on each switches and how big was the general noise floor levels on them? It would be nice to see spectrum analyzer screen shots from each switches how then behaved.
If you login via Patreon, you can see all the measurements.
really interesting test!
What surprises me is that there is no clear correlation between phase noise and jitter – since they are just two sides of the same coin.
I believe that the Pura the AMMONITE switches perform well because they have better phase noise in the Ethernet signal by using the OXCO clocks. That would at least be a coherent explanation that would also agree with John Swenson’s thesis. Then we would have two things affecting the sound: common mode noise and phase noise.
Have you found a dependency regarding the frequency range of the common mode noise to worsen the streamer clock – I would be interested. I could not clearly assign this in the measurements.