The sound of network switches


I have had a number of reactions again on
the review of the fantastic UpTone Audio EtherREGEN audiophile ethernet switch, especially from
people with knowledge of computer networks. They state, simply put, that if the sound
differs, the switch must have altered the data. I agree, but in what way. Before I get into the subject at hand, let
me first try to explain what I consider to be my function in the audio universe. I am no researcher or designer, I am a tech
journalist and reviewer, specialised in audio. Like a motoring journalist is no mobility
researcher or car designer. His technical knowledge is clearly limited
compared to these researcher and designer. But he does have a broad overview over all
makes and types of cars. When a new technology is developed, the manufacturer
invites journalists to a press meeting and have one or more technicians explain why the
new technology is such a big improvement – and they almost always exaggerate. When then another brand organises a press
meeting, the journalist asks technicians of that manufacturer what they think of their
competitor’s new system. A good journalist listens carefully to the
arguments in favour and against and then tries to see if he finds resulting behaviour during
test drives. Since he test drives dozens of cars a year,
he is in a very good position to compare. Especially if he has done this job for years. And a good motoring journalist is also able
to explain new technologies to his audience comprehensively. He should tell his audience that more valves
in a cylinder head in general increases horsepower as claimed in the advertisement but that this
often goes at the expense of torque. Unless a well dimensioned turbo is added. And a good motoring journalist does not only
drives a car in top gear sideways but also reports on normal use, economics and so on. He does simple measurements like acceleration,
top speed and fuel consumption but far more important is de feel of the car and its practicalities
like boot space. I consider myself in a more or less equal
position as a decent motoring journalist. I do basic measurements just to check if something
is wrong, get information from manufacturers or their technicians and if considered interesting
report on it. But my added value is in test-driving audio
equipment, so judging by ear. If you don’t agree, do what I do with tv-programs
I don’t like: I don’t watch them. And if you want to save the world from people
like me: start your own YouTube channel. OK, back to the subject. Can a switch have so much influence on the
sound quality? In the end the only way to establish this
is to listen. Trained listeners certainly don’t need double
blind tests, if the stereo setup used has sufficient audiophile qualities. See my view on double blind tests in a separate
video. I was rather sceptic when I heard about the
‘sound of switches’. So I approached AQVox to see if I could have
a review sample on loan. They were holding back a bit since they had
negative experiences with some members of the press. But in the end they send me the AQ Switch
Special Edition. Watch my review and you’ll see I was rather
surprised by the improvement it brought. In retrospect I even think I underplayed the
outcome because of unbelief. When I reviewed the SOtM sNH-10G switch I
had accepted the influence of switches on the sound a lot more. So much more that I accepted the sound difference
between a network cable and a fibre cable between the switch near the Intel NUC Roon
server on the third floor and the SOtM switch in my living at ground level. Again from a network specialist’s standpoint
fully ludicrous. Total cable length is 15 meters, slightly
under 50 ft, even for CAT5 cable no problem at all. When I backup my video editing files of several
hundreds of gigabytes from my editing computer to the big Synology NAS at the third floor,
the files remain perfectly in tact when copied back to the editing computer. So the connection is completely lossless and
more than fast enough for audio. Even 384 kHz 24 bit files only do 18.4 Mbit/s
while modern networks do 1000 Mbit/s. I once had 8 streams running from the Roon
server to 11 Roon Endpoints without any problem. So why should fibre sound somewhat better
than CAT6 Cable. I think that we can agree that all data sent
over a normally functioning network arrives unharmed. Switches are supposed to galvanically isolate
each port by what is called a magnetics module. I wondered if the consumer grade switches
I used conform to this. I opened the Linksys SLM2008 and it indeed
has two 4 port modules. I also opened the Sitecom LN-120v1 that had
four 2 port modules. Since these were rather cheap switches we
might conclude that in general – and certainly in my case – cheap switches offer galvanic
separation. Remains the question how well these high frequency
transformers perform. I seem to recall that the SOtM did not use
magnetic modules and you might wonder why. But the EtherREGEN does use them. Perhaps here again quality of the components
plays a role. The magnetic modules I found use center tap
transformers for use with Power over Ethernet – abbreviated to PoE. Only a slight mismatch between the two halves
of the transformer windings might already cause a DC offset. That offset might even be within IEEE specifications
for it won’t disturb normal network use. I’ll come back to this offset later on. All digital techniques work with stepped processes. Like a mechanical clock that steps through
a minute in 120 half-second movements of the escapement. Digital techniques have a crystal oscillator
as escapement which makes sure that going through the steps happens at a regular interval. In this graph those intervals are drawn by
vertical dotted lines and a digital signal then looks like this. At least in theory. In practice those intervals are not really
as constant as we would like. That becomes visible when on an oscilloscope
the traces are superimposed. This animation shows what happens. We see the vertical lines get wider since
they vary in timing. Sometimes the vertical movements, both up
and down, are earlier or later. This phenomenon is know as jitter. Within limits this is no problem for data
transport. Even if it goes wrong and a bit is missed,
the error correction is able to repair that bit or even more than one bit. And when the error correction is confronted
with too many faults to repair, the receiving side simply asks the sender to resend that
package. Up till now we have seen perfect square waves
but these never happen for it would require unlimited bandwidth. They will at least have rounded corners but
to explain the problem at hand I drew what can happen in a very exaggerated way. I have drawn two horizontal lines that indicate
the thresholds for the electronics to decide if there is a polarity change. The upper vertical line is for up going voltage
and the lower for down going voltage. The vertical dotted lines show that the moment
in time the voltage change is detected has moved to the right – thus later in time. This again is no problem if it is as constant
as shown here. Problems get more serious when we combine
the distorted wave form with phase noise. Now the detection of the voltage change can
be at the moment of the dotted vertical red line but just as good at the second red line
and the difference between them again shows the jitter. If we then add a less stable power rail, you
see the wave form als very in amplitude, the vertical axis. Because the rounded shape of the wave form,
not only variations in the time domain – the horizontal axis – but also variations in amplitude,
the vertical axis, change the moment the voltage change is detected. Still for data transport this would not be
a problem if these errors are within the defined limits. Let’s make the deviation more visible by
copying the distances between the vertical white lines to these green dotted lines and
move them to about the middle of the jittered signal of the first half cycle. Now look at the other cycles and you see the
nice stable clock is gone. And this is a relatively static representation. In real live all parameters can constantly
change. But normally this will not harm data integrity
during transport or storage. So what can go wrong then? Jitter – for that is what we talk about – can
only harm at two places in the audio chain: during analog to digital conversation and
during digital to analog conversion. The analog to digital conversion is done in
the studio and thus beyond our control. But the digital to analog conversion is, so
let’s see what happens there. The digital player, being a cd-player, streamer,
network bridge and the like, reads series of voltages that are registered on cd or in
an audio file in binary form and renders these in voltages of an analog signal. I use here a straight line as analog signal
for educational reasons, normally a straight line is most unlikely in audio. You can see that when the plotted voltages
are connected by a line, it results in a perfectly straight line. This is the function of the reconstruction
filter that evens out voltage changes faster than a bandwidth of half the sampling rate
would allow. Watch my video ‘Digital audio: Pulses are
not real world. Sine waves are’ for more information. As soon as the timing is off, like so, the
line is no longer straight. This is how jitter causes distortion in the
analog signal. Depending on the spectrum of the jitter – at
what frequencies the jitter modulates the digital stream – the impact on the sound quality
will vary. If the jitter is at general consumer equipment
level, harshness is quite common. Also deep lows can fully disappear. The stereo image is usually quite poor while
transients sound rather poor. With streamers at the level of my setup 2
harshness is clearly less to a level it doesn’t irritate anymore. Lows are more defined, voices start sounding
more natural with only sibilance being audible. Digital sources at the level of my setup 1
have instruments and voices more separated and projected in a virtual space, transients
make for instance a picked guitar sound in front of the speakers and a piano sounds far
more dynamic. When were is no sound, there really is no
sound, described in the audiophile world as ‘black background’. This phenomenon is hard to explain other than
that it can shock people. I link this to time smearing, where the energy
of a pulse is smeared over a longer time span. Quote: “If digital transport is so robust
you can easily buffer it and all is fine.” This is the comment I frequently get from
the Digital Inquisition, waving with their Malleus Nyquistum. If that was the case, how then can I hear
a difference between digital sources? And I am not alone. The Digital Inquisition can’t hear a different
or often haven’t even bothered. I have seen it all before in the 80’s when
we didn’t know about jitter. Me and my colleagues clearly heard a difference
but since it was not yet understood technically, we were told we heard thing that weren’t
there. L’histoire se répète. I’ll give you possible reasons why buffers
don’t solve the problem but first a real life example: I have used a fantastic Chord
QDB76HDSD digital to analog converter for years. It had a large buffer built in that could
store up to 3 seconds of 44.1 kHz audio. I have extensively compared the buffer lengths
offered by that DAC and didn’t hear a difference. I did hear differences when changing the digital
source, though. Regardless of the length of the buffer. How can that happen? This is what designers told me: after the
buffer the signal has to be sent to the chip that does the digital to analog conversion. During this transport jitter can be introduced
again. I already mentioned a noisy ground pane and
varying voltages on power rails. That is why the EtherREGEN has the secondary
side completely isolated from the primary side, including the power supply. I am also told that buffers can – or will
– generate noise that can vary depending on how hard the buffer has to work. Logic introduced intermodulation distortion
was also mentioned, although I can’t imagine how and where. Common mode noise on ethernet is mentioned
too as cause for a noisy ground pane.. I sincerely hope research will find definitive
answers although definitive answers don’t seem to go together with serious audio very
often. One reason being that many researchers rather
believe their theories and measurement gear than their ears. Those researchers with ears to trust often
work at smaller companies that keep their findings to themselves since applying for
a patent is far too costly. If possible at all. Only a very limited number of small companies
are open about their products and designs, UpTone Audio being one of them. You don’t need to trust my ears but please
don’t write me long comments without researching the phenomenon yourself. What you need is a proper stereo setup of
at least € 4,000. The better the setup, the easier it is to
hear. If you do want to write a comment, start with
a full list of equipment and the conditions you did your listening evaluation with. If you can’t understand why a switch influences
the sound quality to the degree I have described, I appreciate that. But any discussion is fruitless if you haven’t
researched it yourself. There is no use in fighting prejudice. I have seen that UpTone Audio allows for a
30 day return period, so there is no excuse for trying. And to those that reacted with hands-on reports
on the reviews of audiophile ethernet switches, I agree completely: they’re great. That brings us to the end of this video. There will be a new video, as always at Fridays
at 5 PM central European time. If you don’t want to miss that, subscribe
to this channel or follow me on the social media so you’ll be informed when new videos
are out. If you liked this video, give it a thumbs
up. Many thanks to those viewers that support
this channel financially, it keeps me independent and thus trustworthy. If that makes you feel like supporting my
work too, the links are in the comments below this video on Youtube. I am Hans Beekhuyzen, thank you for watching
and see you in the next show or on theHBproject.com. And whatever you do, as always, enjoy the
music.

, , , , , ,

Post navigation

31 thoughts on “The sound of network switches

  1. Well, you just lost a subscriber…. if you can copy a ton of video data via the switch UNALTERED, that is not enough proof to you that the switch is simply not altering the data on any way?

  2. Thank you Hans for sticking your neck out. It is not because for many this remains a controversial subject that we should not talk about it and try to explain. For me the deepest digital secrets still haven’t emerged. I too was very skeptical regarding audiophile switches, but I’m glad I did at least tried it. Many don’t even bother to give it a try, but nevertheless have strong opinions. That is beyond comprehension to me. And yes just as you explained, it has nothing to do with data integrity. Even a 20€ switch will manage data integrity just fine. To me it is about not upsetting the DA converter with a dirty signal.

  3. OMG Hans and everyone. Please come to your senses! The reason any digital tweak or gadget affects the sound of a proximate DAC is due to radiated or conducted RF noise. The data through a digital chain is always 100% perfect. It's only deep in a DACs d/a section that all the RF noise becomes manifest as subtle SQ differences.
    Hans I have a 90dB RF isolation chamber I am selling and when the digital chain in placed inside there is NO sound quality difference with any tweak.

  4. Network Engineer of some 25yrs here….. So as ive said before Switches do not alter Data. Their job is to purely Route the data based on the packet info, MAC tables, and ARP lookup tables. However, CAT5 is a copper cable, and as such liable to electrical noise. In an Audio world this could be made even worse by using STP. Screened Twisted Pair since RF noise could be transferred from a cheapo network switch with a terrible Chinese Switching Power Supply, along the screening to the device at the end. And if the ground planes are not isolated then that noise just got transmitted across the Ethernet network. Even though the IP packets remain unaltered. Apart from any added routing information that is. As you have noted, there is a difference with a Fibre based ethernet network switch. Why… because there is no copper, and so no RF interference impacting on the cable.

  5. If the IP Packet is identical comming from a cheap switch as from the EtherREGEN, then all your techno-baloney doesn't make a difference. The only place the data can be distorted is on the network client after the IP Packet is decoded and sent to the DAC. If you believe otherwise, then I hear the emperor got a new set of clothes that would suit you.

  6. Hi Hans, really appreciate this video and agree to most of the statements.

    Only might have one different perspective on what we call the impressive "black background".
    I think this is caused by listening to a really natural sound with minor distortion ("maybe you mean the same with without time smearing ?");
    In this case our brain do not have to proces and compensate for the artifacts (of distorted sound) we are (unconsciously) hearing.
    In case the music stops our perception is "black" because our brain compensation system was not working so there suddenly is no hearing (compensation) activity anymore.
    No brain activity for artifact compensation leads to black background…………

    It's your money or mine;
    just stating my thinking without scientific proof (at this moment) as we both dare to do this

    thx again for your inspirational reviews.

  7. You know once a degree qualified mechanical engineer would say all a turntable needs to do is turn at 33 1/3 rpm. We now know for a fact there is a lot more to it than that. I really don’t care what a “network” expert tells me. All they are saying is they don’t understand. I now have the EtherREGEN in my system feeding a very high end steamer. I’m not deaf and I don’t have money to burn. The difference is overwhelming. Who the hell are these “experts” to tell me it’s not better based on absolutely zero first hand experience! Well done Hans, you have made the enjoyment of music I love much better. Thank you and UpTone Audio.

  8. Ok i have no idea in digital transport but i was skeptical on rca wires for a long time and just used $3 to $5 rca cables for as long as i can remember. With watching ps audio, this channel, and steve, i tried a cable thats well received (qed profile) with a modest price. I was shocked how well it improved in detail, returned to my old cable and the difference is really there. I had the option to return it but i got one more set for my rca cable. Needless to say, im a convert.

    Now with this topic of digital noise, i will try the ifi purifier 3 and see if that cleans it up more than my idac2 can deliver. I am very skeptical about this purchase and dont have any hope that it will change how my music sounds. If it does, i am inclined to give this video a benefit of a doubt on digital transmission does make a difference. The illustration made in the video does make sense but im no expert. It seems to me that it presents the extremes on both sides that is why im a bit reluctant to 100% agreement on the subject.

  9. Those suckers do sound – a little. A friend and I recently had two Zyxel GS108 and the Bonn – which is actually a modded Zyxel – for a test session. We modded one of the Zyxel step by step with capacitors (POSCAPs) mostly. We agreed that our modded Zyx did not fall behind the Bonn for less money. My older modded Zyxel GS108 again sounded a little different again. Zyxel puts into the switches whatever they get from OEMs, it seems. The name and case looks the same, the boards are totally different. I highly doubt that anything has to do with data errors. That does make no sense at all, because the IP protocol does not let anything go wrong and the devices have buffers for a number of seconds. My dCS Network bridge keeps on playing for 5 seconds or more after I unplug the switch. Meaning: The switch cannot have any influence at this moment. There are no data errors nor jitter induced by the switch. It is all about high frequency noise travelling across the cables. That's my theory.

  10. Apologies if this has already been answered but I couldn’t see it anywhere.

    If audiophile switches sound better than normal switches because of attention paid to electrical noise etc, and galvanic isolation of data and clock lines help the sound, why not just use WiFi instead of an expensive switch?

    I’d be interested in your opinion, thanks 👍

  11. Hans, I thoroughly enjoyed this video. It explains a complex subject in a digestible manner. Although I consider myself an audio sceptic, i have also heard for myself that the tiniest details CAN matter. As an audiophile, I must investigate for myself. The one thing I cannot get behind is cable elevators. Couldn't hear a difference in my highly resolving system.

  12. Two things to consider:

    Ethernet will split the digital sound into multiple packets (fragmenting), so according to your theory, you would have extremly hard jitter at least every 1500 bytes, as there will be a header of the next package and no Audiodata in that timeframe.
    Therefore buffering has to be implemented in any Ethernet DAC, there is no way arround it! DAC conversion from Buffer -> Analog will always be done with the clock of the DAC device!

    Galvanic isolation for ethernet ports isn´t just present in the switch, but all Ethernet devices! Common mode on ethernet cable pairs won´t affect the ground plane of a device as it is "galvanically isolated!"

  13. Which is more likely- The underlying technology of the entire digital world is fundamentally flawed, or you think you hear things that you definitely do not?

  14. Hans, I recently purchased an ER and am trying to see if there is an improvement. I’m also using aN UPTONE LPS power supply. My DAC is a PS audio DSD Jr. My amp is an STA 200. My speakers are Tekton double impacts and I have DIY cables. My music is streamed via Qobuz/Tidal from a Mac mini running Audirvana. USB is a curious evolved cable. I am having difficulty hearing a difference. I heard clear differences when trying different USB cables, interconnects, and speaker cables. I also heard a difference with an ISO Regen that I recently sold. Any thoughts on why I’m not hearing a difference? I’ve completed about 50 hours of burn in as well. Thanks for your help.

  15. We appreciate your work and devotion, Hans. Your standards are high. Not accepting audible differences by digital specialists is like medieval astronomists continuing to argue the earth is flat. Based on knowledge that will be proven wrong afterwards.

  16. L'histoire se repete, indeed! Reminds me of the never ending battle against pops, clicks, skips, wow and flutter during the Age of Vinyl!

  17. Wouldn't be the right conclusion to avoid switches and anything that follows and just put the music files u want to play onto a ram drive (either virtual or physical). I also don't understand why advocate USB reclockers instead of direct I2S output

  18. How you relate jitter of the Ethernet bus to an i2s bus for example ? They are 2 different signals, from 2 different busses and 2 different clocks.
    Except I'm missing something here

  19. Perhaps I'm misunderstanding this – How can a network cable or network switch have an impact on audio quality?

    Most digital formats/protocols employ checksums to ensure data integrity, including TCP. The only way something like this could conceivably happen is if all the audio is being sent via UDP, where the source-device sends traffic, but requires no acknowledgement from the destination, so its can easily get dropped/lost etc.

    TCP is incredibly reliable, if data was being corrupted, the destination-device would see that packet checksum and contents don't match, and the source would be asked to re-transmit it. Likewise, a packet is a packet, a switch does not know (or care) if its part of a word document or audio file, if its alternating/corrupting the audio, it would also be doing the same to all content on your network – the net result being, either all your content ends up corrupted, or your network will saturated by all the retransmits (An it would feel more like a 10Mb network, rather than 1Gb or even 10Gb).

    I've 20 years experience managing networks, working with everything from $50 desktop switches, to $5m chassis switches for backbones, so my scepticism doesn't come from ignorance, but experience – I often call BS when I read an article claiming a $1000 network cable gives better audio than a $30 CAT6a/CAT7 cable. But I would seriously like to know, why people think it is and the science behind it? also has anybody tried to a packet-capture to see if there is a difference?

  20. Ok i posted in this topic a day ago and went to the store to purchase a ipurifier 3 to see if the theory of digital signal does not change or suffer from degradation. The people there were kind enough to let me try all their ifi filter gear and try any of them on any of their wide varieties of dacs and amps.

    They advised me to try the idefender and isilencer first then the purifier 3. So i did… tried the idefender with sony, fiio, ifi, and chord dacs and various amps they had lying around And yes the difference is quite noticeable. More definition, separation, and clarity. I was a bit skeptical about the mercury wire and tried that too and my o my everything improved. The bass was not just low end but it has definition… thats the most noticeable. Needless to say Im a true convert now that cables matter. Not everything about cables and filters are snake oil.

    Ended up with the idefender and isilencer for now as i dont have additional funds for ipurifier 3 and mercury cable. For sure im going to save up for those 2 in the future. Those little improvements add up to a very clean sound.

  21. In economics scientists turned inquisitors during the eighties and ninieties when big data analyses showed their trusted theorie unsufficient, but they kept their trust in their tried and tested. Their teachers werd the best, they could't be wrong. Right? Additional input from psychologists and biologists gave economics the broader view that restored relevance. Neuroscience, chemistry and even psychology are now filtering through in audio. So inquisitors, turn inquisitive. If you want to describe what can be heard, be VERY sure you understand how hearing works. Hans does, and he is inquisitive.
    Hans, spot on. Like always. Chapeau.

  22. Hi Hans nice video again. I received my EtherREGEN this Friday and it is really a big improvement on different levels. 3d, depth, width, blacks. It is even a bit louder. Very happy. I asked some questions on their forum about adding a second EtherREGEN at the internet router. The router has my Roon ROCK directly connected and I wonder what would be the benefit of using this B-port on mij Roon server. The EtherREGENs would be interconnected by glass fiber of course. I will update if something interesting comes out of that discussion…

  23. Absolutely outstanding. The best and most accurate explanation of why network switches have an audilble impact I've seen so far.

Leave a Reply

Your email address will not be published. Required fields are marked *