A (very!) deep dive into network switches- listening and measuring

Intro

Readers and viewers of Alpha Audio know that the platform has been involved with streaming audio for quite a long time. And with that comes the influence of infrastructure: cables, switches, settings of routers, et cetera. We have tried to measure switches before. But not yet on a large scale, allowing us to look for some connection between measuring and listening. Well… Until now!

The test we conducted consisted of two parts: one part includes measuring and one part includes the the – blind! – listening. Let’s go through the measurement procedure for a moment:

The measurements

The measurements were done on our new measurement setup, specifically made for switches. (well… and streamers). This consists of a CON-power CDN-T8 coupling/decoupling network to measure noise on the Ethernet cable and a TEKBOX LISN to measure power supply noise. Both devices decouple the device so we measure only the device. Both devices can also send the measured electrical noise via a 50 Ohm BNC cable to our Rigol Spectrum Analyzer and Prism dScope III audio analyzer.

In addition, we tested power supplies using our Rigol function generator. This can output all kinds of signals. In this case, we send out a 1 kHz sine wave and a 1 KHz square wave. We inject these into a power supply to see the response. To do this, the Rigol must send the signal into an amplifier (we use only 50 mV) which then amplifies the signal, so that the power supply can respond to that signal. We capture the signal and amplify it with a small, simple pre-amp (based on a TI OPA2134) after which it enters the Rigol scope (otherwise the signal would be too small). There we see the response. That way we get an idea of how clean a power supply is. The differences are remarkable. But you will see that later.

We also purchased a lab power supply and electronic load for this test to measure power supply stability. We measure within the range specified on the adapter. The electronic load can load a power supply and write down the results so that we can make a nice graph. The results are shown in the power supply tests.

We measured the switches “loaded” and unloaded. Loaded means we injected white-noise through the function generator (500mV rms, 100 Ohm output resistance). That way we can see how well the switch is isolated. We can see to what extent the ports are “leaking” and how the power supply reacts. This is important because, of course, we want as little leakage as possible. Also, we can see to what extent the “galvanic disconnection” does its job. Spoiler: they are not completely isolated. And yes: there are some big differences here, too.

What do the measurements say?

That, of course, is the big question. How much impact does ‘port-noise’ have on audio quality? And power supply noise? We’re going to find out in the listening tests by linking these results to the measurement results.

But what you see in the measurements is how solidly a switch performs in terms of isolation. We ran a few tests:

  • Switch noise – no load (no other cables connected) – how quiet is the switch?
  • Switch noise – loaded: white noise injection via the function generator – how good is port isolation?
  • Power supply noise (PSU noise) – loaded – leakage from the port through the power supply to the outside grid.
  • Power supply response – adapaters only – how well does the adapter follow the input? How clean is the signal?
  • Power supply stability – adapters only – how stable does the voltage remain with increasing load?

These measurements definitely say something about the quality of the product in terms of component selection. After all: better galvanic isolation reduces white noise injection leakage. And a neat power supply response can only be achieved with decent power adapters. The same goes for the stability of the power supply.

The listening test

The listening test was done blind on December 18, 2022. Your author made the measurements and thus did not actually listen in to exclude bias. His job was to change the switches. This was done in a – not visible to the listener on the couch – “separate” part of the room so that the listening ears could not “see” which switch was playing. You can watch the livestream back on our Youtube channel.

This test used the same ‘main cable’ each time. This network connection simply came from a standard switch in Alpha Audio’s server room. So no tweaked, audiophile-grade connection. That’s intentional, because no one has a tweaked, managed switch at home. And this test is all about the various switches and their impact. In this way, we hope to hear more of the differences between the test candidates.

Furthermore, the familiar Alpha Audio reference system was used, with the exception of the streamer. We now put in the Metrum Acoustics Ambre, instead of the AlphaPC. The reason is that the AlphaPC contains a Jcat XE network card. That’s not representative. The Ambre is a neat, affordable streamer that fits better with an average home streamer.

Type test
Multitest
Features
  Cisco SG110D-05
  • Build: Simple, small and light
  • Sound: Restless, sharp
  • Imaging: Bit flat
  • Overall: Not very nice. Bit sharp. Too expensive
  • Price: 71
  Cisco Meraki YETI
  • Build: Sturdy, bit boring.
  • Sound: Lively, airy, detailed
  • Imaging: Large, sharp image
  • Overall: Nice sounding switch. Decent upgrade
  • Price: 929
  Dlink 1210P
  • Build: Decent, sturdy
  • Sound: Restless, detailed, but sharp
  • Imaging: Okay, but not special
  • Overall: Bit sharp sounding switch. Lacks bass
  • Price: 199
  Dlink DGS-108
  • Build: Decent, no-nonsense
  • Sound: Very nice. Fluid, detailed, nice image
  • Imaging: Large and sharp image
  • Overall: Very nice switch with a decent price
  • Price: 39
  Edimax 5500G V3
  • Build: Plastic, cheap
  • Sound: Average. Nothing special. Doesn't 'click'.
  • Imaging: Bit flat and boring
  • Overall: Not a very nice product. Cheap build and dull sound
  • Price: 30
  LHY SW8
  • Build: Very nice build. Sturdy
  • Sound: Very detailed presentation. Light footed
  • Imaging: Very large image. Impressive
  • Overall: Bit fresh sounding, but loads of detail and air
  • Price: 849
  Netgear GS108E
  • Build: Sturdy, no-nonsense
  • Sound: Nice balance. Robust and coherent
  • Imaging: Stable image, tight focus
  • Overall: Very nice product. Sounds coherent and nice! Measures extremely well
  • Price: 27
  Pura Amonite
  • Build: Sturdy, industrial look
  • Sound: Airy, large image. Bit fresh and sharp sometimes.
  • Imaging: Large and airy
  • Overall: Nice product, but tad sharp sometimes.
  • Price: 1900
  TP-Link 1008D
  • Build: Plastic, cheap
  • Sound: Sounds fun! Airy, not really precise, but good.
  • Imaging: Large image. Airy
  • Overall: Weird. Sounds decent, but measures bad...
  • Price: 17.50
  TXE064
  • Build: Metal, but ugly
  • Sound: Sounds really bad via PoE ports. It's okay with normal ports.
  • Imaging: Flat.
  • Overall: Sounds bad. Measures bad.
  • Price: 39
  Delock Fiber
  • Build: Sturdy, no-nonsense
  • Sound: Tight, detailed, but bit dry
  • Imaging: Wide, but low image. Feels like 16;9 TV
  • Overall: Nice calmness, but tad dry sounding
  • Price: 90
 
 

14 Comments

  1. I have received and read to post re the testing of various Ethernet switches. The comments were useful and interesting. The switches tested all seem to be at the lower end of the what is available especially for audio networking purposes.

    It would be of more interest to me and probably other readers of the Forum if you undertook a similar evaluation of some higher quality more expensive audio network switches. I suggest the following Brands as examples. There are others
    Melco
    SOtM
    Paul Pang dual or quad
    Silent Angel Bonn Pro
    Waversa
    Renolabs

    John

    1. Dear John,

      De idea of this test was to give more insight in why switches matter. That is why we had some cheap ones, and some more expensive ones. It gave us a good idea of what really matters. We will do a test with more expensive ones and compare it to a ‘stock’ one with a good power supply. The question then is: are they really better? An if yes… why?

  2. Thanks Jaap,

    I await the next testing with considerable interest. The what is happening and then the why is it happening (if it exists) fundamental. For me there are definite listening benefits with audio rated quality switches, cables and accessories. Those with data centre enterprise networking experience say no way.
    John

  3. On fiber – I have a similar finding. Quieter but also a bit of a hard edge to the music that is uncomfortable (much like the digital glare of old). This is improved with a better fiber converter on the streamer end (Sonore Optical Module Deluxe) – but even improved, that hardness was still there. I have to think it is a noise problem of converting from optical back to copper. My testing was with a Bricasti M3 DAC with network renderer. Copper with a passive noise isolator (Network Acoustics ENO) was less quiet (blackground) than the fiber but didn’t have the hard edge.

    That said, when I upgraded to a Bricasti M21 DAC with the same network rendering technology as the M3, I could use the fiber with the Sonore module and didn’t have the hard edge. Must be better noise isolation in the M21, although the network card is supposed to be same between the two.

    1. Hi Bill,

      Thank you for sharing your insights. Fiber is something very different and the effects also differ per system. I like the quietness a lot. But there is a trade off… Martijn does not really like it.

  4. Hi,

    Thanks for the awesome switch test. I acquired an LHY SW-8 switch some weeks before your test. I use it with an R26 DAC and a Mac Mini M1 serving Roon/HQP. It brought great improvements to my system that were consistent with what you described, particularly in Martijn’s separate review of the SW-8: a broader and deeper soundstage, longer reverberations and sense of space, whilst being a more natural sound with less glare. Amazing what a good switch can do.

    So just last week based on your test findings & recommendations as an experiment I got a Netgear GS108E switch which I paired with a spare Ifi Power X PS and placed in series with the LHY I.e. Generic router > Netgear GS108E > LHY > R26.

    Wow. The soundstage focus tightened considerably with a seemingly quieter background, more detail and improved micro-dynamics, a bit leaner, but still retaining the characteristic LHY natural and easeful sound.

    Flicking back to the LHY alone it was warmer and smoother with a more generous if slightly bloomed bass. It took a few minutes to get used to the shift – the two switch combo is definitely more resolving and feels balanced so is staying. Oh, and adding a small grounding tube to the Netgear further relaxed and focussed the sound (doing this with the LHY had a negative effect, perhaps due to its already well sorted power section).

    I am constantly amazed how upstream improvements in the digital chain, even seemingly minor ones, can seriously and cumulatively improve a streamer’s & DAC’s performance.

    Will try adding FMCs next, further down into the rabbit hole I go! 😅

    Keep up the great work chaps.

    Jake

    1. Jake, thank you for posting your experience! I was just about ready to pull the trigger on another SW-8. I have one on order, but looking at my network, a few more connections are needed. I was lucky to stumble upon this switch shootout and just ordered the Netgear router based upon your experience, to use with the SW-8. We are running similar setups, so I am hoping my experience is similar to yours. Thanks again! You saved me $500+. 😉

Leave a Reply

}