|Remember folks, this is what an "ethernet frame" looks like. It's data. There is no "music" in there until decoded and prepared for the DAC! Notice the CRC bytes for error detection. (The numbers represents how many bytes for each segment.)|
0. PreambleHey folks, over the years, I have been critical of high-end audio cables... Previously, I have shown that RCA analog interconnects can result in measurable differences with channel crosstalk changes with long lengths. But the digital interconnects themselves do not result in measurable differences even in terms of jitter (TosLink SPDIF, coaxial SPDIF, or USB). Although my HDMI receiver's DAC isn't as accurate or jitter-free, different HDMI cables don't seem to make any measurable difference either. The only caveat to this being that a digital cable can just plain fail, in which case the distortion to the audio signal has a particular (annoying) characteristic which is clearly audible and not a subtle change (eg. a poor USB cable sound).
So far, I have not seen any further measurements to suggest my conclusions are inaccurate. I have seen audiophile reviewers and forum posters still claim digital cables make an audible difference and when questioned they provide lots of words but no actual empirical evidence. It has been awhile since I've seen any articles claiming objective evidence for cable measurements - haven't come across new ads or audiophile articles although of course I may have missed some.
However, as computer audio expands, there will be opportunities to "brand" more hardware as somehow "audiophile approved" and companies that make audio cables likewise will naturally capitalize on new lines of interconnects / cables... And as expected, cost of these things will be commensurate with "premium" products.
Which brings us to the concept of "audiophile ethernet cables" (see here also, and recent mainstream press exposure of the "madness"). Let me be clear. If I have issues with USB cables, or SPDIF cables, making any significant contribution to audible sound quality (assuming again essentially error-free transmission of data), there is no rational explanation whatsoever that ethernet cables should make any difference. The TCP/IP protocol has error correction mechanisms that allow for worldwide transmission integrity (otherwise Internet financial transactions should be banned!), and is asynchronous so there is no temporal dependence on exact timing mechanisms (jitter not an issue with adequate buffer to reclock and feed the DAC). So long as the "protocol stack" is functioning as it should between the devices, there will not be any issue. Systematic errors causing audible distortion either means hardware failure or poorly implemented communication software. Therefore the expectation if we were to test or "listen to" different ethernet cables is that there would be no difference.
Since I like to make sure objectively, let us at least run a few tests to see if indeed evidence can be found to support the hypothesis.
I. Test SetupFirst, we must decide where to place the ethernet cables to test... You see, in any server/streamer system, we expect that there would be a few network cables in the data path. For the sake of ease in measurements and assuming the same thing as audiophile beliefs in power cables, let us place the test cables as the last leg of the data path between the streamer and the last ethernet switch (this guy also thinks the last leg is important). Here then is my setup:
Server PC <--> 6' Cat 6 UTP patch cable <--> 20' Cat-6 Gigabit generic cable (in wall) <--> NETGEAR Nighthawk R7000 router <--> 30' Cat-6A STP Gigabit cable (in wall) <--> 6' Cat 6A STP patch cable <--> Test switch <--> Test cable <--> Logitech Transporter streamer
As you can see above, if we trace the route the data takes between server and streaming device, we're usually looking at quite a bit of cable! In a typical "wired" house, much of the cable exists in the wall and would not be amendable to easy rewiring. Since I just did some renovations last year, I made sure to run high quality Cat 6A STP from router to the sound/media room. I am going to not just test a few cables, but I'm also going to try a different ethernet switch! Here are some details:
Server PC: AMD A10-5800K quad core PC, stock 3.8GHz speed, running Windows Server 2012 R2, Logitech Media Server 7.9.0 build 1420794485 [Jan 12, 2015], 16GB DDR3 RAM, built-in Realtek PCIe ASUS motherboard gigabit ethernet interface.
NETGEAR Nighthawk R7000 router: running dd-wrt.com custom firmware "kongac build 24345". Very stable with >100days uptime currently, underclocked to 800MHz just because I never needed the 1GHz speed.
Streamer/DAC device is the venerable Logitech Transporter. Remember that the Transporter only runs at 100Mbps whereas the rest of the system is capable of gigabit (1000Mbps) speeds.
The "Test switches": for the most part, I will use the inexpensive gigabit TP-LINK TL-SG1008D which I bought at a local computer store slightly more than a year ago (<$30). It's got 8 ports and fast enough for 100MB/sec (that's 100 megabytes/sec) file transfer through it from server to my HTPC:
|The white thing underneath is just my LP carbon fibre brush to lift it a little to photograph the front easier. Ahem... Pardon the dust... :-)|
In comparison, for a couple of the tests I will use this little guy:
A LinkPro SOHOHub 4-port 10/100Mbps switch which I believe is about 10 years old (the TC3097-8 interface controller inside came out around 1998). I found it in the attic of the house I bought, powered by a Canon AC adaptor which provided adequate juice.
For both these switches, I will keep my HTPC computer connected to one of the other ports.
The "Test cables":
So, I rummaged around my pile of computer parts and found these cables to test. Note that I was shopping at NCIX.com when I was doing some renovations and getting my network system up. I "standardized" on some rather inexpensive Cat 6A cables on sale there - hence the nGear brand which they carried.
The top picture, from the left we have a 1-foot length of Cat 6A STP (<$3.00) - presumably the "best" cable given the short length and excellent shielding. Note that Shielded Twisted Pair (STP) cables are not necessarily better than UTP (Unshielded...); one must make sure the shield is properly connected at each end. Next we have presumably the "worst" cable of the bunch - a generic "freebie" 3-foot length of Cat 5E UTP patch cable that has been sitting around for the last 5 years in my pile of parts. The blue plastic jacket is loose and quality so flimsy that I can probably pull it apart easily without much strength needed. Then we have a 10-foot length of Cat 6A (<$6.00), and finally, a much longer 50-feet length of Cat 6A STP (~$15.00). Cables from the same brand will allow us to see if length makes a difference.
The green cable in the lower picture was one I found in my office. It's a 12-year old 20-feet generic Cat 6 UTP cable that has been in daily use for the last 12 years... I guess you can call it "burned in"!
Sorry folks, I don't have any Cat 7 cables here. At this point, I don't see any reason to use these since I'm only running a 1 Gbps network. Anyone out there running a 10 Gbps network at home requiring Cat 7 cables? Realize that even Cat 6 is potentially capable of 10 Gbps up to 50m (>160 feet) or so.
I will measure with RightMark (newest 6.4.1) to look at the usual dynamic range, noise floor, distortion along with the Dunn J-Test signal to see if there's any evidence of jitter anomaly in the Transporter's RCA DAC output (rather than the XLR for the sake of convenience). Some well shielded 6' RadioShack interconnects used (Cable C here). As usual, my E-MU 0404USB device was used for the measurements. All measurements done in 24/96 (high resolution) or 24/48 for the jitter test.
Let the measurements begin...
II. RightMark Audio Analyzer (24/96)Here's the summary table of all results with 5 cables with the TP-LINK gigabit switch and 2 other measurements with the old 100Mbps LinkPro switch:
As you can see, there are no significant differences in the audio output at all. Analogue output was measured all the way to 48kHz - well beyond the audible spectrum. It didn't matter whether the cable was 1-foot all the way to 50-feet. Likewise, Cat 5E, Cat 6, Cat 6A, UTP or STP made no difference whatsoever. In 2 of the tests (50' CAT 6A & 3' CAT 5E + LinkPro), I was playing 20Mbps 1080P MKV video concurrently on the HTPC connected to the switch to increase the data rate coming from the server - no difference in background noise or anything else.
A few graphs from which the calculated data were derived:
|Frequency Response: Exact overlay.|
|Noise level: slight 60Hz hum measured down at -115dB, everything else even further below this.|
|IMD: Again, essentially perfect overlay with the different cables.|
|Stereo Crosstalk: Would be very bizarre to see any anomaly here!|
III. J-Test (24-bit)Instead of showing 7 individual J-Test graphs, I decided to overlay each one to create a composite image:
As you can see, there is some normal variability in the noise floor around the 12kHz primary frequency but otherwise, nothing sticks out. There's some low-level jitter around 12kHz, some of which I'm sure related to the E-MU device itself rather than just the Transporter.
No evidence that any of the cables / switch changes resulted in any anomaly using the 24-bit Dunn jitter test. None of the sidebands exceeded -110dB from the primary frequency peak at 12kHz. Note that the peak itself is at -3dBFS, but I measured it a bit lower to avoid the use of the E-MU's input amplifier which would add some noise. Again, no change observed (ie. worsening of noise floor or stimulated jitter sidebands) even when the HTPC was concurrently streaming a 20Mbps movie from the server.
IV. Summary / ConclusionI believe if there indeed is an ethernet audio device that "sounds different" because of different cables being used, then that device should be returned because it is obviously defective. Remember folks, it is like accepting that the earth is spherical or that 2+2=4 - because that's just the way it is. Ethernet communication is an engineered system, the parameters and capabilities of this system is not only understood but designed to be the way it is by humans! You really cannot claim to have "discovered" some combination of dielectric or conductor or geometry that works "better" within an already errorless digital system unless you're claiming improved performance outside technical recommendations (in the case of Cat 6 for gigabit networks, it's 100m or 328 feet lengths within a reasonable ambient electrical noise environment).
It's also worth remembering that audio data bitrates are quite low. Today, I hope nobody is running anything slower than 100Mbps "fast ethernet". Although my music is generally streamed out as compressed FLAC, even if you stream uncompressed WAV files, standard stereo 16/44 CD-quality audio requires <1.5Mbps, 24/96 requires ~4.6Mbps, and stereo 24/192 ~9.2Mbps. Even if we went uncompressed multichannel, 5.1 24/96 would only use up <14Mbps. Considering how cheap gigabit (1000Mbps) networks are, there's no reason not to build upon the gigabit standard these days. There's generally no reason to complain about decent Cat 5E cabling, but splurging a little on Cat 6+ isn't a big deal. The Transporter device used in these tests is almost 10 years old at this point and limited to 100Mbps. I would certainly be surprised and disappointed if a modern audio streaming device measured differently with various cables these days with even faster ethernet interface hardware!
Ultimately, I'm not suggesting anyone use the cheapest ethernet cable he/she can find. If you like the esthetics and build construction, go for it! Just realize that it's essentially impossible to argue that a functioning (free of data transmission error) ethernet cable will "sound" any different or worthy of significant cost differential based on sonic quality. The idea of specialized "audiophile" ethernet cables (or "ethernet switches" for that matter) is plain nonsense.
For the record, subjectively, I have never heard a difference between ethernet cables on my system. For fun I did have a listen to Shelby Lynne's Just A Little Lovin' (2012 Analogue Productions SACD ripped to 24/88) - sounded great to me even with the Cat 5E freebie cable and cheap LinkPro switch while a 20Mbps movie was playing off my HTPC. I have never tried those expensive cables from AudioQuest or Chord, but seriously, why bother when there's no logical rationale based on understanding of an engineered product and the lack of empirical evidence? Must a person try out or demo every claim made or testimonial uttered when some things are self-evident? Must I double check when someone comes up to me and tells me the world is flat or the sun rises in the west? Should I also try Snake Oil if someone in a crowd around the traveling salesman yelled out that it "Works for me!" without any other evidence?
Well, it looks like Chord got their hands slapped for claims about sound quality with their ethernet cable ads determined to be "misleading advertising", lacking in "substantiation", and "exaggeration" in November 2014. Bravo to the UK's Advertising Standards Authority. Truth is important.
Bottom line: There's no evidence that any of the digital cables make an audible difference be it TosLink, coaxial, USB, or now ethernet within an error-free system.**
As usual, if anyone feels that I am in error, please demonstrate and leave a link to the evidence.
Okay... Now I really have to go do some real work :-). Enjoy the music!
** I was reminded about this post I made using the Squeezebox Touch and EDO plug-in awhile back. In it, I was able to demonstrate measurable differences using an unshielded cheap 3' "zip-chord" RCA cable instead of a proper coaxial cable (I'm sure it's nothing close to the 75-ohm impedance spec). It is a reminder that we of course should be using *proper* cabling and that extreme situations like in this post will allow demonstration of noise phenomena that otherwise would be highly unlikely. Notice also how this poor RCA cable degraded sound quality when pushed to 24/192 which is also outside the usual Squeezebox Touch specification but available thanks to the Triode plugin.