The setup
Contents
The following setup was made in the listening room in Haarlem:
- All computers, the Raspberry Pi, the Shuttle PC and the Gigabyte PC, were plugged into an unfiltered Supra socket.
- All computers are connected to a DLink DGS108 switch using the same standard Cat6a copper network cable.
- The network cable of the Alpha Audio network was also plugged into the DLink switch. This part of the network is decoupled with a fibre network to the back-end systems, but the connection to the DLink is a regular network cable.
- The Dlink DGS108 switch is powered by an iFi iPowerX adapter plugged into the same Supra socket.
- The Volumio Motivo is used as a streaming bridge. Audirvana is used to stream to the Motivo using UPnP. The Motivo was chosen because it performs very well with its own adapter and the digital out is clean as well.
- The Motivo is connected to the DLink switch using the same Cat6a cable.
- The Sonnet Pasithea DAC is connected directly to the Motivo on the SPDIF input (coax).
- The Mutec was not used as an interface between the Motivo and the Pasithea DAC. The Mutec can mask the influence of the computers, OS and their power supplies by reclocking the signal and we do not want that for this test.
Choices made in how we test
A conscious decision was made to use Audirvana through UPnP and not connect the Pasithea DAC directly to the computers with a USB cable. The quality of the USB output on the different computers varies and thus becomes a variable. We want to exclude that variable. Ethernet ports are also of influence, but that influence is neutralised equally for all computers with the DLink switch.
The clever reader will now note that the same applies to USB ports and using the Mutec as in intermediary. That is true. However, reclocking the audio stream, as the Mutec performs, is not desirable for this test.
Despite the Volumio Motivo being an interlink between the computers and the Pasithea DAC, the impact or influence of the Motivo is the same for all connected computers. The Motivo itself does not have to do much, Audirvāna passes on the stream as RAW PCM, a format the DAC can read directly. The Motivo only has to pass the RAW PCM signal through the SPDIF connection.
How we prepared this test
We used the Shuttle PC and Gigabyte PC because they are available at Alpha Audio and do get used for tests quite often. Martijn did the preparatory work at home and came up with the test setup. Jaap hears in Haarlem what has been prepared for this test for the first time. Martijn has heard the music and computers (ad nauseam) at home, but hears everything for the first time in the set in Haarlem. We combined our notes for this article.
Music used
We use the following tracks, all played from the computer’s SD card or hard disk:
- Mammal Hands – Labyrinth
- Bell X1 – The Upswing
- Monteverdi Choir conducted by John Eliot Gardiner – Brahms: Gesänge, op.104: No. 1 Nachtwache
Hi,
When I started USB Async (UAC 1.1) back in 2003 I thought here it is digital freedom from jitter and other problems with SPDIF. Was I more wrong.
In 2010 I was at Rocky Mountain and a bunch of us were talking about why software sounds different. When I returned to the office that next week I cleared my schedule and put the following test setup:
MacBook Pro (with bootcamp Windows)USB Analyzer | both a Tektronix 4K series with USB and I2S plug ins and a Beagle Protocol Analyzer | Wavelength Wavestream module (basically a test set XMOS for companies using my USB to I2S software) with an I2S header.
Using flat WAV files (just made it easier to see samples) I ran that into the test set captured the USB and I2S and compared it to the original file. Actually told a few companies they were not bit true, but they had fixes really quick.
So after I had the new software and I determined everything was bit true I looked at Jitter. Ok well my Symetricon didn’t really show any difference in jitter on the I2S. I used my modified Stanford SR770 (good for noise testing down to 10nVrms at 1Hz) and saw a little difference in the noise on the power rails even though the unit was self powered. This was one of the reason’s at that point I started to think of isolation in my products.
I swapped out the Wavestream for several dacs I had on hand from my company and others I had done software with in the lab system and tried a bunch of different software. I then ask Atkinson and Charlie Hansen (RIP) from Ayre. Both of them said your nuts it could be anything.
I took the setup down and went on my way, not really happy with the outcome.
Years later I had this problem were my fan on my MacBook Pro was spinning like crazy. I asked my buddies at Apple and they suggested I load iStat Menus which keeps track of everything, super utility. Found a rogue device driver for my battery backup was taking up one of my CPU’s to 100% deleted it, reboot and everything was fine.
I was designing the AudioQuest Cobalt and I was listening to music on my AKG702 and was happy with what I heard but I went back through the software wheel and then it hit me!!! Products like Audirvana pre-load and decode the file into memory and then when interrupted by the system it just returns with a pointer to the next block of samples. Audirvana system usage 0.3%. I set the default sample rate for the file (88.2) and ran iTunes and it was 43% ok maybe we have something here.
I went through all the software and go some interesting numbers. Then I said ok, JTEST to the Prism dScope III with the Cobalt. Sure enough some differences. I have a break out USB board by Tektronix for testing USB and on that is VBUS and Ground so I hooked the Standford up and could see a difference in the power supply noise to the Cobalt.
Ok but why a difference when I use a self powered or even an isolated USB with reclocker? These computers have a boat load of clocks and switching supplies regardless of who makes them. I started measuring noise on the mains and sure enough the % of cpu usage and noise correlated. That noise goes back to the panel and into the system amps, preamps and so forth.
Yes isolation helps really well but there are other things I learned.
1) WIFI which we knows is really a bad network connection with audio in mind.
2) Your library interface should never be the same as your dac interface. Your basically doubling up on the transmission and reception and it’s all synchronous. Not good!
3) Memory the more you have the better off you are!
4) Laptops better sounding than desktops. FCC standards for laptops are far better than desktops and also the engineering going into laptops over desktops is almost 10:1. Mainly for energy usage which from above can be a real difference.
5) The OS makes a difference but the application makes more.
6) Sample rate… funny but upsampling to really high rates is not a good idea. The dac chips really don’t sound better at 768 compared to 88.2 or 96. I think that is heat related plus how much work the CPU (yes they are processors now) can do between samples!
The big thing is…. computer audio is not a slam dunk. Trial and error can really make a big difference in the sound.
Have fun, and thanks great topic!
Gordon
Wavelength Audio, ltd.
Hi Gordon,
Thank you for that long comment, much appreciated. I edited it slightly for better readability by adding some white lines in between. I think that will help more readers to actually read what you wrote 🙂
There’s one thing you say that caught my eye:
“Your library interface should never be the same as your dac interface.”
How should I interpret that, if you have a computer networked through ethernet where Audirvana is running and a streamer/DAC that is connected on the network as well, and you use UPnP to stream from Audirvana to the streamer/DAC?
Correct, in that case you have 2 synchronous streams running against each other. This would not be an optimal setup. A better one would be local storage and then stream via Ethernet. Or USB DAC and pull from files or server over Ethernet. WIFI of course has 4x the traffic so I don’t suggest that to users.
Thanks,
Gordon
I´m getting the feeling that you are concluding, or working from the hypothesis, that noise in the digital domain is mainly the problem when it comes to the variation in sound quality in the digital domain. The source of this noise could come from many different places, even software generated CPU noise, right?
What i find interesting is then the talk about taste, in this matter. Taste is of course taste… and we can´t change what people like and perceive. But at the same time, the best sound should be created if we have NO noise getting into to the DAC/Clock, from what i understand. (even if there always will be some level of noise) Then the DAC conversion has the best possibility to recreate that digital feed exactly correct as it was captured in the ADC process.
My thought is that we should at all cost (if we want or care) remove the noise in the digital domain and this should ideally never be driven from matter of taste. Otherwise it feels a bit like saying that i prefer that my vinyl records and stylist is dirty and therefore i keep them dirty, which is of course ok, but not something we want to suggest doing because you like it.
If we want to match the sound with our taste, that should ideally be done somewhere on the analogue side, at least not before the DAC, in my mind.
Just some philosophical thoughts 😀
If there’s something noisy, both mechanical and electrical, it is vinyl. The noise floor is significant higher than any digital piece of equipment. With vinyl, we know that changing tone arms, elements, the material on the platter and pucks influences the sound. We never debate if that’s ’as intended’ or not.
We once did a live stream where we compared an unmixed digital recording with a direct to disk one in the Artone studio, featuring Tim Knol. I’d recommend reading / watching that one to get a feel for what is going on. Spoiler alert: the analogue recording sounded so much better, more like reality than the digital one: https://www.alpha-audio.net/background/digital-versus-analogue-a-true-and-fair-comparison/
In the digital domain, noise is something which introduces problems in the time domain, also known as jitter. There are different forms of jitter, but the most important one is where the reconstruction of the samples is not spot on. Hans Beekhuyzen has made several videos about this, explaining why it matters and what is happening. You can find those on his YouTube channel.
So, as with vinyl, there’s more than just noise which drives the tonal output of a digital signal, where taste matters. There’s not any magic going on with digital recordings or digital playback where everything would sound equal if it was devoid of any form of disturbance. Just as there isn’t with vinyl or tape (reels).
Thanks!
But the difference with digital is that it is always perfect. You can only degrade the sound on the digital side and never make something that is perfect even more perfect.
I agree that it is very simple and no magic: -Just keep the digital signal as clean as you possibly can, then your DAC will perform at its best, to its limitations.
This is similar to Vinyl, in concept. The pickup can only pickup what is in the groove and never make it better than the quality of that Vinyl. You can only degrade what is already there, not make it even better.
Yes, i have watched all Hans videos and i fully agree with them.
“ But the difference with digital is that it is always perfect.”. I don’t know what perfect means in this context, but it is not true. Just as with analogue, a lot can go wrong in the recording process and the playback process.
Marketing in the CD age has put the idea that it is infallible in our heads, but it is simply marketing and not the truth.
Yes, i agree that a lot can go wrong in the conversion steps. That is where the problem is. If someone is saying that they prefer the sound when there is more measured noise on the clock, as you have shown, then i would argue that it is similar to saying that you prefer the sound when the Vinyl is dirty, which people normally don´t do.
It´s not actually a matter of taste what happens on the digital side just as it is not what happens on the playback medium/hardware when it is different levels of dirty.
I think many people are so used to the digital sound that they think it should sound like that and therefore sometimes say that it doesn´t make a sound improvement, or that they prefer something, even though it actually is less accurately reproduced in the DAC step. This is more of a fact then a taste thing, i am thinking.
You can also “split the processing” among the cpu cores of a single computer, which is what I’ve been doing with a 8Gb RPi4 running Audiolinux as a Roon endpoint. Starting with Allo’s SBC-based DigiOne and USBridge transports six years ago, I’ve played with several stripped-down OSs (DietPi, Volumio, PiCore Player, VitOS and, now, Audiolinux) and different players (Roon, AV, Squeezelite) to find what sounds best to me through my Meridian active speakers.
Isolated cores can make a difference.
Thank you for that addition.
Amazing work Martijn and very interesting observations! As you have alluded to earlier I really like that you take the approach of “not knowing anything” when going into your tests and don´t assume anything. I make some assumptions my-self but its good that you don´t!
Maybe this question is out of topic, but i can´t get away from wondering what the plug´n-play streamer sounds like in comparison with fiddling around with these software´s yourself. I get that the Raspberry Pi is much cheaper, which is an interesting prospect if it sounds similar, or better, than the Motivo on its own, but does it?
Is it worth going down the software rabbit whole, from a sound quality perspective, if you already own a streamer, similar to the Motivo, with good power supply for example?
To make my self more clear. Let´s say i use the Tidal App in Motivo which is not connected to any other computing device in the signal chain. Is it then worth exploring other software alternatives using a Raspberry Pi instead, if all the connected hardware in the signal chain is the same? (again, from a pure sound quality perspective)
Hi Tobias. Good questions.
First of all, we don’t create our own streamer in this test. We add a music server to the chain, and still use a streamer. We don’t use much of the Volumio software in this case, but we still do and utilise all the work Volumio has put into designing the signal chain within the Motivo.
We could have used the Primare NP5 Mk2 instead of the Motivo, or the Volumio Rivo, but those streaming bridges would need a very good power supply, the Dodo in our case, to prevent them from becoming the weakest link in the chain of this test and thus influence the outcome.
We needed the Dodo power supply for the test setup. That’s why we chose the Motivo: we know it is optimal with its own power supply and it is a very high quality streamer. I think a lot of people underestimate how good it is.
To return to the question: comparing the Motivo on one hand and a Raspberry Pi running Audirvana Core Player with the Motivo on the other hand, will result in a different sound signature, because of Audirvana software. If you compare like this, it will be hard to tell what attributes to the sound and draw any conclusion.
But what would happen if we add a Raspberry Pi running Volumio software next to the Motivo and compare to using the Motivo stand-alone? Because that would be the right test to see what the effect of adding hardware and splitting tasks across computers will add. I don’t know the answer.
I will try to see if I can use the Volumio software on a Pi to stream to a Volumio device, just like Audirvana can. I’m not sure if the Volumio software will be able to work this way, but there’s only one way to find out! If we compare like this, we would be able to conclude something. That will be a fun test, thank you for bringing up the suggestion.
If you look at the high end market, to for example Antipodes music servers: they use different computers for different functions: storing music files, running the music library software and processing the sound.
Also Grimm uses two computers in their players: one running Roon and one for the processing of sound. Of course, these expensive machines have a lot of custom hardware, it is not just tying together a few computers and a streaming board. From visiting Grimm (to be found in the archives of this website) we know they research and engineer a lot when creating their players. But, given that manufacturers like these split the music processing across different computers, each dedicated to specific tasks, you can deduct that doing this yourself has added benefits. But you also introduce new sources of noise as well, so it is not that easy to do it right.
Thanks for the detailed answer. I really had concentrate and read several times what you where saying since the complication of the noise problem is really intricate and all the variables involved makes it hard to compare apples with apples, i see that now. Thanks again for your work for the audio community!
I’ve tried to see if I can stream from a Pi running the Volumio OS to a Volumio Motivo. The answer is ‘no, you cannot’. Volumio OS cannot send a signal to another device using UPnP (like Audirvana or any other music player communicating over UPnP, being the UPnP server), it can only work as a UPnP renderer (meaning being the receiver of music streamed through UPnP).
When connecting on USB, the Motivo isn’t seen as a USB DAC by the Volumio OS running on the Pi.
So, unfortunately, I cannot test this scenario.