A novel way to massively improve the SQ of computer audio streaming

[Superdad]

17 minutes ago, austinpop said:

I have high hopes for the EtherRegen. It would be great if it simplified network topology and improved SQ.

 

You're not the only one with high hopes my friend!   It darn well better do both 'cause between development and first production run close to 6 figures in $$ will have gone into the launch...

[Dev]

1 hour ago, elan120 said:

I bridged my setup in the following way.  Not sure how close this is to your bridged topology, but the impression is similar where the bridged configuration sounds better.

 

 

I have a similar setup with JCAT and bridged topology ever since I started using the NUC couple of months back. Interesting that you are able to mod the JCAT with sCLK-Ex. Did Sotm did the mod ? Btw, which mobo are you using to host the JCAT PCIe ?

[austinpop]

56 minutes ago, ray-dude said:

 

Thank you Rajiv!  I'll snag one to play with as I work on WiFi/Ethernet bridging (frustratingly close)

 

To your question about why it makes a difference, does anyone have a sense of what happens on bridged segments at the physical layer?  That is, is the only electrical signal that gets sent on the bridged port packets intended for the the device on the other side of the bridge?

 

Yes, but this is true of both hardware switches and software switches - the bridge0 interface created in Linux, for example, is nothing more than a software (or virtual) switch. On the Ethernet (layer 2), they're not packets, but are instead called frames.

 

Switches use Address Learning to learn which MAC addresses reside on which port of the switch, and then they use Traffic Filtering to only forward frames to a port if the destination MAC address is a match in its learned address table. See https://www.oreilly.com/library/view/ethernet-switches/9781449367299/ch01.html for details.

 

Of course, things get more complicated when switches are connected to other switches, when other mechanisms like spanning trees come into play.

 

Still, the bottom line is - whether connected to a software switch (what we on this thread call bridging) or to a regular switch, after some initial learning, the Ethernet adapter on the endpoint should only see frames directly addressed to it, multicast frames on a group to which it belongs, and broadcast frames.

 

Finally, keep in mind that even if the occasional frame arrives at the Ethernet adapter of the endpoint that is not addressed to it, the adapter will not raise an interrupt with the CPU, but instead silently discard that frame. 

 

56 minutes ago, ray-dude said:

Pure speculation, but our experiments with large buffers with SqueezeLite on the end point NUC clearly points to ethernet traffic having a very noticeable impact on SQ on the end point.  My hypothesis (untested) is that the end point responding to interrupts from the ethernet controller is somehow impacting the connection to the DAC over USB (data and audio have different messaging interfaces in USB, so there may be some contention in the USB controller?)

 

There is some network-related mechanism underlying the SQ difference we hear with large buffers.  To narrow it down, on my near infinite "to test" list is see if copying a large file to the large buffer SL end point during playback causes SQ to fall back to what we heard with small buffers.  If it does. next test would be to see if pushing heaving traffic across a shared switch (non-bridge config) impacts SQ on a large buffer SL end point connected to that switch.  

 

Alas, I'm a couple weeks away from having time for the next series of tests, but it is great to read that bridging still has an impact with the NUCs.

 

Let's be careful here. It's tempting to go down a rabbit hole speculating about causality. As per my own OP guidelines, I'm going to curtail that discussion in this thread. However, if you do try listening experiments as you suggest above, by all means, please report the results here!

[the_doc735]

5 minutes ago, Dev said:

 

I have a similar setup with JCAT and bridged topology ever since I started using the NUC couple of months back. Interesting that you are able to mod the JCAT with sCLK-Ex. Did Sotm did the mod ? Btw, which mobo are you using to host the JCAT PCIe ?

what is your definition of bridging/non  bridging?

 

i am thinking of something like this CISCO Linksys WES610N

 

cheers! 

[austinpop]

25 minutes ago, Superdad said:

 

You're not the only one with high hopes my friend!   It darn well better do both 'cause between development and first production run close to 6 figures in $$ will have gone into the launch...

 

Godspeed, my fine, feathered friend!

[Dev]

1 minute ago, the_doc735 said:

what is your definition of bridging?

 

see the post above you...its reasonably well explained 😀

[austinpop]

1 hour ago, Dutch said:

@austinpop, @elan120, both of you are using an optimized switch in the path music server - endpoint according to your diagrams. Did you try it without (not asking you to, just curious) or do you have the experience from a previous setup that the switch still helps things forward. I did see Romaz writing the same so it would not surprise me but I’d rather not make assumptions.

 

Thanks for sharing your experiences BTW, very helpful and interesting!

 

Hi @Dutch

 

I always encourage people - most of all, myself! - to revisit their optimizations from time to time, just to confirm that they are still beneficial in their current setup.

 

Your question came at a fortunate time where I could easily switch alternate between the direct connection from NUC to endpoint vs. with the optimized switch in the path. I just did the experiment.

 

Yes, the TLS OCXO switch still makes a noticeable improvement in the SQ.

[elan120]

1 hour ago, Dutch said:

@austinpop, @elan120, both of you are using an optimized switch in the path music server - endpoint according to your diagrams. Did you try it without (not asking you to, just curious) or do you have the experience from a previous setup that the switch still helps things forward. I did see Romaz writing the same so it would not surprise me but I’d rather not make assumptions.

 

Thanks for sharing your experiences BTW, very helpful and interesting!

I started with no OXCO clocked switch when I just starting to test AL loaded NUC endpoint, and was only concentrate on finding the best configuration with this new setup.  Once things are more settled, I added the switch in place as well as the 3-way bridging, and it was an obvious improvement, so after couple times of switching back and forth, switch is now stay in the chain.

[elan120]

1 hour ago, rickca said:

Beautiful setup.  My question is whether the EtherREGEN will reduce the need for such a sophisticated server.  I'm hoping.

 

1 hour ago, austinpop said:

 

I have high hopes for the EtherRegen. It would be great if it simplified network topology and improved SQ.

 

If. 🤞

 

58 minutes ago, Superdad said:

 

You're not the only one with high hopes my friend!   It darn well better do both 'cause between development and first production run close to 6 figures in $$ will have gone into the launch...

 

I am also have very high hope this switch will make elaborated server a thing in the past, although it will make my current server setup less valuable, but will make any future setup simpler.  I am rooting for the success of EtherRegen when it is ready.

 

[elan120]

45 minutes ago, Dev said:

 

I have a similar setup with JCAT and bridged topology ever since I started using the NUC couple of months back. Interesting that you are able to mod the JCAT with sCLK-Ex. Did Sotm did the mod ? Btw, which mobo are you using to host the JCAT PCIe ?

I am using X399 Taichi mobo, and took the easy way out on this mod by sending them including the JCAT to SOtM for the clock mod instead of doing the mod myself.  Since I have unused output ports available on the sCLK-OCX10, I also connected it to JCAT Net Card Femto as well.

[seatrope]

Other potential thoughts on adding additional interfaces to the nuc7i7DNBE board:

 

there are 2 M.2 slots which are unused... one the standard storage key M and one for the WiFi card with key E. We could use this to the key E slot:

http://www.commell.com.tw/Product/Peripheral/M.2 (NGFF) card/M2-210.htm

 

Or

https://www.amazon.com/Ableconn-M2-NW-107-Gigabit-Ethernet-Module/dp/B07599LZ5S

 

and potentially replace the clocks on these. Might work better keeping the USB bus to audio purposes only?

 

[User471]

On 1/17/2019 at 8:01 PM, davide256 said:

Has anyone found a device with a better  hardware implementation than NUC which can be used with AL as Roon endpoint?  I do find it freaky/worrisome that just removing the USB  drive after boot to RAM makes such a difference. And intriguing that an RPI can come so close to  Pentium NUC as endpoint. Still hoping that Sonore makes AL an option for xRendu...

 

If you have the skills it's very straightforward to build one

 

So for example I have an MSI mini-ITX board in a passively cooled HDPlex case. The board takes its power from a Teddy Pardo PSU directly to mobo, and I have a SoTM USB PCI-E card with its own Teddy Pardo (so it takes only data from the mobo via an internal header, not power)

 

If you want to add tricked out Ethernet cards or even SFP slots it's all possible (although my particular case has only one back slot, bigger silent options are available)

 

The trickiest bit was finding the correct adapter between the TP cable and the mobo DC in

 

AL extreme in RAM sound miles better than the previous generic Linux I used to run

[Dutch]

I believe many of us were using DIY built computers using a range of components before trying the NUC’s. Many used LPS’s to power the whole computer or multiple to power separate parts. Some used high end clocks replacing the out of the box ones, low power CPU’s, SLC SSD’s, etc. etc.

 

There’s no silver bullets I’m afraid. For us, end consumers, it’s a ‘hit or miss’ thing. Companies like Innuos have stated they’ve tested many different components and selected the best sounding. This is of course a (too?) time consuming and expensive venture for us. Perhaps the thing that got closest to centrally designed solution comparing different parts was the CAPS initiative by Computer Audiophile.

[davide256]

6 hours ago, User471 said:

 

If you have the skills it's very straightforward to build one

 

So for example I have an MSI mini-ITX board in a passively cooled HDPlex case. The board takes its power from a Teddy Pardo PSU directly to mobo, and I have a SoTM USB PCI-E card with its own Teddy Pardo (so it takes only data from the mobo via an internal header, not power)

 

If you want to add tricked out Ethernet cards or even SFP slots it's all possible (although my particular case has only one back slot, bigger silent options are available)

 

The trickiest bit was finding the correct adapter between the TP cable and the mobo DC in

 

AL extreme in RAM sound miles better than the previous generic Linux I used to run

I guess what I'd like to see is a comparison of a PC using something like a JCAT USB  FEMTO as endpoint machine with AL vs AL/NUC endpoint using built in USB. I don't see anything special about the hardware quality in an endpoint NUC,  just good hardware integration. My suspicion is that an independently powered well made USB card in a decent endpoint machine will best NUC.  Hopefully we can get a CA member who already has one of these or an alternative to give it a shot and report in.

[RickyV]

I believe, but not 100% sure, that with the nuc 7i7dnbe usb data lines go straight to the cpu. With my mini itx board they first go to a xmos chip and then to the connector.

With the the jcat usb card you go also, via pci lanes,  to a xmos chip but at least it clean powered and clocked.

[Johnseye]

40 minutes ago, davide256 said:

I guess what I'd like to see is a comparison of a PC using something like a JCAT USB  FEMTO as endpoint machine with AL vs AL/NUC endpoint using built in USB. I don't see anything special about the hardware quality in an endpoint NUC,  just good hardware integration. My suspicion is that an independently powered well made USB card in a decent endpoint machine will best NUC.  Hopefully we can get a CA member who already has one of these or an alternative to give it a shot and report in.

 

Done and done with SOtM hardware. The NUC sounds better. Look up my previous posts in this thread. 

[auricgoldfinger]

25 minutes ago, Johnseye said:

 

Done and done with SOtM hardware. The NUC sounds better. Look up my previous posts in this thread. 

 

Are you using your REF10 for anything now?

[Johnseye]

2 minutes ago, auricgoldfinger said:

 

Are you using your REF10 for anything now?

 

Yes, the Core server for system and ethernet clocks. I also have the SOtM switch on loan so it's clicking that. 

 

When Uptone's switch comes out I'll likely buy that and it has an external master clock connector. 

[Cornan]

A while ago @seeteeyou recommended the WD Passport Wireless Pro as a great SSD. It was a HDD, but that’s not at all my point here. I have’nt got the patience to look for that post and quote it. My actual point is that I have just bought the 2tb version to backup my SD card while I’m out photographing. I will receive it in a couple of days. The thing I did’nt really realize properly before (but maybe @seeteeyou did?) is that it is a uPnP server/Plex server and a hot-spot as well plus all-day battery powered (approx 9-10 hrs). I can surely see that this could prove to be a game changer as a server/HDD close to the audio gears. I intend to try it out and let you guys know how it compares to direct OpenHome Tidal streaming when listening to the same album.

[davide256]

3 hours ago, Johnseye said:

 

Done and done with SOtM hardware. The NUC sounds better. Look up my previous posts in this thread. 

thanks, I've book marked the post. It will make it easier to part with some gear and not worry that I might regret it later 😥

 

[seeteeyou]

Speaking of wireless storage, Pioneer made something called Stellanova and basically it's kinda like Wi-Fi dock of some sort but you must install their own apps in order to access USB drives and DACs via its wireless network

 

https://darko.audio/2015/10/everyman-hifi-at-fujiya-avic-2015-pioneer-stellanova/

 

Even third party products like Chord Hugo could be connected

 

 

There's a much cheaper model with only 1 USB 2.0 port but we could actually power that with a pair of A123 LiFePO4 plus LT3045

 

https://www.amazon.co.jp/Pioneer-パイオニア-IEEE802-11a-対応ワイヤレスドック-APS-WF01J-2/dp/B01L0ZRI2G

https://www.ebay.com/itm/LT3045-four-parallel-regulated-power-supply-5V-2A-Upgrade-LT3042/273135946356

[FelipeRolim]

18 hours ago, davide256 said:

I guess what I'd like to see is a comparison of a PC using something like a JCAT USB  FEMTO as endpoint machine with AL vs AL/NUC endpoint using built in USB. I don't see anything special about the hardware quality in an endpoint NUC,  just good hardware integration. My suspicion is that an independently powered well made USB card in a decent endpoint machine will best NUC.  Hopefully we can get a CA member who already has one of these or an alternative to give it a shot and report in.

 

I haven't tested any kind of NUC recently, but I compared AudioLinux against Windows on my computer, which at the time had SOtM tx-USBexp (today I have the JCAT USB Card FEMTO). In my opinion, listening on the same computer, Windows has played considerably better. I cannot tell if a NUC would be even better, but considering at least the dedicated USB card, I don't think so. I also have linear power supply on this computer, use RAMdisk and did several other correct adjustments.

 

Another detail is that I didn't exhaust the possibilities of AudioLinux. I haven't tested everything that it can offer, like HQPlayer, for example. At the end, I cannot give any constructive opinion. I really don't like Linux. It isn't good for audio purposes...

[lmitche]

For those of you using a USB stick to boot Audiolinux and Roonbridge on your NUC endpoint, version 1.6 of Roon is available today with a new build of Roonbridge. The update will install per normal Roon operations into the ramdisk in the endpoint and play music as normal.

 

To permanently save this build on your USB stick, boot the NUC leaving the USB stick in place, login to the NUC and run Save System from the Audiolinux menu. Depending on your version of AL, this can be done from a browser, terminal emulator or console terminal via keyboard and monitor.

 

If you are running an older pre-menu version of Audiolinux, typing the ramsave command from the root prompt will save the changed contents of the ramdisk to the usb stick.

 

Make sure that the Roonbridge software has updated to version 1.0 build 169 before doing the ramsave operations described above. To check, in Roon go to Settings: About  to see the currently running version numbers. Relaunch the Roonbridge if it hasn't been done yet.

 

Enjoy,

 

Larry

 

 

[mourip]

Thanks Larry.

 

If you have been running in ramroot with the USB stick unplugged can you plug it back in to save your new settings or do you need to make all of your changes and the save after a fresh boot with USB stick left in place?

[luisma]

Thanks Larry for the update