Wavelength Ethernet Spacelator

[Miska]

2 hours ago, mansr said:

Everything except the lowly UART has some kind of clock transmitted along with the data.

 

Well, I wouldn't say most wireless transmissions transfer clock as such, but more just modulation patterns over one or more carriers (typically multi-carrier these days), which are of course driven by a clock. Bunch of frequencies and levels encoded at some (sometimes variable) bitrate. 256-QAM or similar, mixed with randomized center frequency hopping or such. Receiver can just hook into changes that can happen at varying rate. Of course you can have some "clock" that defines channel band center(s). Or band division like IIRC ADSL has that uses low frequency band for the uplink and high frequency band for the downlink.

 

1 hour ago, jabbr said:

Basically as of 10GBase-X compliance with eye-pattern for end-to-end jitter in picosecond range rather than specs for individual components.

 

"Eye pattern" sounds a bit inefficient because you have lot of "holes" in it. You could fill those with different waveforms/frequencies. I'd expect efficient transmission look more like noise on a scope. Unless you don't think about "eye pattern" in the traditional meaning. Keysight seems to have some nice scopes for dealing with modern comms (good screenshot of QAM too):

https://www.edn.com/electronics-products/electronic-product-reviews/other/4461130/Oscilloscope-sets-bandwidth-record--on-all-four-channels

 

1 hour ago, jabbr said:

Of course 10G speeds are not necessary for audio — I’m just pointing out that the really old 10mbs and 100mbs networks tolerated/used poor jitter components — 1Gbe less jitter but when 10G was introduced, the jitter specs became vastly better and the 10G switches are already very low jitter. If an audiophile switch is introduced it should be compared — end to end — against 10G equipment not 100M equipment

 

Ahh, yes, sure. I was kind of anyway expecting comparison at least within the same speed class. Gigabit switch against gigabit switch at least. I would expect all gear to be at least gigabit these days anyway.

 

 

Only case I know where any of the jitter/latency have any relevance to audio, is AES67/Dante/Ravenna and sometime AVB, which use PTPv2 for clock transmission.

[plissken]

1 hour ago, Arpiben said:

What I wanted to point is that the switch is not hitless (free of bit errors).

By unplugging one cable you trigger a Loss Of Frame/Input that allow to switch faster-> a few ms

By waiting for LACP packets loss the switching is much longer-> less than 1s

 

 

 

 

Didn't say it was. Error's are recoverable and modern playback devices have enough buffer. 

[Miska]

3 hours ago, Arpiben said:

For anyone interested in Ethernet synchronisation aspects:

- SyncE or synchronous Ethernet for transfer clock purposes (ITU-T G8261/2/4)

- 1588 PTP (Precision Time Protocol)

 

Of which only latter has relevance to audio and even then only for pro audio...

 

[mansr]

29 minutes ago, Miska said:

Well, I wouldn't say most wireless transmissions transfer clock as such, but more just modulation patterns over one or more carriers (typically multi-carrier these days), which are of course driven by a clock. Bunch of frequencies and levels encoded at some (sometimes variable) bitrate. 256-QAM or similar, mixed with randomized center frequency hopping or such. Receiver can just hook into changes that can happen at varying rate.

There's still some fundamental unit interval the receiver locks onto in order to make sense of the signal. That is, in fact, what you say in the last sentence.

[Arpiben]

1 hour ago, Miska said:

 

Of which only latter has relevance to audio and even then only for pro audio...

 

Fully agreeing.

Dealing with Wireless transmission (Microwave) nowadays the modulation schemes are reaching the 4096 QAM ( Quadratic Amplitude  Modulation) rather than 256 QAM.  Clocking  is retrieved from symbol rate.

Dealing with Latency, even if with little/no relevance in audio, the advertised values are around 5 ms for 5G networks. For such, in wireless, you need to have huge Bandwidth or Channel Spacing (500MHz/1000MHz/1500MHz).This means using frequency bands around 80 GHz and above.

Otherwise in cables or optical fibers the latency is around 5 ns / meter.

Nothing to worry about or in other words no need for MQA even with mobile phone audio streaming☺️

[Miska]

2 hours ago, Arpiben said:

Dealing with Latency, even if with little/no relevance in audio, the advertised values are around 5 ms for 5G networks. For such, in wireless, you need to have huge Bandwidth or Channel Spacing (500MHz/1000MHz/1500MHz).This means using frequency bands around 80 GHz and above.

Otherwise in cables or optical fibers the latency is around 5 ns / meter.

Nothing to worry about or in other words no need for MQA even with mobile phone audio streaming☺️

 

I know there are even bigger constellations than 4096 too. Was just an example...

 

Since for streaming we have enough buffers and fairly low bandwidth usage compared to for example Netflix 4K video, there are certainly no issues doing these things over the internet. I'm writing this and streaming Tidal over fixed 4G (LTE) internet connection. I have ADSL too, but it is slow and disappearing already. Operators want to get rid of cables here... My pings to computeraudiophile.com over this LTE connection are in 25 ms range, "jitter" about 5 ms (there's quite a bit of traffic there).

 

Quote

--- www.computeraudiophile.com ping statistics ---
29 packets transmitted, 29 packets received, 0.0% packet loss
round-trip min/avg/max/std-dev = 17.135/26.590/36.495/5.402 ms

 

So far, maximum LTE speed I've reached in practice here was 350 Mbps. I have also one router capable of LTE Advanced CAT16 (1 Gbps), but network is not yet up to it.

 

Doing things over LAN is certainly non-problem.

[jabbr]

4 hours ago, mansr said:

There's still some fundamental unit interval the receiver locks onto in order to make sense of the signal. That is, in fact, what you say in the last sentence.

 

Right, so the totally allowable jitter can be allocated between transmitter and receiver -- there is lots written but to start: https://www.keysight.com/upload/cmc_upload/All/669-10GigEthernet-08-01-02.pdf

https://www.sitime.com/api/gated/AN10026-10GbE-RJbudget.pdf

[jabbr]

6 hours ago, Miska said:

Only case I know where any of the jitter/latency have any relevance to audio, is AES67/Dante/Ravenna and sometime AVB, which use PTPv2 for clock transmission.

 

802.3ae (2002) as I've said, specifies end to end jitter and because the input frame is locked by PLL, jitter on the input results in a degree of jitter on the receiver. Consequently the specification of end to end jitter: both the transmitter and receiver together have to meet the specification.

 

The premise of Ethernet reclocking is that the transmitter or switch has too high levels of jitter/phase noise.

 

Not necessary when using a good 10 Gbe switch or NIC as these already have femtosecond clocks i.e. https://www.sitime.com/api/gated/AN10026-10GbE-RJbudget.pdf

 

So ... this thread is about a device that is intended to improve Ethernet SI ... I'm saying that fiberoptic Ethernet 10G (802.3ae) already does this ... that's what I use. They are giving away very high quality equipment on eBay because the pros are going 100 Gbe. It may not make a difference but at the prices who cares?

[Miska]

5 hours ago, jabbr said:

802.3ae (2002) as I've said, specifies end to end jitter and because the input frame is locked by PLL, jitter on the input results in a degree of jitter on the receiver. Consequently the specification of end to end jitter: both the transmitter and receiver together have to meet the specification.

 

The premise of Ethernet reclocking is that the transmitter or switch has too high levels of jitter/phase noise.

 

The only effect of too high jitter would be packet loss? If the packet loss is already 0, what would be the improvement of improved clocking?

 

Quote

Not necessary when using a good 10 Gbe switch or NIC as these already have femtosecond clocks

 

I would stay away talking about "femtosecond" jitter figures and such. So totally different for those GHz clocks compared to audio clocks. From the presentations you linked to, one can see that they look at jitter / phase noise figures above 4 MHz point. For audio clocks one looks at figures above 10 Hz point or even lower.

 

Higher the clock frequency gets, worse the low frequency phase noise figures gets. For example 100 MHz clocks used with ESS DAC chips have about 20 dB worse phase noise at 10 Hz than 22.5792/ 24.576 MHz clocks you need for direct DSD512. That is one problem when decoupling DSP and D/A conversion under same synchronous clock domain on one chip. One of the reasons I rather run all DSP asynchronously from the actual D/A conversion process...

 

5 hours ago, jabbr said:

So ... this thread is about a device that is intended to improve Ethernet SI ... I'm saying that fiberoptic Ethernet 10G (802.3ae) already does this ... that's what I use. They are giving away very high quality equipment on eBay because the pros are going 100 Gbe. It may not make a difference but at the prices who cares?

 

I'm still sticking to 1G switches although thought about 10G to improve speed of connection between switches. Now one can get good managed 1G switches at reasonable prices (new ones). Since I'm using HPE gear, I recall eying some model that had like two 10G SFP ports for uplink and rest were 1G.

 

[Arpiben]

14 hours ago, Miska said:

 

I know there are even bigger constellations than 4096 too. Was just an example...

 

Since for streaming we have enough buffers and fairly low bandwidth usage compared to for example Netflix 4K video, there are certainly no issues doing these things over the internet. I'm writing this and streaming Tidal over fixed 4G (LTE) internet connection. I have ADSL too, but it is slow and disappearing already. Operators want to get rid of cables here... My pings to computeraudiophile.com over this LTE connection are in 25 ms range, "jitter" about 5 ms (there's quite a bit of traffic there).

 

 

So far, maximum LTE speed I've reached in practice here was 350 Mbps. I have also one router capable of LTE Advanced CAT16 (1 Gbps), but network is not yet up to it.

 

Doing things over LAN is certainly non-problem.

 

The latency values were dealing with 5G Networks! I agree there are not so many at present time (few trials) ?

Bigger than 4096 QAM in Bell Labs/Huawei R&D department->yes. Under exploitation not yet as far as I know.

Anyhow I am out of thread topic.

Rgds. 

[plissken]

7 hours ago, Miska said:

 

The only effect of too high jitter would be packet loss? If the packet loss is already 0, what would be the improvement of improved clocking?

 

You know how many times I've pointed that out and it falls on deaf ears?

 

I've even pointed out you can inject the worst jitter possible by pulling the cable and plugging it back in and the music never stopped in non-realtime systems.

[Account Closed]

10 hours ago, Miska said:

 

 

 

I'm still sticking to 1G switches although thought about 10G to improve speed of connection between switches. Now one can get good managed 1G switches at reasonable prices (new ones). Since I'm using HPE gear, I recall eying some model that had like two 10G SFP ports for uplink and rest were 1G.

 

 

Care to elaborate on the good (new) managed switches?

[Miska]

2 hours ago, bobflood said:

 

Care to elaborate on the good (new) managed switches?

 

Recently I've been buying HPE 1820 series:

https://www.hpe.com/us/en/product-catalog/networking/networking-switches/pip.switches.7687976.html

and 1920S series:

https://www.hpe.com/us/en/product-catalog/networking/networking-switches/pip.overview.1009689650.html

 

And after spending a little bit of effort on configuration, like making sure flow control (802.3x) is enabled, things work nicely. I also utilize VLANs on my setup, but it is not necessary for typical home configurations.

 

I'm rarely a brand loyal person, but for network switches, I've been using HP for a very long time in various different kinds of environments and never had much to complain.

 

My WiFi access points are also HPE, powered from the switch using PoE.

 

[jabbr]

15 hours ago, Miska said:

The only effect of too high jitter would be packet loss? If the packet loss is already 0, what would be the improvement of improved clocking?

 

Not entirely simple answer but if the question is whether the jitter on the transmitting interface can affect jitter on the receiving interface, the answer is yes.

http://www.ece.tamu.edu/~spalermo/docs/rx_jitter_tracking_fwd_clk_ragab_jece_2011.pdf

 

Similarly both common and differential mode noise: http://www.ece.tamu.edu/~spalermo/docs/rx_jitter_tracking_fwd_clk_ragab_jece_2011.pdf

 

Now a leap of faith would say that jitter on the Ethernet interface thus could affect jitter on the USB stream. This has not been demonstrated to my knowledge but based on the above (and other data) the answer is that its possible.

 

Not to say that this makes its way into the DAC, or is relevant for audio  but if the question is whether the upstream clock can affect the downstream interface, the answer has been known to be yes for some time.

[jabbr]

Let me expand slightly on the mechanism.

 

The Ethernet transmitter doesn't transmit a clock per se, rather the clock is recovered from the incoming packet via PLL. The PLL is necessary to phase lock the incoming packet with the receiver interfaces so that the bits can be latched and reclocked at the receiver. The PLL tracking is affected by source jitter.

 

In an ideal world, a FIFO isolates and reclocks the Ethernet buffer. In the case of a high quality 10G fiber interface, the receiver CPU undoubtedly has higher jitter than the Ethernet stream, so this might be moot. This is the reason, though, that 802.3ae requires the test of end-to-end jitter, rather than individual components. I focus my efforts on the terminal receiver i.e. good PSU, good hardware knowing that my network will give me a measurably good signal up until that point.

[Miska]

29 minutes ago, jabbr said:

Let me expand slightly on the mechanism.

 

The Ethernet transmitter doesn't transmit a clock per se, rather the clock is recovered from the incoming packet via PLL. The PLL is necessary to phase lock the incoming packet with the receiver interfaces so that the bits can be latched and reclocked at the receiver. The PLL tracking is affected by source jitter.

 

In an ideal world, a FIFO isolates and reclocks the Ethernet buffer. In the case of a high quality 10G fiber interface, the receiver CPU undoubtedly has higher jitter than the Ethernet stream, so this might be moot. This is the reason, though, that 802.3ae requires the test of end-to-end jitter, rather than individual components. I focus my efforts on the terminal receiver i.e. good PSU, good hardware knowing that my network will give me a measurably good signal up until that point.

 

I understand that, but if the end effect at ethernet packet level is zero? Since I'm getting zero errors at ethernet packet level, playing with jitter optimizations is moot. Even if there would be 50% packet loss due to excessive jitter on 10G ethernet, still it wouldn't make a difference for audio because you would have 5G effective bandwidth. While DSD512 over NAA requires that you manage to push 50 Mbps through within +-500 ms TCP level jitter.

 

[jabbr]

7 minutes ago, Miska said:

 

I understand that, but if the end effect at ethernet packet level is zero? Since I'm getting zero errors at ethernet packet level, playing with jitter optimizations is moot. Even if there would be 50% packet loss due to excessive jitter on 10G ethernet, still it wouldn't make a difference for audio because you would have 5G effective bandwidth. While DSD512 over NAA requires that you manage to push 50 Mbps through within +-500 ms TCP level jitter.

 

 

I can’t disagree with any data. Devil’s argument says that jitter at the receiver somehow makes it into the DAC. Not technically impossible but not proven. I picked up some Solarflare 7322f cards at like $120-150 so ... some day when my very long stack of things to test out dwindles 99%, then I’ll sit down and tell you if I can hear “SQ” ??? 

[Arpiben]

10 hours ago, Miska said:

 

I understand that, but if the end effect at ethernet packet level is zero? Since I'm getting zero errors at ethernet packet level, playing with jitter optimizations is moot. Even if there would be 50% packet loss due to excessive jitter on 10G ethernet, still it wouldn't make a difference for audio because you would have 5G effective bandwidth. While DSD512 over NAA requires that you manage to push 50 Mbps through within +-500 ms TCP level jitter.

 

Jitter specs at Rx side define the bath tube 10exp-12 BER ( Bit Error Rate)

together with the eye aperture ones.

In other words the transport media minimum specs in order to carry data at max rate with a low probability of errors.

 

In case of degraded BER you don't know which packet will be affected audio or not. You have to think in terms of probability of audio bandwidth vs total bandwidth (10Gbps) error.

[jabbr]

14 hours ago, Miska said:

Even if there would be 50% packet loss due to excessive jitter on 10G ethernet, still it wouldn't make a difference for audio because you would have 5G effective bandwidth. 

 

Yes! That is a great way of looking at it — and why I use 10Gbe actually — it is so over spec’d for the job. Jitter is really low, fiber blocks common mode noise, Ethernet blocks differential mode noise. 

 

Its also really cheap, so way over spec & cheap, and I don’t need to worry about the network   ... plus I like the color of the cables ?

[Miska]

3 hours ago, jabbr said:

Yes! That is a great way of looking at it — and why I use 10Gbe actually — it is so over spec’d for the job. Jitter is really low, fiber blocks common mode noise, Ethernet blocks differential mode noise. 

 

Its also really cheap, so way over spec & cheap, and I don’t need to worry about the network   ... plus I like the color of the cables ?

 

I think overall Ethernet and other networking technologies are good fit for audio. It certainly increases complexity though, that's the cost. Optical Ethernet gives full isolation and lot of bandwidth and flexibility for all kinds of use.

 

I was trying to check 5.1 channel DSD256 over WiFi today, but exaSound driver on Windows machine was acting up so I couldn't. Maybe it's time to try the Mac driver instead.

 

[jabbr]

On 11/9/2018 at 10:00 AM, plissken said:

 

You know how many times I've pointed that out and it falls on deaf ears?

 

I've even pointed out you can inject the worst jitter possible by pulling the cable and plugging it back in and the music never stopped in non-realtime systems.

Regarding jitter and it’s potential effects:

 

Jitter generally consists of phase noise. There is also amplitude noise. Noise can be differential and it can be common mode. 

 

Noise of all the above types can be transmitted from the transmitter to the receiver. There are various mitigation strategies and circuits for each of the above noise types. 

 

Without circuit details and measurements, one cannot be sure about how much noise may or may not be transmitted between systems. I suppose if you leave the cable unplugged that you have some assurance that noise won’t tend to pass. For those of us who don’t want to keep plugging and unplugging the cables there are different isolation techniques. 

[plissken]

15 hours ago, jabbr said:

Regarding jitter and it’s potential effects:

 

Jitter generally consists of phase noise. There is also amplitude noise. Noise can be differential and it can be common mode. 

 

Noise of all the above types can be transmitted from the transmitter to the receiver. There are various mitigation strategies and circuits for each of the above noise types. 

 

Without circuit details and measurements, one cannot be sure about how much noise may or may not be transmitted between systems. I suppose if you leave the cable unplugged that you have some assurance that noise won’t tend to pass. For those of us who don’t want to keep plugging and unplugging the cables there are different isolation techniques. 

 

Ya, like WiFi.

 

Also the phase noise is going to happen around the signalling rate and not audio-band. You seem to want to put a square peg into a round hole.  When all you need to do is measure the output of the DAC.

 

But again, if Gordon wants to prove any of this is audible, I welcome him to sitting down to a test rig like I outlined with the LACP config I posted or he can provide an isolation device to Amir at audiosciencereview.

[jabbr]

33 minutes ago, plissken said:

 

Ya, like WiFi.

 

Also the phase noise is going to happen around the signalling rate and not audio-band. You seem to want to put a square peg into a round hole.  When all you need to do is measure the output of the DAC.

 

Why bother measuring when we have resources here such as yourself that know everything.

 

Phase noise is curious, and I agree that mere logic would question whether phase error at 25 Mhz (BTW the crystal for Ethernet tends not to be the signalling rate but no matter) should translate down to audio frequencies.

 

The reason that power supplies make a big difference in crystal phase error is that there is a 1/f nonlinearity which converts 1/f voltage noise into 1/f offset phase noise. This phase noise will widen the peak.

 

I don't know that the nonlinearity is unidirectional.

 

33 minutes ago, plissken said:

 

But again, if Gordon wants to prove any of this is audible, I welcome him to sitting down to a test rig like I outlined with the LACP config I posted or he can provide an isolation device to Amir at audiosciencereview.

 

Well yes I think folks who are promoting a product or certain capabilities of a product have some incentive to demonstrate benefits.

 

I am using WiFi and pointing out where COTS hardware, albeit pro grade, already deals with issues that have been raised here, so can serve as a starting point for folks who are interested in upgrading network hardware. I have been advocating optical isolation for a few years now, and its great to see products such as the Lumin X1, the Wavelength spacelator and upcoming Uptone switch incorporate optical elements,  and even SFP capabilities (there are copper as well as fiber SFP modules so this is very flexible).

 

Wifi can be a bit glitchy still but getting there.

[plissken]

17 minutes ago, jabbr said:

 

Why bother measuring when we have resources here such as yourself that know everything.

 

Well I know what I've measured and Wired or WiFi has no audible effect on the output of my Emotiva DC-1.

 

Using 1Khz input nothing but a bit of mains noise way beyond perceptible hearing and that went away with both optical and WiFi.

 

I don't know everything but I do know some things.

Quote

 

Phase noise is curious, and I agree that mere logic would question whether phase error at 25 Mhz (BTW the crystal for Ethernet tends not to be the signalling rate but no matter) should translate down to audio frequencies.

 

It doesn't because if it did you would have posted something crowing about it.

Quote

 

The reason that power supplies make a big difference in crystal phase error is that there is a 1/f nonlinearity which converts 1/f voltage noise into 1/f offset phase noise. This phase noise will widen the peak.

 

I don't know that the nonlinearity is unidirectional.

 

If the non-linearities propagated down into the human hearing band you would have posted a resource on it.

Quote

Well yes I think folks who are promoting a product or certain capabilities of a product have some incentive to demonstrate benefits.

 

Actually the more they can market without any peer review the better.  They will let the gullible with gold ears do the selling.

Quote

 

I am using WiFi and pointing out where COTS hardware, albeit pro grade, already deals with issues that have been raised here, so can serve as a starting point for folks who are interested in upgrading network hardware. I have been advocating optical isolation for a few years now, and its great to see products such as the Lumin X1, the Wavelength spacelator and upcoming Uptone switch incorporate optical elements,  and even SFP capabilities (there are copper as well as fiber SFP modules so this is very flexible).

 

Wifi can be a bit glitchy still but getting there.

 

Wifi is downright reliable actually. I have 248GB wireshark capture with 0 dropped packets and only 4 re-transmits around here somewhere.

 

Heck, you can pickup Cisco 2960-8-TC's left and right for $25 with SFP GB. That will provide every ounce of isolation.

 

 

[jabbr]

24 minutes ago, plissken said:

 

It doesn't because if it did you would have posted something crowing about it.

...

24 minutes ago, plissken said:

 

If the non-linearities propagated down into the human hearing band you would have posted a resource on it.

 

Don't make assumptions, and don't twist what I have already posted, including references. My own interests lie more in DAC and amplifier topologies. where certain nonlinearities are clearly audible ... if you are of the opinion that all DACs and amplifiers which have the same THD sound the same, then I don't have any interest in proving anything to you and certainly no interest in crowing. I've posted many many resources which anyone can read if they want, or not.

 

24 minutes ago, plissken said:

 

Actually the more they can market without any peer review the better.  They will let the gullible with gold ears do the selling.

 

Wifi is downright reliable actually. I have 248GB wireshark capture with 0 dropped packets and only 4 re-transmits around here somewhere.

 

Heck, you can pickup Cisco 2960-8-TC's left and right for $25 with SFP GB. That will provide every ounce of isolation.

 

That's my point.