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

[JohnSwenson]

3 hours ago, Johnseye said:

 

Just got mine and it was the 12v from Amazon.  I found a used v2 7.5v one on ebay. 

Why are you guys going after the 7.5V ones? The 12V ones work just as well.

 

John S.

[JohnSwenson]

15 hours ago, Johnseye said:

 

John,

 

1. The reasons are that it becomes possible to use an LPS-1 and that there is a grounding screw on v2.  Is there a benefit from using a linear PSU even when shunting with this switch?  Is there any benefit of using the grounding screw?
2. If one has multiple switches with multiple devices connected to them throughout a home network, do you see a benefit of using these models, shunted to contain the noise?

Thanks

There is no advantage from a leakage standpoint using an LPS. The grounded output of the SMPS already gets the leakage to essentially zero. There MAY be an advantage to a low noise supply, but I don't know for sure, I haven't spent any time looking at that.

 

The grounding screw is only useful if it directly connects to the ground plane of the switch. Just grounding the metal chassis is not sufficient. Grounding the negative of the supply guarantees that the ground plane of the PCB gets grounded. I just put a meter on mine, the ground screw does NOT connect to the ground plane, so connecting to it is not sufficient.

 

There should be no need to use these models anywhere else in the system. That is the whole purpose of these, IF the supply is grounded they completely block any leakage coming in from anywhere else on your network, period.

 

John S.

[JohnSwenson]

9 hours ago, Summit said:

 

If I get a switch with one of FS105 and FS108 and grounding the negative of the output of the SMPS I will get rid of both the leakage from the SMPS and the network leakage, ok. Is it best to have the switch on the same outlet as the AE or can I have it on another outlet next to the router or does it not matter?

The switch can be on a different strip than the AE, BUT the grounding plug needs to be on the same strip as the SMPS.

 

John S.

[JohnSwenson]

3 hours ago, lmitche said:

John, you findings with small SMPSes are great. Is it possible to divert leakages currents to ground with an ATX SMPS power supply in a similar way?

 

My assumption is that these supplies use a common ground, so shouldn't grounding any neutral wire from any arbitrary modular ATX PSU output plug accomplish the same?

I haven't tried that aspect. IF the ATX supply has output "gnd" wires connected to the safety ground of the power cord then that should be sufficient. IF the back wires are not connected to the power cord gnd, then yes grounding any of the output "gnd" wires should be sufficient.

 

John S

[JohnSwenson]

4 hours ago, Johnseye said:

 

Thanks John.  It would be interesting if this shunt blocked most of the noise from an SMPS making an LPS unnecessary.  I tried your shunt recommendation yesterday.  Still trying to determine whether I can "hear" a difference.  I do use LPS's so my noise level may already be low.  Need to do some A-B listening tests.  It's good for others doing this to know that the space where the wire plugs into the 5.5 x 2.1 mm adapters is small.  If you're using 14 gauge solid cable to connect between the two, your cable running to the wall plug needs to be pretty thin.

 

I didn't see this specifically mentioned, but I assume it's safe to run grounding wires from more than one adapter shunt to a single wall plug?

Yes, you can use one gnd plug to ground several SMPS, as long as they are all plugged into the same strip as the ground plug.

 

John S.

[JohnSwenson]

The chips in all these devices run at 3.3V or less, this is supplied by a switching DC-DC converter. This means you should be able to run it from a supply at the rated voltage down to 4.5V or so.

 

The manufacturers go with higher voltage supplies so they can use a very cheap lower current supply.  So you almost certainly can run from a lower voltage supply as long as it can handle the current at the lower voltage.

 

Exactly how much current that is will depend on how many connections you have to the switch, whether they are Gigabit or 100Mbit, how much traffic there is on each port etc.

 

John S.

[JohnSwenson]

2 hours ago, austinpop said:

 

This is an interesting question, and worth unpacking. If you look at where the SQ improvements due to clocks in my (and others') system have come from, it's from 2 waves:

1. First wave: the use of clock points from an sCLK-EX board. The key features here are:
   • A 4:1 benefit, as one board can drive as many as 4 independent clock frequencies, and potentially 4 independent components.
   • This is a frequency synthesizer design, that uses an internal 10MHz reference oscillator, that can be overriden by an external reference oscillator of even higher quality
   • We do not know the phase noise characteristics of the sCLK-EX, as SOtM has never published. It is clearly "great," based on sonic impressions, but we don't have a number.
2. Second wave: the use of a 10 MHz reference clock. In my case, my OP-14 has a -114.9 dBC/Hz @ 1Hz, in others', the Ref 10 is rated at -116 dBC/Hz @ 1Hz. These excellent numbers translate to another quantum leap in SQ.

Now, some interesting points to ponder:

1. The architecture of using a single reference clock to drive a 4:1 frequency synthesizer like the sCLK-EX gives a very nice way to multiply and distribute the "goodness" of a single reference clock, to components that were not designed to accept reference clocks.
2. We don't know how much that phase noise increases (i.e. degrades), once you factor in the cable length, cable quality, connector quality, impedance mismatches if any, etc. So there are losses, and you just have to evaluate the end result. I would love to see measurements of how phase noise is degraded by these factors. 
3. In other words, when you start with a reference clock with a -116 phase noise, what is the effective phase noise at the destination of the distribution chain? And could you replace this distribution chain with a higher phase noise clock that was internal to the device? What does that number need to be? -110? -105? -100? Fascinating question.
4. Could we replace this architecture with clocks in each device? Certainly yes, but I think the cost would be prohibitive!
5. I see two approaches:
   • a) Fixed frequency oscillators: OCXO's in every box in the chain - router, switch, server/player, regenerator, DAC, with phase noise numbers of -114 or better! Maybe if you factor in the short cable lengths and direct connections, you could get away with lesser specs to achieve equivalent SQ.
   • b) Frequency synthesizers with reference 10 MHz oscillators: I suspect this was what you were asking. What if each component incorporated a frequency synthesizer, a sort-of simpler sCLK-EX with just the number of taps needed for that device? And if each device had an input for a 10MHz reference clock, that could override the internal reference?

Frankly I would love to see a move to 4b) design. If I remember correctly, @JohnSwenson had observed in one of his informative posts, that the price of frequency synthesized clocks had come down substantially in price to where such designs are more cost effective. I am parapahrasing, of course.

 

Sorry if this is a long winded way of saying yes to your question: this means you could reduce the spaghetti by having a clock board per device right?

 

 

First off, very good summation of the situation.

Second the SOtM clk board is in fact one of these clock synthesizers chips fed by a single oscillator. Most of the board is actually power supply to feed those two parts.

 

The type of single board you are talking about does not have to be terribly expensive. Clock synthesizer chip, decent oscillator (something like a 575 is probably all you need for most parts of the chain) and a pair of 3042 regulators is all you need. It would be a small fairly simple board with not very much on it. You also need some form of connector to program the synthesizer.

 

Some comments on distribution of clocks.

 

A cable by itself cannot increase the phase noise of an oscillator, the cable CAN degrade the Signal Integrity (SI), which can cause the receiving chip(s) to increase the phase noise, but if the connections are done right this will be a very small effect. The problem is the "doing it right".  A LOT of what we deal with today is CMOS circuitry with CMOS clocks. CMOS signals are NOT terminated, ie they are not routed as a transmission line with a proper termination  resistor at the end. This works as long as the parts are close together. If there is more than a couple inches between parts, the above does not hold true and bad things can happen to SI. This has a very significant implication for clock distribution. You cannot just take a CMOS clock and run it through a couple of feet of cable (ANY type of cable, it doesn't matter) and solder the end onto a board where an oscillator used to be. This completely violates the "couple inches between parts" rule. If you want to preserve that really good clock you have to do something else. Some options are:

 

Use 75 or 50 ohm coax, and use a clock SOURCE that has the matching impedance, and put a matching resistor across the coax at the END where you connect it to the board. The problem is you have now built a resistive voltage divider, so you need enough voltage at the source so you have the voltage you need at the destination board. On top of that the source has to be able to provide a LOT of current to support this, normal chips can't support this much current.

 

Use one of the low voltage differential signaling schemes such as LVDS, send it over a differential cable, and have an LVDS to CMOS converter at the destination. This works very well, but it means you have to power that converter at the destination end.

 

The clock synthesizer chips do not support a high enough voltage to provide a 3.3V output at the end of the cable if you are doing a properly terminated cable. Their LVDS outputs are specifically designed for driving long cables so they may be a better choice.

 

So how bad is it if you just send the clock through the coax without termination? The SI can look terrible at the other end. Fortunately many CMOS inputs will still work with this. The big issue is what happens to the phase noise on the output of the receiver. This can range from hardly any degradation to quite a lot of degradation, it depends very much on exactly what is inside the chip, there is no easy way to tell.

 

A synthesizer at each end point cuts down on these issues, its output can be right where it needs to be. You don't have to worry about any of these thorny clock distribution issues.

 

As a side point to all this I am just starting some fundamental research into how phase noise from clocks in different parts of a system interact and move around a system. The ultimate goal is to figure out how to prevent any of this from getting into a DAC, which make all of this irrelevant. At this point I don't know how long any of this will take, but that is where I am headed.

 

John S.

 

 

[JohnSwenson]

19 hours ago, lmitche said:

John, isn't this the design of the power and clock circuits in the ISO Regen?

The ISO REGEN does not have a clock synthesizer chip. The hub chip is directly connected to the output of the 575. Custom 575 chips are used which have the correct frequency for the hub chip. There are 5 LT3042 regulators powering different circuits.

 

John S.

[JohnSwenson]

19 hours ago, rickca said:

John, do your comments apply equally to something like the Mutec Ref 10 output going into an sCLK-EX card, or is that a different situation than the sCLK-EX output going into an SOtM clock connect PCB that replaces a motherboard clock?

I don't know what SoTM does in any of these situations. The Ref10 has clock output of the appropriate impedance (some 50 some 75) so they will drive a coax properly. The ref inputs on the clock synthesizer chips are designed for low voltage, so they can handle reduces signal amplitude from the properly terminated source. I presume the ref input on the SoTM board has the proper termination resistor for the connector being used.

 

I have no idea what is on the clock interface board, but it is probably some form of termination and maybe a conversion chip. I presume they did it right so it will work driving standard CMOS inputs.

 

I was writing the post to people that want to do their own clock distribution scheme to let them know it is not so simple to do it right.

 

John S.

[JohnSwenson]

1 hour ago, Em2016 said:

 

Hi John @JohnSwenson

 

While it's complicated stuff for sure, it's kind of semi intuitive for a layman like me to understand that it's possible phase noise (or something else that isn't leakage currents) along the chain can interact and move around the system into your DAC. How phase noise moves around and how it interacts is the complicated stuff, for smart people like you.

 

BUT, from my layman perspective - how does a wireless (WiFi) networked USB endpoint fit into this picture? Is the phase noise of this wireless/mobile USB endpoint now the only phase noise that matters feeding the USB DAC input, since there is no electrical connection with the upstream components?

 

I guess the obvious silly question I'm asking is, does this problematic phase noise you're studying, travel over WiFi? LOL it sounds like such a silly question even reading it back to myself but I need to ask. If a WiFi USB endpoint is running off batteries, then you have no leakage currents involved in that path. No such thing as a free lunch of course, since you have potential RF/EMI issues.

 

But for these questions, just putt aside potential for RF/EMI issues with wireless transmissions and assume bit perfect playback of course.

 

Cheers!

 

The answer is "I don't know". That is why I am doing the research, to find out what is going. Before I do that, ANYTHING is purely a guess.

 

Six months to a year from now I will be able to answer that question, but as of now, no idea.

 

John S.

[JohnSwenson]

7 hours ago, Johnseye said:

 

Right.  If I incorporate a NUC as endpoint, which I'm likely to try now, I would use a 6" or 1' ethernet cable.  Likely 360 shielded.

I would recommend against using such sort Ethernet cables, I have been doing a LOT of experimentation with Ethernet interfaces recently and found that many PHYs need a certain amount of capacitance on the pairs in order to work properly, a 6" cable just doesn't have enough. I had a situation where a 3ft cable worked fine, a 1ft was marginal (significant packet loss) and 6" did not work at all. I soldered some small value caps across the pairs and presto the 6" cable worked great. This tends to support the "not enough capacitance" theory.

 

So going for very short Ethernet cables may not be a good thing. I personally would not go shorter than 2ft of cable between PHYs.

 

Of course not ALL devices are going to have this problem, but I found it on a high percentage of the devices I was testing.

 

John S.

[JohnSwenson]

11 hours ago, austinpop said:

 

LMS has been around for quite some time, so is a pretty well grooved ecosystem. Just like Roon uses RAAT for its underlying communication, LMS uses SlimProto on top of TCP. As best I can tell, SlimProto is used for control, but the actually audio streams are sent over HTTP/HTTPS.

 

Point is - this is a pretty mature platform. It's just not as slick as Roon. Not by a long shot.

LMS definitely does NOT use HTTP for sending audio data. SlimProto is actually a very simple protocol and is used for sending the audio data. It supports several "native" formats, which are essentially the audio data in common file formats. (WAV, FLAC, MP3, probably a couple more). The rest of the file formats get converted in the server into one of these.

 

The only thing HTTP is used for is talking to the web server GUI that is part of the server. If you are using any of the other controllers (hardware or software) HTTP is not used at all.

 

LMS also uses UDP multicast to coordinate servers, controllers and players finding each other on a network, but after they are found the rest happens through SlimProto.

 

John S.

[JohnSwenson]

4 hours ago, austinpop said:

 

Good point, Alex. Low output impedance is a prerequisite for a PSU's ability to deliver instantaneous current bursts. Indeed, I wonder if a PSU's inability to respond instantly to current demand can introduce another flavor of induced jitter in the signal.

 

Maybe @JohnSwenson's white paper addresses this aspect as well?

No it's not in the white paper, but I can give some insight into measuring it.

 

First off is what you are measuring. You can use test equipment to measure the supply response, this means a controlled current load on the supply then measure what the supply does. Some electronic loads let you vary the load with periodic patterns then you look at the voltage with a scope and see what it does. These are usually square wave changes to the current. This is nowhere near good enough to really characterize a supply.

 

Another method is to hook the supply up to a real load (computer etc) and see what you get. Scopes are fairly useless for this, the patterns in load current from a computer are so all over the place that no matter how you set a scope you will miss important stuff.

 

Then there is how you look at the results, you can look at it in time or frequency. Exactly the same thing as using a scope and a spectrum analyzer, they are different ways to to look at the same thing. A scope gives you the timing view and an output impedance plot gives you the frequency view.

 

Finding the right test equipment to measure this is not easy. You need to measure small changes sitting on top of DC voltages. The usual way to do this is with AC coupling (the signal goes through a cap) that works fine for periodic high frequency signals, but we are not measuring that. The variations from digital loads (computers etc) are not very periodic and some of the most interesting events are low in frequency which will not pass through the cap. When doing an impedance plot we also want to measure low frequencies which again will not pass through the cap. If you use DC coupling a scope does not have enough resolution to measure the small voltage changes at the DC voltage of the supply.

 

Many years ago I tried doing this with a custom AC coupled amplifier, I first tried high value electrolytic caps, which couple the low frequencies fine, but are very non-linear at high frequencies. So I tried large electrolytic in parallel with small film caps. well the parasitic inductance and capacitance reacted with each other creating massive resonances which again made the results useless. Then I tried high value film caps which are way more linear, but by the time you get enough capacitance are large and with enough inductance that the high frequency response is terrible. OK so now it is non-inductive film, this might have worked but the prices were insane, I could not afford it. So I dropped this project at the time.

 

To make an impedance plot you have a voltage controlled load and you feed it a swept frequency sine wave. This will cause small changes in the voltage of the supply, the higher the impedance at the particular frequency the higher the voltage change. Part of the problem here is that NONE of the test equipment manufacturers actually make such a device. The frequency range, voltage and current of the supply vary so much that no company can produce a universal load that will work for most people.

 

A year and a half or so ago I built my own voltage controlled load, it works pretty good, but it is not calibrated, there is no single voltage change to current change number, unfortunately the delta varies depending on the supply voltage  and how much current you are pulling with the load. I used this with my spectrum analyzer which has a "tracking oscillator" which produces a sine wave at the center of the instantaneous frequency of analyzer. This DID actually work. I get a nice amplitude verses frequency graph, but it is not calibrated and because of the non-linear behavior of the load you can't just look at the graph, there have to be corrections made. In addition the spectrum analyzer can only go down to 10Hz which is not low enough to capture some of the most important ranges.

 

So after all this I decided the best way to do this is to use a high resolution ADC and DAC, the DAC generates the variable frequency sine wave to drive the load, the ADC is set so the DC of the supply is close to the top of the range, then the high resolution can properly measure the small changes in output voltage. The data from the ADC can then be post processed to account for the non-linearities in the load and put in graph form (either time or frequency. It turns out there is a fairly inexpensive piece of equipment on the market today which is perfect for this called the Red Pitaya, it has high resolution DACs and ADCs in a data collection system with a lot of memory and an FPGA and ARM processor to control things. I have one of these but have not had the time to write all the code necessary to make this work. It is doable but is going to take a fair amount of work to get it right.

 

Well that is probably more than anybody wanted to know about measuring power supplies!

 

John S.

 

 

[JohnSwenson]

17 hours ago, sandyk said:

Clearly., you do  not have a problem like some, as 40.0~41.1 degree C is quite a low temperature (comfortably warm)  for most electronic devices You also need to take into account the Ambient Temperature at different locations.

 Unless it uses internal switching voltage regulators, then it should run noticeably warmer at the higher voltage unless it is in a well ventilated or Air Conditioned area when it is drawing >1A, even if the current draw is a little less at 12V.

7V at >1A is quite a bit of heat to dissipate in a small case (>7W)

 

The EtherREGEN does in fact use a whole bunch(12) of very highly shielded very low emi switching DC-DC converters, each of which drives a carefully designed passive filter then into a linear regulator. This was the only way to run the power networks. Most of the current drawn is in the 1.0-1.2V range which would burn up linear regulators trying to drop the full range.

 

Thus the power dissipation of the EtherREGEN stays almost constant as you change the input voltage.

 

John S.