[JoshK]

Also, just to be clear, you are suggesting that the best solution shown above is the fast driver, long cable and proper built in pad right?   I see the differences, but wasn't 100% clear which you thought was better.   It seemed like the slower driver and 10db of attentuation resulted in an OK outcome too. 

[jneutron]

Yes, it should show a negative going spike, but as you have deduced, whatever is after it is causing that spike. My guess is that this unit has the SPDIF  transformer that everyone (well, almost everyone) thinks they must use. Guess I should take the lid off, and verify.

Most really have no idea how to control the current path to keep inductance down.  Admittedly, it isn't a piece of cake.  My experience is with 250 pS 100 volt signals, so I've been shall we say, bitten before.

(Stuck at home today............ice everywhere. So, something to do.)

Yah, I'm running out of places to shovel the global warming, the piles are getting too high..32 inches in january..  If it gets any warmer, I'll have to just shovel tunnels to the doors.  At least we have second floor windows just in case.

The small downward step is indeed the cable. Its impedance is a bit less than the other piece.

Cable?  You mean the one on the other side of the rca on the inside of the case heading to the inductive termination?  I'd think this the most likely smoking gun.

Cheers, John

[jkeny]

Thank you Pat for a very informative thread.

Can you say something about the benefits of a faster rise time apart from being able to use shorter cables? i.e. is it one means of dealing with intersymbol interference & the resultant associated deterministic jitter - I've heard it said that this is the most objectionable type of jitter?

[art]

No, there is a small piece of coax, about 1', between the output of the attenuator, and the DAC. That is the step I am referring to. You may have to take my word on this one, as I know it will show up on our TDR that way. Actually, if I were to do this, on the TDR, it would show the jack, the PCB trace, the transformer..........it would all show up.

Pat

[art]

Also, just to be clear, you are suggesting that the best solution shown above is the fast driver, long cable and proper built in pad right?   I see the differences, but wasn't 100% clear which you thought was better.   It seemed like the slower driver and 10db of attentuation resulted in an OK outcome too.

Not quite............

If you want to use a short cable, you need to have a fast rise time driver. You also need to pay more attention to connectors and other minor factors.

If you don't know, you are better off with a longer cable.

Attenuators help. They won't fix ringing, overshoot, and a lot of other crud, that has been shown on other sites. If there is an extreme mismatch, and that exacerbates the problem, it will lessen that. But, the ringing and all will still be there. Just maybe not as bad. Attenuators are not a magic bullet. If you put a garbage signal into it, you will get a garbage one out.

(Translation: if the guy who designed your USB-SPDIF thingie copies the TI data sheet, you should expect problems. Because the data sheet is wrong, and he/she/they didn't figure that out. Obvious, to some of us old farts.)

As you can see, with a DAC that lots of folks use, there is a big enough reflection problem, to indicate an attenuator will help. (It took 8-10 dB, before it lost lock.)

Pat

[art]

I'd love to know what the draw backs to using I2S are (whatever cable format) too?  I do recall reading that sending the signal requires really fast drivers.  Is this really difficult to design?  Are reflections worse?

Just how long do you think you can run those signals? Ain't gonna be 40'. I2S was designed as an IC-IC interface. If you are serious about it, you would need terminated drivers and receivers.

Since you are not trying to recover the clock, from a compound signal, reflections won't have the same sort of effect.

What you would need to think about is how much the rise time is lowered, as the clock signal gets to the far end. As the rise time gets slower, the signal is more susceptible to jitter caused by noise and interference. Since it will most likely be Gaussian in nature, it won't muck things up as bad as data-correlated jitter would.

(It is a dv/dt thing. Lower dv/dt = lower slope. More time the signal exists in the "twilight zone", where it is susceptible to jitter.)

Maybe I should point out one thing, that many of you many not have thought about:

If you take a typical CMOS output stage, its PSRR is 6 dB, at the midpoint. As both the top and bottom MOSFET is on, at roughly the same conductance, and crud on the supply is only attenuated by 6 dB. 50 mV of noise on the rail means 25 mV of noise, in the signal. Right at the midpoint, which is where the decision point is.

There is the main cause of jitter. Which is why clean rails are important, isolating all critical stages, and not shoving as many functions as possible, into one chip. Especially the important functions. Like the clock.

Pat

[JoshK]

Well I don't know what the big deal many audiophiles have against 3m digital cables.  It seems that could only help placement of gear. 

[Joseph K]

Hi all,

As respect to that "strange" response from the dac measured by Jocko.
I mean with that "inductive" peak.
I think it's just a healthy big fat example of this phenomenon, discussed at diyhifi:
http://www.diyhifi.org/forums/viewtopic.php?p=43739#p43739

That is, that dac probably has a high speed video opamp, with feedback, at it's input.

Ciao, George

[Joseph K]

here it is, the response of a video amp, with feedback, when "pinged" with a TDR.
That is, the same way like Pat has done it here with that Dac.
(For those who don't have access to the Diyhifi site - which is a pity  )
It's a little bit different, because this TDR shot contains also the effect of the saturated amplifier.
Without that it would "relax" back all the way to the baseline.

[jkeny]

Aha, JosephK, thank you for this post - so video op-amps may not be the best termination method as has been promoted?

[jneutron]

here it is, the response of a video amp, with feedback, when "pinged" with a TDR.
That is, the same way like Pat has done it here with that Dac.
(For those who don't have access to the Diyhifi site - which is a pity  )
It's a little bit different, because this TDR shot contains also the effect of the saturated amplifier.
Without that it would "relax" back all the way to the baseline.

Agreed, they kind of look the same.  However, Pat's scope is clearly more consistent with an inductive termination, whereas your scope doesn't have the exponential decay.

It would be great if Pat could repeat with the power off, and if your picture had the same cabling setup.

Cheers, John

[Joseph K]

High John,

I will try to clarify:
My setup, resulting in that shot, was: TDR - 50ohm (Lemo) cable 8nsec - Lemo conn. - Circuit under test direct at conn.,on PCB, 50ohm trace.

Circuit under test:



R1=50ohm (smd) R2=1kohm (smd)

The TDR sends a ~ 90psec rise time, 400mV magnitude step signal
Horizontal resolution is 5nsec. as you see, the reflection arrives at 16nsec later, 8nsec RG174 cable.
We are looking at what is reflected from the "input".
That first pic was taken with full size incidental signal - the output should have gone to -8V, but was saturated at < -5V.
Attenuating the input, the circuit gets out of saturation, and shows this reflection:



Which is more like that shown by Pat.

[Joseph K]

What happens is: in theory, the C.U.T. above should be like the input is terminated in 50ohm. The 50ohm resistance is connected to a virtual ground.
And it is, once all transitions are settled down - that is, the output had reached it's final value. But it takes time, the "settling time" of the amplifier, to reach this equilibrium.
During this transition, as the current starts to flow slowly, we will see an input impedance slowly decreasing from ~infinite (R1+R2+stray inductance)
to the value in equilibrium, that is 50ohm. This may look like an inductive spike, but is "output settling reflected".

I think what can help to distinguish between the two is that it would take a real big stray input inductance in the circuit, to reach that monster timing in the shot of Pat: tau is something like 20 nseconds?
As amplifier settling time, it's even quite nice.

In my case, tau is ~ 7nsec, and that is a 1.5 Ghz GBW amplifier.   

[Joseph K]

Ok, finally I did get what your objection was about:

In my case, the amp is slew rate limited, not bandwith limited. (400V/usec; .4V/nsec - and had to slew ~ 5V)

If the amp in Pat's case was reasonably designed, the output slew had been chosen up to task..

I had only grabbed the first proto board well at hand, for that demo.

[art]

I already did that...........by accident. It is exponential, for a reason, and that reason is inductance.

Don't make go out to my cold garage, and take the lid off of that DAC. (Hint: It was made by a company that starts with "B". There, that will give you guys something of real value, to speculate over.)

Pat

[art]

Well I don't know what the big deal many audiophiles have against 3m digital cables.  It seems that could only help placement of gear.

If you ever figure out what audiophiles want, and why, please let us know. We have not been able to figure that out, in over 20 years, in business.

Pat

[art]

Ok, lucky for you guys, it hasn't gotten that cold. Yet.

Yep, transformers. Didn't see any op-amps.

Well, may have been some for the audio output, but they don't count.

Pat

[rfluongo]

(Translation: if the guy who designed your USB-SPDIF thingie copies the TI data sheet, you should expect problems. Because the data sheet is wrong, and he/she/they didn't figure that out. Obvious, to some of us old farts.)

Pat

Lucky for me the guy who designed mine knew what he was doing 
Ron

[jkeny]

(Translation: if the guy who designed your USB-SPDIF thingie copies the TI data sheet, you should expect problems. Because the data sheet is wrong, and he/she/they didn't figure that out. Obvious, to some of us old farts.)

It's only wrong if using the old TI datasheet, methinks!

[jneutron]

I've not seen this level of understanding often..thank you both.

What happens is: in theory, the C.U.T. above should be like the input is terminated in 50ohm. The 50ohm resistance is connected to a virtual ground.
And it is, once all transitions are settled down - that is, the output had reached it's final value. But it takes time, the "settling time" of the amplifier, to reach this equilibrium.

Agreed.  That node is indeed virtual ground.    My thinking:  Given your initial waveform, the starting slew looks like the internal circuitry of the amp is not fully engaged yet...this could be supply layout, decoupling, or ground layout.  As the circuit inductive reactance decays, the amp gains more and increases slew..  Your followup is clearly more the classical amp thing.

During this transition, as the current starts to flow slowly, we will see an input impedance slowly decreasing from ~infinite (R1+R2+stray inductance)
to the value in equilibrium, that is 50ohm. This may look like an inductive spike, but is "output settling reflected".

One must also take care of the inductive loop of the current being used to keep the virtual node.  The current has to traverse the fb resistor, the chip outputs, through the rails to the actual ground the input line sees.

I think what can help to distinguish between the two is that it would take a real big stray input inductance in the circuit, to reach that monster timing in the shot of Pat: tau is something like 20 nseconds?

A twisted pair of wires runs about 200 nh per foot.  15 nH per inch roughly..  Sometimes this parasitic is ignored.  Sometimes a designer will break t-line impedance in stripline to active, stripline to coax, coax to pc board, or strip to xfmr..  So many rules, so little time...

In my case, tau is ~ 7nsec, and that is a 1.5 Ghz GBW amplifier.

Rather fast, nice work.  I've worked circuits with tau in the nano range as well as roughly 50 pico, and, Toto...we ain't in Kansas anymore...

Ok, finally I did get what your objection was about:

Hmmm...I'll have to remember to be careful in what I say around you....there's no dust settling on you...sheesh.

Cheers, John

ps...actually, I've never seen this level of understanding on any forum anywhere.  nice..