If it were data-related, it would be a disaster. As long as it's clock-related, it is harmless.
What about in terms of jitter, say, due to a shifting clock switching threshold related to noise on Vdd? Sorta seems like @ThorstenL said something about that?
I never trust the ground. What happens with the probe directly connected to the same ground point? This is the baseline of the comparison though not necessarily all else that can happen to produce inaccurate results.
Last edited:
The probe was connected right across the capacitor using a spring ground accessory, if you know what that is. Looks kind of like this:
So you get sort of a differential measurement across the cap, at least at high enough frequency. And you don't get all the probe ringing due to ground wire inductance.
So you get sort of a differential measurement across the cap, at least at high enough frequency. And you don't get all the probe ringing due to ground wire inductance.
This looks very good but may still not reflect zero signal output with the probe tip connected to the ground point as connected to the circuit at that point. The circuit is powered from some AC source and so is your scope. You are working at RF frequencies often very difficult to get a handle on. Basically thats why your active probes exist. You may be alright though not sure as there seems bound to be something left.
You could use two probes both attached to the same ground point, with one tip to ground and the other to the cap and subtract them digitally.
You could use two probes both attached to the same ground point, with one tip to ground and the other to the cap and subtract them digitally.
With an active differential probe, there is also a loop from the circuit to the probe, through the probe and back that could magnetically couple to something. There is not much you can do about it but keep the loop area small (like Mark did) and hope for the best. As a sanity check, you can measure between two ground points, as Mark also did.
Sure, similar to using x10 settings.
I've used active probes when I was still working in Engineering at UWO. After retiring I tested numerous conventional I/V networks across their differential inputs by embedding a 100x differential amplifier into the network on the same supplies. This is similar to the testing of I/V converters of the AD811 engaged in by Walter Jung. You still need to be careful but the results were considerably more believable.
I've used active probes when I was still working in Engineering at UWO. After retiring I tested numerous conventional I/V networks across their differential inputs by embedding a 100x differential amplifier into the network on the same supplies. This is similar to the testing of I/V converters of the AD811 engaged in by Walter Jung. You still need to be careful but the results were considerably more believable.
The question is, how exact to we need the measurements to be in this case? I mean, we don't have figure of merit numbers for scope waveforms. So we can't do a numbers game like THD so easily. If there is a particular need for exactness then we need to know what it is. Its kind of like if we have a tire swing hanging from a tree in the back yard, do we need .0001" dial indicator to measure the swing motion repeatability? If so, why exactly? For what purpose?
In the case here, IMHO this is sort of an eyeball it and see what you think type of measurement. To me, it says I might be able to get a little better sound if I put the shift registers on a daughterboards and clean up the power as best as I can using more PCB layers to help keep supply impedances low, as well as keeping bypass impedance low inductance (all of which could likely be better than what we have now).
In the case here, IMHO this is sort of an eyeball it and see what you think type of measurement. To me, it says I might be able to get a little better sound if I put the shift registers on a daughterboards and clean up the power as best as I can using more PCB layers to help keep supply impedances low, as well as keeping bypass impedance low inductance (all of which could likely be better than what we have now).
Last edited:
Sure Mark4. In my experience many networks used by myself and others didn't have ground planes, whereupon the output signal observed was no different from a probe connection alternatively applied to the ground point of the probe. This is to say the signal never existed at all, as a fiction, often with ringing too. My concern is in the chasing of nothingness.
Actually, I didn't want to get to what I am sort of considering in the back of my mind so soon in the conversation. I really wanted to engage with Marcel first and see what he thinks, as I value his opinions. Maybe I am jumping to conclusions or would change my mind. Don't know.
That's a heck of a lot of LIM (Logic Induced Modulation) right there.
Well, C48 is 1210 format and 22uF, ideally we adjust decoupling to use more smaller capacitors with their minimum impedance near the shift register clock. I think 0603 1uF X8R would do.
I usually prefer C0G as they are not microphonic, which means 0.47uF/1206 and a big budget for passives. I just ordered a reel for some didital stuff I'm working on. Cheaper than buying a few 100. I wish someone made readily available 0.47uF C0G on 0612 footprint (reverse geometry).
Stacked Film Caps in 1206 and 1210 have limited utility, they are good for old slow IC's like TDA1541, which I happen to like. But so are C0G in 1206.
For bulk decoupling the Tantalum Anode Polymer Cap's (aka Panasonic PosCap) look much better than high K ceramic.
I recently got more into Elna DynaCap super Capacitors. The 8mm X 11mm DU series 3,000,000uF/2.7V measure 60mOhm ESR with a resonance above 100kHz, impressive. Just add a PosCap with 1MHz resonnace and 1206 C0G with > 10MHz, very good decoupling, if very slow PSU rise time (minutes in most cases). It means adding a voltage monitor and a low resistance P-MOS between the supercapacitors and the rest of the PSU, to get fast enough rise time for most circuits to start up clean.
Also, solid ground and power planes should be used, with triple vias directly at the Vss/Vcc pin's. Thermal reliefs are devils work. They are not needed for reflow or hot air rework on hot plate.
Thor
Well, the ripple appears after the clock edge I think. But I think the resonant circuit still rings when the next clock edge arrives, that could throw out thresholds.
More critical, the on chip triggering of the individual inverters making up gates making up flip flops will be affected. How much, hard to say.
I think Ed Meitner had an estimation for CMOS logic how much Logic noise equates to how much jitter. I cannot recall the exact number, but it was rated in mV/ps...
Correct parts selection, layout and decoupling can easily get single digit numbers for logic noise on power pins.
Amusing fact, a product where I had closely overseen the PCB design for was upgraded and the whole layout redone (and the USB processor changed), without me having a hand in it. Parts and schematics for DAC, Power, reclocking, Clock etc. identical.
After measuring ~ 10pS jitter for the original, the redone version was measured by the same magazine at nearly 1nS jitter. That's just PCB layout.
Thor
There is another dac I have which uses tantalum for bulk decoupling, SMD film for output flip flops, and ceramic (likely X7R) for incidental logic (shift registers, etc.) prior to the output flip flops. It also used ferrites on the logic power input filter. It was getting to the point it didn't sound as good as Marcel's dac (given evolving improvements to Marcel dac's support circuitry) so I decided to do an experiment. Replaced all ceramic caps on the other dac with .1uf, 805, SMD film caps. Replaced all bulk decoupling tantalums with 47uf, 50v leaded electrolytics (soldered to the SMD pads). Removed all ferrites and replaced them with jumpers. Let it run a day to settle in, then listened. Everyone agrees it now sounds much closer to Marcel's dac (except for a little sound stage depth), and that it sounds clearly better than it did before (still suspect there may be a remaining unfixed issue with a little clock skew on that other dac).
Thus I remain a bit leery about some of the standard approaches to decoupling. Not sure what is best though, so will probably have to do some prototyping to see what I can find out.
NOTE: Listening comparisons were done with both dacs using identical output stages (DC blocking, then into transformers). Volume levels were also adjusted to account for differences in Vref voltage.
Thus I remain a bit leery about some of the standard approaches to decoupling. Not sure what is best though, so will probably have to do some prototyping to see what I can find out.
NOTE: Listening comparisons were done with both dacs using identical output stages (DC blocking, then into transformers). Volume levels were also adjusted to account for differences in Vref voltage.
Last edited:
Just for comparison my take on RTZ dac has about half the noise on shift register decoupling caps.
I have 10uF X7R || 100nf C0G (both 1206) on shift registers. No doubt decoupling could be made even better but whether or not it brings something else than peace of mind is another thing. Many moons ago I tried a tantalum polymer instead of the X7R but that did not make any difference on dac output (did not look with scope back then). I believe it was this cap: https://www.mouser.fi/ProductDetail/KEMET/TF08A226M016APE200?qs=W/MpXkg%2BdQ565wPnGhuDzQ==
I have 10uF X7R || 100nf C0G (both 1206) on shift registers. No doubt decoupling could be made even better but whether or not it brings something else than peace of mind is another thing. Many moons ago I tried a tantalum polymer instead of the X7R but that did not make any difference on dac output (did not look with scope back then). I believe it was this cap: https://www.mouser.fi/ProductDetail/KEMET/TF08A226M016APE200?qs=W/MpXkg%2BdQ565wPnGhuDzQ==
Tantalum <> Tantalum Polymer
Tantalum Polymer ard more like Os-Con, except chipscale.
Traditional Tantalum capacitors are high ESR and pretty useless for decoupling and highly nonlinear in coupling applications, only having small size in their favour.
Tantalum Polymer are low ESL and low ESR and linear, in the precise way traditional Tantalum capacitors are not.
I don't do "standard decoupling".
As a rule, scale the Resonance frequencies (not values) by a factor 10.
So you could get 6,000,000uF and ~100kHz resonance with 30mOhm ESR, in parallel with a 2220 size pos-cap 100uF with a resonance at ~ 1Mhz and 10mOhm ESR and a group of ceramics (C0G preferred) of (say) 1.94uF total (0.47uF x 4 per IC) with a resonance > 10MHz and < 10mOhm for a passive decoupling combo.
This would give 0.03 Ohm or less from around 1Hz to well over 30MHz.
Without specific listening, in terms of noise on PSU pins with good layout, this set-up is extremely low.
As a general rule, I think we can agree that less noise is better on principle. This is ultimately down to basic electrics.
X7R or Y5U etc are significantly microphonic.
If applied like idiots, following bad advise, would, instead of a contiguous power plane for Vss and Vcc applying ferrites or inductors in the Vcc plane, which now is now not a plane, the microphonics will cause serious problems.
The only fix without power planes is to get rid of them. In a system with proper design, the microphonic capacitors have an extremely low "look back" impedance, with any microphonic noise sunk by local low impedance non-ceramic bulk decoupling capacitors and regulators.
So what should have been done was to simply remove ferrites and short them out, leave X7R decoupling capacitors in place and replace bulk decoupling capacitors with stable low impedance types.
By changing too many variables at the same time, it becomes hard to nail down what was the main mechanism of improvement.
Thor
General note if considering noise measurements: Didn't measure ringing frequency on my Marcel dac bypass caps yet; maybe will next time. If ringing is up around 100-200MHz could be some attenuation on some scopes due to scope bandwidth. Maybe a few dB down down? For the scope I've been using here the passive probes and scope are rated for 600MHz bandwidth (need to check how they define it). Active probes rated for 900MHz. Haven't calibrated frequency response up there, but maybe I should.
Yeah. The way I told the story was a bit condensed ("Reader's Digest" version). Ferrites came out first, then I listened. Sound was better.
Later ceramic caps were replaced with SMD film; after a day of running I listened. A little better again, using Marcel dac as the unchanging reference.
Later tantalums (don't have a part number but they were yellow rectangular in appearance) were replaced with aluminum electrolytics. Better but a bit rolled off in the sound.
Added .01uf MKP across across each electrolytic. That fixed rolled off sound. At that point I had something pretty close to the sound of Marcel's dac. This dac was slightly warmer in the midrange and slightly worse in sound stage depth, as compared to Marcel's dac. (Both dacs were running the same PCM2DSD modulator, both had the same output stage.)
Then invited other listeners to evaluate. For them it was all the changes at once as compared to Marcel's dac at that state (like it is now). I don't count the listening as done until the group opines on it. However I don't convene the group until I think I have something to listen to worth their time.
Last edited:
Hmmm, 100mOhm ESR is kinda high, around 10 times what you get from Pos-Cap.
They are probably a poor choice to replace X7R ceramics.
Even generic Electrolytic capacitors offer lower ESR.
Thor