About 2000 posts ago I commented on your layout regarding those decoupling capacitors. I'm using the "recommended" approach (vias only) but whether or not that results in improvements is another question. Also I'm using X7R decoupling caps which may be slightly more linear than X5R.
^ 1+
When designing stuff like this (look at it as a low power but very high frequency "switching supply"), what I typically do goes like so:
This is pretty much a brute force approach but tends to give an excellent starting point for fine-tuning, looking at all the high dI/dt loops and high dV/dt surface areas.
It may be sometimes beneficial to partly localize some of the GND and power plane layers, often by making them peninsulas with a well-defined bonding point to the main copper sheet. The idea being here that any fast and strong transient currents don't flow accross a larger area which carries any kind of reference potential.
In this particular case one might also consider placing the two shifters on opposite sides of a board for smaller/shorter high-current loops and more inner symmetry.
When designing stuff like this (look at it as a low power but very high frequency "switching supply"), what I typically do goes like so:
- 6 layer board preferably with a single core and 3 layers on each side, separated by pre-pregs, 35u copper, total thickness as thin as possible. All layers are solid GND fills and stitched together with a via every 5mm or so. Basically a solid sheet of ~0.2mm copper where I then "carve out" the circuit.
- power planes extend the solid copper sheet by treating them just as if they were GND, which they actually are for AC. That means you replace / extend the stitching vias with stitching caps, strictly via-connected (several pcs. per pad). Same for chips. Top GND is used for signals only but larger areas may have isolated GND fills with vias
- always keep in mind that for RF the two sides of a plane are basically isolated from each other (skin effect) and any fast current path (of signal or supply) which switches sides needs a bunch of connecting via holes.
This is pretty much a brute force approach but tends to give an excellent starting point for fine-tuning, looking at all the high dI/dt loops and high dV/dt surface areas.
It may be sometimes beneficial to partly localize some of the GND and power plane layers, often by making them peninsulas with a well-defined bonding point to the main copper sheet. The idea being here that any fast and strong transient currents don't flow accross a larger area which carries any kind of reference potential.
In this particular case one might also consider placing the two shifters on opposite sides of a board for smaller/shorter high-current loops and more inner symmetry.
To the extent possible, I try to use the inner layers as shields between the data-handling circuitry and the clock and reference circuitry. That's incompatible with putting one 74LV574A on the top side and the other on the bottom side.
If anyone wants to make a version with more layers, just go ahead and report the results, preferably with the design and layout databases included.
The past extensive discussions concerning RTZ, NRZ, Interleaving, 4//4 or 4+4 SR may be revealing, but the only thing that really matters is the effect on sound perception.
Isn’t the point reached to start investing some time in trying one or more of these options ?
Probably the easiest one is putting the shift registers in series to get a 4 bit Firdac instead of the current 2 bit version.
Hans
Isn’t the point reached to start investing some time in trying one or more of these options ?
Probably the easiest one is putting the shift registers in series to get a 4 bit Firdac instead of the current 2 bit version.
Hans
Maybe I missed it in this long thread but in short: NRZ: sinc filter, less HF and lower jitter sensitity, sensitive to inter symbol interference (ISI). RTZ: low sensitivity to ISI but more HF and hence jitter sensitive. interleaved RTZ with half clock cycle: sinc filter (low jitter sensitivity) and low ISI sensitivity.
It still works to some extent, you are basically high-pass filtering it.
Suppose that each 74LV574A has some data-dependent supply current variations due to systematic mismatch due to different internal supply and ground wiring of its flip-flops (or whatever other cause). Without extra delays, the data-dependent supply current of the second 74LV574A with swapped sd and sdn will then be equal but opposite to that of the first one with non-swapped sd and sdn.
Calling the (Laplace transfer of the) data-dependent part of the supply current of the first 74LV574A Idatadep and the sample time (bit clock time) T, the sum of the data-dependent supply current variations of both 74LV574As will be:
No delay:
Idatadep - Idatadep = 0
Perfect suppression, if the 74LV574As and the components and traces around them would match perfectly (which they don't).
Half bit clock cycle delay:
Idatadep - esT/2 Idatadep
High-pass filter.
Two bit clock cycle delays, so four clock2f cycles delay, so with the four-tap shift registers cascaded (the Hans variant):
Idatadep - e2sT Idatadep
Comb filter.
Using s = j2πf, the comb filter will have notches at f = k/(2T) for all integer k, so exactly at all multiples of half the sample rate. That means that with the Hans variant, at least anything that could mix idle tones around half the sample rate down to audio frequencies would still be suppressed. This is a very interesting feature of the Hans variant, I didn't realize it has this feature until today.
Yes, that holds for any frequency far above 277 Hz (assuming the 22 uF capacitors have about 80 % of their nominal capacitance at 5 V).
Last edited:
Yes, that's been mentioned several times in several phrasings (but the thread is indeed long). Do you regard an even-order uniformly-weighted RTZ FIRDAC with half-bit-clock-cycle delays between its taps as interleaved?
Well, you could build it. The data inputs to the second shift register are hardwired, so no one has the hardware to try your variant without damaging a PCB.
I just found an interesting feature of your variant, see post #3428.
Last edited:
Indeed, at 1210 format these capacitors have a resonance around 1MHz. I would have used TDK C3225X7R1C226M250AC but they all look very similar)
Still 100mOhm @ 15MHz and ~600mOhm @ 100MHz is not too shabby.
Better decoupling is possible.
1uF/X7R in 0603 (e.g. C1608X7R1E105K080AE) X 4 at all four corners of each IC with 3 vias per pad (as per H. Ott). For two DAC's this would give us a decoupling network with 10mOhm @ 15MHz and < 60mOhm @ 100MHz (or a 20dB improvement in performance).
Together with the parasitic capacitor formed between grounds and power planes we should have a clean low impedance supply to >> 100MHz.
IF possible, using 0402 formfactor is better, but no way I hand solder these. 0603 is absolutely minimum. For 0402 machine assembly is mandatory.
Using 0612 "reverse geometry" parts with larger values is a good way to get further improvements.
I'd also like to see a solid ground plane directly under the IC (IC side) that links both IC's and links both IC's with a solid copper area and a power plane that mirrors that. The ground plane under the IC will lower impedance for the decoupling.
I'd then add a pair of global ~500uF Os-Con L/R of each DAC section (two IC's) (~1,000uF in total).
Anyway, we were not discussing layout and decoupling which would need a total PCB redesign. At that point many other options open up.
Thor
There is no need to damage the PCB.
Just lifting pins 2 and 3 from one SR and connecting them to pins 12 and 13 from the other SR will do.
Timing of these two signals is not critical, so two short connecting wires will do the job.
Hans
Good point. I can lend you my prototype again if you want to try it. It would be interesting to see what it does to the noise floor and the conversion of stuff around fs/2 to audio frequencies.
That would be fine, but not within a few weeks.
I’m currently involved in a PP tube OPT project .
Maybe someone else could try ?
If nobody does, I will accept your offer.
Hans
To some extent I think you guys may be missing the forest for the trees.
I know a very good and successful high end analog audio designer who takes a very different approach to his work. When we first met a few years ago he listened to my 2nd ES9038Q2M dac. He said its good enough to make into a product, which he said I should do. I didn't want to commercialize it, but he decided to try take me under his wing anyway and teach me his way of design. Being an EE by background and being very measurement oriented, I didn't get it. He tried mind-dumping on me faster than I could absorb it all. Eventually he became frustrated with me for not growing faster into being someone with skills like his. I told him I need to take the time to learn on my own and develop my own ways of his approach to audio design. However he saw it as a waste of time that he had tried to help me bypass.
One of the most remarkable things about this guy is that he knows how to listen only to what's real. Its very difficult to master; he must have been sort of a natural to begin with. Listening along with him, I get better at it slowly over time though. I also get better at my way of his approach to audio design as opposed to the standard EE way. Listening, analyzing sound, recalling experiences with similar circuits, and much more is involved. I have been asked not to speak of it in more detail.
Anyway here is my view, you guys don't know what's wrong with the sound of Marcel's dac now. Don't know what it needs most. Maybe Thor knows the most about such things. IME when Marcel's dac is surrounded by good support circuitry and played through a very accurate reproduction system (again, I have been asked not to go into detail about what that entails), and when it is listened to without bias, just for what the sound really is, then it starts to become clear what needs attention most right now. I call it "peeling away the layers of the onion," because one small problem tends to mask the next smaller problem. IMHO all those layers of problems have to be found and fixed to make a top-notch world class design. The problem is that some of what needs fixing isn't so easy to measure and some areas are not rich in terms of theory relating audio design to the physics of what influences sound.
So it seems to me focusing on areas where theory is best and meter readings can be chased is fine only if the end goal is to satisfy some meter (which it most often is for EEs). However, for audio the goal should be to please the ear with a wonderfully realistic listening experience. If that is accomplished properly then IME the meters will read quite acceptably good anyway, yet other things not so easily metered or theorized about will be quite good too. IOW, a different approach leads to a different corner of the optimization space. Meters used exclusively or excessively don't always lead to an especially good local optimum in that space.
I know a very good and successful high end analog audio designer who takes a very different approach to his work. When we first met a few years ago he listened to my 2nd ES9038Q2M dac. He said its good enough to make into a product, which he said I should do. I didn't want to commercialize it, but he decided to try take me under his wing anyway and teach me his way of design. Being an EE by background and being very measurement oriented, I didn't get it. He tried mind-dumping on me faster than I could absorb it all. Eventually he became frustrated with me for not growing faster into being someone with skills like his. I told him I need to take the time to learn on my own and develop my own ways of his approach to audio design. However he saw it as a waste of time that he had tried to help me bypass.
One of the most remarkable things about this guy is that he knows how to listen only to what's real. Its very difficult to master; he must have been sort of a natural to begin with. Listening along with him, I get better at it slowly over time though. I also get better at my way of his approach to audio design as opposed to the standard EE way. Listening, analyzing sound, recalling experiences with similar circuits, and much more is involved. I have been asked not to speak of it in more detail.
Anyway here is my view, you guys don't know what's wrong with the sound of Marcel's dac now. Don't know what it needs most. Maybe Thor knows the most about such things. IME when Marcel's dac is surrounded by good support circuitry and played through a very accurate reproduction system (again, I have been asked not to go into detail about what that entails), and when it is listened to without bias, just for what the sound really is, then it starts to become clear what needs attention most right now. I call it "peeling away the layers of the onion," because one small problem tends to mask the next smaller problem. IMHO all those layers of problems have to be found and fixed to make a top-notch world class design. The problem is that some of what needs fixing isn't so easy to measure and some areas are not rich in terms of theory relating audio design to the physics of what influences sound.
So it seems to me focusing on areas where theory is best and meter readings can be chased is fine only if the end goal is to satisfy some meter (which it most often is for EEs). However, for audio the goal should be to please the ear with a wonderfully realistic listening experience. If that is accomplished properly then IME the meters will read quite acceptably good anyway, yet other things not so easily metered or theorized about will be quite good too. IOW, a different approach leads to a different corner of the optimization space. Meters used exclusively or excessively don't always lead to an especially good local optimum in that space.
Last edited:
Underneath for symmetry is an obvious idea, don't you think? There is no other way to get full symmetry given that many components are asymmetrical by nature. I proposed the same point to my analog audio designer friend some time ago. He said underneath may be closer to the chassis which may may then cause its own asymmetry, and another set of problems.
IMHO and IME he had a point. For example, IME radiated EM fields and their interactions are not necessarily a trivial matter for very high performance dacs.
Quite some time ago you reported results of extensively testing Marcel’s Firdac in SE transferring the signal to a non disclosed transformer
At a later stage you mention to have found ways to improve sound even further, but you can’t tell all details.
In no way wanting to be rude, but IMO either disclose what you think can be considered as steps forward, or keep these secrets for yourself without referring to.
Maybe your intention is to commecialize a Firdac ?
Hans
Hans,
I recently designed and gave away some open source boards in another thread. There are pics showing how I hooked things up to Marcel's dac. I have just given away a lot more than people will realize until a few people start building Marcel's dacs with that type of support circuitry. Its a whole new world there. Also there are some hints in the thread and in the designs that should give rise to taking another look at Marcel's already very good dac. However, I see more schematics for the reclocker are being downloaded than design file downloads; some people may not understand how the layout can sometimes be just as important if not more so than the schematic.
Anyway, please try that stuff first then we will see what might come next.
Also please be aware that everything needs at least running overnight to settle in. Iancanada clocks take almost a week of full time running to pretty much fully settle.
In addition, please see comments in post #144: https://www.diyaudio.com/community/threads/general-purpose-dac-clock-board.413001/post-7729060
Mark
P.S. I am investigating to see if I can find any off the shelf transformers that can do the dac justice. Don't know how that will turn out.
P.P.S. Also I am trying to walk a very fine line here. I want to see diy dac building advance, however I am limited in terms of what I can say. The basic agreement I have is that I can give hints, suggestions to experiment with, but not final designs. I am already probably stepping over the line a bit much on favoring diy progress, so will need to try to be patient for awhile and see if the seeds I planted will grow. I know that might sound puffy, and if so then sorry about that. Just try the boards please, then see.
I recently designed and gave away some open source boards in another thread. There are pics showing how I hooked things up to Marcel's dac. I have just given away a lot more than people will realize until a few people start building Marcel's dacs with that type of support circuitry. Its a whole new world there. Also there are some hints in the thread and in the designs that should give rise to taking another look at Marcel's already very good dac. However, I see more schematics for the reclocker are being downloaded than design file downloads; some people may not understand how the layout can sometimes be just as important if not more so than the schematic.
Anyway, please try that stuff first then we will see what might come next.
Also please be aware that everything needs at least running overnight to settle in. Iancanada clocks take almost a week of full time running to pretty much fully settle.
In addition, please see comments in post #144: https://www.diyaudio.com/community/threads/general-purpose-dac-clock-board.413001/post-7729060
Mark
P.S. I am investigating to see if I can find any off the shelf transformers that can do the dac justice. Don't know how that will turn out.
P.P.S. Also I am trying to walk a very fine line here. I want to see diy dac building advance, however I am limited in terms of what I can say. The basic agreement I have is that I can give hints, suggestions to experiment with, but not final designs. I am already probably stepping over the line a bit much on favoring diy progress, so will need to try to be patient for awhile and see if the seeds I planted will grow. I know that might sound puffy, and if so then sorry about that. Just try the boards please, then see.
Last edited:
There are very good and successful high end audio designers who take a very different approach to what you described. Some here may prefer that approach instead of yours.