Because we are addressing a very different problem.
The problem is the same, no matter if the regulator following is a Germanium Transistor with Germanium Zener diode, 7805, LM317 or ultra supa dupa complex self biasing 10A CCS with 10Mohm flat impedance up 10Mhz followed by a 10A capable shunt with 100nOhm Impedance flat to 10MHz.
Of course, the circuit supplied will react differently to each regulator, but that is not our concern for transformer/rectifier interactions.
Leakage into chassis, mains, earth, ground.
The transformers secondary inductance damped by mains impedance plus leakage inductance will ring with all the capacitances in the transformer each time the rectifier diodes switch off.
A resistive load will damp it.
A subber can damp it.
If we make a PI snubber with R 2 pcs 1/2 effective DCR at secondary and set the output capacitor so that we get a cutoff for harmonics above the 10th mains harmonic, we get damping and also isolation plus softer switching from the rectifier.
If we follow the rectifiers with a similar snubber / filter we will absorb reverse recovery glitches.
Say I have effective 3.3R secondary impedance at mains frequency, I set my resistors to 2 pcs 1.5R, our C would need to be 100uF, so Nichicon ES, or we use film and relax the 10th harmonic requirement.
Input capacitor then becomes at least 1uF film.
Add 1uF film at the rectifier output then again 1.5R 2pcs (or chokes with sufficient current rating and ~ 1.5R DCR) and follow with the first reservoir capacitor (1uF film bypass in case).
Add multiple RC (or RC/LC) cells, usually after the 4th cell noise remaining is below 1mV (if not, recalculate) and has no upper harmonics, being almost pure sine.
In this rectifier circuit now all current changes and voltage changes are "soft" and all ringing is well damped while HF noise is filtered well.
This is a specific example the actual sizing is effective DCR (includes primary), allowable drop on the DC output determines resistor, then size capacitors to suit.
I prefer this over the bad habit of using unnecessary complex regulators with extreme performance, to cope with the noise created by extremely basic rectifier circuits.
Solve the problem at the root, instead of maximising the problem and then trying to fix it.
An example of someone partially implementing my style for a 48V phantom power supply for Microphones, max 60mA expected. Implementation as I do, based on what is in the junk boxx:
Overkill for 48V / 60mA? Definitely.
Thor
Last edited:
For those who wants to design/layout themself a digital 4 layers pcb , here is a link to the Altium Academy on youtube
they have tons of good vids , especialy those from Zach Peterson , a very good learning place 😉
on some of them they talk about thermal settings and teardrops , very important too 😉
.
they have tons of good vids , especialy those from Zach Peterson , a very good learning place 😉
on some of them they talk about thermal settings and teardrops , very important too 😉
.
Last edited:
Brijac, you are liked by the most subjectivist here with big onanism thd measurements showing more about the guy brain than the sound, trying soon to explain us he understood everyrhing.
Okay, I do the same as you do about datasheet, Parts, blah that bring nothing everyone already does, plus I believe knowing that chip more rhan you just as you beginn with.
Your words about snake oils jugement show you know few. Yeah you putt kz everywhere in the ad1862 thread. I have not that said subjectivité approach, sorry. Soon you will learn us what is à D3A and so on... We are not born yesterday. I wait more from you here.... please. In the other threads some talk about Russie good tech about tubes that are poor copies of ECC88 or those last was best fab in east Europe on tired tools. We hope not going that low here.
I hope we are not ending with pictures boards and management project, uh!
Do the pcbs, I will be happy to buy you one if I have no time to do it myself, and resist the lord of the ring effect, I know here à guy from Aussie that did like that, and serious experienced guys, which have long expérience about tda1541a think it is poor designed (and alas sounding one). Better to follow the good guys here and resist to the ring effect. Imoh, ymmv.
Okay, I do the same as you do about datasheet, Parts, blah that bring nothing everyone already does, plus I believe knowing that chip more rhan you just as you beginn with.
Your words about snake oils jugement show you know few. Yeah you putt kz everywhere in the ad1862 thread. I have not that said subjectivité approach, sorry. Soon you will learn us what is à D3A and so on... We are not born yesterday. I wait more from you here.... please. In the other threads some talk about Russie good tech about tubes that are poor copies of ECC88 or those last was best fab in east Europe on tired tools. We hope not going that low here.
I hope we are not ending with pictures boards and management project, uh!
Do the pcbs, I will be happy to buy you one if I have no time to do it myself, and resist the lord of the ring effect, I know here à guy from Aussie that did like that, and serious experienced guys, which have long expérience about tda1541a think it is poor designed (and alas sounding one). Better to follow the good guys here and resist to the ring effect. Imoh, ymmv.
Last edited:
Geez, malicious much 😅 Subjectivism in physics, how. You know what, not gonna even go othere with a loony 😉 Ignore option is a bliss.
Thanks for the explanation. I suppose it's not straightforward to trace all possible paths for noise / parasitic caps / leakage, that would allow a complete understanding of this.
Btw, I do not have mains earth in the old apartment I live in (very nice home otherwise). So I cannot employ the three caps shunting noise to mains earth. I use AC filters with common mode choke before the transformers, but not hooked to mains earth.
To add one nonsense reason - with low -Ug value of Rk become small and there is a need for BIG Ck value... 🙂
If we want full amplifivation and non local feed back. That Ck value determine LF phase and roll-off. And gaving sound signature. And it is not derived as independent RC net with simple Fo formula... It is a part of wider system, with a tube, power C at anode etc. Should be detemine by simulation regard to -db roll-off and phase in LF.
🙂
So the bigger negative grid bias - bigger value if Rk - smaller value of Ck.
Simulations have to be compred with direct negative grid bias, with tha same value of anode C in PS. Because in that case we dont have RC on the cathode.
.
I will draw sch and post transfer to ilustrate this. But it is pretty obvious?
These tubes have open anode. Huge space... And they high transconductance types. They have specific sound signature.
We want opposite, close-box anode, high voltage, preferably DH, with low Ri (as it could be 🙁 ). And available.
These types are super linear.
But they dont come together with low Ri (to avoid buffer and have chance for simple OT optionally) and high mju...
.
Like 5687, closed box anode, medium-low mju, and lower Ri...
.
You can use classic and available EL84 in triode mode, to additionally amplify signal after BJT IV, and correct phase.
The main aim is to use a normal unshielded (no screens between windings) transformers and have minimal parasitic leakage into the Audio "ground" (remember ground isn't) by closing current loops back into the mains (PEN) prior to ever really entering the device.
The second aim is to minimise Di/Dt and Dv/Dt which minimises both conducted and radiated EMI.
Ignoring for a moment the fact that this a serious death trap and against code anywhere in the civilised world and should be rectified, as this "earth" is not for safety, it can often be improvised.
In such cases it is worth checking for metal water pipes. They can provide an earth. If the floors are reinforced concrete, a large sheet of copper on the bare concrete often makes a passable "earth".
Another option is to find the "cold" mains wire and use a few 100 nF to create a return to that "cold" that wire which is also "earth". In many systems back in the 80's the safety earth was created by simply connecting "cold" to "earth" in the mains outlet, the entire mains ring was 2-wire. This is no longer permitted, as a wiring fault that swaps life and neutral would be fatal.
The problem is, current flows in loops. Always. If you do not give the current an easy path to equalise potentials, all sorts of mayhem ensues.
Thor
Not at all. Capacitiors in Tube circuits for example are subject to microphonics. You will not measure it in a lab in normal test settings, but it still exists.
Most "exotic" capacitors share a few similar characteristics, namely thicker than average dielectrics, metal foil electrodes and often being encased in a rigid case, with some form of potting compound.
All these characteristics mitigate microphonics compared to a generic old style rolled up capacitor.
I found an extreme case of this with the AMR PH-77. I had a large stash of very nice NOS 1% polystyrene capacitors to use for part of the EQ. The DC voltage at the point where were used was only a few volt DC, and signal levels already had passed the first (J-Fet folded cascode) stage, so signal levels where robust. Nevertheless the capacitors were extremely microphonic and had to be replaced by custom Silver Mica to overcome this problem.
Of course, we could also use WIMA FKP with an elevated voltage rating. But they pretty much cost as much as the lower end exots.
The same issues apply to electrolytic capacitors, with the added issues around electrolyte formulation and aggressiveness of foil etching. Thicker, less etched foils and suitable damping spacer material reduce microphonics, but make for bulkier capacitors for a given size. Smooth foil electrolytic capacitors have disappeared, but had the best behaviour. Bipolar electrolytic capacitors are in effect two in one case wound together and thus again are more bulky, more linear and less Microphonic.
Ceramic, except NP0/C0g are a complete disaster for linearity and microphonics, though their high levels of H2 can be exploited to create a "Maxx Bass" like effect in a simple high pass, bypass with C0G for clean mid/high..
I do, in my own chain, minimal. In the studio, it really, really, really depends.
In my studio Mixer (modified Yamaha 01V-96II VCM) two, on the way into the box, only one coupling cap (it's balanced Mic-in to ADC-in - so in principle each Op-Amp is dual for two phases but both circuits are fully differential amplifiers made from a dual Op-Amp (Birt Circuit), and DC coupled except the coupling cap's blocking the Phantom Power for the microphones. Op-Amp's now are OPA1652 and OPA2156. No Op-Amp's in the microphone.
So "How many Op-Amp'?" can be from zero (I might convert the ADC in to transformer coupled, microphone pre to discrete with transformers and transformers only on the DAC out) to "how many NE553X can dance on the pin of a needle?".
That is like "pick a good car".
I think we need to be specific what is "Good" and why, in a technical sense, otherwise we are like Winnie the Poh: "Hunny is good!"
Thor
I'm not an expert.
I'm not able to perform instrumental or literary verifications.
I've nothing against anyone here.
I can only say what I think, for what it is worth.
I would like to say that this post should be framed as a manifesto for past generations and especially future ones.
In particular for those who use logic thinking that it can replace knowledge (as I've humbly said here).
What sane people with no personal interests at stake share with others dispassionately about their listening experiences are not snake oil just because it seems illogical, there is often an explanation (even if yet unknown).
Although sometimes they are, they possibly are exceptions, not the rule.
The human being typically, and especially when he has personal interests at stake, is almost never pure.
Last edited:
Measurements are the tool for those who do things in the real world and wants to get over their cognitive bias , without measurements you can listen crap without knowing it , just because you like it 😒
you have no idea how your brain is able to fool you , giving you what your ego likes , audio is way more than that
the point here is building a DAC based on the Thorsen and Zoran idea , it's time for you to demonstrate to us all what your are able to ........
.
Last edited:
I never said "Cap is no good". The CD-77 stock used Industrial Tinfoil Capacitors, which both objectively (HD, Microphonics etc.) and subjectively were exemplary. For some reason people insisted replacing them with often inferior "audiophile" parts.
The final version used a Ebony wood box within which a MKP, Tinfoil and a group of Silver Mica Capacitors were potted in resin.
Everything is a trade off. Capacitor coupling is more simple, DC coupling can be sonically more pure (CAN BE, not must be).
Mainly because it is an easily tractible part. Almost everyone can "roll" capacitors and tubes.
Not at all silly. Oscillators microphonic result in a different type of fidelity impairment compared to (say) Ceramic Capacitors in power supplies, Coils in power supplies and microphonics in coupling/decoupling capacitors.
Yes, in my more experience, the more simple a system the more complex the problems it evidences, while the on the other hand the more complex the system the more complex problems of simple systems are hidden from clear view, making people assume they do not exist (out of sight out of mind).
That is unnecessary.
You can disagree with people nicely, no need to be insulting.
Thor
It wasn't you. You popped into quote of the post out of nowhere, without being mentioned nor addressed to, into a convo line you didn't follow.
Quote the whole thing. I said in this regard. Again, you popped in a convo not related to your writing, and you clearly didn't read it trough.
As for the rest, i wrote about the whole thing microphonic. Read again, in this regard of the topic mentioned you popped in without reading it all.
Yet again, you popped in, speaking of something you didn't read trough, nor you know history of. It is merely a statement, a fact, and insults were firstly thrown in my direction.
OK, let's go back on topic.
Following all the debates and my inductive/deductive analyses of the likely makeup of logic inside the TDA1541, I concluded that we have two distinct sets of current loops.
First, +5V/DGND/-5V are one power section with +5V = Logic GND (ECL/CML). This is essentially logic only. There should not be signal dependent current, as ECL/CML is differential, non-saturating, AS LONG as the inputs are slew rate limited and have limited voltage swing.
Thus decouple -5V to DGND and use completely separate +/-5V supply (7805/7905 or LM317/337 probably ok).
As currents on +5V & -5V are about equal, using a single 10V regulator (7810 will be ok) and a TL431 to pass the excess current.
Normally this means TL431 clamps the +5V line to DGND at 5V, unless we want to allow for PECL re-clocking etc. which needs a lot of current. In this case we need a ballast resistor to draw most current and a TL431 across DGND to -5V.
When not using PECL relocking etc., placing TL431 in the -5V line and using a ballast resistor across +5V & DGND takes advantage of the 8dB larger PSRR on the -5V line. Note, I have seen zero enthusiasm to integrate PECL based re-clocking and signal conditioning into the TDA1541 DAC's, which I think is a shame. But so be it. The power supply will allow as much as 1A @ 10V, so it can power a bunch of ECL.
Further +5V/AGND/-15V are another power section with -15V = GND AND +5V = GND in AC terms. Yes, this seems to make no sense, but it is what it is.
This section is the actual signal current circuit.
I still think a low impedance shunt is best here. The TL431 at ~0.15 Ohm (or less) typical impedance is a good choice, with the upper resistor in the feedback divider bypassed for lowest noise. Here the target of 1.1k & 100uF gives ~ 1.5Hz turnover, adequate I think.
So +5V & -15V link via decoupling capacitors and the shunt regulator string. For this loop, AGND & +5V decouple to -15V. We need to remember that +5V carries the mirror image of the current flowing in the output pints.
With a 0.3R loop and 8mA P-P error voltages on +5V with respect to -15V are limited to 2.4mV P-P. In light of the respective PSRR on these pins feed trough into the output will likely result in a few 100nV worst case error and as this error is the reverse polarity of the signal with L/R summed, the effect is an ephemeral reduction in output level and some crosstalk at levels > 120dB below full scale.
So a fairly simple, complete and way past adequate supply is below.
Note that short return path are provided for high frequency common noise passing through our doughnut transformers (R-Cores, O etc are all the same here, without shield terrible noise leakage from all the SMPS's in anything from lightbulbs to TV's, Aircon etc.
Further, a direct current path from transformer / rectifier are carefully avoided.
To match this power supply, here the last adjusted version of the "Core", including all recommended protection.
If divider logic for DEM clocking is needed, put a a pair of 74XX74 footprints, including TH Options with SMD inside the TH including an option for VQFN-14 for TI 74AUC74.
The Power should be separate 5V or 3.3V (or even 2.5V) depending is using 74F74/74AS74 or CMOSL logic (74LVC74, 74AUC74 etc.) to avoid throwing in unnecessary noise into the TDA1541 supplies. A piggyback I2S to SIM Module is shown here. A discrete TDA1541 focused option should end up on the same PCB if desired.
I really think all this should be just a single PCB, including transformers, SMA, FL and XH input options (XH with micro coax and alternating signal/grounds works really well). It becomes biggish and a little expensive, but it is easily build and integrated with other BIY modules.
If breaking things up into multiple PCB's, place transformers, rectification and filtering on the Power PCB, including our 317 (or 78XX) Regulators on Heatsinks and the 1mH chokes as final elements and 431's and decoupling capacitors on the TDA1541 PCB. Make the Transformers a third "break-off" PCB, so those who want to use Magic Transformers instead of lowly Talema Torrid Toroids can do so.
Analogue stages, knock yourselves out.
Thor
Following all the debates and my inductive/deductive analyses of the likely makeup of logic inside the TDA1541, I concluded that we have two distinct sets of current loops.
First, +5V/DGND/-5V are one power section with +5V = Logic GND (ECL/CML). This is essentially logic only. There should not be signal dependent current, as ECL/CML is differential, non-saturating, AS LONG as the inputs are slew rate limited and have limited voltage swing.
Thus decouple -5V to DGND and use completely separate +/-5V supply (7805/7905 or LM317/337 probably ok).
As currents on +5V & -5V are about equal, using a single 10V regulator (7810 will be ok) and a TL431 to pass the excess current.
Normally this means TL431 clamps the +5V line to DGND at 5V, unless we want to allow for PECL re-clocking etc. which needs a lot of current. In this case we need a ballast resistor to draw most current and a TL431 across DGND to -5V.
When not using PECL relocking etc., placing TL431 in the -5V line and using a ballast resistor across +5V & DGND takes advantage of the 8dB larger PSRR on the -5V line. Note, I have seen zero enthusiasm to integrate PECL based re-clocking and signal conditioning into the TDA1541 DAC's, which I think is a shame. But so be it. The power supply will allow as much as 1A @ 10V, so it can power a bunch of ECL.
Further +5V/AGND/-15V are another power section with -15V = GND AND +5V = GND in AC terms. Yes, this seems to make no sense, but it is what it is.
This section is the actual signal current circuit.
I still think a low impedance shunt is best here. The TL431 at ~0.15 Ohm (or less) typical impedance is a good choice, with the upper resistor in the feedback divider bypassed for lowest noise. Here the target of 1.1k & 100uF gives ~ 1.5Hz turnover, adequate I think.
So +5V & -15V link via decoupling capacitors and the shunt regulator string. For this loop, AGND & +5V decouple to -15V. We need to remember that +5V carries the mirror image of the current flowing in the output pints.
With a 0.3R loop and 8mA P-P error voltages on +5V with respect to -15V are limited to 2.4mV P-P. In light of the respective PSRR on these pins feed trough into the output will likely result in a few 100nV worst case error and as this error is the reverse polarity of the signal with L/R summed, the effect is an ephemeral reduction in output level and some crosstalk at levels > 120dB below full scale.
So a fairly simple, complete and way past adequate supply is below.
Note that short return path are provided for high frequency common noise passing through our doughnut transformers (R-Cores, O etc are all the same here, without shield terrible noise leakage from all the SMPS's in anything from lightbulbs to TV's, Aircon etc.
Further, a direct current path from transformer / rectifier are carefully avoided.
To match this power supply, here the last adjusted version of the "Core", including all recommended protection.
If divider logic for DEM clocking is needed, put a a pair of 74XX74 footprints, including TH Options with SMD inside the TH including an option for VQFN-14 for TI 74AUC74.
The Power should be separate 5V or 3.3V (or even 2.5V) depending is using 74F74/74AS74 or CMOSL logic (74LVC74, 74AUC74 etc.) to avoid throwing in unnecessary noise into the TDA1541 supplies. A piggyback I2S to SIM Module is shown here. A discrete TDA1541 focused option should end up on the same PCB if desired.
I really think all this should be just a single PCB, including transformers, SMA, FL and XH input options (XH with micro coax and alternating signal/grounds works really well). It becomes biggish and a little expensive, but it is easily build and integrated with other BIY modules.
If breaking things up into multiple PCB's, place transformers, rectification and filtering on the Power PCB, including our 317 (or 78XX) Regulators on Heatsinks and the 1mH chokes as final elements and 431's and decoupling capacitors on the TDA1541 PCB. Make the Transformers a third "break-off" PCB, so those who want to use Magic Transformers instead of lowly Talema Torrid Toroids can do so.
Analogue stages, knock yourselves out.
Thor
Agreed.
Also agreed. It will be implemented. If we're making ultimate, we're making ultimate. It can also be made as modular, so one can pick their poison.
Feel that supercap supply is better. I belive it isn't that much big a bump in budget, compared to what we get.
Last edited:
Ok, I will post my supercap version another day, though it should be obvious.
If we do Supercapacitors we should do ECL and local clocks. Make it like this:
Here its a 10MHz OCX veryow phasenoise and clock synthesiser with "femtoclock" phasenoise levels, with a 9-Way clock distribution.
The key is that the whole clock module floats decoupled from the main PCB on rubber O rings and is "weighted down" and the PCB reinforced by the 2-piece machines copper shield (with gold plating - codename for the system was "Heart of Gold" - referencing both H2G2 and Neil Young).
But I think this will be too far for most.
So I propose no Super Capacitors, no "HOG" clock etc, basic design, which still goes very far, compared what you get on AliExpress or what I have seen do far as ready PCB.
Thor
I encourage every contributor here to build a simple measuring tool: a x100 low noise AC coupled audio voltage amplifier. I built one around a NE5534AN, powered from 2x 9V batteries. It comes very handy for checking PSU hum and noise, analog output, DEM pins superimposed ripple, and more. The output can go to a headphone amplifier (hearing is believing!) or to an oscilloscope with 1 mV to 5 mV sensitivity (seeing is believing!). Alternatively, it can go to a soundcard and in turn to an FFT analyzer.
I used it in the other thread together with an I2S test generator by creating constant analog output levels between -4 mA and 0 mA, superimposed +/- 1 LSB steps (about 61 nA). Any deviation from the ideal, ground bounce, noise, glithes can be seen on the screen of the oscilloscope. You instantly know what should be improved on. Or if you have a bunch of TDA chips, you can sort out the best performing 😉
I used it in the other thread together with an I2S test generator by creating constant analog output levels between -4 mA and 0 mA, superimposed +/- 1 LSB steps (about 61 nA). Any deviation from the ideal, ground bounce, noise, glithes can be seen on the screen of the oscilloscope. You instantly know what should be improved on. Or if you have a bunch of TDA chips, you can sort out the best performing 😉
Better, use a 100MHz CFB OPA, they also come in lower noise that 5534. Use with a 100MHz+ oscilloscope that opens a whole new window.
Then build a second unit with 3pcs as instrumentation Amp, so you can probd differentially.
I designed these as active probes PCB's many years back. 300MHz+. Very useful.
Thor
Shouldn't be too expensive to gold plate copper, already have a company for that that did several jobs for me, but much larger pieces. On cnc copper shield is doable, not a large surface, should be cheap as well, i can make enough to give out to people building the dac. I like that concept (i've done similar vibration decoupling with my smd nutube priject). We can dampen it further with vibrofiltr below the board, it works wonders. I've used mmcx connectors for the last year, and liking it more than sma and ufl, i think it will do very well here, feels like middle ground between the two. NDK2520SDA (probably best bang for buck) are only available in 20-50mhz and inbetween range. So other affordable replacement in 10mhz should be found.
Last edited: