Cx=0.47uF is enormous; 47 times greater than the recommended starting point (p.7 of Quasimodo design note). Equation A.11 on page 17 tells you what happens to the optimum value of Rsnub, when you increase Cx by a factor of 47. It is not completely impossible that 10R might actually be the optimum value of Rsnub in this zany scenario.
You could double check that your Quasimodo is giving the right answer, by applying it to a known fixed inductor instead of a transformer secondary. Choose a factory-made inductor (±10% or better yet ±5%) whose inductance is between 150 microhenries and 200 microhenries. Use the same zany value of Cx (0.47 uF) and dial in the optimum Rsnub on your QM+scope setup. Then double check this Rsnub value, by using math to calculate the theoretical optimum Rsnub. This calculation is possible because you know all component values, including the inductance value. {with a transformer you don't know the inductance so you can't calculate Rsnub; "no math" Quasimodo to the rescue}
Here is a candidate inductor at Mouser
And here is a candidate inductor at DigiKey
If experiments with QM give a value "Rsnub1" , while theoretical calculations using the known L and the known Cx give a value "Rsnub2", and if Rsnub1 is close to Rsnub2, then you can feel comfortable that Quasimodo is doing its job correctly. You can feel comfortable that it gives the right answer in the zany case when Cx is 47 times bigger than recommended.
You could double check that your Quasimodo is giving the right answer, by applying it to a known fixed inductor instead of a transformer secondary. Choose a factory-made inductor (±10% or better yet ±5%) whose inductance is between 150 microhenries and 200 microhenries. Use the same zany value of Cx (0.47 uF) and dial in the optimum Rsnub on your QM+scope setup. Then double check this Rsnub value, by using math to calculate the theoretical optimum Rsnub. This calculation is possible because you know all component values, including the inductance value. {with a transformer you don't know the inductance so you can't calculate Rsnub; "no math" Quasimodo to the rescue}
Here is a candidate inductor at Mouser
And here is a candidate inductor at DigiKey
If experiments with QM give a value "Rsnub1" , while theoretical calculations using the known L and the known Cx give a value "Rsnub2", and if Rsnub1 is close to Rsnub2, then you can feel comfortable that Quasimodo is doing its job correctly. You can feel comfortable that it gives the right answer in the zany case when Cx is 47 times bigger than recommended.
Last edited:
Thanks, Mark! I'll check the values with inductor. As I said, Cx is already in the amplifier - so it wasn't my choice to set Cx so high.
BTW, I'm estimating Rs (R1 on scheme) power dissipation using LTspice scheme attached. Am I doing it right, or should I omit snubber from the scheme completely and place Rs in series with V1 to correctly estimate dissipation?
Because currently simulation gives me values around 6mW, and if I make Rs even lower - dissipation lowers too. So if we know that ζ rises when Rs lowers, why not keeping Rs as low as possible for overdamping? What is the "stop factor"?
If I place Rs in series with V1 in simulation, it becomes current limiter for the smoothing capacitor and so it's power dissipation will be pretty high (lower resistance - higher dissipaton) until capacitor fully charges.
BTW, I'm estimating Rs (R1 on scheme) power dissipation using LTspice scheme attached. Am I doing it right, or should I omit snubber from the scheme completely and place Rs in series with V1 to correctly estimate dissipation?
Because currently simulation gives me values around 6mW, and if I make Rs even lower - dissipation lowers too. So if we know that ζ rises when Rs lowers, why not keeping Rs as low as possible for overdamping? What is the "stop factor"?
If I place Rs in series with V1 in simulation, it becomes current limiter for the smoothing capacitor and so it's power dissipation will be pretty high (lower resistance - higher dissipaton) until capacitor fully charges.
Attachments
I boldly claim (and you should verify!) that at 50 Hertz, the magnitude of the impedance of C2 is 677 ohms. This capacitor C2 is in series with R1.
You can use Ohm's Law to calculate an approximate value of the current thru these two series elements, when driven by a 17 volt, 50 Hz sinewave. Then the power dissipated in R1 is just
Iapprox * Iapprox * R1
The calculated current is only an approximation, because we ignored the phase angle of the impedance. Fortunately, in this case the error is small and the approximation is very good. The arctangent of (10/677) is a tiny: less than one degree.
By the way, I promise you that God will not smite your household and burn your crops and strangulate your goats, if you remove the large Cx that's already in your amplifier, and replace it with a lower valued capacitor.
You can use Ohm's Law to calculate an approximate value of the current thru these two series elements, when driven by a 17 volt, 50 Hz sinewave. Then the power dissipated in R1 is just
Iapprox * Iapprox * R1
The calculated current is only an approximation, because we ignored the phase angle of the impedance. Fortunately, in this case the error is small and the approximation is very good. The arctangent of (10/677) is a tiny: less than one degree.
By the way, I promise you that God will not smite your household and burn your crops and strangulate your goats, if you remove the large Cx that's already in your amplifier, and replace it with a lower valued capacitor.
Alright, I've checked the Quasimodo with 180µH inductor in place of transformer - calculated ζ=1 Rs and eye-matched Rs are almost identical. So it's working properly even with so large Cx/Cs. Actually I understand that current Cx in amplifier is an attempt of original designers to place a snubber, so if I'm doing my own snubber and calculations are ok - I have all the rights to replace Cx for the one with smaller capacity. But, is there really a good reason to keep Cx/Cs capacity small? So what are the downsides of big capacity Cx/Cs in relation to snubber performance?
Is there a good reason to design CRC snubbers with smaller values of Cx and Cs, rather than larger? What are the negatives about using large Cx and Cs? Does snubber performance suffer?
Several people who read this thread, are Circuit Theorists. I invite them to respond to the query above; probably their analysis is better than mine. Blowhards Who Have All The Answers (whether correct or otherwise) will of course respond with or without an invitation. To avoid biasing the discussion or bumping it in any particular direction, I'll wait a couple of weeks before weighing in with my own thoughts.
Somehow, you can set the display to show 50 posts per page - should be 100/page for easier reading.
If you ignore the redundant posts and stick to the technical ones, it reduces to just a couple of hundred specific posts plus a bit of separate reading (and the other associated thread)- a really good & clear, non-redundant thread so far
If you ignore the redundant posts and stick to the technical ones, it reduces to just a couple of hundred specific posts plus a bit of separate reading (and the other associated thread)- a really good & clear, non-redundant thread so far
Is there a good reason to design CRC snubbers with smaller values of Cx and Cs, rather than larger? What are the negatives about using large Cx and Cs? Does snubber performance suffer?
I'm surprised that no opinions and analyses have been put forward. Maybe this means nobody has an idea of their own, they're just waiting to read someone else's idea and then proclaim it to be "obvious" (or "wrong") ??
I'm surprised that no opinions and analyses have been put forward. Maybe this means nobody has an idea of their own, they're just waiting to read someone else's idea and then proclaim it to be "obvious" (or "wrong") ??
I'll Play
We want a damping factor >1. We also know that damping factor is inversely proportional to Q. Therefore we want a capacitor with a low Q. I would guess, therefore, that larger capacitors have higher Q. In looking at the DS of a couple of capacitors, however, it appears that, typically, larger capacitors have higher Qs. (In comparing capacitors, I used Dissipation Factor, which I believe is the inverse of Q).
George
We want a damping factor >1. We also know that damping factor is inversely proportional to Q. Therefore we want a capacitor with a low Q. I would guess, therefore, that larger capacitors have higher Q. In looking at the DS of a couple of capacitors, however, it appears that, typically, larger capacitors have higher Qs. (In comparing capacitors, I used Dissipation Factor, which I believe is the inverse of Q).
George
IMO, the values are pretty universal as they are - I've used similar values as per the previous Hagerman method and I generally use styrenes or propylenes - you can also get quite a selection of high voltage caps at these values.
Curiously, just using a single cap across the transformer winding can have a most significant effect on the sound but a real PIA to get even close to optimum value - hardly ever used these days - there maybe still an archived thread here about 'tuning a transformer' -maybe 10 years ago now - fascinating 'stuff'
Curiously, just using a single cap across the transformer winding can have a most significant effect on the sound but a real PIA to get even close to optimum value - hardly ever used these days - there maybe still an archived thread here about 'tuning a transformer' -maybe 10 years ago now - fascinating 'stuff'
Is there a good reason to design CRC snubbers with smaller values of Cx and Cs, rather than larger? What are the negatives about using large Cx and Cs? Does snubber performance suffer?
Reasons someone might prefer smaller Cx,Cs rather than larger
- Size
- Cost
- Power
- Fear
Especially if you're buying higher voltage (WVDC > 80V) film capacitors, more nanofarads means more size AND more dollars. If money is tight, smaller Cx,Cs is preferred.
Those readers who are good at algebra will have no difficulty proving that snubber power dissipation @ 50Hz / 60Hz, rises as (Cx raised to the power 1.5). That's not a misprint: power is a super-linear function of the capacitance Cx. And since Cs = k*Cx for some constant value k (I recommend k=15), the same is true of Cs.
Those not-so-good-with-algebra can demonstrate super-linearity to themselves, using Excel. Assume a constant Lsecondary, a constant Csecondary, and a constant Crectifier. Then make a spreadsheet that varies Cx from 1nF to 10uF in steps of 1.1X (each row's Cx is 1.1 times larger than the preceding row's Cx) and calculate the optimum snubber resistance which gives zeta=1. Finally calculate the power dissipated in the snubber at 50Hz or 60Hz depending on your mains. Plot power versus Cx on log-log axes and fit an Excel trendline. Voila! Increasing Cx by a factor of 100, increases Power by a factor of 1000. Power rises as the (3/2) power of Cx. You can see it yourself on your own computer, in a spreadsheet you built yourself, with no outside influences from anyone.
Fear is very real. Read post #719 in this thread. Then read posts #702, #230, et cetera. When DIYers see a low value of optimum Rsnub, it frightens them. Even though they see with their own eyes that the ringing is completely damped out, they are still frightened. Larger Rsnub is less frightening. Since Rsnub is inversely proportional to sqrt(Cx), decreasing Cx by a factor of four will increase Rsnub by a factor of two. This reduces fear by a factor of two.
In email correspondence with Morgan Jones, author of Valve Amplifiers 4th ed and other texts, he points out that the ringing frequency decreases when Cx increases. Dr. Jones prefers the ringing frequency to be higher, because (a) the farther away from the audio band the better; and (b) the higher the frequency, the easier it is to eliminate by simple RC or LC filtering {which are common in valve amplifier power supplies}. Respectfully I disagree with Dr. Jones; in my opinion, if the ringing is damped it makes no difference what its "natural frequency" would have been; the waveform is bloody well damped after all. There's no ringing for downstream filters to remove. And the larger the Cx the more confident you can be, that Cx is much MUCH greater than (Csecondary + Crectifiers). It swamps them out completely. So the Quasimodo test jig, which assumes (Csecondary + Crectifiers) is a negligible capacitance, becomes a better and better approximation when Cx becomes larger and larger.
But don't let all of the above chatter obscure the one central, CRUCIAL fact: a test jig such as Quasimodo (or CheapoModo, or Quasimodo ExtraLight, or ...) lets you SEE that your snubber has completely damped out the transformer ringing. When you SEE a snubber that does its job optimally -- USE IT. And don't fall into petty or picayune arguments about whose optimum is even optimumer than who else's optimum -- unless you simply enjoy flapping your gums about minutae. If you do, why then go right ahead and entertain yourself, but try not to confuse newbies with your arcane pirouettes and obscure backflips.
Mark, this was a good coke spurter. I'm still laughing. Where's those paper towels before my keyboard shorts out ...
And that is the bottom line.
mlloyd1
Recently, while helping a colleague from another (non-english speaking) forum determine the value of the snubber resistance for his power supply (PS) I made an interesting observation which I think is worth sharing.
The problem was that the colleague did not have any means for exciting the ringing by external means, i.e. neither Quasimodo nor a function generator, but he possesses an oscilloscope. I suggested doing it in a running power supply, and put together a step-by-step instruction by performing the procedure on an improvised PS utilizing a transformer I already measured with Quasimodo, obtaining Rs=160R for Zeta=0.707 (see my table in post #643 - it is the Triad FS24/800 with two 12V secondaries). In my makeshift PS it feeds an ordinary bridge rectifier to produce a nominal +-12V PS. The secondaries are connected in the center tap fashion, each of them having a Cx=10nF capacitor between the center tap and the respective rectifier bridge connection. The filter section consists of two 1000uF electrolytics.
The first figure shows the waveforms at the respective secondary ends with the PS unloaded. There are no parasitic ocsillations, or rather they are so weak as to be not visible with my scope's resolution. The subsequent figure shows the waveforms when one branch of the PS is loaded with 60mA; the oscillations at the diode shut-off are clearly visible on both secondaries, the spike amplitudes being some 6-7V.
Loading the other half of the PS increases the spikes to about 10V because there is now more energy available for the excitation; the subsequent plot shows the oscillations' details. The resonance takes place at about 26kHz (never mind the scope display, I measured the frequency with cursors placed four periods apart). Comparing this value with the one of 86kHz measured with Quasimodo in the standard setup (the primary and one secondary shorted) this may come as a surprise at first sight, but bear in mind that the two circuit configurations are quite different.
Now I connected a CR snubber (Cs=150nF, Rs=1000R - a trimmer) across one secondary only, and was struck by seeing the oscillations at both secondaries weakened by the same amount (the fifth figure). Once again, it is a snubber across one of the two secondaries attenuating the parasitic oscillation at both secondaries.
Adjusting now the trimmer resistance for the damping factor Zeta=1.0, I obtained the waveforms in the sixth figure; the trimmer resistance at this setting measured 330R, i.e. twice the value obtained with Quasimodo and the primary and one of the two secondaries shorted.
So I had two surprises in this exercise. The first is that a single snubber quenches parasitic oscillations in two connected secondaries; and the second one is that this happens with a larger resistance value Rs than in the idealized measurement, although the latter was to be expected given the lower resonance frequency in the configuration with both secondaries active.
By extension, one could now expect that a single snubber between the two outside legs of a center-tapped secondary may also do the job, and it is indeed so: with Quasimodo I measured Rs=625R for Zeta=0.707 in such a configuration, but I have to scrutinize this result yet, and will describe it in another post.
This behaviour seems to be the case with transformers having two or more secondaries. Last year I determined Rs for a single secondary transformer both with Quasimodo and in a running PS; both values were nearly the same. On the other hand, prompted by the above findings, another colleague from the same forum investigated the PS in his amplifier, consisting of 2x22V secondaries for the power stage, and 2x14V one for the small-signal section. He found that a single snubber across one of the 22V secondaries quenched the oscillations in all four secondaries. Unfortunately, he only posted figures for the 2x22V secondaries, but I might be able to persuade him to repeat the measurements and provide the figures.
Regards,
Braca
The problem was that the colleague did not have any means for exciting the ringing by external means, i.e. neither Quasimodo nor a function generator, but he possesses an oscilloscope. I suggested doing it in a running power supply, and put together a step-by-step instruction by performing the procedure on an improvised PS utilizing a transformer I already measured with Quasimodo, obtaining Rs=160R for Zeta=0.707 (see my table in post #643 - it is the Triad FS24/800 with two 12V secondaries). In my makeshift PS it feeds an ordinary bridge rectifier to produce a nominal +-12V PS. The secondaries are connected in the center tap fashion, each of them having a Cx=10nF capacitor between the center tap and the respective rectifier bridge connection. The filter section consists of two 1000uF electrolytics.
The first figure shows the waveforms at the respective secondary ends with the PS unloaded. There are no parasitic ocsillations, or rather they are so weak as to be not visible with my scope's resolution. The subsequent figure shows the waveforms when one branch of the PS is loaded with 60mA; the oscillations at the diode shut-off are clearly visible on both secondaries, the spike amplitudes being some 6-7V.
Loading the other half of the PS increases the spikes to about 10V because there is now more energy available for the excitation; the subsequent plot shows the oscillations' details. The resonance takes place at about 26kHz (never mind the scope display, I measured the frequency with cursors placed four periods apart). Comparing this value with the one of 86kHz measured with Quasimodo in the standard setup (the primary and one secondary shorted) this may come as a surprise at first sight, but bear in mind that the two circuit configurations are quite different.
Now I connected a CR snubber (Cs=150nF, Rs=1000R - a trimmer) across one secondary only, and was struck by seeing the oscillations at both secondaries weakened by the same amount (the fifth figure). Once again, it is a snubber across one of the two secondaries attenuating the parasitic oscillation at both secondaries.
Adjusting now the trimmer resistance for the damping factor Zeta=1.0, I obtained the waveforms in the sixth figure; the trimmer resistance at this setting measured 330R, i.e. twice the value obtained with Quasimodo and the primary and one of the two secondaries shorted.
So I had two surprises in this exercise. The first is that a single snubber quenches parasitic oscillations in two connected secondaries; and the second one is that this happens with a larger resistance value Rs than in the idealized measurement, although the latter was to be expected given the lower resonance frequency in the configuration with both secondaries active.
By extension, one could now expect that a single snubber between the two outside legs of a center-tapped secondary may also do the job, and it is indeed so: with Quasimodo I measured Rs=625R for Zeta=0.707 in such a configuration, but I have to scrutinize this result yet, and will describe it in another post.
This behaviour seems to be the case with transformers having two or more secondaries. Last year I determined Rs for a single secondary transformer both with Quasimodo and in a running PS; both values were nearly the same. On the other hand, prompted by the above findings, another colleague from the same forum investigated the PS in his amplifier, consisting of 2x22V secondaries for the power stage, and 2x14V one for the small-signal section. He found that a single snubber across one of the 22V secondaries quenched the oscillations in all four secondaries. Unfortunately, he only posted figures for the 2x22V secondaries, but I might be able to persuade him to repeat the measurements and provide the figures.
Regards,
Braca
Attachments
I suggest that you find a colleague who owns both a Quasimodo jig, an oscilloscope, and a center tapped transformer. I suggest the two of you perform the following experiment:
Attach one snubber to the secondaries however you prefer. Either between Sec1 and CenterTap, or between Sec2 and Sec1. Or any other way you like.
Short the primary and connect Quasimodo between Sec1 and GND. Power up Quasimodo and take a photo of the ringing (if any). Now connect Quasimodo between Sec2 and GND. Power up Quasimodo and take a picture of the ringing (if any). Carefully look at the two scope photos: did a single snubber damp both? Answer: no.
The reason why you get worse behavior (much more ringing) with Quasimodo than with the supply itself, is that Quasimodo is designed to be an EXCELLENT bell-ringer. Rectifier diodes vary all over the place; the worst diodes ring 20X more than the best diodes ring. {I published an article in Linear Audio v.10 which detailed this behavior, for 48 different semiconductor diodes}. If the ringing in your power supply with your diodes was small-to-not-observable: congratulations. Your transformer/diode combination is one that has relatively low-Q and relatively little ringing. However don't become confused: the reason for no-observable-ringing is not that one snubber is somehow magically able to snub two secondaries.
Proposition X: One snubber cannot simultaneously damp both halves of a center tapped secondary, whether you connect the snubber between terminals 1&2 or between terminals 1&3.
I am tempted to offer a wager of 100 British Pounds that it is possible to demonstrate and prove Proposition X in an LTSPICE simulation having fewer than 49 components. (49 = 7x7). However I would be slightly embarrassed to take anybody's money on such a sure thing; a non-gamble; so I won't make the offer. I will however listen if anyone wants to vigorously insist upon making this bet -- as long as the amount is 100 GBP.
Mark Johnson
Attach one snubber to the secondaries however you prefer. Either between Sec1 and CenterTap, or between Sec2 and Sec1. Or any other way you like.
Short the primary and connect Quasimodo between Sec1 and GND. Power up Quasimodo and take a photo of the ringing (if any). Now connect Quasimodo between Sec2 and GND. Power up Quasimodo and take a picture of the ringing (if any). Carefully look at the two scope photos: did a single snubber damp both? Answer: no.
The reason why you get worse behavior (much more ringing) with Quasimodo than with the supply itself, is that Quasimodo is designed to be an EXCELLENT bell-ringer. Rectifier diodes vary all over the place; the worst diodes ring 20X more than the best diodes ring. {I published an article in Linear Audio v.10 which detailed this behavior, for 48 different semiconductor diodes}. If the ringing in your power supply with your diodes was small-to-not-observable: congratulations. Your transformer/diode combination is one that has relatively low-Q and relatively little ringing. However don't become confused: the reason for no-observable-ringing is not that one snubber is somehow magically able to snub two secondaries.
Proposition X: One snubber cannot simultaneously damp both halves of a center tapped secondary, whether you connect the snubber between terminals 1&2 or between terminals 1&3.
I am tempted to offer a wager of 100 British Pounds that it is possible to demonstrate and prove Proposition X in an LTSPICE simulation having fewer than 49 components. (49 = 7x7). However I would be slightly embarrassed to take anybody's money on such a sure thing; a non-gamble; so I won't make the offer. I will however listen if anyone wants to vigorously insist upon making this bet -- as long as the amount is 100 GBP.
Mark Johnson
Last edited:
The rectifier I used in the above test is a Semikron B250 C1500, an ancient (ex-equipment) part that happened to be the first one I saw in the parts box.
While it might be possible to try snubbing a transformer without the 10nF capacitors across the secondaries, I prefer to work with them. The reason for this is that they bring structure into the system, i.e. they swamp out the small capacitances in the power supply, making thus the performance of the latter much more repeatable.
Regards,
Braca
The extra capacitance produces an oscillating circuit with constant natural frequency; without it the parasitic oscillations' frequency varies with the rectifier capacitance and other internal capacitances, which affects fixing the correct snubber value. And in comparison with the secondary current in operation, this additional current loop is negligible.
I'm now enclosing the results obtained with two more recitifiers. The first two figures below refer to the measurements made with a W01 rectifier bridge; the first one shows the oscillations with the secondary unsnubbed, and the second one with a 150nF/760R snubber across both secondaries at a load of 94mA each. Both secondaries appear to be correctly snubbed.
Since in the Mark's paper in Linear Audio Vol. 10 the 1N4005 came out as the worst of the "terrible" diodes in the 1A current group, I tested the Triad transformer with a bridge made up of four 1N4004 diodes.
The results are shown in the three figures that follow, the first one referring to the configuration without snubbers, the second one snubbed with 150nF/680R across both secondaries, and the third one snubbed with 150nF/170R across a single secondary.
Both secondaries appear to be correctly snubbed in the above two configurations. However, the Zeta value in the last plot is a little lower than in the second one (150nF/680R) because at Rs lower than 170R, the Cs capacitor starts becoming a part of the oscillator and ringing starts appearing at the frequency determined by (Lt, Cx+Cs),
The above method of snubbing appearrs to work with this transformer. I don't know whether it would work with other transformers, but I intend to look into this matter further. The next step will be to repeat the measurements with another transformer with two secondaries.
Regards,
Braca
