If there were big resonances AND big spikes in the music to trigger those
resonances, you'd see them.
Bass tones are poor at causing the "ringing" we're looking for, if the ringing
would be at high frequencies. And treble tones are poor at causing base ringing.
To search for high-freq ringing, you need
(1) a sin-sweep generator providing perhaps 1volt peak-peak
or
(2) a fast-edge generator, and a 555 "rectangle wave" IC at 1KHz period
would be fine; the edges of 555 are faster than 1microSecond, thus serving
to stimulate lots of candidate ringing; the fast edges are just like the slap
of a hand on the belly of a fat guy---the belly joshes around and "rings"
and soon stops because of dampening.
or
(3) music with fast edges, such as (your call) high trumpet notes? they don't
sound like pure tones to me, so probably are good waveforms to search for
"ringing".
The lowest resonance/ringing frequency would be two of the 2,200uF in a loop
with some of the wire. In series in the loop, they become 1,100uF. Assume
couple pieces of 4" (0.1meter) wire, which has inductance approximately of
MUo * length of wire in meters, or [ 4*pi*10^-7 ] * (0.1 meter) = 4*pi*10^-8
or 12.6 * 10*-8 ===~~~ 10^-7 Henry.
Since this is a nice round number, we'll assume the total loop inductance is 10^-7.
Resonate freq of 1,000uF (10^-3 Farad) and 10^-7 Henry is 0.159/sqrt(10^-3-7)
or 16KHz.
All other resonances will be higher.
I think you should try some loud high-freq unpure-sine material,
or build up a 555 oscillator running at 10,000Hertz. And use the scope
to check. The OPA627 has good power-supply rejection at 20,000 and lower
http://www.ti.com/lit/ds/symlink/opa627.pdf
Examine the third row of plots in the section "Typical Performance Curves";
depending on how you've got the opamp connected (don't you hate those
weasel words? check out the schematic! and the compensation cap inside)
you should have at least 60dB (1000:1 reduction) and maybe 80db (10,000:1)
at 20,000Hertz. Thus 1milliVolt of rail-ringing would appear as 1 microvolt
between input to opam, to be amplified by whatever gain configuration you
are using (Unity?).
If we aim for damping factor of approx 1, which also approx Q=1,
then Rloss = Z(L at 16KHz) = Z(C at 16KHz). One is easy to remember. Use one.
But...what if Q=5, because Rloss varies with temperature or aging or brand_name
or Vdd? I like a Q=5, because its dampening, once the ringing has been triggered,
is 6dB per cycle. Thus if the initial half-sin wave is 10milliVolts, the following
opposite polarity half-sine will be 10mV * 0.70 ,
and the following same polarity half-sine will be 10mv * 0.7 * 0.7 = 5milliVolts.
I can easily predict how fast a Q=5 resonance will decay, and Q=5 does last
a while, so coloration is possible IF your PowerSupplyRejection is poor.
Back to Q=1. No ringing, thus no sin-response, no coloration, but you may see
the actual music or voice waveforms on the VDD, accurately appearing, if
the caps are small. [ hmmm I hope this does not become "folklore". ]
What Rdampen to use? Rloss = Z(L @ 16KHz) = Z(C @ 16KHz)
== 2*pi*16KHz * 100 nanoHenry = 10^+5 * 10^-8 ==> 10^-3
or
0.001 ohm.
I expect the wiring resistance, the Caps ESR, or the solder!
to provide enough dampening.
Happy listening
tank
resonances, you'd see them.
Bass tones are poor at causing the "ringing" we're looking for, if the ringing
would be at high frequencies. And treble tones are poor at causing base ringing.
To search for high-freq ringing, you need
(1) a sin-sweep generator providing perhaps 1volt peak-peak
or
(2) a fast-edge generator, and a 555 "rectangle wave" IC at 1KHz period
would be fine; the edges of 555 are faster than 1microSecond, thus serving
to stimulate lots of candidate ringing; the fast edges are just like the slap
of a hand on the belly of a fat guy---the belly joshes around and "rings"
and soon stops because of dampening.
or
(3) music with fast edges, such as (your call) high trumpet notes? they don't
sound like pure tones to me, so probably are good waveforms to search for
"ringing".
The lowest resonance/ringing frequency would be two of the 2,200uF in a loop
with some of the wire. In series in the loop, they become 1,100uF. Assume
couple pieces of 4" (0.1meter) wire, which has inductance approximately of
MUo * length of wire in meters, or [ 4*pi*10^-7 ] * (0.1 meter) = 4*pi*10^-8
or 12.6 * 10*-8 ===~~~ 10^-7 Henry.
Since this is a nice round number, we'll assume the total loop inductance is 10^-7.
Resonate freq of 1,000uF (10^-3 Farad) and 10^-7 Henry is 0.159/sqrt(10^-3-7)
or 16KHz.
All other resonances will be higher.
I think you should try some loud high-freq unpure-sine material,
or build up a 555 oscillator running at 10,000Hertz. And use the scope
to check. The OPA627 has good power-supply rejection at 20,000 and lower
http://www.ti.com/lit/ds/symlink/opa627.pdf
Examine the third row of plots in the section "Typical Performance Curves";
depending on how you've got the opamp connected (don't you hate those
weasel words? check out the schematic! and the compensation cap inside)
you should have at least 60dB (1000:1 reduction) and maybe 80db (10,000:1)
at 20,000Hertz. Thus 1milliVolt of rail-ringing would appear as 1 microvolt
between input to opam, to be amplified by whatever gain configuration you
are using (Unity?).
If we aim for damping factor of approx 1, which also approx Q=1,
then Rloss = Z(L at 16KHz) = Z(C at 16KHz). One is easy to remember. Use one.
But...what if Q=5, because Rloss varies with temperature or aging or brand_name
or Vdd? I like a Q=5, because its dampening, once the ringing has been triggered,
is 6dB per cycle. Thus if the initial half-sin wave is 10milliVolts, the following
opposite polarity half-sine will be 10mV * 0.70 ,
and the following same polarity half-sine will be 10mv * 0.7 * 0.7 = 5milliVolts.
I can easily predict how fast a Q=5 resonance will decay, and Q=5 does last
a while, so coloration is possible IF your PowerSupplyRejection is poor.
Back to Q=1. No ringing, thus no sin-response, no coloration, but you may see
the actual music or voice waveforms on the VDD, accurately appearing, if
the caps are small. [ hmmm I hope this does not become "folklore". ]
What Rdampen to use? Rloss = Z(L @ 16KHz) = Z(C @ 16KHz)
== 2*pi*16KHz * 100 nanoHenry = 10^+5 * 10^-8 ==> 10^-3
or
0.001 ohm.
I expect the wiring resistance, the Caps ESR, or the solder!
to provide enough dampening.
Happy listening
tank
Yes, I have previously gone that path trying to use LTSpice to model all possible RLC in the power supply caps and wires. The pictures were never pretty unless I padded resistance here and there. But in reality, I never like the sound of a power supply with padded resistance. Any resistance added creates ripples which are distortions.
I have built more than a few types of line level regulators. So far the best sounding one is Ikoflexer's shunt reg, but it is quite difficult to build. From learning from and tuning the Iko reg, I gained a bit of experience.
Now I want to sell one of the active speakers I previously built. I don't want to build a shunt reg in the active crossover, so this time I built a LM317/337 reg. With all the experience I gained from building the Iko reg, this time the LM317/337 regs seem to work fine, actually, very fine.
Finding no trace of ringing from my limited scope up to 10Mhz, and hearing the very clean sound from the speakers, I am pretty much sure that even if there are resonances above 10MHz, they would be so mild that don't really affect the sound.
I guess keeping the rail and ground paths short, using ground plane or thick wires, is the first step reducing possible resonances. I guess the 3 large 2,200uF has sufficient ESR to form the damping circuit which may reduce the possibility of ringing.
I have built more than a few types of line level regulators. So far the best sounding one is Ikoflexer's shunt reg, but it is quite difficult to build. From learning from and tuning the Iko reg, I gained a bit of experience.
Now I want to sell one of the active speakers I previously built. I don't want to build a shunt reg in the active crossover, so this time I built a LM317/337 reg. With all the experience I gained from building the Iko reg, this time the LM317/337 regs seem to work fine, actually, very fine.
Finding no trace of ringing from my limited scope up to 10Mhz, and hearing the very clean sound from the speakers, I am pretty much sure that even if there are resonances above 10MHz, they would be so mild that don't really affect the sound.
I guess keeping the rail and ground paths short, using ground plane or thick wires, is the first step reducing possible resonances. I guess the 3 large 2,200uF has sufficient ESR to form the damping circuit which may reduce the possibility of ringing.
Last edited:
Question on power supply caps
Hello,
Have been adding a 10 or so microfarad film cap slightly before the main supply caps. Reason I am thinking this would be a good thing is that any voltage spikes get taken by the film caps (more reliable and usually a higher voltage rating I've chosen for this reason).
Thinking this would make life easier and perhaps lengthen the life of the electrolytic caps.
Any idea if this idea has any merit or problems??? So far it has not caused any difference in sound quality that I can tell.
Hello,
Have been adding a 10 or so microfarad film cap slightly before the main supply caps. Reason I am thinking this would be a good thing is that any voltage spikes get taken by the film caps (more reliable and usually a higher voltage rating I've chosen for this reason).
Thinking this would make life easier and perhaps lengthen the life of the electrolytic caps.
Any idea if this idea has any merit or problems??? So far it has not caused any difference in sound quality that I can tell.
No merit, unless you have an unusually high resistance/inductance path between the film cap and the main electrolytics.
Minor possible problem: any HF voltage spikes which do make it past the transformer and rectifier will now see a lower impedance path so the HF current could be greater. This means that is it even more important to ensure that the relevant circuit loop has a very small area to reduce inductive coupling into signal circuits - not always easy to do with large PSU components.
If you want a bypass, the right place for it is as close to the amplifier output circuit as possible i.e. as far away from the PSU as possible.
Minor possible problem: any HF voltage spikes which do make it past the transformer and rectifier will now see a lower impedance path so the HF current could be greater. This means that is it even more important to ensure that the relevant circuit loop has a very small area to reduce inductive coupling into signal circuits - not always easy to do with large PSU components.
If you want a bypass, the right place for it is as close to the amplifier output circuit as possible i.e. as far away from the PSU as possible.
No merit, just problems.
If you can provide quantitative data about your configuration, I can show you the difference (just a pic of your set-up would already be sufficient to make a rough estimate).
You're probably better off adding RC snubbers to the rectifier, which would reduce a major source of HF noise (ringing on diode cutoff) and also bypass a certain amount of external crud.
Here's a previous discussion:
http://www.diyaudio.com/forums/powe...rectifier-diodes.html?perpage=10&pagenumber=7
Here's a previous discussion:
http://www.diyaudio.com/forums/powe...rectifier-diodes.html?perpage=10&pagenumber=7
Tom,
I have two different choices here
I am placing capacitance on my board building something similar to Doug Self's Opamplifier.
He simply put a .1uf PET across the rails on each opamp. Actually he had the .1uf going from the negative of one opamp the the positive of the next. Shortest distance traces I imagine is the reasoning there and easier layout.
So I can do that but was going to use a midsize tantalum and small ceramic OR I can also use power ground so I would have two tantalums and two ceramics. Your work seems really to incorporate ground here but the opamps never touch ground except through the load.
I did see your discussion on the signal passing through the caps.
What do you think is the best option. Ground involved or rail to rail with opamps?
Uriah
I have two different choices here
I am placing capacitance on my board building something similar to Doug Self's Opamplifier.
He simply put a .1uf PET across the rails on each opamp. Actually he had the .1uf going from the negative of one opamp the the positive of the next. Shortest distance traces I imagine is the reasoning there and easier layout.
So I can do that but was going to use a midsize tantalum and small ceramic OR I can also use power ground so I would have two tantalums and two ceramics. Your work seems really to incorporate ground here but the opamps never touch ground except through the load.
I did see your discussion on the signal passing through the caps.
What do you think is the best option. Ground involved or rail to rail with opamps?
Uriah
Found some likely good information. Maybe you can confirm and if its not already been discussed in this thread this is a nice read. The 6th, last, installment of this article series is more to the point but all 6 are good information that help develop a skeptical view of opamp datasheets, models and extremely low ESR ceramics.
Yet More On Decoupling, Part 6: Simulating the complete op amp/power supply circuit
Yet More On Decoupling, Part 6: Simulating the complete op amp/power supply circuit | EE Times
Here are the takeaways the authors present at the end of the article that I thought were most relevant to this thread:
Don't use low-ESR regulator capacitors just because the regulator vendor says they are OK for 'stability'. Your system performance is likely to be significantly better with the additional losses that tantalum capacitors can provide in this role.
A bit of slew-rate limiting in the op amp isn't necessarily bad; the resonances on the supplies are excited less strongly the slower the amplifier output is slewing. But if there are some fast-slewing, wide-bandwidth amplifiers affecting the supply rails, then low-bandwidth amplifiers, with their consequently poor supply rejection, will be particularly sensitive to the resulting noise. Careful in a mixed system!
Use the highest value local ceramic decouplers you can get, in the largest package that will fit, and don't use Y5V dielectric material unless you've fully allowed for the value change with bias voltage. But the volumetric advantage of the high-k dielectric may be completely lost with this allowance.
Adding another small ceramic capacitor in parallel with the main decoupler has a small effect on the amplifier output " but it's often a negative one, so test carefully; most times, it may be unnecessary.
Yet More On Decoupling, Part 6: Simulating the complete op amp/power supply circuit
Yet More On Decoupling, Part 6: Simulating the complete op amp/power supply circuit | EE Times
Here are the takeaways the authors present at the end of the article that I thought were most relevant to this thread:
Don't use low-ESR regulator capacitors just because the regulator vendor says they are OK for 'stability'. Your system performance is likely to be significantly better with the additional losses that tantalum capacitors can provide in this role.
A bit of slew-rate limiting in the op amp isn't necessarily bad; the resonances on the supplies are excited less strongly the slower the amplifier output is slewing. But if there are some fast-slewing, wide-bandwidth amplifiers affecting the supply rails, then low-bandwidth amplifiers, with their consequently poor supply rejection, will be particularly sensitive to the resulting noise. Careful in a mixed system!
Use the highest value local ceramic decouplers you can get, in the largest package that will fit, and don't use Y5V dielectric material unless you've fully allowed for the value change with bias voltage. But the volumetric advantage of the high-k dielectric may be completely lost with this allowance.
Adding another small ceramic capacitor in parallel with the main decoupler has a small effect on the amplifier output " but it's often a negative one, so test carefully; most times, it may be unnecessary.
Local decoupling is a way to supply near instantaneous current into the current consumer. That could be an opamp, or a digital circuit, or a power amplifier.
The faster the change in current, the lower the inductance of the SUPPLY to meet that changing demand.
Low inductance effectively means the same as, very close to the power pins.
The ceramic X7R needs to be virtually on the power pins to meet the fastest demand for changing currents.
Those X7R ceramics have a limited charge available to supply current.
They NEED to be recharged in time for the next pulse of current demand.
That is the job of the local decoupling that is located a little bit further away (more inductance) from the Power Pins.
The faster the change in current, the lower the inductance of the SUPPLY to meet that changing demand.
Low inductance effectively means the same as, very close to the power pins.
The ceramic X7R needs to be virtually on the power pins to meet the fastest demand for changing currents.
Those X7R ceramics have a limited charge available to supply current.
They NEED to be recharged in time for the next pulse of current demand.
That is the job of the local decoupling that is located a little bit further away (more inductance) from the Power Pins.
Paul Brokaw's AN-202 is an excellent discussion and reference on this aspect:
http://www.analog.com/static/imported-files/application_notes/AN-202.pdf
http://www.analog.com/static/imported-files/application_notes/AN-202.pdf
goof ref, always follow the current loops for intuition, analysis
if you don't know how the currents flow in closed, zero sum loops then you don't understand your circuit, ps, source and load
the classic 30 year old datasheet through hole sintered Ta Electrolytics are way behind current technology - today's better smt parts are cheaper, can be more reliable be designed for correct damping as well
designing for good ps bypass performance needs appreciation of the current paths, frequency dependencies - including parallel resonance possibiities
if you don't know how the currents flow in closed, zero sum loops then you don't understand your circuit, ps, source and load
the classic 30 year old datasheet through hole sintered Ta Electrolytics are way behind current technology - today's better smt parts are cheaper, can be more reliable be designed for correct damping as well
designing for good ps bypass performance needs appreciation of the current paths, frequency dependencies - including parallel resonance possibiities
Martin
This pdf is going in my Audio Papers folder. Its really instructive and I'm learning a lot from it. Thanks!
I think I found my answer. I checked Self's book on Small Signal circuits since he is obviously going to talk about decoupling a 5532. He says to place a .1uf across the rails only a few mm's from the pins. Specifically across the rails to avoid injecting rail noise into ground or ground noise into the rails. Self warns that "It is not necessary, and often not desirable, to have two capacitors going to ground; every capacitor between a supply rail and ground carries the risk of injecting rail noise into the ground."
I'm sorry if my last few posts seem a bit off topic. To me they were quite on topic as my design first started with a 100uf electrolytic paralleled by a .33uf film paralleled by a .1 ceramic, both across the rails. I second thought myself after reading this thread and many of the articles linked in it. My most recent layout includes a few uf of tantalum from each rail to ground each paralleled with a .1uf ceramic. This of course is against what Self was doing with his opamplifier, but on the other hand he had some distortion problems which apparently were solved by removing some electrolytics and replacing them with one large one across the rails. So, I felt a bit in a quandary as to which way was most correct.
Anyway, this is still a reference thread for me and I hope that there is more development on it. Its been great so far.
This pdf is going in my Audio Papers folder. Its really instructive and I'm learning a lot from it. Thanks!
I think I found my answer. I checked Self's book on Small Signal circuits since he is obviously going to talk about decoupling a 5532. He says to place a .1uf across the rails only a few mm's from the pins. Specifically across the rails to avoid injecting rail noise into ground or ground noise into the rails. Self warns that "It is not necessary, and often not desirable, to have two capacitors going to ground; every capacitor between a supply rail and ground carries the risk of injecting rail noise into the ground."
I'm sorry if my last few posts seem a bit off topic. To me they were quite on topic as my design first started with a 100uf electrolytic paralleled by a .33uf film paralleled by a .1 ceramic, both across the rails. I second thought myself after reading this thread and many of the articles linked in it. My most recent layout includes a few uf of tantalum from each rail to ground each paralleled with a .1uf ceramic. This of course is against what Self was doing with his opamplifier, but on the other hand he had some distortion problems which apparently were solved by removing some electrolytics and replacing them with one large one across the rails. So, I felt a bit in a quandary as to which way was most correct.
Anyway, this is still a reference thread for me and I hope that there is more development on it. Its been great so far.
"rail noise injection" shouldn't be a problem with regulated op amp supplies and your analog signal ground - if it is you have a big problem - with your ground
if your load is ground referenced then you have to have op amp supply pin C to load gnd in order to reduce that current loop
rail to rail C fails the current loop test - what loop is it improving? ( hint - that would be only with balanced/floating load on a fully differential output op amp like a OPA1632 )
I think Self is wrong there - at least as you have quoted - if you read further he gives the better answer for bypass C rail noise injection in ground impedance - "...the cure is to correct the grounding..."
if your load is ground referenced then you have to have op amp supply pin C to load gnd in order to reduce that current loop
rail to rail C fails the current loop test - what loop is it improving? ( hint - that would be only with balanced/floating load on a fully differential output op amp like a OPA1632 )
I think Self is wrong there - at least as you have quoted - if you read further he gives the better answer for bypass C rail noise injection in ground impedance - "...the cure is to correct the grounding..."
Last edited:
you must follow the current routes.
That's what that reference was showing us how to do.
It also said that the decoupling varies with different opamp internals.
Loop Areas rule !
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.