Power Supply Ripple Calculations deleted
- Thread starter adamasd
- Start date
Help Support GroupDIY Audio Forum:
TomWaterman
Well-known member
Stupid question - what is Rload, the load the regulator provides?
Cheers Tom
Cheers Tom
TomWaterman
Well-known member
Aha!
Makes a bunch of sense...
So you just plug that value back into the formula and pick a required design goal for ripple and off you rock?
Can I just take this moment to say thanks for posting some really useful stuff mediatechnology...I've been lurking lately (busy) but still check in and see a lot of stuff from you that I'd like to spend more time reading when I get the chance.
Very good - its forum members like you that make me feel like learning harder so one day I can hopefully contribute the same nuggets of info/interesting ideas.
I thought your matrix pan circuit was cool as hell too.
Thanks Tom
Makes a bunch of sense...
So you just plug that value back into the formula and pick a required design goal for ripple and off you rock?
Can I just take this moment to say thanks for posting some really useful stuff mediatechnology...I've been lurking lately (busy) but still check in and see a lot of stuff from you that I'd like to spend more time reading when I get the chance.
Very good - its forum members like you that make me feel like learning harder so one day I can hopefully contribute the same nuggets of info/interesting ideas.
I thought your matrix pan circuit was cool as hell too.
Thanks Tom
bcarso
Well-known member
There's a good article in the back of the old National Semiconductor Audio Handbook. It reproduces a number of graphs from the definitive article of O. H. Schade, Proc. IRE, vol. 31, 1943, plus some more.
Possibly this material has re-appeared in later National publications, as I know the phono preamp stuff has.
Possibly this material has re-appeared in later National publications, as I know the phono preamp stuff has.
TomWaterman
Well-known member
Thanks Wayne and Brad (you rock also!)
Thats a good document...
-T
Thats a good document...
-T
jdbakker
Well-known member
One caveat:
So, applying this formula it looks like I can make ripple arbitrarily small just by increasing capacitance, right ?
Well, not exactly. The rectifier that goes between the transformer and the cap can only recharge the cap if the transformer voltage is higher than the capacitor voltage. All other things (including load current) being equal, the fraction of time that the transformer voltage is > the cap voltage gets smaller as output ripple decreases. However, the amount of energy drawn by the load is the same, so the transformer and rectifier have less time to transfer this same amount of energy to the cap, which again leads to higher peak current. This produces all kinds of badness:
- rectifiers and (to a lesser extent) transformers die faster with increasing peak to average current ratios.
- the 'sharper' recharge current spikes produce more EMI/RFI, and have more high-frequency content which is harder to filter out by subsequent linear regulators (and those bulky input caps are useless at higher frequencies, too).
Just wanted to point out that 'infinite input capacitance' is not a goal that should be pursued. The NatSemi appnote touches upon this point, too.
JDB.
[I expect Wayne and Brad know this, too; this caveat is meant for the casual (ab)user of the quoted formula]
So, applying this formula it looks like I can make ripple arbitrarily small just by increasing capacitance, right ?
Well, not exactly. The rectifier that goes between the transformer and the cap can only recharge the cap if the transformer voltage is higher than the capacitor voltage. All other things (including load current) being equal, the fraction of time that the transformer voltage is > the cap voltage gets smaller as output ripple decreases. However, the amount of energy drawn by the load is the same, so the transformer and rectifier have less time to transfer this same amount of energy to the cap, which again leads to higher peak current. This produces all kinds of badness:
- rectifiers and (to a lesser extent) transformers die faster with increasing peak to average current ratios.
- the 'sharper' recharge current spikes produce more EMI/RFI, and have more high-frequency content which is harder to filter out by subsequent linear regulators (and those bulky input caps are useless at higher frequencies, too).
Just wanted to point out that 'infinite input capacitance' is not a goal that should be pursued. The NatSemi appnote touches upon this point, too.
JDB.
[I expect Wayne and Brad know this, too; this caveat is meant for the casual (ab)user of the quoted formula]
Svart
Well-known member
One of the reasons to move to "soft/fast" diodes. you can ultimately use less bulk cap as well too and saves a bit of time at the end during FCC testing as well..
TomWaterman
Well-known member
Prior to getting to the point where everything melts down...if you use some increased capacitance (but still at sensible proportions) are you guys suggesting and outlining an advantage from an emissions and safety point of view in using those heatsinked bridges with a specific recitifer diode type?
Cheers Tom
Cheers Tom
Samuel Groner
Well-known member
Indeed. Personally I've settled to 1 Vpp ripple at full load. Makes calculation of the required capacity simple--10 uF per mA.
Samuel
Svart
Well-known member
10000uf/amp? isn't that about 10 times excessive? usually 1000uf/amp is fine, at least for me.
The acceptable amount of ripple varies a lot based on several design factors, like high line, low line, breakdown voltage of reservoir capacitors or even VR breakdown. In audio power amps, inadequate reservoir capacitance can cause clipping sooner at LF (less than 100-120 Hz) than for MF (>120 Hz). In commercial designs there is cost/size motivation to not make these larger than needed, and some value amps may ignore LF power droop.
In unregulated circuits this ripple can get into the audio via PSR, so in principle reducing ripple voltage could reduce noise floor, but as often as not ground contamination from charging current and PS layout will dominate, so as always, life is a simultaneous equation with many parts. Optimize one at the expense of another.
JR
PS: an obscure trick to reduce dissipation in a marginal VR application is to go down in reservoir cap value. Increased ripple means lower average voltage across the subject VR. Certainly a cheaper running fix than adding a heatsink, or changing the design. Not optimal of course, but in large scale production you don't have the luxury of do-overs.
In unregulated circuits this ripple can get into the audio via PSR, so in principle reducing ripple voltage could reduce noise floor, but as often as not ground contamination from charging current and PS layout will dominate, so as always, life is a simultaneous equation with many parts. Optimize one at the expense of another.
JR
PS: an obscure trick to reduce dissipation in a marginal VR application is to go down in reservoir cap value. Increased ripple means lower average voltage across the subject VR. Certainly a cheaper running fix than adding a heatsink, or changing the design. Not optimal of course, but in large scale production you don't have the luxury of do-overs.
bcarso
Well-known member
Look out for cap ripple current being exceeded when downsizing bulk caps. There's a class D amp outfit that brags about how small of bulk caps they get away with, but fail to mention ripple current issues.
Another gotcha on charging current spikes: look closely at the way the transformer may saturate on peaks. Toroids, often used because of their intrinsic self-shielding properties, can go very awry when they saturate and spray magnetic field junk all over the place. I've resorted to E-I core including opposing the fields of two of them in tandem; you can also get what Deane Jensen once recommended, double-bobbin hum-bucking* constructions.
I was going around for days after saying double-bobbin hum-bucking to frieds and passerby.
Another gotcha on charging current spikes: look closely at the way the transformer may saturate on peaks. Toroids, often used because of their intrinsic self-shielding properties, can go very awry when they saturate and spray magnetic field junk all over the place. I've resorted to E-I core including opposing the fields of two of them in tandem; you can also get what Deane Jensen once recommended, double-bobbin hum-bucking* constructions.
I was going around for days after saying double-bobbin hum-bucking to frieds and passerby.
Samuel Groner
Well-known member
Man you make me think--and realise that I've used a wrong formula for years. I'm back after some more thinking...
Samuel
Samuel Groner
Well-known member
A few thoughts later I think I'm not that wrong, at least http://sound.westhost.com/power-supplies.htm#capacitor-value comes to about the same conclusion--7 uF per mA for 1 Vpp ripple. Didn't study the differences between my result and the one given above but I guess my formula calculates Vrms and not Vpp, hence the need for more capacity.
Samuel
> The peak currents can be very high as C is made overly large. To the point of destruction.
Peak current is limited by ripple depth and transformer resistance. Not by C, not by load. Pick your diodes bigger than your transformer rated current, you will usually be fine.
Rectifier RMS current rises fast from small to medium caps, but slow from medium to huge caps. In almost any audio situation, your RMS is nearly as high with a "good enough" cap as with an over-size cap.
Duncan's Power Supply Calculator is the modern way to figure this stuff. The hard part is getting real-world values for transformer parasitics.
Hyper-low ripple just won't happen. It is diminishing returns even with ideal parts. With real parts, "negligible" resistances become real problems. An inch of cap lead in common between rectifier circuit and load circuit, 0.01 ohms taking 100A peaks, will inject 1V of ripple to the load even if the internal cap voltage could be "dead-steady". You will almost never get to 1% ripple (maybe on very small projects where over-kill is cheap). For simple tube-work, you get a few-% ripple and then add an R-C. This not only doubles the rejection slope, it rejects the parasitic ripple on cap leads. To go further, add another R-C for every 20dB-30dB of ripple reduction you need; trying to get 40dB in one R-C is usually frustrating and also more expensive than two 20dB R-C networks.
I have a 30V 60mA D-C-R-C-R-C supply which feeds directly to a low line-level input (simplex audio+power). One more R-C to the mike capsule. There's not a lick of hum/buzz in it. To do this on one stage would need a 60F (yes F) cap with 10 microOhm lead and internal resistance. Bigger than a milk-crate and over $10,000. My way was $13.
I'm puzzled by your blown-open transformer event. Yes, the one winding was taking double surge current. But unless the cap-bank was too big to fit through the door, the surge should not have lasted the many seconds it takes to melt heavy copper. I wasn't there, but from a distance I wonder if it was just connected wrong, or very marginally designed. I've been around similar events and sometimes you just can't tell after the smoke clears.
Peak current is limited by ripple depth and transformer resistance. Not by C, not by load. Pick your diodes bigger than your transformer rated current, you will usually be fine.
Rectifier RMS current rises fast from small to medium caps, but slow from medium to huge caps. In almost any audio situation, your RMS is nearly as high with a "good enough" cap as with an over-size cap.
Duncan's Power Supply Calculator is the modern way to figure this stuff. The hard part is getting real-world values for transformer parasitics.
Hyper-low ripple just won't happen. It is diminishing returns even with ideal parts. With real parts, "negligible" resistances become real problems. An inch of cap lead in common between rectifier circuit and load circuit, 0.01 ohms taking 100A peaks, will inject 1V of ripple to the load even if the internal cap voltage could be "dead-steady". You will almost never get to 1% ripple (maybe on very small projects where over-kill is cheap). For simple tube-work, you get a few-% ripple and then add an R-C. This not only doubles the rejection slope, it rejects the parasitic ripple on cap leads. To go further, add another R-C for every 20dB-30dB of ripple reduction you need; trying to get 40dB in one R-C is usually frustrating and also more expensive than two 20dB R-C networks.
I have a 30V 60mA D-C-R-C-R-C supply which feeds directly to a low line-level input (simplex audio+power). One more R-C to the mike capsule. There's not a lick of hum/buzz in it. To do this on one stage would need a 60F (yes F) cap with 10 microOhm lead and internal resistance. Bigger than a milk-crate and over $10,000. My way was $13.
I'm puzzled by your blown-open transformer event. Yes, the one winding was taking double surge current. But unless the cap-bank was too big to fit through the door, the surge should not have lasted the many seconds it takes to melt heavy copper. I wasn't there, but from a distance I wonder if it was just connected wrong, or very marginally designed. I've been around similar events and sometimes you just can't tell after the smoke clears.
Svart
Well-known member
was it a toroidal transformer? Most of those have a thermal fuse on the primary windings deep down in the coils themselves to prevent a catastrophic copper failure. I know because i had one open like you mentioned and then I actually unwound one and found it. I removed it and rewound it and it works fine to this day.
Unquestionably too much of a good thing.. worthwhile object lesson against brute force over finesse. If the caps weren't properly formed in they could present a nastier than usual load. On my home (brew) amp I have a two position on switch where I first power it up through a soft start resistor, shunted in normal mode.
I wonder if this whole exercise isn't complicating a relatively simple exercise (ripple voltage estimates). Most PS current draw looks like a simple current source (sink?). Ripple calculations are back of the envelope simple. Amps/farads=volts per second. There will be small errors for conduction angle and such but how much precision is needed in ripple calculation?
At one point I considered looking at ripple voltage as a noninvasive measure for transformer winding temperature (imputed from winding resistance, imputed from peak voltage vs, ave current). It was cheaper to just drop in a thermal sensor that time, but with some spare computer power available who knows?
JR
I wonder if this whole exercise isn't complicating a relatively simple exercise (ripple voltage estimates). Most PS current draw looks like a simple current source (sink?). Ripple calculations are back of the envelope simple. Amps/farads=volts per second. There will be small errors for conduction angle and such but how much precision is needed in ripple calculation?
At one point I considered looking at ripple voltage as a noninvasive measure for transformer winding temperature (imputed from winding resistance, imputed from peak voltage vs, ave current). It was cheaper to just drop in a thermal sensor that time, but with some spare computer power available who knows?
JR
Similar threads
