Svart
Well-known member
rail/ground bounce. Still happens and is the reason for keeping analog and digital sources and sinks separate. Most think the analog portion affects the digital more but it is the other way around.
Help Support GroupDIY Audio Forum:
rail/ground bounce. Still happens and is the reason for keeping analog and digital sources and sinks separate. Most think the analog portion affects the digital more but it is the other way around.
bcarso
Well-known member
Regarding the issues of current transients from digital gates, I seem to recall some discussions in Howard Johnson's books, which I have at least one of...somewhere.
Recently I revised a design that had a simple D-flipflop divide-by-two using a 74HC74. Initially the other section was unused and had the inputs referenced appropriately to keep them from wandering about and having things draw excess current.
I noticed that there was a way of hooking the unused FF up to be an inverter, and now have it driving a resistor to be a compensating load for another section of the circuit. Probably a small effect, and it does double the current consumption in that section, but also makes the rail quieter. Since that same rail, with additional filtering, is used for an IR receiver chip sensitive to that clock frequency, it seemed like a good thing to do.
EDIT: Note however that I did this to make the ~steady-state drain on the rail more constant. In the process I roughly doubled the glitch energy associated with the internal flip-flop shoot-through currents. But the hope is that proper local bypassing and perhaps series-R or series-L decoupling will control this effect. As Chris indicates, you're not likely to do much better, in terms of response speed, than a proper capacitor.
I get a lot of freebie magazines and try to at least glance through them as they come in. In a uW&RF I just noticed an ad for an American Technical Ceramics 100nF cap in an 0402 package that has low insertion loss characterized out to 40GHz. Now that's a fast cap---of course to realize its benefits the rest of the circuitry has to be of comparably small geometry. I had to work with 0402 on the last major thng I did for Harman, the JBL-branded On Tour. I did not enjoy it, but that I could manage at all, with marginal tools and presbyopia coming on strong as I push 60, was fairly amazing. If a client isn't super-pressed for board space I tend to go no lower than 0805 packages.
Recently I revised a design that had a simple D-flipflop divide-by-two using a 74HC74. Initially the other section was unused and had the inputs referenced appropriately to keep them from wandering about and having things draw excess current.
I noticed that there was a way of hooking the unused FF up to be an inverter, and now have it driving a resistor to be a compensating load for another section of the circuit. Probably a small effect, and it does double the current consumption in that section, but also makes the rail quieter. Since that same rail, with additional filtering, is used for an IR receiver chip sensitive to that clock frequency, it seemed like a good thing to do.
EDIT: Note however that I did this to make the ~steady-state drain on the rail more constant. In the process I roughly doubled the glitch energy associated with the internal flip-flop shoot-through currents. But the hope is that proper local bypassing and perhaps series-R or series-L decoupling will control this effect. As Chris indicates, you're not likely to do much better, in terms of response speed, than a proper capacitor.
I get a lot of freebie magazines and try to at least glance through them as they come in. In a uW&RF I just noticed an ad for an American Technical Ceramics 100nF cap in an 0402 package that has low insertion loss characterized out to 40GHz. Now that's a fast cap---of course to realize its benefits the rest of the circuitry has to be of comparably small geometry. I had to work with 0402 on the last major thng I did for Harman, the JBL-branded On Tour. I did not enjoy it, but that I could manage at all, with marginal tools and presbyopia coming on strong as I push 60, was fairly amazing. If a client isn't super-pressed for board space I tend to go no lower than 0805 packages.
Svart
Well-known member
A balanced flip flop? Now that is something very cool. For some reason I believe that I would have likely never thought of that...
0402 parts.. This past week I was messing with 0402 parts on an RF tuner trying to debug an erroneous peak at 33mhz.. The engineer(s) previously in charge of this design(thankfully I'm not usually working with them) decided that putting a 33mhz clock near a very low impedence input was a good thing..
:roll:
But those damn parts were just a nightmare to fish in and out of a shielded area about .5" wide and .6" deep with tweezers and an iron. I almost had the soldertech take the whole damn fence off the board for the testing but the coffee wore off and my hands steadied a bit.
Now they have 0201 parts! I can't even work with them without a microscope and precision tweezers and I have 20/15 vision!.
Svart
Well-known member
Hey Brad, would you say that a lot of the HF noise on the power rails is byproduct of diode snap? I was thinking about this after buying a batch of power diodes, supersoft-superfast etc.
It's interesting because I was thinking about this thread again and I still can't shake the notion of linear Vregs being pretty good at what they do, especially some of the pricey LDOs.
It seems that we need to look at ALL aspects of the DC generation process here, not just the "what we do to fix the problem" discussion. A lot of what folks do to suppress noise is nothing more than band-aids on a larger issue, much like paralleling caps with rectifiers to "fix" noise issues.. we all know this doesn't actually "fix" the issue but move the problem into another area.
What do you say?
It's interesting because I was thinking about this thread again and I still can't shake the notion of linear Vregs being pretty good at what they do, especially some of the pricey LDOs.
It seems that we need to look at ALL aspects of the DC generation process here, not just the "what we do to fix the problem" discussion. A lot of what folks do to suppress noise is nothing more than band-aids on a larger issue, much like paralleling caps with rectifiers to "fix" noise issues.. we all know this doesn't actually "fix" the issue but move the problem into another area.
What do you say?
bcarso
Well-known member
Recovery spikes are abominable because they are so damn broadband. Really effective suppression at the source is tough. For very cost-sensitive designs I throw on the ceramic discs to shift most of the energy down to lower frequencies where it doesn't radiate as efficiently, but as one does so you are introducing more coupling from the line, and potentially from your circuit back into the line.
Too bad the SiC diodes are so much higher voltage drop and so pricey. Everyone I know who has played with them says they really do have nothing but a pure C in reverse and no trace of nasties.
Of course the recovery spike emissions do discipline one to make the rest of the circuit less sensitive, so in a way this can produce benefits in terms of immunity from other things. And we know there's plenty of other stuff spewing out there, some of which we cavalierly push next to our brains all the time. It's an interesting race to see what the new scourge will be, with my money on the prions but cellphones coming up on the inside...
Too bad the SiC diodes are so much higher voltage drop and so pricey. Everyone I know who has played with them says they really do have nothing but a pure C in reverse and no trace of nasties.
Of course the recovery spike emissions do discipline one to make the rest of the circuit less sensitive, so in a way this can produce benefits in terms of immunity from other things. And we know there's plenty of other stuff spewing out there, some of which we cavalierly push next to our brains all the time. It's an interesting race to see what the new scourge will be, with my money on the prions but cellphones coming up on the inside...
Svart
Well-known member
Digikey has/had them but I have not looked in a while because they were $$$$$ and you had to buy in bulk.
EDIT:
Digikey: search for prefix SDT.
Infineon makes em, cheapest is 6$ a piece, minimum order of 10
EDIT:
Digikey: search for prefix SDT.
Infineon makes em, cheapest is 6$ a piece, minimum order of 10
bcarso
Well-known member
[quote author="Gus"]Who sells SiC diodes?
and a highjack
prions
Have you read "Deadly Feasts" By Richard Rhodes?[/quote]
Haven't read that yet but I believe I listened to an interview with him, if it is the one I'm thinking about, wherein he points out that there may be a lot of variant Creuzfeldt-Jakob disease in seniors misdiagnosed as Alzheimer's. Scary scary stuff.
and a highjack
prions
Have you read "Deadly Feasts" By Richard Rhodes?[/quote]
Haven't read that yet but I believe I listened to an interview with him, if it is the one I'm thinking about, wherein he points out that there may be a lot of variant Creuzfeldt-Jakob disease in seniors misdiagnosed as Alzheimer's. Scary scary stuff.
[quote author="JCMaudio"]I have been reading about voltage regulation recently and have been considering building the sulzer regulator (schem below) as an alternative to the 3 term regulators. I've found very little info. on the practical differences between the two when used in an audio application. Can anyone with any experience with these comment on audible differences? I'm assuming that audible differences would very somewhat with amplifier design... I know the best solution is to use my ears, but I'm just wondering what others have experienced.
http://www.alw.audio.dsl.pipex.com/sulzer_circuit.htm
Thanks[/quote]
Since I'm a new guy, I'll try to be careful about my choice of words.
IMO if you can hear your PS your circuit has inadequate PSRR. While some pretty popular opamps are suspect in one rail or the other this is a bit of a "lower the water, raise the bridge" tradeoff. In general complexity is better invested in the circuitry doing the heavy lifting (operating at high noise gains).
I will concede only one incidence of my "hearing" an obvious PS flaw. A batch of 3 terminal regulators in a good sized recording console that used a pair of 3 terminal regs on each strip exhibited a output shot noise (probably the result of a die shrink and process purity) that with some 65 pair of regulators in a single chassis all dumping their noise spikes into a common ground worked their way into the sensitive L/R bus. The vendor (one of the majors) argued the parts were good, thanks to a data sheet noise specification you could drive a truck though. So in sensitive apps it can make a difference.
Back in the '80s when people were still listening to those round black vinyl discs, I used a PS design in a phono pre that looks a little like that Sulzer. I used a popular bi-fet opamp instead of 5532 since there was a discrete pass element so the opamp didn't have to source much current and was there primarily for GBW and modest input noise. I also just used a common zener for my voltage reference since I didn't care what the absolute output voltage was, only that it was in the proper range, clean, and low impedance over the useful bandwidth.
While this PS worked fine, it was IMO overkill. Asking you to do what I say not what I do a far more practical approach is to consider a 3 terminal reg but with a simple tweak. If you look at the popular GP 3-terminal series, they are quite good for most important metrics. One perhaps important shortcoming (besides noise floor) is a rising output impedance with frequency which I attribute to the 741 level performance internal gain circuitry. If you look at a graph of the output impedance it starts rising from a respectable level well down in the audio range. I found by simply adding a 1,000 uFd electrolytic in parallel with a typical 3-terminal regulator the falling impedance of the capacitor with frequency would nicely balance the rising impedance due to limited GBW. Of course even this will not insure low source impedance at very high frequencies where the electrolytic's ESL would dominate but that is better managed by local decoupling caps as appropriate.
I am not advocating arbitrarily cutting 1,000uF caps into existing designs although it should be relatively harmless. The main physical consideration besides size is that the ground lead of these added capacitors have a clean shot to the reference ground.
Using a high performance PS does not eliminate the need to engineer good PSRR into high gain stages, but I guess it could be used as a band-aid after the fact.
JR
http://www.alw.audio.dsl.pipex.com/sulzer_circuit.htm
Thanks[/quote]
Since I'm a new guy, I'll try to be careful about my choice of words.
IMO if you can hear your PS your circuit has inadequate PSRR. While some pretty popular opamps are suspect in one rail or the other this is a bit of a "lower the water, raise the bridge" tradeoff. In general complexity is better invested in the circuitry doing the heavy lifting (operating at high noise gains).
I will concede only one incidence of my "hearing" an obvious PS flaw. A batch of 3 terminal regulators in a good sized recording console that used a pair of 3 terminal regs on each strip exhibited a output shot noise (probably the result of a die shrink and process purity) that with some 65 pair of regulators in a single chassis all dumping their noise spikes into a common ground worked their way into the sensitive L/R bus. The vendor (one of the majors) argued the parts were good, thanks to a data sheet noise specification you could drive a truck though. So in sensitive apps it can make a difference.
Back in the '80s when people were still listening to those round black vinyl discs, I used a PS design in a phono pre that looks a little like that Sulzer. I used a popular bi-fet opamp instead of 5532 since there was a discrete pass element so the opamp didn't have to source much current and was there primarily for GBW and modest input noise. I also just used a common zener for my voltage reference since I didn't care what the absolute output voltage was, only that it was in the proper range, clean, and low impedance over the useful bandwidth.
While this PS worked fine, it was IMO overkill. Asking you to do what I say not what I do a far more practical approach is to consider a 3 terminal reg but with a simple tweak. If you look at the popular GP 3-terminal series, they are quite good for most important metrics. One perhaps important shortcoming (besides noise floor) is a rising output impedance with frequency which I attribute to the 741 level performance internal gain circuitry. If you look at a graph of the output impedance it starts rising from a respectable level well down in the audio range. I found by simply adding a 1,000 uFd electrolytic in parallel with a typical 3-terminal regulator the falling impedance of the capacitor with frequency would nicely balance the rising impedance due to limited GBW. Of course even this will not insure low source impedance at very high frequencies where the electrolytic's ESL would dominate but that is better managed by local decoupling caps as appropriate.
I am not advocating arbitrarily cutting 1,000uF caps into existing designs although it should be relatively harmless. The main physical consideration besides size is that the ground lead of these added capacitors have a clean shot to the reference ground.
Using a high performance PS does not eliminate the need to engineer good PSRR into high gain stages, but I guess it could be used as a band-aid after the fact.
JR
bcarso
Well-known member
Welcome to the forum. Please don't be too concerned about expressions of dissent or floating controversial proposals! As long as it doesn't descend to personal attacks we're a pretty resilient bunch.
Some of the minimalists feel better when their gain stages are very simple. I don't tend to agree with them unless they can justify their prejudices beyond hearsay and superstition, but neither would I trust a design just because it was complex. You have to know what you are doing. There are pernicious half-truths out there.
But when a minimalist stage has poor PSR then the onus is on the supply, and some care needs be taken.
On the subject of adding lots of output C to three-terminal regs, one does need to be mindful of the potential of the regulator to be damaged by a dump of output C energy in the event that the unregulated input is suddenly collapsed. Now, if one can assure that never happens, not to worry. Nonetheless, a reverse-biased diode from output to input is cheap insurance against this contingency.
Some of the minimalists feel better when their gain stages are very simple. I don't tend to agree with them unless they can justify their prejudices beyond hearsay and superstition, but neither would I trust a design just because it was complex. You have to know what you are doing. There are pernicious half-truths out there.
But when a minimalist stage has poor PSR then the onus is on the supply, and some care needs be taken.
On the subject of adding lots of output C to three-terminal regs, one does need to be mindful of the potential of the regulator to be damaged by a dump of output C energy in the event that the unregulated input is suddenly collapsed. Now, if one can assure that never happens, not to worry. Nonetheless, a reverse-biased diode from output to input is cheap insurance against this contingency.
[quote author="bcarso"]
On the subject of adding lots of output C to three-terminal regs, one does need to be mindful of the potential of the regulator to be damaged by a dump of output C energy in the event that the unregulated input is suddenly collapsed. Now, if one can assure that never happens, not to worry. Nonetheless, a reverse-biased diode from output to input is cheap insurance against this contingency.[/quote]
Thanks for the welcome. :guinness:
Agreed, if reverse biased these parts can be damaged. I have seen those diodes on app notes but never personally had an application where the unreg decayed faster than the regulated. I guess it is possible if there are other circuit blocks pulling directly from the unreg and the regulated rail is lightly loaded.
If the regulator is the only discharge path for the unreg supply this shouldn't be an issue. Of course they wouldn't put the diodes on the app notes if somebody somewhere wasn't toasting parts... I guess another reason for the diode is to protect against inductive spikes coming from switching relays or turning inductive loads on and off, however the 1000uF cap might help more than hurt in that case.
To reinforce what I said in my earlier post, I am not advocating wholesale addition of massive capacitors to regulated rails everywhere. I'm just offering this "3-term +1kuF" as a simple, cheap way to generate a reasonably low impedance rail over the full audio bandwidth. Adding a diode that you don't need will do far less harm than not having one that's needed.
JR
On the subject of adding lots of output C to three-terminal regs, one does need to be mindful of the potential of the regulator to be damaged by a dump of output C energy in the event that the unregulated input is suddenly collapsed. Now, if one can assure that never happens, not to worry. Nonetheless, a reverse-biased diode from output to input is cheap insurance against this contingency.[/quote]
Thanks for the welcome. :guinness:
Agreed, if reverse biased these parts can be damaged. I have seen those diodes on app notes but never personally had an application where the unreg decayed faster than the regulated. I guess it is possible if there are other circuit blocks pulling directly from the unreg and the regulated rail is lightly loaded.
If the regulator is the only discharge path for the unreg supply this shouldn't be an issue. Of course they wouldn't put the diodes on the app notes if somebody somewhere wasn't toasting parts... I guess another reason for the diode is to protect against inductive spikes coming from switching relays or turning inductive loads on and off, however the 1000uF cap might help more than hurt in that case.
To reinforce what I said in my earlier post, I am not advocating wholesale addition of massive capacitors to regulated rails everywhere. I'm just offering this "3-term +1kuF" as a simple, cheap way to generate a reasonably low impedance rail over the full audio bandwidth. Adding a diode that you don't need will do far less harm than not having one that's needed.
JR
pstamler
Well-known member
The diode, though, has a hidden gotcha: it has enough capacitance to act as a nice conduit for RF noise around the regulator and onto the regulated rails. Ben Duncan has a nice fix: use a 1N4048 instead of a power diode (much lower capacitance). Add 22 ohms in series with it for current limiting, since the 1N4048 can handle much less current than, say, a 1n400x.
Peace,
Paul
Peace,
Paul
[quote author="pstamler"]The diode, though, has a hidden gotcha: it has enough capacitance to act as a nice conduit for RF noise around the regulator and onto the regulated rails. Ben Duncan has a nice fix: use a 1N4048 instead of a power diode (much lower capacitance). Add 22 ohms in series with it for current limiting, since the 1N4048 can handle much less current than, say, a 1n400x.
Peace,
Paul[/quote]
At the risk of wandering further off topic, how many have actually lost regulators due to not using a protection diode? Most 3-term failures I've seen were only after multiple cycles of severe over temp abuse (like after losing heat sink). They will generally tolerate quite a bit of that too before finally giving up the ghost.
I'm not familiar with 4048 but if it is similar to 914/4148 series they will actually handle 1A peak. They are dissipation limited due to package size but to simply discharge a single electrolytic cap into whatever, they will probably hang.
Again I don't have a clear picture of what this low impedance path to ground is in the unregulated PS other than perhaps a fault condition like shorted rectifier diode or reservoir cap. At some point you might be better off throwing your pennies at bigger rectifier diodes or higher voltage caps.
The coupling of HF noise from the unregulated PS to output of the regulator is yet another symptom of the stock 3 terminal's poor output impedance at high audio frequencies. The 1,000 uFd cap helps address that and of course HF decoupling caps located at the ICs are always good design practice.
JR
Peace,
Paul[/quote]
At the risk of wandering further off topic, how many have actually lost regulators due to not using a protection diode? Most 3-term failures I've seen were only after multiple cycles of severe over temp abuse (like after losing heat sink). They will generally tolerate quite a bit of that too before finally giving up the ghost.
I'm not familiar with 4048 but if it is similar to 914/4148 series they will actually handle 1A peak. They are dissipation limited due to package size but to simply discharge a single electrolytic cap into whatever, they will probably hang.
Again I don't have a clear picture of what this low impedance path to ground is in the unregulated PS other than perhaps a fault condition like shorted rectifier diode or reservoir cap. At some point you might be better off throwing your pennies at bigger rectifier diodes or higher voltage caps.
The coupling of HF noise from the unregulated PS to output of the regulator is yet another symptom of the stock 3 terminal's poor output impedance at high audio frequencies. The 1,000 uFd cap helps address that and of course HF decoupling caps located at the ICs are always good design practice.
JR
bcarso
Well-known member
I've never lost a regulator that way that I know of. I suspect the most likely mode however would be when one of us inadvertently shorts the unreg bus. Another possibility is when a reg fails with a short to common and other regulators tied to the same unreg bus are reverse-biased.
By the way, tangentially related to the thread topic, I just found something I'd been hunting for, a reference to what I call the driven-rail approach, which I've used and thought I invented---except Funasaka and Kondou stole my idea several years before I had it: see J. Audio Eng. Soc., Vol. 30, No. 5, 1982 May, Feedforward Floating Power Supply.
By the way, tangentially related to the thread topic, I just found something I'd been hunting for, a reference to what I call the driven-rail approach, which I've used and thought I invented---except Funasaka and Kondou stole my idea several years before I had it: see J. Audio Eng. Soc., Vol. 30, No. 5, 1982 May, Feedforward Floating Power Supply.
bcarso
Well-known member
[quote author="JohnRoberts"]
I'm not familiar with 4048 but if it is similar to 914/4148 series they will actually handle 1A peak.
JR[/quote]
I think Paul meant a 4148. The 4048 according to circuitmaker is a 275A part :shock:
I'm not familiar with 4048 but if it is similar to 914/4148 series they will actually handle 1A peak.
JR[/quote]
I think Paul meant a 4148. The 4048 according to circuitmaker is a 275A part :shock:
pstamler
Well-known member
Yep -- sorry, typo.
Peace,
Paul
Peace,
Paul
Svart
Well-known member
everytime you guys mention something I always have a story about it.. :green:
a motor controller had a 12v regulator to power the PWM ic and the FETdrivers, randomly the 12v reg would die and short full source rail directly to the PWM ic and fet drivers. the fet drivers would die but the pwm ic was actually rated higher than the rail but I digress.
Turns out someone thought it a good idea to add a lot of cap on the output of the reg to keep part of the circuit up longer for a digipot to store it's settings.
The vreg was being reversed biased during poweroff and dying a quiet death, however very randomly between regulators and manufacturers. Seems that some manufacturers add some degree of protection to their models of regulators, but the bean counters wanted CHEAP so adding .50$ more was actually being denied.. again i digress. We thought about many fixes including a protection diode but ultimately went with a zener shunt and a power resistor which worked better and was cheaper.
a motor controller had a 12v regulator to power the PWM ic and the FETdrivers, randomly the 12v reg would die and short full source rail directly to the PWM ic and fet drivers. the fet drivers would die but the pwm ic was actually rated higher than the rail but I digress.
Turns out someone thought it a good idea to add a lot of cap on the output of the reg to keep part of the circuit up longer for a digipot to store it's settings.
The vreg was being reversed biased during poweroff and dying a quiet death, however very randomly between regulators and manufacturers. Seems that some manufacturers add some degree of protection to their models of regulators, but the bean counters wanted CHEAP so adding .50$ more was actually being denied.. again i digress. We thought about many fixes including a protection diode but ultimately went with a zener shunt and a power resistor which worked better and was cheaper.
bcarso
Well-known member
Your stories are always appreciated Svart. Nothing like the real world to calibrate our conceptions.
Your last one also is grist for the mill of minimalism and MTBF (mean time between failures) analysis. Quality Assurance types often start off in life with a mantra of minimizing component count. And they can be encouraged by short-sighted management, who see parts that could be eliminated as money to be saved. But when understanding of circuits and systems is lacking, both of these laudable goals often get pursued at the expense of the bigger picture.
Half-truths abound. They are maybe better than no truths at all, but when emblazoned on banners and paraded about they can be disasterous. There is no substitute for the combination of experience and thoughtfulness, despite the pronouncements of Taguchi and his cult followers.
Your last one also is grist for the mill of minimalism and MTBF (mean time between failures) analysis. Quality Assurance types often start off in life with a mantra of minimizing component count. And they can be encouraged by short-sighted management, who see parts that could be eliminated as money to be saved. But when understanding of circuits and systems is lacking, both of these laudable goals often get pursued at the expense of the bigger picture.
Half-truths abound. They are maybe better than no truths at all, but when emblazoned on banners and paraded about they can be disasterous. There is no substitute for the combination of experience and thoughtfulness, despite the pronouncements of Taguchi and his cult followers.
Svart
Well-known member
:thumb:
Just as your wealth of knowledge that covers just about everything electronic is greatly appreciated here too!
Just in case anyone was wondering, TI and Motorola Vregs faired the best, almost never failing, ST and fairchild were worse off. We even got a bunch of some brand that I had never heard of, rough plastic casing and an even rougher metal tab... :shock: they HAD to be cheap knockoffs because they would fail at a rate that was astounding. Out of 10, 2 would die withing a few minutes, a few more would die within weeks, and eventually they would all fail without reason. This led to a lot of RMAs.. :roll:
Just as your wealth of knowledge that covers just about everything electronic is greatly appreciated here too!
Just in case anyone was wondering, TI and Motorola Vregs faired the best, almost never failing, ST and fairchild were worse off. We even got a bunch of some brand that I had never heard of, rough plastic casing and an even rougher metal tab... :shock: they HAD to be cheap knockoffs because they would fail at a rate that was astounding. Out of 10, 2 would die withing a few minutes, a few more would die within weeks, and eventually they would all fail without reason. This led to a lot of RMAs.. :roll:
Similar threads
