"PSRR" is usually reserved for active circuits which have an output node distinct from the power supply node(s). PSRR(dB) = 20 log10 (dVout/dVsupply)
In this case I think you are talking about "line rejection" which is (dVout/dVin) for a power supply. If so, try not to call it PSRR.
I suspect you are seeing "the output pole" increase in frequency, as output current increases. I suspect this is because the emitter follower's output resistance re = (kT/q)/Ic falls as collector current rises. re forms a pole with the total capacitance on the output. Larger Ic means smaller re, means smaller RC timeconstant at the output, means higher frequency for the output pole.
In this case I think you are talking about "line rejection" which is (dVout/dVin) for a power supply. If so, try not to call it PSRR.
I suspect you are seeing "the output pole" increase in frequency, as output current increases. I suspect this is because the emitter follower's output resistance re = (kT/q)/Ic falls as collector current rises. re forms a pole with the total capacitance on the output. Larger Ic means smaller re, means smaller RC timeconstant at the output, means higher frequency for the output pole.
Yes, I meant line rejection - thanks for the correction to proper terminology.
But for whatever explanation for the increase in the gain crossover frequency (I hope it is correct to use this term in this situation - I am referring to the frequency at which loop gain falls to unity/0dB) for the source follower with load current (see graphic in post #258) is it not the case that, as it does so, there's at least the same or more negative feedback at any given frequency up to it? Intuitively I would have thought that for a regulator such as this that more feedback => better line rejection.
But for whatever explanation for the increase in the gain crossover frequency (I hope it is correct to use this term in this situation - I am referring to the frequency at which loop gain falls to unity/0dB) for the source follower with load current (see graphic in post #258) is it not the case that, as it does so, there's at least the same or more negative feedback at any given frequency up to it? Intuitively I would have thought that for a regulator such as this that more feedback => better line rejection.
I stumbled across Linear Technology's Application Note AN104 today. It contains some useful ideas; have a look at this figure. It shows a voltage regulator testing circuit in action. The load current has a slow and linear ramp-up, then a very fast step-down. Surprise! The regulator under test responds differently to the two stimuli.
_
_
Attachments
I read a method of using the RC time constant to determine the effective leakage. Can't remember where.FYI the 470u 6V3 FR capacitor that I charged to 4V a few days ago and let sit on my desk is now at 3.12V. Not bad for leakage
A long time for voltage to fall = low leakage. A couple of weeks to fall to half voltage is good. A month or longer is better.
BTW,
testing the voltage pulls out some charge and effectively stops the test from continuing, one needs to "start again", if one wants a longer test.
Thanks guys.
I'm going around in circles with respect to output cap selection. As noted in #258 I had concluded to simply use two of the 'low impedance' EEU-FMC1C471L on each rail. However, my conclusion that adding another cap had little impact on output impedance didn't sit well and so I did the analysis again. Of course, the series addition of ESR did help and lowered overall output impedance. It also seemed to help stability margins a little. So why not make it four caps - I have room...
Mark's post reminded me that I hadn't looked at transient response again since adding series inductance to the output caps. Attached are two charts plotting the response while stepping the load from 0.25A to 4.75A in 15ns (the step down in load is easier on the regulator). Transient 1 is with four 470uF caps each with 30mR ESR and 4.5n LSR, in parallel. Transient 2 is with just two caps.
A 4.5A step in load is extreme but useful for as analysis tool. The 4-cap waveform suffers less dip in voltage than the 2-cap. Neither seem particularly pretty, although note the timescale is (a somewhat miniscule?) 50ns per division.
Application note AN-104 which Mark referred to discusses combining 0mR ESR ceramics with higher ESR capacitors "to achieve both good high frequency bypassing and fast settling time". However, adding a plain 10uF 'bypass cap' just seemed to throw me back into poor stability. Transient response 3 shows the oscillation that occurs by adding a 10uF cap. It also shows the impact of an increasing level of ESR for the main output caps (red=30mR, blue=50mR and green=100mR if you can see it). The amplitude of error is less but all cases seem very "un-damped." Note the substantial difference in timescale.
While I think Mark presented AN-104 principally for its test circuits, it has lead me to go around in circles regarding output caps. Thoughts?
I'm going around in circles with respect to output cap selection. As noted in #258 I had concluded to simply use two of the 'low impedance' EEU-FMC1C471L on each rail. However, my conclusion that adding another cap had little impact on output impedance didn't sit well and so I did the analysis again. Of course, the series addition of ESR did help and lowered overall output impedance. It also seemed to help stability margins a little. So why not make it four caps - I have room...
Mark's post reminded me that I hadn't looked at transient response again since adding series inductance to the output caps. Attached are two charts plotting the response while stepping the load from 0.25A to 4.75A in 15ns (the step down in load is easier on the regulator). Transient 1 is with four 470uF caps each with 30mR ESR and 4.5n LSR, in parallel. Transient 2 is with just two caps.
A 4.5A step in load is extreme but useful for as analysis tool. The 4-cap waveform suffers less dip in voltage than the 2-cap. Neither seem particularly pretty, although note the timescale is (a somewhat miniscule?) 50ns per division.
Application note AN-104 which Mark referred to discusses combining 0mR ESR ceramics with higher ESR capacitors "to achieve both good high frequency bypassing and fast settling time". However, adding a plain 10uF 'bypass cap' just seemed to throw me back into poor stability. Transient response 3 shows the oscillation that occurs by adding a 10uF cap. It also shows the impact of an increasing level of ESR for the main output caps (red=30mR, blue=50mR and green=100mR if you can see it). The amplitude of error is less but all cases seem very "un-damped." Note the substantial difference in timescale.
While I think Mark presented AN-104 principally for its test circuits, it has lead me to go around in circles regarding output caps. Thoughts?
Attachments
For low current regulators, a 74AC14 makes a nice test pulse generator. It has lots of drive current and very fast edges (two things that make 74AC possibly the worst possible choice to use as a logic IC). One gate will happily drive 50-100mA. I sometimes "overclock" mine to 6 or 7 volts VCC, it works fine although way past maximum ratings and gets even meaner. If it burns, it costs like 20c, so no problem.
2ns 100mA current step makes sure that if something wants to ring or oscillate, it will, and you'll see it. Next step up would be a properly driven MOSFET, if you're willing to invest about a buck you can wire some driver like ADP3120 and a dual MOS like AO4840 and you'll get 10-20ns switching time and a few amps. The di/dt isn't much faster than an old 74AC gate though, so for testing purposes a 74AC and a resistor is a very simple and useful tool.
> I'm going around in circles with respect to output cap selection.
The whole point of ceramics is
1) low inductance
2) they're cheap so you can put an whole army of them in parallel and get lower inductance
So if you use ceramic caps, put them at the load. In your case, there are already a few hundred ceramics on your mobo, you don't need to put more on your boards.
If your regulator is connected to the load via wires, then you don't care too much about its output inductance. Just design it so it is a bit less than the wire inductance. So, don't put any ceramics on the output. This also makes it a lot easier to stabilize.
> However, adding a plain 10uF 'bypass cap' just seemed to throw me back into poor stability
MLCC have very low ESR which makes the impedance curve a sharp "V" shape and, more important, makes a sharp 180° phase shift at the tip of the V. Sharp phase shifts make control loops difficult to design.
Last edited:
Hi. But what about the transient response...are you saying not to sweat it because the burden of this will be borne by mobo capacitance? (I also wonder about other applications of this type of reg which may not have the benefit of bypass capacitance on the load...) Should I just leave it at the 4x470uF 'low impedance' caps and move on?
There would appear to be a heavy trade-off between output stage ESR being high enough to allow for some bypassing to speed transient response and output impedance (and stability).
FWIW here is the latest draft of the board. I have ordered PCBs for the power control / measurement pre-amp board - hopefully they will ship ahead of Chinese New Year. I have also ordered all the parts for it (plus spares). I have likely missed the deadline for sending this board off and so have a couple of weeks to complete it. Any comments appreciated.
(I've not yet done the silkscreen but it would be largely based on what you see here with obvious changes for those parts fitted under the board.)
(I've not yet done the silkscreen but it would be largely based on what you see here with obvious changes for those parts fitted under the board.)
Attachments
You sure your traces will carry 5 amps ?
You could put a copper pour for Vout in the lower left corner on top layer. for example. This way, the impedance of the trace from Q4 to output connector would be much lower. You can visually estimate the resistance of a trace, it is about 0.5 mOhm per square. A trace 3mm wide and 30mm long is 10 squares, ie 5 mOHm. Check your copper thickness.
Also the entire bottom layer could be GND, with just a slit along the line of C7 C6 C9.
> But what about the transient response...
1st rule of transient response is make sure there are no resonances or instabilities. Just calculate damping factor.
Then, if your regulator is connected to the load via traces or wires, there is no need to make the regulator have lower output impedance than the wires/traces. I mean, yeah, you can do it, but it will only be for glory.
You could put a copper pour for Vout in the lower left corner on top layer. for example. This way, the impedance of the trace from Q4 to output connector would be much lower. You can visually estimate the resistance of a trace, it is about 0.5 mOhm per square. A trace 3mm wide and 30mm long is 10 squares, ie 5 mOHm. Check your copper thickness.
Also the entire bottom layer could be GND, with just a slit along the line of C7 C6 C9.
> But what about the transient response...
1st rule of transient response is make sure there are no resonances or instabilities. Just calculate damping factor.
Then, if your regulator is connected to the load via traces or wires, there is no need to make the regulator have lower output impedance than the wires/traces. I mean, yeah, you can do it, but it will only be for glory.
Yes two ounce copper for all boards (soft start, Power Control Board, and this one). The main traces are 100 mil. However 2x100 mil traces at the pass transistor would violate minimum distances between traces so the trace from C5 to Q4 is currently 70 mil. Seems a bit silly to have fat traces that are then small at one point. Nonetheless, it's easy to do a pour from source to Vout. Earlier, we discussed the need for traces to sequence the large filter caps rather than a big pour - does the same not apply here?
Re GND, ""could or "should"? I can easily make the change, but harder to know his/whether it really helps.
Thanks for all the guidance.
Re GND, ""could or "should"? I can easily make the change, but harder to know his/whether it really helps.
Thanks for all the guidance.
The cut plane is only on the bottom layer. C7 is only connected on the top layer - there's no connection to ground. It is, however, easy to shift the cut higher although I don't see what difference it can make when there's no connection to the bottom board planes. C4 and C6 pins are in there respective ground plane areas. Does it really matter if their cans overlap planes? If so, the comment would have been due re post #270. C5 has its ground pin in the right plane. It's positive trace goes, as before, to Q4 - there's no connection to the lower plane (except via the channel joining each plane). I can shift bleeder R2 if need be.
(blue=bottom layer; red=top layer)
(blue=bottom layer; red=top layer)
Last edited:
> I don't see what difference it can make when there's no connection to the bottom board planes.
If the traces were carrying highspeed signals then they would need to hug their reference plane. In this case I dont think you gonna have a problem.
Now you can get rid of the left slit in the ground plane, which has no real use except redirecting the load current through the sensitive parts of your regulator.
About the 3 high-power resistors : you can replace or enhance them with a high-current ferrite bead, just a piece of wire through a bead, you can buy that readymade or just buy the bead and supply your own wire
Anyway, a resistor will provide resistance down to DC, but the downside is dissipation. An inductor (or ferrite) will provide better filtering (higher impedance) at HF, much lower dissipation, but of course, no filtering at low frequencies like 50 Hz. Fortunately the opamp has huge loop gain at LF, which translates into huge PSRR.
If you don't care about the dissipation, and you like your resistors, you can put a ferrite bead on the resistor lead for some extra HF filtering almost for free (well, about 20c each...)
If the traces were carrying highspeed signals then they would need to hug their reference plane. In this case I dont think you gonna have a problem.
Now you can get rid of the left slit in the ground plane, which has no real use except redirecting the load current through the sensitive parts of your regulator.
About the 3 high-power resistors : you can replace or enhance them with a high-current ferrite bead, just a piece of wire through a bead, you can buy that readymade or just buy the bead and supply your own wire
Anyway, a resistor will provide resistance down to DC, but the downside is dissipation. An inductor (or ferrite) will provide better filtering (higher impedance) at HF, much lower dissipation, but of course, no filtering at low frequencies like 50 Hz. Fortunately the opamp has huge loop gain at LF, which translates into huge PSRR.
If you don't care about the dissipation, and you like your resistors, you can put a ferrite bead on the resistor lead for some extra HF filtering almost for free (well, about 20c each...)
Hi
I can shift the slit easily so that C7's pins are both over the lower plane. On your second point, I'm now confused. I thought the point was to isolate from the rest of the circuit the area in which the pulsing currents from the rectifier diodes and filter caps (C3-C5) circulate; that these should circulate 'locally', amongst themselves, and away from other parts of the circuit. In an amplifier context, Cordell and Self recommend bringing the ground points for the diodes and filter caps together with a heavy bus bar and then taking a spur of this for the grounding point for the rest of the amplifier. My attempts re this began with post #206.
About the 3 high-power resistors : you can replace or enhance them with a high-current ferrite bead, just a piece of wire through a bead, you can buy that readymade or just buy the bead and supply your own wire
Anyway, a resistor will provide resistance down to DC, but the downside is dissipation. An inductor (or ferrite) will provide better filtering (higher impedance) at HF, much lower dissipation, but of course, no filtering at low frequencies like 50 Hz. Fortunately the opamp has huge loop gain at LF, which translates into huge PSRR.
If you don't care about the dissipation, and you like your resistors, you can put a ferrite bead on the resistor lead for some extra HF filtering almost for free (well, about 20c each...)
Mark first mentioned the possibility of adding some ferrite beads in post #180. However, there was concern regarding whether beads such as those could withstand the heavy in-rush and narrow pulses associated with a capacitive input regulator such as this. The beads have a current rating but I was not sure if this was an RMS rating and what their peak capability might be. I reached out to Wurth but to no avail and so dropped the idea of including them. Perhaps you are familiar with their deployment in this area?
Thanks for the help
Steve
Re ground plane, see attached images with load current return in blue, it is better to keep it away from the reg's ground pour.
> However, there was concern regarding whether beads such as
> those could withstand the heavy in-rush and narrow pulses associated
> with a capacitive input regulator such as this. The beads have a current
> rating but I was not sure if this was an RMS rating and what their peak
> capability might be.
Fair-Rite 2743009112
This one has a real wire in it, about 3 mOhm, no chance of overheating in your case.
Kemet 80-B-20F-46
This one has no wire, just a hole. You can slip this style of ferrite on a resistor to make it more badass at HF.
I haven't checked if those two are OK for your design, though.
The manufacturer doesn't spec saturation current. That's annoying. I've used them at around 1 amp and they seem to work alright.
> However, there was concern regarding whether beads such as
> those could withstand the heavy in-rush and narrow pulses associated
> with a capacitive input regulator such as this. The beads have a current
> rating but I was not sure if this was an RMS rating and what their peak
> capability might be.
Fair-Rite 2743009112
This one has a real wire in it, about 3 mOhm, no chance of overheating in your case.
Kemet 80-B-20F-46
This one has no wire, just a hole. You can slip this style of ferrite on a resistor to make it more badass at HF.
I haven't checked if those two are OK for your design, though.
The manufacturer doesn't spec saturation current. That's annoying. I've used them at around 1 amp and they seem to work alright.
Attachments
An example ferrite bead: the Wurth 742 792 910 has a peak impedance at 19MHz and is rated for 5.5A (and conveniently has a model that comes with LTspice). See attached clipping from the data sheet. The data sheet warns not to exceed this limit else it will result in excessive heat and may damage the component...
I could squeeze this between the first two filter caps but I am unsure re peak currents, especially when the caps are first charged.
Re ground plane, I think the attached at least alleviates the points Andrew highlighted.
EDIT: ah I was typing as you were posting...
I could squeeze this between the first two filter caps but I am unsure re peak currents, especially when the caps are first charged.
Re ground plane, I think the attached at least alleviates the points Andrew highlighted.
EDIT: ah I was typing as you were posting...
Attachments
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.