Question about Using Remote Sense Leads:
I have been playing around with some LTspice simulations, trying to improve some of my positive series-pass voltage regulator designs, which have remote sense lines for the load's power supply and ground pins. Note that I do model the series inductances and resistances of both the main and the remote sense power and ground leads, between the regulator and the load.
I need some sort of a "sanity check" regarding the design goals: For the load ends of the remote sense lines, shouldn't a design goal be to make the load power supply pin voltage, MINUS the load ground pin voltage, constant?
If so, doesn't that imply that the negative input of the error amplifier should be fed from the output of a differential amplifer for which the two remote sense lines are the inputs?
Please correct me if I'm wrong, but, I think (IIRC) that most or all of the discrete regulator designs that I have seen, so far, seem to more-or-less ignore the load ground's remote sense line, in that way. Simulations of that type of topolgy, with highly-dynamic load currents, can make the power supply pin voltage of the load device look very, very quiet, relative to the "main" ground. But then the differential power supply voltage, i.e. directly across the load device, is NOT looking very quiet, because the load's ground pin voltage is "bouncing", due to the load's ground conductor's inductance and resistance. Using a differential remote sense amplifier tends to fix that problem (assuming I am correct about it being a problem).
Am I on the right track?
- Tom Gootee
http://www.fullnet.com/~tomg/index.htm
I have been playing around with some LTspice simulations, trying to improve some of my positive series-pass voltage regulator designs, which have remote sense lines for the load's power supply and ground pins. Note that I do model the series inductances and resistances of both the main and the remote sense power and ground leads, between the regulator and the load.
I need some sort of a "sanity check" regarding the design goals: For the load ends of the remote sense lines, shouldn't a design goal be to make the load power supply pin voltage, MINUS the load ground pin voltage, constant?
If so, doesn't that imply that the negative input of the error amplifier should be fed from the output of a differential amplifer for which the two remote sense lines are the inputs?
Please correct me if I'm wrong, but, I think (IIRC) that most or all of the discrete regulator designs that I have seen, so far, seem to more-or-less ignore the load ground's remote sense line, in that way. Simulations of that type of topolgy, with highly-dynamic load currents, can make the power supply pin voltage of the load device look very, very quiet, relative to the "main" ground. But then the differential power supply voltage, i.e. directly across the load device, is NOT looking very quiet, because the load's ground pin voltage is "bouncing", due to the load's ground conductor's inductance and resistance. Using a differential remote sense amplifier tends to fix that problem (assuming I am correct about it being a problem).
Am I on the right track?
- Tom Gootee
http://www.fullnet.com/~tomg/index.htm
how about a single sense wire on one side of the supply with a 2:1 correction ratio? volt drop on one leg must equal volt drop on the other leg if they are the same size.
OzMikeH said:
Thanks, OzMikeH! Off the top of my head, it sounds like that should be theoretically correct. If you just thought of it, that's impressively analytical of you.
I can try to simulate it. But, in looking, just now, at the + and gnd pin voltages for a simulated load (connected to a regulator model with differential sensing, however) consisting of a mosfet driven by a squarewave, with an 8 Ohm resistive load for the mosfet, with the mosfet and series 8 Ohm resistor combination bypassed by a 3300uF cap with 0.013 Ohms ESR and 50nH ESL, with approx 30V regulator output voltage, giving a load current that's a square wave varying between about 2.352A and 0.877A with rise and fall times of about 5us, I see that there are slight (mV range) differences in the voltage waveforms at the + and gnd pins of the load, mainly in terms of transients, which also vary with the length of the supply conductors. Unfortunately, unless I'm missing something, these differences do appear to be significant-enough to still warrant differential sensing, since I'm (hoping to be) interested in single-digit-microvolt and sub-microvolt variations of the load device's supply voltage.
- Tom Gootee
It just seemed to make sense to me, why try to do two things when one bigger one gives the same result.
trying to stabilize 30 volts to within microvolts with a square wave load like that is going to be incredibly difficult. A correction circuit with enough sensitivity and gain to deliver such steep transients is likely to oscillate if you sneeze near it.
I don't know what your application is, Have you considered using a low internal resistance NiCd battery (like those used to start jet engines) to see if brute force can slow the transients down to give you a chance to compensate.
Is it for a measurement application? It might be far simpler to log the power supply errors and record them along with your measurements. Afterward you can do the maths to normalise your measurements.
What you're trying to achieve is way above my head. You need one of the power supply engineers who worked in Motorola's two way radio division in the 1980s. Those guys cooked up some diabolical power supplies that make any technician's brains run out their ears. I'm sure they were just trying to get a higher component count than the Radio Transmitter guys.
trying to stabilize 30 volts to within microvolts with a square wave load like that is going to be incredibly difficult. A correction circuit with enough sensitivity and gain to deliver such steep transients is likely to oscillate if you sneeze near it.
I don't know what your application is, Have you considered using a low internal resistance NiCd battery (like those used to start jet engines) to see if brute force can slow the transients down to give you a chance to compensate.
Is it for a measurement application? It might be far simpler to log the power supply errors and record them along with your measurements. Afterward you can do the maths to normalise your measurements.
What you're trying to achieve is way above my head. You need one of the power supply engineers who worked in Motorola's two way radio division in the 1980s. Those guys cooked up some diabolical power supplies that make any technician's brains run out their ears. I'm sure they were just trying to get a higher component count than the Radio Transmitter guys.
OzMikeH said:
Well, that's just one of the somewhat-sadistic "test loads", for these circuit-development simulations. But initial testing, with a differential sense amplifier in one of my regulator circuits, with the described load conditions (about 1.5A p-p squarewave), shows the voltage across the load varying by about 15 uV p-p, except for the transients, which are also affected by the load's bypassing, and were still 2 to 3 mV p-p in this case. For 2.5A p-p through the load, the voltage across the load varies about 24 uV p-p, with about 4mV p-p transients (with almost no ringing).
The differential load voltage, during the flat parts of the squarewave, had no detectable noise or oscillation. That was with about 40VDC input voltage that also had 5V p-p of 120 Hz sine and 100mV p-p of 100 kHz sine. Changing to 1V p-p of the 100 kHz sine on the input, I still could not detect any 100 kHz in the output voltage, with my screen-plot resolution of about 800 nV.
Oops. Looks like I was sort of cheating, there. I had a ZTX1048A model for the pass transistor. Substituting a D44H11 model, but not re-optimizing the frequency responses, I got up to 1.6 uV p-p of 100 kHz in the differential output voltage, for 1V p-p 100 kHz in the input. The other output variations were approximately doubled, then, too.
Yup. That's one of the problems. Lead-length is a big factor there, too. So far, it looks like I can either trade off some performance to make it stable for some range of lead lengths, or, have better performance but require setting a trimmer or three based on lead lengths. If I get more serious about it, I'll also have to model the impedances of all of the internal connections, and everything might go to ****, then.
Good thoughts. But I'm "just playing around", with no particular application in mind, really. I've only been seriously looking at regulators' internal circuits for a few weeks. I thought maybe I'd see how it sounded when powering an audio amplifier!
My design goals so far are basically to see how stable I can make the load voltage, when faced with medium-amplitude, dynamic current loads, and to see how low I can get the load's differential voltage noise to be.
<Grin!> I really like that technician imagery, heh-heh. Maybe that should be required for design validation. I like the "diabolical" adjective, too. Seriously, though, I usually strive for minimum component count. Some of the guys I worked with in mil aerospace, back in the early 1980s, spent their lives trying to make everything smaller and lighter. They even had company-wide mottos, like, "Not a pound for air-to-ground.".
I'm still looking for a really-great pass transistor! Anyone?
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
gootee said:
Doesn't anyone have any thoughts about this? Or has anyone else tried this, yet? (Or maybe everyone is too busy, trying it, to answer? 🙂
I've made a bit more progress:
Simulating with two MJD44H11 NPNs in parallel, as the pass transistor (to better-handle the power dissipation), with the same MOSFET/8-Ohms/bypassed load pulling 1.5A p-p with 5us edge times, the voltage across the load now varies by only 6.4uV p-p. And another small bypass cap, across only the mosfet, has made the largest transients about 2.4 mV p-p, although I haven't attempted to optimize that, since it's just a test load, anyway. That was with 1-inch power and sense lead lengths, +30V output, 40.77V input, AND with, added to the input, 5Vp-p of 120 Hz sine and a full 1Vp-p of 100 kHz sine (no trace of either sine in the output, with 1-inch-length power and sense leads).
I haven't simulated any other regulator designs, besides my own. So I don't know how any of my results might compare to similar simulations of any other regulators. Any comments about whether my results are good or bad, or meaningless, would be appreciated.
I did try it with other pass transistors, including MJL3281A, MJE15032, MJF15030, MJE15028, and an IRF7343N Mosfet. They all worked quite well, with the diff amp's compensation adjusted. But none of them worked as well as the MJD44H11. (Any suggestions on what others might be good to try? Through-hole NPN would be nice, >=40V or more, capable of more than 20W power dissipation, but performing like a ZTX1048A or MJD44H11.)
My regulator's circuit is much like that of a "standard" series-pass regulator circuit, with a well-filtered voltage reference and an AD825 error amplifier, EXCEPT that the other (neg) input to the error amplifier is fed from a heavily-bypassed resistive divider (still "standard"), which is fed in what appears to be a not-so-standard way: from another AD825 that is configured as a differential amplifier, with the two remote sense lines as its inputs, in an attempt to minimize the amplitude of variations of V(load_+) pin MINUS V(load_gnd) pin. (If this is not a correct regulator optimization goal, I hope that someone will comment on that. Most other regulators for which I've seen schematics seem to attempt to minimize variation of Vout (or V(load+)) minus Vmain_ground, even when they use remote sense.)
The AD825 differential remote-sense amplifier uses four 9.1k resistors, in the standard differential amplifier configuration (see AN-31 at national.com). The feedback resistor has 0.3 pF (parasitic) in parallel with it. The two input resistors have 8.2pF/.023 Ohms ESR in parallel with them. Note that more than 8.2pF could/should be used with the bypassed mosfet-and-resistor test load, for better performance. But then the regulator would not be stable with a plain resistive 8-Ohm load. However, more than 8.2pF is required for stability with, for example, 4-inch power/gnd and remote sense leads to the load.
Some Zobel/snubber networks to gnd are used just after the pass transistor, for HF compensation.
There is NO capacitor across the output.
The main input + and gnd leads are 4 inches long, modeled with 100nH in series with 0.004 Ohms. There is then a 0.1 Ohm resistor in series at the + input, followed by some caps to gnd.
Note that ALL capacitors' ESRs are modeled, with a linear ESR versus frequency model for all electrolytics, with ESR settable for any particular frequency, which was set to 100 kHz for all of my tests.
The two opamps' power pins are fed through an LT1086 regulator (which feeds nothing else), with the regulator's input taken from just before the pass transistors (needed the headroom for using a standard transformer's 30VAC secondary with 1.5Vp-p ripple under "low line" conditions, so couldn't take it from after the pass transistor), with a small resistor and large-ish cap as a pre-filter for the LT1086, which also has 56uF from ADJ to gnd, and 330 uF and a snubber from OUT to gnd.
The reference's input is taken from the output of the differential remote sense amplifier, which seems to be the quietest signal available, on-board. There are beefy RC pre and post filters for the voltage reference.
I know that I could use an overall pre-regulator before the input. But I wanted to see how well it could do on its own, first.
I also have not yet tried it with buffers for the differential remote sense amp's inputs (or with a true instrumentation amp configuration, there), and have not tried it with a buffer for the reference voltage feed's pickoff.
To try to measure the output impedance, I connected an ideal 40.77VDC source to the input, and an ideal 1A (AC) current source across the output. I set the current source's DC "bias" value to 1A, at first, so that the 1A (2A p-p) AC would all be above 0A, since this is only a positive supply.
To get Zout, I plotted V(load_+)-V(load_gnd)/I and selected a linear scale, so that LT-Spice would give a vertical axis labeled in Ohms.
The plotted Zout was 115 uOhms at 10 milliHertz, 11 uOhms at 10 Hz, 9.5 uOhms from 100 Hz to 1 kHz, 20 uOhms at 10 kHz, 37 uOhms at 20 kHz, and 220 uOhms at 100 kHz. There are some nasty-looking peaks of about 3.8 kOhms at 464 MHz and 480 Mhz, then a dip to 1.5 Ohms at 668 MHz, followed by a steady rise to about 1.5 kOhms at 10 GHz. (I realize that the high-frequency stuff may not mean much, without every conductor's parasitics modeled, etc.)
For higher (than 1A) DC current source bias-current settings, the Zout improves, slightly. With a zero DC current bias setting, the Zout plot's shape is similar to the above, but starts at 25 mOhms, drops to 2 mOhms, and then rises to about 40 mOhms at 100 kHz. I don't know if I should use zero for the DC portion of the current source (or anything less than the AC amplitude), or not, since this is a positive-output-only voltage regulator.
To look at something like PSRR, I connected an ideal voltage source to the input and, at first, open-circuited the output. The input source was 40.77VDC with 1VAC swept-frequency. I plotted V(load_+) minus V(load_gnd).
The output level was -108 dB at 10 mHz, -135 dB from 10 Hz to 100 Hz, -142 dB at 1 kHz, reached a minimum of -149 dB at 5 kHz, then rose to -147 dB at 10 kHz, and -127 dB at 100 kHz. There was a -110 dB peak at 295 kHz, then a drop to -125 dB at 2.5 MHz. It stayed between -124 dB and -125 dB until it started to rise again at about 115 MHz, then formed a sharp -91 dB peak at about 480 MHz, and a similarly-shaped -168 dB trough at about 680 MHz, then continued from -132dB between 1.5 Ghz and 2.3 GHz, and was falling through -149 dB at 10 GHz.
Then I tried the same setup, but with an 8-Ohm resistive load, and things got a bit worse. At VLF the output was -94 dB. The minimum at 5 kHz was only -130 dB. AT 100 kHz it was -111 dB. There was a -71 dB peak at about 5.64 MHz (but that dropped to -94 dB with 10 pF instead of 8.2 pF, temporarily, in parallel with the diff amp's input resistors [The PCB design might be "fun".]). After 10 MHz, it was dropping at about 26 dB per decade, except for blips at 480/680 Mhz, and was falling through -186 dB at 10 GHz.
Just out of curiosity, I tried the same thing but with the active MOSFET/resistor/bypassed load in place. Plot shape was about the same, but with -104 dB at low f, -128 dB at 100 Hz, -141 dB minimum at 5 kHz, -122 dB at 100 kHz, -91 dB peak at 2.1 MHz, a small -111 dB peak at 16 MHz, then falling to -250 dB at 10 GHz (with some blips at 480/680 Mhz).
That's all I've got, so far.
P.S. Before anyone gets their hopes up, it looks like using a differential amp for the remote sense lines has already been patented.
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
The way I typically do remote sensing is the control ground is tied to the load return and the pos sense is tied to the load positive.
Isn't that the only right way to do it?
Isn't that the only right way to do it?
switchmodepower said:
Well it sounds right, to me, if you're connecting pos sense to load's positive pin, and gnd sense to load's gnd pin. I don't really know what you mean by "control ground". But I assume it means that it's an input or gnd for the remote sense processing circuitry of the regulator.
My experiments, described above, mainly have to do with what is done with the gnd remote sense lead, on the regulator end of it.
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
"how about a single sense wire on one side of the supply with a 2:1 correction ratio? volt drop on one leg must equal volt drop on the other leg if they are the same size."
I was wrong, of course, about that being theoretically correct.
Since the v(load_+) and v(load_gnd) voltages are varying in the same direction, with almost the same amplitude, the correct differential "rough estimate" correction factor should have been 0X, not 2X.
gootee said:
I was wrong, of course, about that being theoretically correct.
Since the v(load_+) and v(load_gnd) voltages are varying in the same direction, with almost the same amplitude, the correct differential "rough estimate" correction factor should have been 0X, not 2X.
OK. Having done all of that, I'm now back to something like the original question.
I have noted that if I lift the connection from the load ground sense line to the differential amplifier, and ground that input of the amplifier, then this thing behaves like a more-conventional regulator, and I get about the same specs for the voltage between the load + pin and main ground as I did above for the voltage between the load + pin and the load gnd pin.
I guess the original question was something like "Which way is more desirable?". And maybe that depends on the application, and on whether or not there are more circuits after the initial load, and whether voltage output or current output is needed for them, or something. (Just speculating, there, since I have no idea.). Maybe it's even _always_ better to try to hold the load + pin's voltage constant relative to the main ground (instead of relative to the load's gnd pin), as most conventional regulators with remote sensing seem to try to do (Please correct me if I'm wrong about them doing that.). I don't know.
But now there's a maybe-better question:
Can we have it both ways at once?
I guess that would involve trying to actively regulate the voltage at the load's ground pin, relative to the main ground, using remote sense lines from both.
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
I have noted that if I lift the connection from the load ground sense line to the differential amplifier, and ground that input of the amplifier, then this thing behaves like a more-conventional regulator, and I get about the same specs for the voltage between the load + pin and main ground as I did above for the voltage between the load + pin and the load gnd pin.
I guess the original question was something like "Which way is more desirable?". And maybe that depends on the application, and on whether or not there are more circuits after the initial load, and whether voltage output or current output is needed for them, or something. (Just speculating, there, since I have no idea.). Maybe it's even _always_ better to try to hold the load + pin's voltage constant relative to the main ground (instead of relative to the load's gnd pin), as most conventional regulators with remote sensing seem to try to do (Please correct me if I'm wrong about them doing that.). I don't know.
But now there's a maybe-better question:
Can we have it both ways at once?
I guess that would involve trying to actively regulate the voltage at the load's ground pin, relative to the main ground, using remote sense lines from both.
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
OK. I gave it a try.
The answer is "yes". We CAN regulate the "ground bounce", too.
My quick-and-dirty test implementation was a little costly, in terms of power dissipation, with the ground regulation circuit dissipating a little more power than the load, and almost as much power as the the original regulator and load combined. And, for convenience, I also assumed that a -30V supply was available.
But, with 1-inch power and remote sense leads, I was able to lower the load gnd pin's voltage variation from 1.5 mV p-p to 0 uV p-p (not counting transients), for the same 1.5 A p-p squarewave @ 30V load conditions as described in my previous posts. (I didn't try very hard. So I'm pretty sure that the ground regulation circuit could be improved, significantly, so that it might be able to be done with much lower additional power dissipation.) With 4-inch power/gnd and remote sense leads, it was back to a few hundred uV p-p load gnd voltage variation, but could be "tweaked back into submission" (at a cost of more power dissipation, with this particular circuit).
So now, with the load gnd's remote sense line NOT connected to the differential amp referenced in my previous posts, and that amp input grounded (so it's acting more like a conventional regulator), and using the same 1.5 A p-p squarewave load current (0.873A to 2.355 A) with 5 us edge times, the 30.1V at the load's + power supply pin varies by 11.3 uV p-p (for the two levels of the load-current squarewave), relative to the main ground, mostly flat like the squarewave but with 1.2 mV p-p maximum combined amplitude (separate pos + neg going total p-p) transients at the square wave edges. And the load's gnd pin voltage, relative to the main gnd, varies by 0 uV, with positive- and negative-going transients each at about 1.3 mV (2.6 mV p-p total).
That can be compared to the same regulator without the "ground regulation", which had the same 11.3 uV p-p variation of the load + pin voltage versus main ground, with 1 mV p-p transients, while the load's gnd pin was varying by 1.5 mV p-p (versus 0 uV p-p w/gnd reg), with positive- and negative-going transients of 5.8 mV each for 13.1 mV total p-p (versus 1.2 mV each for 2.4 mV total p-p w/gnd reg).
(Note that the 6.4 uV p-p load+ pin voltage variation figure in an earlier post was incorrect. It was actually 11.3 uV p-p. I somehow had two 40.77 DC sources in series on the input, when I got the 6.4 uV p-p figure.)
I tried a sort-of "brute force" approach for the ground regulator circuit, just to quickly see if anything could be done with it. There are probably much better ways to do it: I hung an AD825 opamp below ground, with its + pwr pin to gnd, and its - pwr pin to -30V. I configured the opamp as a very-high-gain differential amplifier with an NPN power booster, with 2.7 Ohms from the load gnd remote sense line to the opamp's + input and 2.7 Ohms from main gnd "reference" to opamp's - input, and 100k from the opamp's + input to -30V, and 100k from opamp's - input to the collector of an NPN power transistor (I used several parallel MJD44H11, since they were already used on the LT-Spice schematic, but can only dissipate 20W each, max, with this test case dissipating about 28W avg, total, in the transistor(s).). From the transistor's collector, there is a 6.177 Ohm 50W resistance (w/about 29W avg dissipation for this test) connected to the load's gnd pin. And there's a 1 Ohm 10W resistance (w/about 4.7W avg dissipation for this test) from the emitter to -30V. That's all there was to it. (Oops, I wonder if the AD825 can dissipate 400 mW. Looks like only about 250 mV/us slew rate is needed, for this test at least. So maybe some other amp chip, or discretes, would work better. Or maybe two amplification stages would be better.)
It looks like the entire "ground regulator" circuit was dissipating about 62 Watts average, during this test, while the mosfet/resistor load was dissipating about 49 Watts average. The 40.77V main input source had about 64 Watts average dissipation.
I ran the Zout vs frequency test again, but for the load + pin voltage divided by current (instead of load + pin volatge minus the load gnd pin voltage, since the load gnd's remote sense wasn't connected to the diff sense amp), and the Zout was almost the same as it was before, from 10 mHz to 20 kHz. But at 100 kHz it had dropped from 220 uOhms to 191 uOhms, and was a couple uOhms lower than before, for frequencies <= 20 kHz. Results for load + voltage minus load gnd voltage, with the load gnd remote sense re-connected to the differential amp, were very similar, but with about 200 uOhms at 100 kHz, instead of 191 uOhms.
I also checked input rejection at the output, versus frequency, as before, with the output open-circuited. Results were up to -0.5 dB better, below 100 kHz. At 100 kHz, it was -2 dB better. At the 295 kHz peak (now at 279 kHz), it was -5 dB better. Above that, it was about -6 dB better until at least 2 or 3 GHz, but was +3 dB worse at 10 GHz.
Oops. Note that I had changed the differential remote sense amplfier's resistors to 10k, from 9.1k, but still had 8.2pF in parallel with each of the input resistors.
Sorry to have blathered on for so long, again.
At any rate, it looks like it CAN be done. BUT, is it worth doing?
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
The answer is "yes". We CAN regulate the "ground bounce", too.
My quick-and-dirty test implementation was a little costly, in terms of power dissipation, with the ground regulation circuit dissipating a little more power than the load, and almost as much power as the the original regulator and load combined. And, for convenience, I also assumed that a -30V supply was available.
But, with 1-inch power and remote sense leads, I was able to lower the load gnd pin's voltage variation from 1.5 mV p-p to 0 uV p-p (not counting transients), for the same 1.5 A p-p squarewave @ 30V load conditions as described in my previous posts. (I didn't try very hard. So I'm pretty sure that the ground regulation circuit could be improved, significantly, so that it might be able to be done with much lower additional power dissipation.) With 4-inch power/gnd and remote sense leads, it was back to a few hundred uV p-p load gnd voltage variation, but could be "tweaked back into submission" (at a cost of more power dissipation, with this particular circuit).
So now, with the load gnd's remote sense line NOT connected to the differential amp referenced in my previous posts, and that amp input grounded (so it's acting more like a conventional regulator), and using the same 1.5 A p-p squarewave load current (0.873A to 2.355 A) with 5 us edge times, the 30.1V at the load's + power supply pin varies by 11.3 uV p-p (for the two levels of the load-current squarewave), relative to the main ground, mostly flat like the squarewave but with 1.2 mV p-p maximum combined amplitude (separate pos + neg going total p-p) transients at the square wave edges. And the load's gnd pin voltage, relative to the main gnd, varies by 0 uV, with positive- and negative-going transients each at about 1.3 mV (2.6 mV p-p total).
That can be compared to the same regulator without the "ground regulation", which had the same 11.3 uV p-p variation of the load + pin voltage versus main ground, with 1 mV p-p transients, while the load's gnd pin was varying by 1.5 mV p-p (versus 0 uV p-p w/gnd reg), with positive- and negative-going transients of 5.8 mV each for 13.1 mV total p-p (versus 1.2 mV each for 2.4 mV total p-p w/gnd reg).
(Note that the 6.4 uV p-p load+ pin voltage variation figure in an earlier post was incorrect. It was actually 11.3 uV p-p. I somehow had two 40.77 DC sources in series on the input, when I got the 6.4 uV p-p figure.)
I tried a sort-of "brute force" approach for the ground regulator circuit, just to quickly see if anything could be done with it. There are probably much better ways to do it: I hung an AD825 opamp below ground, with its + pwr pin to gnd, and its - pwr pin to -30V. I configured the opamp as a very-high-gain differential amplifier with an NPN power booster, with 2.7 Ohms from the load gnd remote sense line to the opamp's + input and 2.7 Ohms from main gnd "reference" to opamp's - input, and 100k from the opamp's + input to -30V, and 100k from opamp's - input to the collector of an NPN power transistor (I used several parallel MJD44H11, since they were already used on the LT-Spice schematic, but can only dissipate 20W each, max, with this test case dissipating about 28W avg, total, in the transistor(s).). From the transistor's collector, there is a 6.177 Ohm 50W resistance (w/about 29W avg dissipation for this test) connected to the load's gnd pin. And there's a 1 Ohm 10W resistance (w/about 4.7W avg dissipation for this test) from the emitter to -30V. That's all there was to it. (Oops, I wonder if the AD825 can dissipate 400 mW. Looks like only about 250 mV/us slew rate is needed, for this test at least. So maybe some other amp chip, or discretes, would work better. Or maybe two amplification stages would be better.)
It looks like the entire "ground regulator" circuit was dissipating about 62 Watts average, during this test, while the mosfet/resistor load was dissipating about 49 Watts average. The 40.77V main input source had about 64 Watts average dissipation.
I ran the Zout vs frequency test again, but for the load + pin voltage divided by current (instead of load + pin volatge minus the load gnd pin voltage, since the load gnd's remote sense wasn't connected to the diff sense amp), and the Zout was almost the same as it was before, from 10 mHz to 20 kHz. But at 100 kHz it had dropped from 220 uOhms to 191 uOhms, and was a couple uOhms lower than before, for frequencies <= 20 kHz. Results for load + voltage minus load gnd voltage, with the load gnd remote sense re-connected to the differential amp, were very similar, but with about 200 uOhms at 100 kHz, instead of 191 uOhms.
I also checked input rejection at the output, versus frequency, as before, with the output open-circuited. Results were up to -0.5 dB better, below 100 kHz. At 100 kHz, it was -2 dB better. At the 295 kHz peak (now at 279 kHz), it was -5 dB better. Above that, it was about -6 dB better until at least 2 or 3 GHz, but was +3 dB worse at 10 GHz.
Oops. Note that I had changed the differential remote sense amplfier's resistors to 10k, from 9.1k, but still had 8.2pF in parallel with each of the input resistors.
Sorry to have blathered on for so long, again.
At any rate, it looks like it CAN be done. BUT, is it worth doing?
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
Just to wrap this up:
With 1-inch power and remote sense leads, and the mosfet/resistor load pulling a 1.5-Amp P-P 20 kHz sine wave from the supply, the nominal (approx) 30.09V load_+ pin voltage has about 50 uV p-p of 20 kHz sine in it.
Unfortunately, even with the "ground regulation" circuit in place, the load_gnd pin voltage has almost 800 uV of the 20 kHz sine in it.
With the same load conditions, but without the "ground regulation" circuit in place, the load_gnd pin voltage has about 3.7 mV p-p of the 20 kHz sine in it, and the 30.09 load_+ pin voltage has about 37.5 uV of 20 kHz sine.
I guess a different ground-reg circuit topology might work better, or else more speed and/or power would be needed for it. (Of course, the load bypassing might be able to have a big effect. And I didn't try changing that, for this case.)
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
With 1-inch power and remote sense leads, and the mosfet/resistor load pulling a 1.5-Amp P-P 20 kHz sine wave from the supply, the nominal (approx) 30.09V load_+ pin voltage has about 50 uV p-p of 20 kHz sine in it.
Unfortunately, even with the "ground regulation" circuit in place, the load_gnd pin voltage has almost 800 uV of the 20 kHz sine in it.
With the same load conditions, but without the "ground regulation" circuit in place, the load_gnd pin voltage has about 3.7 mV p-p of the 20 kHz sine in it, and the 30.09 load_+ pin voltage has about 37.5 uV of 20 kHz sine.
I guess a different ground-reg circuit topology might work better, or else more speed and/or power would be needed for it. (Of course, the load bypassing might be able to have a big effect. And I didn't try changing that, for this case.)
- Tom Gootee
http://www.fullnet.com/~tomg/index.html
- Status
- Not open for further replies.