Thanks - makes sense that they are available!
So Rgate forms a low pass filter with MOSFET's input capacitance of 10nF. The pole of this filter is given by:
f-3db = 1/(2 x Pi x CISS x Rgate)
Where is a sensible starting point for the pole? Without the gate resistor AolB hits 0dB at circa 977kHz (phase margin of 80 degrees). 100R => a pole at circa 159kHz. 1MHz => Rgate closer to 15.9 Ohms. Rgate > 0 shifts AolB=0 lower in frequency. It can also cause phase to roll off more sharply thereby compromising phase margin. (Rgate = 100R implies AolB=0dB at 416kHz and phase margin collapses to 20 degrees.)
Varying Rgate and watching phase margin suggests the maximum Rgate can be is somewhere between 15 and 20 Ohms. That of course assumes Infineon have modeled parasitic inductance and capacitance properly.
It would seem to me that the best assumption is 0 Ohms but leave a space just in case however it is noteworthy that adding Rgate reduces modeled stability (as determined by predicted phase margin) rather than increasing it.
So Rgate forms a low pass filter with MOSFET's input capacitance of 10nF. The pole of this filter is given by:
f-3db = 1/(2 x Pi x CISS x Rgate)
Where is a sensible starting point for the pole? Without the gate resistor AolB hits 0dB at circa 977kHz (phase margin of 80 degrees). 100R => a pole at circa 159kHz. 1MHz => Rgate closer to 15.9 Ohms. Rgate > 0 shifts AolB=0 lower in frequency. It can also cause phase to roll off more sharply thereby compromising phase margin. (Rgate = 100R implies AolB=0dB at 416kHz and phase margin collapses to 20 degrees.)
Varying Rgate and watching phase margin suggests the maximum Rgate can be is somewhere between 15 and 20 Ohms. That of course assumes Infineon have modeled parasitic inductance and capacitance properly.
It would seem to me that the best assumption is 0 Ohms but leave a space just in case however it is noteworthy that adding Rgate reduces modeled stability (as determined by predicted phase margin) rather than increasing it.
Last edited:
But the point I am making is that they claim in one part this is an 88W device and in another part that it is a transient one shot device with a 95W limit to a maximum of 10ms.
That is the anomaly.
10ms value is almost equal to the claimed DC value.
In my blinkered view it is not an 88W device, maybe more like a 10W to 20W device, but that just a guess because they don't tell you/me.
That is the anomaly.
10ms value is almost equal to the claimed DC value.
In my blinkered view it is not an 88W device, maybe more like a 10W to 20W device, but that just a guess because they don't tell you/me.
This gate stopper is still bugging me. It seems like it could easily be a "death spiral" along the lines of:
- start with 0 Ohms
- notice instability
- add gate stopper
- instability gets worse
- unclear if gate stopper not big enough or C1 needs to be increased
- bigger gate stopper makes things worse
Anyway...I guess one has to try. Phase margin can be helped a little by a resistor in series with C1 but that little trick only goes so far. (see post 115 for circuit) Perhaps it makes sense to leave a spot on the PCB for this additional resistor as well.
I have a question regarding SMD resistors as I have only purchased axial before. I see than thick film seem to have a broader range of Ohms but less accuracy. My presumption is that 1% is good enough for all here with perhaps R6 and R7 being the only exceptions. So thick film 1% in 1206 case (e.g. ERJ-8ENF series from Panasonic) generally okay?
(I've checked dissipation and all are well under 100mW, if not 50mW, and so I'm just looking at 1/4W rated resistors.)
- start with 0 Ohms
- notice instability
- add gate stopper
- instability gets worse
- unclear if gate stopper not big enough or C1 needs to be increased
- bigger gate stopper makes things worse
Anyway...I guess one has to try. Phase margin can be helped a little by a resistor in series with C1 but that little trick only goes so far. (see post 115 for circuit) Perhaps it makes sense to leave a spot on the PCB for this additional resistor as well.
I have a question regarding SMD resistors as I have only purchased axial before. I see than thick film seem to have a broader range of Ohms but less accuracy. My presumption is that 1% is good enough for all here with perhaps R6 and R7 being the only exceptions. So thick film 1% in 1206 case (e.g. ERJ-8ENF series from Panasonic) generally okay?
(I've checked dissipation and all are well under 100mW, if not 50mW, and so I'm just looking at 1/4W rated resistors.)
Last edited:
Continuous dissipation around 25% to 50% of max does no harm.
Peak dissipation getting close to 100% of max should be considered your limit.
Metal oxide generally accept more abuse than metal film, but even then a 100% overload on Pmax is really only a 41% overload on current !
Metal oxide do have a bit more noise than metal film, but in most audio this is unlikely to be measurable and certainly not audible.
However there are a few situations where low noise and low tempco become more important (NFB resistors) and probably some others, where audibility may become an issue. These few should use metal/thin film and preferably the better end.
Peak dissipation getting close to 100% of max should be considered your limit.
Metal oxide generally accept more abuse than metal film, but even then a 100% overload on Pmax is really only a 41% overload on current !
Metal oxide do have a bit more noise than metal film, but in most audio this is unlikely to be measurable and certainly not audible.
However there are a few situations where low noise and low tempco become more important (NFB resistors) and probably some others, where audibility may become an issue. These few should use metal/thin film and preferably the better end.
Last edited:
AD797 Bypassing and Inverting Configuration
I presume I should add bypassing caps for Vs of the AD797 (see attached). The datasheet also provides guidance in the section regarding "The Inverting Configuration" which suggests keeping the feedback resistors low and adding a small capacitor across the 'uppermost' feedback resistor. Am I right to add these components to this application?
I presume I should add bypassing caps for Vs of the AD797 (see attached). The datasheet also provides guidance in the section regarding "The Inverting Configuration" which suggests keeping the feedback resistors low and adding a small capacitor across the 'uppermost' feedback resistor. Am I right to add these components to this application?
Attachments
Virtually every amplifier needs decoupling and many show very significant deterioration in performance if the decoupling is poorly implemented.
Have you read Leach's Lo Tim?
He got it wrong and is humble enough to take you through the story of how to do it properly, although a lot of the detail has been omitted.
Decoupling of the power pins of fast opamps is mandatory where the manufacturer mentions it in the datasheet.
Be careful with Cl it can make stability worse.
Have you read Leach's Lo Tim?
He got it wrong and is humble enough to take you through the story of how to do it properly, although a lot of the detail has been omitted.
Decoupling of the power pins of fast opamps is mandatory where the manufacturer mentions it in the datasheet.
Be careful with Cl it can make stability worse.
Thanks. Yes, large CL is a no-go but a cap of the size proposed here (10-20pF) should be fine I think. It doesn't change phase margin materially at all.
If your model is OK and the prediction also says that small pF is OK, then go ahead.
But If I knew how to test the prototype, then I would put all my eggs into the prototype testing basket, not into the sim.
There is too much missing from the model when it comes to parasitic capacitances and parasitic inductances.
But If I knew how to test the prototype, then I would put all my eggs into the prototype testing basket, not into the sim.
There is too much missing from the model when it comes to parasitic capacitances and parasitic inductances.
Whoops, you forgot that when viewed from the gate, the MOSFET is acting as a source follower, an amplifier whose gain from input (gate) to output (source) equals +G, a little less than +1.0. The gate-to-source capacitance Ciss is *bootstrapped* by this amplifier and so its effect upon input capacitance is reduced. (Cgd and Cgb are not bootstrapped.)
The standard analysis procedure is shown in the attached diagram. You squirt a small signal test current into the input and measure the ratio (v/i) which equals input impedance. Comparing the input impedance of the left circuit versus the input impedance of the right circuit, you can calculate the ratio and obtain an effective input capacitance. It varies as the amplifier gain, G. {the circuit on the left is in fact the degenerate case G=0.0000}
Of course the exact same bootstrapping phenomenon occurs in the BJT emitter follower version of a series pass regulator, too.
_
Attachments
Last edited:
I recommend you purchase a few of each of these four opamp types: OP07, TL071, AD817, AD825. Each of them are pin compatible with the AD797 if you leave the weird tweakus maximus pins unconnected. If your actual PCB hardware in your actual application turns out to be unstable or marginally stable, you can do a quick opamp-swap using parts on hand to see whether the problem gets worse or gets better.
OP07 has enormous DC gain, so it will provide the gigantic attenuation of 100Hz ripple that you seek. It also has very modest GBW (1 MHz) which makes frequency compensation a lot more forgiving.
AD817 also has enormous DC gain and relatively high GBW; its cool feature is its spectacularly large phase margin (80 degrees!) at unity gain crossover. This gives your final regulator circuit an extra 20 to 35 degrees of phase margin, for free.
AD825, on the other hand, has quite modest DC gain (only 76 dB), but higher GBW and excellent if not spectacular phase margin (75 degrees).
TL071 is a turkey but you seem to like it for some reason.
You could make it a design goal to have your final circuit exhibit unconditional stability into all loads, with the AD797 spice model AND also with "N" of these other four opamps' spice models. You choose a value of "N" which offers the best tradeoff between design effort and peace of mind.
OP07 has enormous DC gain, so it will provide the gigantic attenuation of 100Hz ripple that you seek. It also has very modest GBW (1 MHz) which makes frequency compensation a lot more forgiving.
AD817 also has enormous DC gain and relatively high GBW; its cool feature is its spectacularly large phase margin (80 degrees!) at unity gain crossover. This gives your final regulator circuit an extra 20 to 35 degrees of phase margin, for free.
AD825, on the other hand, has quite modest DC gain (only 76 dB), but higher GBW and excellent if not spectacular phase margin (75 degrees).
TL071 is a turkey but you seem to like it for some reason.
You could make it a design goal to have your final circuit exhibit unconditional stability into all loads, with the AD797 spice model AND also with "N" of these other four opamps' spice models. You choose a value of "N" which offers the best tradeoff between design effort and peace of mind.
LOL! It merely so happens that I have a regulator which uses this op amp. That's all.
Sensible advice (although I'm not so sure that it is easy to swap out the op amps without a rework station).
For SO-8 packaged ICs, use a Dremel Tool (definition) or equivalent, with a cut-off wheel installed, to cut the epoxy IC body away, leaving only the IC legs soldered to the board. Use a wide tip + rosin-impregnated desoldering braid, to wick most of the solder away from pin1. Then use the soldering iron and tweezers to remove and discard pin 1. Repeat for the other seven pins.
Or you could lay out the PCBoard with both a DIP socket and a SMD footprint. Perhaps overlapping, perhaps not. Start your debug & qual procedures with the DIP socket and find the chip(s) you like best. Then remove the socket and solder the chosen IC onto the SMD footprint.
Or you could lay out the PCBoard with both a DIP socket and a SMD footprint. Perhaps overlapping, perhaps not. Start your debug & qual procedures with the DIP socket and find the chip(s) you like best. Then remove the socket and solder the chosen IC onto the SMD footprint.
I took a look at the AD825 and AD817A. Neither come close to the AD797 re PSRR. More importantly both suffer from the same issue regarding a gate stopper: more than 25 Ohms and phase margin is demolished (less than 45 degrees). I will try to take a look at the OP07 either later this evening or tomorrow afternoon. (All with the IPP037N06L3.)
Perhaps it is about time to decide how much input ripple attenuation is good enough.
If you use an IC preregulator (post #84) it will give at least 60dB of attenuation at 100Hz. If you install series resistors between your smoothing caps, that will give another ~3dB of input ripple attenuation at 100Hz.
If your opamp + series pass transistor circuit gives another 63dB of attenuation at 100Hz, that's a total of 126dB. 0.5 parts per million. How much ripple attenuation is good enough?
If you use an IC preregulator (post #84) it will give at least 60dB of attenuation at 100Hz. If you install series resistors between your smoothing caps, that will give another ~3dB of input ripple attenuation at 100Hz.
If your opamp + series pass transistor circuit gives another 63dB of attenuation at 100Hz, that's a total of 126dB. 0.5 parts per million. How much ripple attenuation is good enough?
And if you heavily bypass the ADJ pin, it seems to give 81dB of attenuation at 100Hz in simulation (12V, 5A).
Now you only need 45dB attenuation in the opamp + series pass xitor, to get 126dB total.
Or I suppose you could simply put two of these IC regulators in series and get 162dB of attenuation at 100Hz. That's 8 parts per billion.
_
Attachments
Anything you do to attenuate the AC signal ("ripple") from the LT1084's output pin, to the ADJ pin, will help. Of course the benefits don't go on forever; simulations suggest that there's very little to be gained by attenuating more than -20 dB or so. According to LinearTechnology's spicemodel anyway. LTSPICE results are presented below; notice the tiny improvement when we go from -14dB attenuation, to -80dB attenuation.
Remember that the final circuit will include input-short-protection diodes which give a ripple injection pathway from IN to ADJ. Thus the lower the impedance at ADJ, the less ripple will be injected here. Running a DC bias of 10mA (1.25V / 120 ohms) seemed a reasonable way to get a headstart on rejecting this injection. It also reduces the voltage error that results from the ADJ pin's ~ 70uA bias current. Since the final application is a 5 ampere regulator, 10mA seemed acceptable.
On the other hand, 470uF SMD capacitors appear to cost USD 0.94 in qty=10 (link 1) , while 2200uF SMD capacitors cost almost 3X more at USD 2.72 in qty=10 (link 2). So maybe a pinchpenny would want to get the same lowpass filter RC timeconstant but using a bigger R and a smaller C. Me, I'd say "I've got the extra two dollars to spend on my hobby."
Consult the LT1084 datasheet; it warns that the bigger capacitor you attach to ADJ, the bigger capacitor you need to attach to OUT. Caveat emptor.
_
Remember that the final circuit will include input-short-protection diodes which give a ripple injection pathway from IN to ADJ. Thus the lower the impedance at ADJ, the less ripple will be injected here. Running a DC bias of 10mA (1.25V / 120 ohms) seemed a reasonable way to get a headstart on rejecting this injection. It also reduces the voltage error that results from the ADJ pin's ~ 70uA bias current. Since the final application is a 5 ampere regulator, 10mA seemed acceptable.
On the other hand, 470uF SMD capacitors appear to cost USD 0.94 in qty=10 (link 1) , while 2200uF SMD capacitors cost almost 3X more at USD 2.72 in qty=10 (link 2). So maybe a pinchpenny would want to get the same lowpass filter RC timeconstant but using a bigger R and a smaller C. Me, I'd say "I've got the extra two dollars to spend on my hobby."
Consult the LT1084 datasheet; it warns that the bigger capacitor you attach to ADJ, the bigger capacitor you need to attach to OUT. Caveat emptor.
_
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.