Enough current to turn on 3 mosfets?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hmm..I'm not understanding properly some of these basic op-amp terms.

With the assistance of AN-581 I got to the schematic below. Three problems. Determining C5 - there seems to be a minimum amount needed above which it has little impact (at least in the time domain) until too large. Second, I'm still not getting the right output, even from the first op-amp. 35x gain (1+R1/R2) for a 10us pulse yields 320uV (on a 6V bias) rather than 350uV. Three, weird looking waveforms - I guess this is the slew rate of the op-amp coming to bear.

Your hint points to not setting Vcm = input voltage range = 0. I can only guess, judging mostly by your style :) that setting the bias at exactly the mid-point of Vs means Vcm = 0. However, even adjusting the bias away from 6V, I still get the error I mention above.

Perhaps I have just been staring at this for far too long today...
 

Attachments

  • Op amp biasing.JPG
    Op amp biasing.JPG
    243.6 KB · Views: 89
Member
Joined 2011
Paid Member
No, Vcm is the Common Mode Voltage, the midpoint around which the opamp inputs fluctuate: Vcm = [(Vin+) + (Vin-)]/2 . Also known as the Bias Point.

Linear Technology tells you there is a huge penalty (increased noise) if you set Vcm to zero volts. So don't set it to zero volts.

Instead, set it to something about halfway between the top supply rail and the bottom supply rail. That way you'll have maximum dynamic range before clipping, and maximum headroom before clipping. You'll also get lower noise from the LT1677.

A standard design error that people make all the time, is to copy an amplifier circuit design from bipolar-power-supplies, to a single-ended-power-supply, without making any other modifications. They forget that "ground" is no longer midway between the power supplies, and so biasing the input and output of an amplifier at "ground" is a terrible idea when there is only a single supply. Eventually it dawns on these folks that "ground" in the original circuit, could have / should have been labeled "Vbias" instead. Then, aha, eureka, the light comes on: disconnect Vbias from ground and connect it to midway-between-the-supplies! All is well once again.
 
ok but isn't that what I did above - set the bias to mid point between 12V and common? i.e. 6V (I realise I made the mistake initially and that, in hindsight, the error was obvious.) If I now have it correct, I don't understand why I have the amplification error


PS: if indeed I did have dual supplies, wouldn't "instead, set it to something about halfway between the top supply rail and the bottom supply rail" clash with "don't set it to zero volts"? Top supply rail =+12V, bottom -12V, halfway = 0V?
 
Last edited:
Well I wish it were a long weekend here. Unfortunately for us our Late Summer holiday weekend has been and gone - next stop is Christmas which seems very distant.

I'm at a loss. AN-581 deals with this exact topic. I don't see where I went wrong. For now just focusing on the first gain section:

- I have a non-inverting amplifier
- I have a single supply of 12V
- "the fundamental problem is the op amp is a dual supply device and so some type of biasing, using external components, must be used to centre the op amp's output at mid supply"
- simple solution is a voltage divider (R8 and R3) and C5 is needed for stability CHECK
- problems: loss of PSRR and instability
- solution: decouple the biasing network from the supply. Add C6 for AC decoupling, R4 for dc return path CHECK
- follow guidance for size of R8, R3, R4, C5 etc

A problem for the LT1677 in the circuit above is the sizes of R1 and R2. Drop them to 15K and 440 and things look better. If R8=R3=R4, then - following the last note to Fig 3 in the AN - they can fall to 22k. Increasing C5 helps further. I can see why AN-581 says optimising everything "can become quite involved."

Attached below is a revised circuit, just focusing on the first amplification stage for now. (I've not looked at it in the frequency domain yet.) Maybe there's an easier way. If so, I don't see it.
 

Attachments

  • Op amp modelling.jpg
    Op amp modelling.jpg
    161.4 KB · Views: 87
The input impedance is 22k + {27k+27k}/2 = 35k5
That is quite low
The input cap @ 22uF is a bit low for a measuring instrument.
The decoupling @ 3uF is very low for a measuring instrument.
How much current is required to match the input offset current of the 1677?
Use that as a guide to the values for r3, r4 & r8
 
Thanks Andrew. I didn't get to spend any time on this since Saturday afternoon. I hadn't focused too hard on optimising the above elements because Mark's comments led me to believe that this circuit construction is completely wrong rather than simply in need of tweaking...
 
Okay, on the presumption that "no, no, and no" related less to the way the first op amp was biased and much more so to the way the second op amp was together with the way they were coupled (all very sloppy), I present another schematic for constructive criticism. Each op amp is biased at 6V (half way between my single 12V supply and common) and the "AC" output of the first is coupled to the input of the second in the same was as the AC of the measured rail is coupled into the first. This circuit has 35x gain for the second op amp (same as first); perhaps this should be reduced to the 20x discussed earlier?

Andrew, with respect to the resistor and cap levels, R1 (R6) seem critical to the stability and responsiveness (transient shape) of this op amp (with R2 and R7 are obviously dependent on them for a given gain). I used the guidance in AN-581 to set the values of R8, R3 and R4 (plus R11). C4 and R4 mean a pole at 1.4Hz (same with C9 and R11). I set C6 large enough to place the pole it creates (with R8 and R3) well lower. C4 at 22uF was originally chosen because I have some Panasonic SEPC 22uF caps already. However, 4 of them alongside 3 220uF is a lot of board space. Perhaps things can be optimised better but I would like to first get a pass mark for the circuit structure. :)

Overall, the circuit as currently configured has modelled gan of 61.8dB with -3dB points of 3.7Hz and 75kHz. The attached shows the circuit and transient waveform.
 

Attachments

  • Measurement amplifier ver2.JPG
    Measurement amplifier ver2.JPG
    285.2 KB · Views: 72
What is R5 doing?
Similarly what is R12 doing?

C3 and C7 seem very high, does the datasheet give guidance? Is this the 75kHz limit?
That seems far too low for measuring noise performance. Wider with a defined passband when that is required should be OK and maybe better.

Both opamps see ~15k on their two input pins. This should reduce input offset. Does the datasheet guide you on this requirement? Is it needed?

If you don't need 60dB of gain, then reducing the second stage gain would reduce noise.
 
Last edited:
R5/R12 - indeed seem redundant (I was thinking low pass filter but clearly all that's need is a short for DC)

C3 and C7 - guidance from datasheet is for R1 (R6) "greater than 2k, a pole will be created with [R1] and the amplifier's input capacitance, creating additional phase shift and reducing the phase margin. A small capacitor (20pF to 50pF) in parallel with [R1] will eliminate this problem." Will drop to 20pF - this helped (some is needed though).

This should reduce input offset. Does the datasheet guide you on this requirement? Is it needed?
I don't follow you here. Very new to op amps...

Will reduce it to 20x second stage.

Thx
 
Last edited:
Member
Joined 2011
Paid Member
I think if you make a rough, piecewise linear, sketch of the frequency response (a "Bode Plot") of that circuit, you'll probably get something like the attached. The three capacitors give rise to three poles and two zeroes (!), and that neglects the opamp's dominant pole. Ugh.

Maybe the circuit would be easier to analyze if C3 and C5 (figure below) were eliminated. Then the frequency response becomes drastically simpler: a single pole (due to C4). And then of course the opamp's dominant pole, which I cheerfully neglected when drawing this Bode Plot.

Speaking of the opamp's dominant pole, Walt Jung selected a GBW=63 MHz opamp (OP37) to build a gain=50 first stage. If you're going to use a GBW=7.2 MHz opamp (LT1677) to build a first stage, what sort of gain do you think is appropriate?

A final lagniappe to chew on: what is the output impedance of the circuit that drives this amplifier's input? What ratio of (Zin_amplifier / Zout_source) do you think is appropriate?
 

Attachments

  • together.png
    together.png
    66.2 KB · Views: 92
I was just reading about gain bandwidth product. I understand the quoted figure is the frequency at which gain falls to 1 and that "bandwidth" decreases as we add gain. Dropping the second stage to 20x already expands the upper 3dB point to 150kHz (gain 1st stage = 35x, 2nd stage = 20x). I don't have a reference point for what bandwidth is required and hence how much gain can be extracted from a pair of 7.2MHz unity gain bandwidth op amps. Obviously with a lower GBW device the figure will be a lot less but I need something to solve for and this I am lacking...

I need to think about your last point some more but presumably the circuit driving this one is the to-be-measured voltage rail with the output impedance I've previously modeled (for the 5V anyway) plus some wiring resistance. I understand that, in general, one wants a driving source to see an input impedance which is a good multiple of its output impedance. I need to think about the input impedance of this amplifier but the output impedance of the modeled 5V rail was only a few milli ohms in the high frequencies.
 
Member
Joined 2011
Paid Member
... the very first stage should have extremely low noise and operate at high gain. That way, the noise of subsequent amplifying stages will be negligibly small -- it's effectively attenuated by the gain of the first stage. There is another benefit: RMS addition of uncorrelated noise sources diminishes the second stage's contribution even further.

The references below, particularly the blue-has-changed-to-purple one, flesh out the computations for this. Study the paragraph that includes
" noise voltages V1, V2 & V3 add to give a result of
An externally hosted image should be here but it was not working when we last tested it.
. ... Thus, if noise voltage V1 is more than 3 to 5 times V2 or V3, it is dominant, and V2 & V3 may generally be ignored, which simplifies noise assessment."
So a cascade of four amplifiers, each of whose gain is 5.63X **, has negligible noise contribution from stages 2-4. Similarly a cascade of three amplifiers, each of whose gain is 10.0X **, has negligible noise contribution from stages 2-3. You are free to trade gain for bandwidth and vice versa, as long as (1st stage gain x first stage ein) is > 5x (second stage ein). In the case where the first stage and second stage are built with identical opamps, ein1=ein2, making the math even simpler.

As for input impedance: Carefully note that noise-on-supply is produced by a circuit whose output impedance is the supply impedance ... i.e., far far less than 1 ohm. For the level of measurement precision you seek here, any amplifier input impedance greater than about 50 ohms will be more than adequate.

**_ Substitute the actual total gain requirement, instead of the phony-baloney placeholder "1000X" used here!
 

Attachments

  • rms_benefit.png
    rms_benefit.png
    79 KB · Views: 79
Last edited:
Okey dokey. I see the attached trade-offs between gain/resolution and bandwidth. My scope goes down to a vertical resolution of 1mV per division and I have calculated the uV/div for each gain pair also. I still don't have a feel for what constitutes adequate bandwidth.

Regarding impedance, I just want to check one thing. You say "is the supply impedance". I thought the relevant output impedance is of the circuit providing the voltage being measured (12V, 5V or 3V3). (I recognise that in one of these cases it will be the same as the supply for the measurement circuit.) In any event, each of these circuits will have very low output impedances. Am I right to say the driving circuit will see a load of C4 in parallel with R4 which for all but very low frequencies looks like R4 i.e. 10k ohms. The relative impedance test is very easily met.
 

Attachments

  • Measurement amplifier resolution.JPG
    Measurement amplifier resolution.JPG
    44.3 KB · Views: 70
Member
Joined 2011
Paid Member
The output impedance of a 3.3V power supply is: that supply's impedance. Far far less than 1 ohm.
The output impedance of a 5V power supply is: that supply's impedance. Far far less than 1 ohm.
The output impedance of a 12V power supply is: that supply's impedance. Far far less than 1 ohm.

If you want your measurement apparatus to change the voltage being measured by less than X%, then the input impedance of your measurement apparatus needs to be

  • (Zin_apparatus / Zout_DUT) > [ (100% - X%) / X% ]
by straightforward circuit analysis.

Plugging in an example X = 0.639%, we find Zin_apparatus > Zout_DUT x 155. So if your 12V power supply's output impedance is 0.28 ohms, Zin_apparatus needs to exceed 43.5 ohms, in order to have less than an 0.639% effect upon the measured value.

You will probably want to tinker around with various possible values of Zout_DUT, and various choices of the permissible measurement error (X%), to see what kinds of impedance ratios (Zin_apparatus / Zout_DUT) you require.

Remember that you propose to "measure" noise by eyeballing the fuzz displayed on an oscilloscope. How precise do you think these "measurements" can be? Don't expect racehorse-quality results from a frog. Don't polish the wrong (excremental word).

The input impedance of your amplifier circuit can be extracted from simulation results, just as the output impedance of your power supply can be extracted. You simply apply an AC input current of 1.0 units, and measure the resulting AC voltage at the input. Ohms Law tells you that Vin = Iin x Zin, thus Zin = (Vin / Iin). You will find that when you ask LTSPICE to plot the ratio (V/I), it automatically displays the data using the units "Ohms". LTSPICE is aware of Ohm's Law. You will see that the input impedance varies with frequency, which should not be a surprise.

At extremely low frequencies, all capacitors are effectively open-circuits (ignoring DC leakage). At extremely high frequencies, all capacitors are effectively short-circuits (ignoring ESR). In between, capacitors smoothly vary between behaving as open-circuits and behaving as short-circuits. Thus circuit response varies with frequency. As your intuition predicts.

The Zin-versus-frequency plot will probably include a broad zone of frequencies where Zin is more or less constant. The magnitude of impedance in the center of this zone (called "the mid-band") is the number that people usually mean when they say "the input impedance is AAAA ohms".
 
The input impedance of your amplifier circuit can be extracted from simulation results

This I did earlier. It matched my expectation from just looking at the circuit schematic but I wanted to check this wasn't merely coincidental.

Unless I am missing something, the input impedance of the measurement amplifier (10k ohms) dwarfs the output impedance of each of the power supply rails (less than a few 10s of milliohms) by a massive margin. Job done - not a significant factor in measurement accuracy.

So I am just left with a bandwidth / amplification choice...

PS: gut instinct suggests 170kHz upper limit is more than enough but this conclusion is only drawn from observing charts on-line that tend to stop at 100kHz.
 

Attachments

  • Measurement amplifier input impedance.JPG
    Measurement amplifier input impedance.JPG
    295.3 KB · Views: 67
Last edited:
Member
Joined 2011
Paid Member
The thing being powered is a box full of digital integrated circuits? Over what range of frequencies is it sensitive to power supply noise?

Presumably your new power supply is designed to produce especially low noise over that range of frequencies.

I assume one of your testing goals is to check whether the supply does indeed produce its design-target value of output noise, over that range of sensitive frequencies.
 
That would be at the right way to approach the subject but I don't have answers to those questions. I started this project to put together the best power supply I could and learn a lot along the way. I'm certainly doing the latter with your great assistance. Perhaps I am chasing fool's gold on the former.

The addition of a measurement amplifier to the board came with two criterion (post 175): not too much cost in board real estate and not too much mental contortion. I'm likely reaching my limit on the latter. I suspect noise to far higher frequencies is of importance. However, I'm not sure that this configuration, constrained in part by a single supply and hence less gain bandwidth product at hand, provides a huge platform for choice. I could add another gain stage (7 or so extra components) but that sounds to me like "polishing a coprolite" to coin your phrase. If measurement at higher frequencies is required I suspect a separate amplifier module would be more appropriate. Lastly, I'm not sure what more, in practical terms, 270kHz covers that 170kHz doesn't.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.