another fast-ramp current source thing

This thread has gone on so long, I forget the raison d'etre for the design. But having done video dacs, it is often best not to disturb the current source. Thus I would suggest a current steering scheme where you either send the current to a sink (ground) or the cap. This all depends on the kind of precision you are trying to achieve.

The current steering elements also provide a cascode to raise the current source impedance.

That's much better, but not quite there yet. haha .

Isn't that what Q2 does?

the

you

depends

No. I think Q2 is for restoring C1 to zero voltage. It's a switch, I think. Of course, I'm just a hobbyist.

Jon

Yeah, it needs at least 2nd order curvature correction to make up for all the nonlinear capacitances in the rest of the semiconductors.

Probably will - without looking up the specs on the P-channel FET I'm assuming it's much higher capacitance than a good bipolar. If the devices don't magically complement/cancel nonlinearity that could get ugly, though I assume you're still after "only" a few % linearity.

What I've done is to use a bipolar as the current output transistor. However rather than regulating the gate/base voltage with the opamp, configure the opamp as a (Howland-style) current source directly feeding the emitter. Now take an additional input resistor from the Howland opamp's inputs to the two terminals of the output transistor's base resistor and (with proper scaling) you can get rid of the base current error term. The CB loading of the Howland configuration provides a nice low impedance load so that you don't need to tweak the resistor network for Howland stability (assuming its designed properly). The differential configuration will also enable using a ground-referenced DAC so you can tweak the scale factor since your load capacitance is not that well controlled.

Perhaps Phil Hobbs' capacitor configuration would help should your PNP RF transistor oscillate. My application (and PNP transistor) was slower, it didn't need that. But with the configuration I am suggesting, low beta becomes much less of a problem (though low Early voltage could be).

-Frank

I see your confusion, mostly because my explanation isn't much of an explanation. In a video DAC, at least in the ECL designs, and some MOS designs, you have a weighted current sources. Each current source is fed to a differential pair (long tail pair) such that one transistor is on at a time. The current is either delivered to the load (resistor for a DAC) or shunted to ground.

Hard to believe this got published in 2002 since the technique was quite old by then. Anyway, a schematic is worth a thousand words. In this case, they fed the current to another resistor, but the idea is the same.

The thermometer code is very typical in video DAC. However, you eventually need to switch in a binary weighted current source, and that is where you get a big glitch. You could see where those video DACs would do the switch to a binary segment back in the bad old CRT monitor days.

It is more the clamp diode that diverts the unneeded current.

Without dimensions for C2 R3 it is hard to say if that makes a bootstrap or just adds a time constant to a integrator that is at least pure by concept.

Note that the AD8009 is not "unity gain stable enough" that they dare to brag about it on page 1 of the data sheet. The Bode plot for G=+1 has ugly overshoot depending on package even w/o capacitive load and C2 is just that.

Gerhard

C2 bootstraps the nonlinear drain capacitance of Q1. The 5LP01C is a remarkable little mosfet; it's only about 5 pF to start with. C2 keeps the voltage across R3 constant during the ramp. C2+R3 is, in fact, an AC bootstrap current source. Q1 only disciplines the average current long-term. Well, I thought it was cute.

C2 doesn't matter; it's just any old big bypass value that's handy,