linear ramp

Very nice...everyone loves a bootstrap.

Yeah, this is 1930's technology. I've seen this done with VR glow tubes, neon lamps, zeners, diode-clamped capacitors, all sorts of ways. MIT RadLab stuff.

I could *almost* charge the cap through just a series resistor+inductor, even simpler, but that's only accurate at very low duty cycles, and my app doesn't quite qualify.

John

I like the current source. Is the driver OK with having current pushed into its output in the high state?

Cheers

Phil Hobbs

Do you pay back with a disgusting falling wave? Wonderful trick, I cannot imagine using it, except it triggers crazy thoughts for doing things.

Grant

Perhaps I'm missing something, but that circuit doesn't appear to me to gen= erate a truly linear ramp. It may appear linear during the first 20 nS of c= harge, but 20 nS is only 1/2 one time constant; the increase of voltage on = the cap is not truly linear. If you want a truly linear charge on the 39 pF= cap, replace the 1 K resistor with a constant current source.

What am I missing?

Probably; it's just a cmos pair. Cool thing is that I don't need any other parts to clamp the ramp when it's done; the LVDS receiver and the diode do it for free.

I can get by with 1% linearity, but I suspect it will be closer to

Every once in a while I have to stop being a grown-up and design a circuit.

John

generate a truly linear ramp. It may appear linear during the first 20 nS of charge, but 20 nS is only 1/2 one time constant; the increase of voltage on the cap is not truly linear. If you want a truly linear charge on the 39 pF cap, replace the 1 K resistor with a constant current source.

The resistor *is* a constant-current source, because we're forcing a constant voltage across it. The top of the resistor starts at +5.3, but ramps up as the cap charges. It's a classic bootstrap sweep generator.

I only expect this to run for 20 ns, then stop, as shown in the waveform. It will get reset later.

John

We're using a similar circuit for a slow precision current source. It's much nicer than something like a Howland source.

Best regards, Spehro Pefhany

The fall will be fast but won't be pretty, but it doesn't matter in this particular application. I have tons of time, 10 usec maybe, before I have to do it again.

John

generate a truly linear ramp. It may appear linear during the first 20 nS o= f charge, but 20 nS is only 1/2 one time constant; the increase of voltage = on the cap is not truly linear. If you want a truly linear charge on the 39= pF cap, replace the 1 K resistor with a constant current source.

Thanks for the explanation. I don't recall seeing that particular scheme be= fore, but I understand it now.

I started with a PNP transistor, with a voltage reference and an opamp servoing the voltage across the emitter resistor. Then add a few parts to keep the opamp loop stable. Then a base resistor to keep the PNP from oscillating. Plus a ferrite bead in the collector to keep C-B junction capacitance from curling the ramp. The ramp cap was grounded/released with a gaasfet, with level shifting into the gate, so I needed a clamp diode and a current-sinking rail to stop the ramp from over-driving the opamp. And a few bypass caps.

This is better.

John

My kids are grown up now, so I don't have to be any more. ;)

Cheers

Phil Hobbs

Much much noisier, though, especially at low frequency.

Cheers

Phil Hobbs

Hmm.. is there an easy way to see that without doing a detailed analysis?

Best regards, Spehro Pefhany

Just the horrible noise of the floating reference, vs. resistors working off a properly filtered one. Voltage references are hard to decouple because they have such a low impedance.

Cheers

Phil Hobbs

In my case, I can easily hang a couple of uF across the reference, and its noise *is* integrated by the ramp R-C itself. LF noise then just makes ramp slope a little different from shot to shot. A Howland needs lots of precision resistors to keep the current source impedance high, and the opamp dynamics would make a fast ramp have all sorts of wigglies.

One can always declare defeat and use an opamp integrator, at least for slow ramps. For fast, high precision ramps, the best technique is probably to charge a grounded cap with a precision opamp+PNP (or opamp+fet) current source, but by the time you do that right, you have a lot of parts.

John

I don't mean that a Howland is always the right answer. It wouldn't be for your circuit, for the reasons you give--it's easy to make a good bypass for a 20 ns transient, and anyway the reference is probably really quiet at that speed anyway, since it's completely run out of loop gain. Spehro was talking about slow and precise, though, which is where the bad behaviour of the voltage reference becomes important.

Cheers

Phil Hobbs

iry,

ng

There is probably a small kick due to the capacitance of the mixer diode. But not much since the ratio is 0.25pf to 39pf. It should be visible on a scope.

The only other solution I see is some current steering scheme through a PNP long tail pair, essentially gating a current source. I evaluated a current steering ECL type dac once and it had the most glitch free current pulse I ever saw. It was so good I needed to borrow a scope to see any glitch. [Occasionally bipolar rules.]

Bipolar usually rules, at least in built-up circuits. You guys don't share your nice integrated FETs with us broad-brush types. ;)

Cheers

Phil Hobbs