Pulse Shaping for 1-10 ns pulses

What chip are you using? What scope?

An RC won't ring if the layout is tight. The capacitor needs low esl and must be grounded hard.

Since it is easy to design a Butterworth filter, I've

A Butterworth inherently rings. A Gaussian or Bessel filter doesn't. It's just a matter of the L and C values. Williams' filter book has all this stuff. Roughly, you could

in-------R------L-------+--------out, hi-z load | | C | | | gnd

where R=50 ohms maybe, R*C = 5 ns maybe, and

L = K * R^2 * C, with K in the range of 0.5 to 1 maybe.

For a ramp, you could...

current source | schottky | diode | in-------|

Yes.

Williams and Taylor's filter book, worth having around.

John

Yes, they are. And there is helper software such as this:

I mostly do it the way John suggested, with the Williams Filter Handbook. Then I simulate and tweak with Spice. And yes, as John mentioned there is a zero tolerance policy for layout. You can't get away with Sauerkraut wiring for something that has to shape a 500psec slope. Meaning a full ground plane, controlled trace impedance, microstrip and all that. This is hardcore RF engineering. And, oh yeah, even 0603 capacitors do exhibit some nasties such as inductance and series resonance.

Look into the old literature on waveform generation, like the MIT Radiation Lab books. Pulse shaping is a venerable tradition.

Triangle pulse is easy; just an RC integrator (which also attenuates, a lot).

Gaussian is impossible, except as an approximation (the tails are infinite, after all, including the ONSET of the pulse).

In the absence of devices that switch much faster than your target times, you will probably want to make multiple-transmission- line filters. A true wideband delay line, after all, is just a bit of stripline on a circuit board...

Pad over ground plane is rather useful to reduce such inductances; tweek the signal pad size on the capacitor, as well as the shape (narrow trace abrupt to middle of square VS triangular transition from trace to square; square pad VS rounded pad).

Thanks John for your suggestions. It has given me many good leads.

I am using Phillips 74AHC08 AND Gates and Toshiba TC7SZ08 AND gates, which are both fairly typical high-speed TTL parts. I have also used a Stanford Research Systems pulse/delay generator, but it produces ugly pulses and can't produce pulses shorter than a couple of ns. I think we might need to send it in for tuning to get it working properly. We typically use a 500 MHz HP digital scope that leaves a few artifacts. We also have a really old HP 12.4GHz sampling scope that is much harder to use and collect data from, but usually shows significantly reduced overshoot and undershoot.

and must be grounded hard.

My layout is with generic copper-plated circuit board and an exacto knife, so I doubt it is tight. I think I will try using two caps in parallel to help reduce problems.

I want to check to make sure I understand the circuit correctly. The schottky diode acts like a capacitor on the first edge of the pulse, with voltage ramping up at a rate determined by the pulse height and the current source. Then, when saturated, it acts like a diode again, and turning the circuit off. Thus, it will produce a right-triangle type pulse like this: ____/|___

The Fairchild FYV0203S series and BAT54 series lists 10 pF of Capacitance and 5ns of recovery time. Are these the specs to look for in choosing a schottky for ramp generation?

Unfortunately, it was already checked out. But, I sure appreciate the reference.

Thanks again.

High order Bessel filters ring. I forget the breakpoint, maybe the 5th order on up rings.

RC filters can't ring. [Their impulse respond is always positive.] Of course, if you have parasitic L in the circuit, all bets are off.

The best passive filter book is that by Zrevev.

No, look at figs 7.9 and 7.10.

That Google Book thing is excessively cool.

John

I am connecting the output of the pulse generators to SMA cables, then to a little board where I have the filter, then to a scope. I am wondering if maybe some of my ringing is coming from impedance mismatches between the boards and wires or something similar.

For my case the attenuation is fine. We use external attenuators to scale down the pulses anyway. I just need to minimize ringing and any junk that happens after the main pulse. I am thinking I will just use an RC filter for ~200MHz, as that will correspond to ~5ns rise and fall times.

If you had more detail in the graphs, you would see the impulse response go negative. it is very small, but high order Bessels do ring. For things where you can't tolerate any ringing (weigh scales), you need a Gaussian.

Looking at my copy of Huelsman&Allen, I'd say the 5th order is where it starts.

You probably have ground bounce. In theory, the TTL has Schottky diodes, but maybe you can clamp the line with an external Schottky. Make that a small diode, not a power Schottky.

I know HP has some trick to make flat pulses using what they called a flat pulse diode. This datasheet mentions it