continuously variable Bessel filter

This is done with current controlled OTAs. TI make/made the part. You are varying the transconductance to shift the corner. Very old school.

But what kind of customer asks you to do something you don't know how to do?

Bingo. See my post with some papers.

OTAs were sort of a fad for a while, mid-80's maybe, but have mostly faded. I haven't used one in decades. The LM13600 would be terrible for my application, way too slow and way too imprecise... BW is 2 MHz and the Gm spec covers a 2:1 range, so the filter coefficients would be all over the place. The circuits get complex, too. A really good OTA would be useful, but they don't exist, as far as I know. Are there any faster ones around?

There are several ways to do the filter I'm considering. I was just wondering is anybody else has done something similar, and if they had any ideas.

The very best kind! Now they want me to put an impossibly fast and precise 6KV pulse into a capacitive load. Another customer called me yesterday and asked me how to program one of our digital delay generators to make a negative 3 millisecond delay, to drive a pump laser for 200 us before the "go" trigger arrives. We figured it out.

John

Even down at John's lower limit (100kHz) I'd be tempted to look at LC. Not sure if good enough but it won't take much time to try capacitor banks plus muxes. Or maybe real esoteric stuff such as gracefully saturating toroids or pot cores where a DC current would control the residual inductance. Q doesn't have to be high in a Bessel filter.

TI has/had a better part. Do you seriously think I'd suggest using anything from the old National. ;-)

Tweaking the transconductance is a very old scheme. I've seen it used in color burst oscillators where you pull the crystal with a varied GM. The basic technology doesn't have a bandwidth wall.

They are more like boutique parts :-)

Besides the two methods mentioned in my other post there is another one. It really requires you to hand out sick bags in the design review: Mix it up to some convenient IF, a few MHz, somewhere nice and quiet. Provide a tuned circuit up there and modify its Q. Resistively, Q-multiplier, whatever. Then, using the very same LO, mix it back down.

If you ever want a super-sharp cut-off of the brick wall kind mix up to a frequency where you can cheaply buy filters with steep slopes and enough bandwidth. 45MHz, 70MHz, whatever is available from the consumer market. Now scoot the LO to cut into one sideband. Use same LO for mixing back down, and voila. For this method the LO for down-conversion must be 0 and 90 degree two-phase because you have to pick out the sideband you are cutting into and throw away the other one. You'd also need a 45-degree phase shifter but over a 100-600kHz stretch that's not so bad because that is only a 6:1 ratio. In literature it is usually called the direct conversion method of SSB demodulation. AFAIK California hasn't yet instituted disposal fees for unwanted sidebands :-)

You can still get air-cooled power tetrodes for UHF transmitters. But they ain't cheap and their filaments alone can use more power than a Chevy Volt.

Ah, they want a time machine :-)

National _is_ TI :-)

Sure, but I bet John needs a scheme where they can get the parts for. Preferably over then next decade or two, and not from scalpers.

I think the new fast LeCroy scopes use a superhet design, mixing the high part of the input bandwidth down to where reasonable ADCs can handle it, then splicing it back together digitally. When Tek was designing the 7104, they considered something like that, all analog of course, but Barrie Gilbert managed to push linear IC technology so that they could do it straight-through. Ft doublers and tricks like that.

Nice, but I really need a simple circuit and tunable Bessel time-domain response. And nanosecond jitter.

Unwanted sidebands and FIFO overflow bits must go in the purple bin.

Triggered spark gaps would be interesting. Or avalanche transistors, if the power dissipation didn't kill them; distinctly a possibility.

Yeah, as Spehro has mentioned, causality causes all sorts of problems. Our delay generators have 20 ns insertion delay, which helps sell them, but I could sell a lot more if it was -20.

John

If they wouldn't be Windows-based I might have bought one.

I have a 7704. Besides being occasionally recalcitrant on start-up it always performs like new. Reminds me of my old Citroen. Crank .. crank .. phutah ... *POOF* ... crank .. crank ... VROOOM.

Simple ... check. Low Jitter ... check. Bessel shape ... maybe not so perfect.

The unwanted sideband goes into resistors and causes global warming :-)

FIFO? Never used one there. Those are for them thar digital dudes.

Avalanche usually means jitter like crazy. You never know exactly at what voltage it'll go off, and how it progresses through the stack. Just like with real avalanches where each one is a little different despite happening in the same location.

For that you'd first have to reverse engineer the tricorder. Then read up on time machines. I've got the Jules Verne series for that, in case one of my clients ever wants a time machine designed :-)

The noise and DC accuracy and linearity will be grim with those

John

Switching among a bank of filters might be OK with my customer, and LC filters might be feasible in this frequency range. A 5-pole Bessel, doubly terminated, is only 7 parts, only two of which are inductors, which is sure appealing.

We have the Nuhertz LC filter design software, which is incredible for designing filters like this. I'll give that a shot and see if it's practical with available surface-mount inductors.

John

The best precision you can buy with catalog inductors at reasonable cost is 5%. Not sure if that's good enough in your case. Below that it's either custom trim or lots of $$$.

Aade Design works also. But I am old school and sometimes use the Williams book. Meaning it looks a bit worn by now, like our bibles. When I used a slide rules at a company (to scoot stuff between catalog values I find those indispensable) the guys there reassured me that they do have Edison electric light and all this newfangled stuff.

Possibly this could also be done with PSoC chips from Cypress. They have switch-cap features and other analog building blocks in there. Nice thing is that stuff tracks well because it's all on the same die. But I don't remember if 600kHz is reasonably feasible, you'd have to check that out.

The ones I was talking about cost $250K or something outrageous like that. 40 GHz or so.

It's the only analog scope anybody still uses at our place. I get a callouses on my thumb from whacking the CRT shutoff button. The microchannel plate has a lifetime limit on coulombs it can output.

The Zetex parts are pretty good. Imagine a SOT23 that will switch 400 volts and 60 amps in a ns or so. Jitter really isn't bad. The old Tek sampling scopes used avalanche transistors.

John

Ouch!

Mine doesn't have a storage tube. If I have to pick up something from the lab late at night and don't turn on the lights you can see that eerie blueish glow and a hint of the last measured trace on there, even if that was hours ago.

A guy at the university in Hannover (Germany) has tried it and the jitter was certainly not in the picosecond range. Quite disappointing, actually. Problem is, you need a beanstalk circuit and the longer that gets the less predictable an avalanche becomes. It's like accelerating a freight train. Klonck .. ker-clunk .. ka-klonk .. thud .. ka-chunk ...

So I suggested TV horizontal deflection tubes. To my utter surprise they could still buy lots of NOS of the PL81, so they did. Cheap, under $2 a pop. Someone must have accidentally discovered a whole warehouse full of those. Designs with that tube were discontinued in the late 50's. Maybe used in "portable" TVs in the very early 60's. European H-final tubes were usually rated at 7kV peak.

Williams is the fundamental filter book; well, along with Don's Active Filter Cookbook. Williams has tables for finite-Q inductors, but not for all forms. The Nuhertz software has got us out of trouble a couple of times, when we needed a high-performance filter to be made from actual, available inductor values with real Qs. I once spent days trying to design a DDS filter, with Williams and Digikey both open, Spice on the screen, getting lost in space. Nuhertz solved the problem in milliseconds. The guy who runs the place is crabby and paranoid, but he sure knows filters!

John

Yep, with some series you can't buy values such as 91uH, or there is no stock. Needs to be pieced together, so at least two inductors per position. But they are usually under 10c. Rotate 90 degrees to each other so they don't couple. If you have the space parallel a non-stuff footprint each, just in case.

A lot of filter stuff ends up being "seat of the pants" after 20+ years on the beat. Sometimes when I change stuff real quick I have a hard time explaining how and why. Like Giuseppe, the guy they always called in when the carburetors on a hot Alfa Romeo had to be adjusted :-)

There's probably an acquired knack for tuning filters, like backing up a double-trailer truck. I apparently don't have it.

John

Wells, not Verne.

Jeroen Belleman

Try this one :-)

The ultimate test would be backing up this rig: