Sallen-Key input Z

r of

l for

nd

le

re

Hmm, I'd be interested to hear any reviews of the 4th edition. (From the amazon reviews it sounds like the 3rd may be better.) And of course I can get the 3rd used at lower cost :^) (I'm living on the trailing edge of technology anyway...)

George H.

Even better.

quoted text -

I used to think that but have changed my mind after using quads with

Mine is sixteen channels (four groups of four, feeding a quad power amp), four per. If the power amps can be crammed into quads, the opamps around them sure can. ;-)

No. The simplest example is an RC high-pass filter, which has a zero at DC. You mention a twin-T elsewhere. Just the twin-T network (without any op-amps to sharpen up the response) has a zero on the imaginary axis.

As John mentioned, replacing some of the caps in an LC low-pass filter with series LC elements creates resonant zeros and makes an elliptical filter.

A pair of complex zeros can be expressed as having a natural frequency and a damping ratio. It doesn't make as much sense as a pair of complex poles, but it still has some direct meaning, as the depth of the notch is directly related to the damping ratio (and inversely to the Q).

It's something that makes the most sense in the frequency domain, although as John alluded to, it's a good idea to refer back to the time domain frequently so you don't end up with an appalling time response.

Yeah, I ALSO read those reviews. That would scare me, as well. Sight unseen, anyway.

Jon

Fourth?

Hmm.. 10% cheaper in Canada

Best regards, Spehro Pefhany

I guess I need to explain this in painful detail.

Let us say this filter is designed for no gain. Now at some intermediate stage, and op amp has 10dB of gain at say 1KHz. Say the filter is a lowpass intended to filter signals with an input bandwidth of 10KHz.

Now we sweep the input from say 1Hz to 10KHz. Now at 1KHz the intermediate stage will be clipping. This means the filter is no longer filtering, but actually creating artifacts. If the intent of a filter is to filter, then one should design the filter to avoid such situations. If you wish to avoid clipping, then you need to reduce the input by

Thus to do a proper filter, you insure that at any frequency an intermediate node will have not have more gain that the output node.

This is what is called good engineering. It is independent of technology. Doing the design optimally costs nothing. When you adjust the nodes for dynamic range, you get the lowest noise.

Fluke did the same design in continuous time for their instrumentation. It takes a root solver and an understanding of root locus to do an optimal design.

Obviously in continuous time, this design can be low noise.

This is the bible. I have a hard copy, but if you search the internet, there is a pirated version. Not that I encourage that kind of stuff.

Jim Williams wasn't the filter guru at LTC. It was Nello Sevastopoulos.

Not if you design it right. For one thing, get the signal level and the supplies right. For another, put the peakiest stage last.

This means the filter is no longer

It would be unusual for an active filter to dominate a system s/n ratio.

But a 6th order Butterworth has two stages of Q

FilterPro is handy because you can tell it to use standard E6...E96 component values. And it tells you what GBW you need in the opamps.

We have the (expensive, dongled) NuHertz software that does the same for LC filters, namely makes them out of standard parts values. It's an incredible time saver. Trying to get a good LC out of standard values is a nightmare to do manually.

I often use FP or NuHertz, or a table from Williams, and Spice the design to make sure everything is good.

Here's a little plug-in anti-aliasing filter module for a precision ADC board. They needed to be field plug-in-interchangable to a few PPM on gain and a few uV on offset. I used chopper opamps.

You said: "Not if you design it right. For one thing, get the signal level and the supplies right. For another, put the peakiest stage last."

But review the example: "Let us say this filter is designed for no gain. Now at some intermediate stage, and op amp has 10dB of gain at say 1KHz. Say the filter is a lowpass intended to filter signals with an input bandwidth of 10KHz. "

Maybe you are confused by the term gain. It is relative to the input of the beginning of the filter. Hence the op amp will clip. This can't be argued away.

As I stated long ago, if you are doing a cascade of 2nd order states, do them in order of increasing Q. However, that is not sufficient to insure there is no peaking in the general case.

When you dynamic range adjust a filter, you put it in a simulator and plot the output of every op amp. You do not study it section by section, presuming you have a cascade of 2nd order systems. You note the frequency and level where each op amp peaks. Presuming you have enough degrees of freedom, you adjust EVERY op amp so that at whatever frequency it peaks, the gain is never more than 0dB. This is filters 101.

Don"t make me go all Jim Thompson on you! I can see why he gets frustrated with you since nothing I am saying here is wrong, yet you try to argue anyway.

of

uoted text -

ide quoted text -

Thanks Jim, I know and love the state variable filter... your circuit is an interesting twist. (I'm totally swamped with work today, so it will have to wait.)

George H.

for

han

ah, the

don?t

ttdesign.com- Hide quoted text -

Thanks Tim, I'll have to think about the highpass as a zero.

The twin T seems a bit of a cheat, in that it sums two out of phase signals to work. (But maybe that's just my pedestrain outlook.)

George h.

Miso, do you know if this approach is discussed in William's Filter Handbook?

Is it done with All pass filters (phase shifts) and summers?

George H.

of

GBW.. as much as you can afford. (I'm always hitting Q-enhancement issues. the only cure I know of is more GBW.)

The other issue I always run into is capacitor matching. (I assume L matching is even worse... I've only done simple, non- critical LC filters.)

George H.

.highlandtechnology.com  jlarkin at highlandtechnology dot com

I was using gain to mean gain per 2nd order stage. When you said that "an opamp has 10 dB of gain" that sure sounded like a stage gain, not the composite gain of all stages back to the input.

If it doesn't clip, there's no problem. If all you are saying is "make sure your opamps won't clip", which is fine by me. But the requirement to have no net gain anywhere is different and more constraining.

You have created a rule that doesn't always make sense. Follow that rule if you like; I prefer to have more freedom to do what works. If I'm filtering a 1 volt max signal with 15 volt supplies, I have plenty of margin for most filters. Lots of useful filter forms have a single stage that gas Q>1, so put it last and it's OK.

You snipped the part about digital filters. You can do them in double floats.

I think it's the mental template of a chip designer vs an instrument designer. In an instrument, you don't put the filter in the front end, because any active filter is horribly noisy. Managing the signal levels and SNR is part of the system design, and the instrument designer has control over that, whereas a chip designer is at the mercy of all the strange things that other people do with his chips.

Your stuff has to work in corner cases that you can't control, whereas John and I and other instrument designers do have that control and so don't need to cover the corner cases. That gives us a lot more freedom in some ways, less in some others. The design space is just different, and so are the right answers.

Cheers

Phil HObbs