Active filter with gain

Dear all,

I need an as high order as possible filter for ~80kHz small signals with good phase linearity (don't care much about the amplitude response flatness). My current choice is to use LTC1562-2 in an 8th order Bessel filter configuration with 5V single supply. The problem is that I also need some gain (36dB looks good) and FilterCAD allows me to enter this value. But how well does this simulation matches reality? Most of the filters I've ever seen are of unity gain and I believe there is a good reason for that. Should I expect oscillations or other forms of instability?

Best regards, Piotr

Reply to
Piotr Wyderski
Loading thread data ...

You need to get hold of a decent text on active filter design.

formatting link

0071471715

is one that I like, as do quite a few people who post here.

It's fairly straightforward to buld an eighth order Bessel filter out of fo ur two-pole Sallen and Keys section, and a bit easier to do it if you allo w the individual sections to have some gain, since this lets you use equal value capacitors (or fudge your capacitor ratio's to the E6 values you can buy off the shell)

9dB per section might just be a bit much - I've not used that much gain in any filter I've designed, but filter design hasn't been ever been my main j ob. Sallen and Keys sections can become unstable if they have too much gain (and too much feedback capacitance), but a single gain stage after the fil ter section shouldn't be too much of a problem.

The LTC1562-2 offers four similar two pole sections (but clearly not Sallen and Keys), with built-in capacitors, and if FilterCAD says that you can co nfigure it to do your job it's applying exactly the same theory as Williams and Taylor (which does tell you how to figure in the finite gain and bandw idth of your op amps).

If you get hold of a decent textbook you will be able to get a much clearer idea of what's going on inside your filter - which probably isn't going to be all that interesting.

--
Bill Sloman, Sydney
Reply to
bill.sloman

Thanks, Bill, but filter design is not and has never been my field of expertise and I would like to keep the current state of affairs. In this particular case it is "just" a block (an antialiasing filter) which must be placed before the block which does the interesting things (from my point of view). I just don't want to do something obviously wrong for an expert, hence my question. I also find this 9dB per stage a bit suspicious, but cannot resolve whether the risk of thing going awry is real or it is just my prejudice. And if the former, then what gain is safe, except of the obvious answer If you get hold of a decent textbook you will be able to get a much clearer idea > of what's going on inside your filter - which probably isn't going to be all that interesting.

Probably, for me what the filter does is important, not how. I can (and will) simulate it in Spice, but I'm afraid the simulation can be too exact, ignoring the feedback paths present in a physical realization of the filter.

Best regards, Piotr

Reply to
Piotr Wyderski

Is this a bandpass? High gain and high Q cost GBW, so looking for both at once may be too optimistic. I'd put as much gain as I could ahead of the filter--just below the point where distortion sets in.

Also with bandpass filters you have to check the phase linearity when you're done. The usual LP to BP transformation is nonlinear in frequency, which messes up the nice Bessel maximally-flat-delay property. Software can fix that, of course.

Cheers

Phil Hobbs Phil Hobbs

Reply to
pcdhobbs

once may be too optimistic. I'd put as much gain as I could ahead of the f ilter--just below the point where distortion sets in.

're done. The usual LP to BP transformation is nonlinear in frequency, whic h messes up the nice Bessel maximally-flat-delay property. Software can fix that, of course.

The phase linearity should be testable in LTSpice, once you've plugged in t he resistor values to get the LTC1562-2 to act like the kind of filter you want. At 80kHz the stray inductrances and capacitances of a real circuit s houldn't be too much of a problem. Spiral-trimmed resistors tend to have ab out 0.3pF of parallel capacitance, but L-trimmed surface mount resistors do seem to have quite a bit less (though ground plane in the vicinity can pus h this up again). Manufacturers ought to document this, and Philips did - f or a while - for some of their spiral trimmed metal-film resistors. The ind uctance their 500R resistor roughly cancelled it's inductance at high frequ encies, which has been exploited here and there.

9dB per section - 36dB over-all - isn't a whole lot of gain, and the LTC156 2-2 is optimised for slightly above 80kHz. If FilterCAD is happy, it ought to work - Linear Technology isn't into deceiving it's customers in the way the TI has from time to time.
--
Bill Sloman, Sydney
Reply to
bill.sloman

Oh, sorry: yes. With BW=2.6kHz. As steep as reasonably possible, but with good phase properties. Actually, the Bessel approximation is way to good and something between it and the attenuation provided by Chebyshev would be even better, but I don't want to earn PhD in filter design and therefore stick to whatever the tool has in its menu.

Their design tool (FilterCAD) doesn't complain, but I am not sure if they check that at all. I'm curious about the simulation result, the LTC part is obviously on their IC list in LTSpice.

Best regards, Piotr

Reply to
Piotr Wyderski

Interesting part. I've not used it. Does it list a maximum frequency for the HP output? (When do the amps inside roll-off). As Phil said there a Gain/ Q/ frequency trade off. I usually figure I want the Gain*Q*Freq to be ~30* times smaller that the GBW of the opamp. You can push that some if you don't mind less than ideal behavior.

80 kHz and gain of ~100 is going to call for a fast amp.

George H.

*There's a Q error term that goes something like G*Q*F/GBW with some factor out in front.. (pi or 2 or 4?) so 30 is ~6% error.
Reply to
George Herold

Huh, so a Q of ~30... usually that is a gain of 30 too, without adding any more. Knowing nothing about the FilterCAD, I'd guess it takes all that into account.

George H.

Reply to
George Herold

Interesting! Looks like the LTC1562-2 is a gyrator-based active filter block. ...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
 Click to see the full signature
Reply to
Jim Thompson

You may be out of luck with analog filters. A Bessel filter is good for group delay, but its shoulders are pretty sloppy.

Anyway, it will be increasingly difficult to keep the sections in tune as you add sections (poles) into the filter.

What is the center frequency of your filter?

--

-TV
Reply to
Tauno Voipio

80kHz is low enough freq that you could probably do an N-path bandpass filter using off-the-shelf CMOS logic:
Reply to
bitrex

Gyrators are the way to go for that sort of thing, because anything else is way too sensitive to component values. Still, you're going to have to use precision components or else tweak it manually.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

Oh, and be sure to put the low-Q sections ahead of the high-Q ones, or the intermediate stages are likely to clip.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

  1. With a Q that high, consider a coupled resonator design.

Q > 30 inductors at 80kHz are pretty uncommon, and you may not have that space anyway; but I imagine an active circuit can do the job just as well. To that end, there ought to exist a suitable transformation of the coupled resonator circuit (e.g., gyrator + capacitor for the resonator, then cap dividers and cap coupling between resonators for setting the peaking).

  1. Equiripple linear phase is a little sharper than Bessel. Like Chebyshev is to Butterworth, the phase isn't smooth, but approximates a linear phase as well as possible, given the ripple parameter. AFAIK, these aren't solved analytically but precalculated in tables (and tweaked afterwards, as others have mentioned).

If you're okay with kinkier phase than that, then you can move towards a Butterworth or Chebyshev design, where the phase response is increasingly exaggerated, but the cutoff is sharper.

You can also add zeroes to the stop band to get a locally sharp cutoff, at the expense of slower asymptotic response (i.e., instead of an 8th order filter being -160dB/dec, each zero adds +20dB/dec).

You can have 0 to N zeroes in a filter. The case for N zeroes, with Chebyshev pass and stop bands (equiripple amplitude), is the Elliptic or Cauer filter.

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Contract Design 
 Click to see the full signature
Reply to
Tim Williams

The topology options available for low-frequency analog bandpass filters with very sharp skirts are definitely few, and results are likely to be disappointing, regardless. :-(

Reply to
bitrex

Nah, you just have to use better op amps, and lots of 'em. ;)

The main problem is tuning. The filter Q is about 30, but to get a flat top , there will be sections with higher Q, maybe 100. With a gyrator, the sens itivity coefficient is 0.5 (like an LC filter, and for the same reason). Th at means that a 2% capacitor is way too sloppy.

To the OP: why do you want to do it this way rather than the usual anti-ali asing filter/digitizer/digital filter?

Cheers

Phil Hobbs

Reply to
pcdhobbs

Because the only ADC not used by other parts of the design is a 16-bit delta-sigma which just can't go high enough to allow the all-digital

4*Fs quadrature mixing/decimation/post-filtering.

The other reason is the noisy environment, so I thought it would be wise to filter out most of the the garbage as close to the input as possible, reducing the risk of amplifier overdrive.

The thing will be a phase modulation decoding DCF77 receiver on a PSoc5LP (or a small subsystem within its responsibility realm, to be precise -- to mysurprise the chip can absorb massive functionality).

Best regards, Piotr

Reply to
Piotr Wyderski

The LTC1562-2 is a precision part - read the data-sheet. You will need precise resistors, but 0.1% parts are available ex-stock at only moderately silly prices.

--
Bill Sloman, Sydney
Reply to
bill.sloman

Would undersampling work in this application? "Theoretically" you don't need your sampling clock to be a multiple of the bandpass center frequency to achieve perfect reconstruction, only a multiple of the bandwidth of the box-of-interest. That is to say if you know by design that your digital signal is going to be aliased but you're going to be digitally bandpassing anyway, you can set up the low-pass analog pre-filter and undersampling frequency such that the aliased components don't step on it.

Reply to
bitrex

op, there will be sections with higher Q, maybe 100. With a gyrator, the se nsitivity coefficient is 0.5 (like an LC filter, and for the same reason). That means that a 2% capacitor is way too sloppy.

An argument for using the LTC1562-2 - the capacitors are built into the par t you buy, and presumably trimmed to respectable precision.

liasing filter/digitizer/digital filter?

Good question - or would be if we didn't already know that Piotr was pretty competent.

--
Bill Sloman, Sydney
Reply to
bill.sloman

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.