Best LP Filter Topology for Minimal Phase Shift

I would like to use a 35Hz low pass filter on the 1Vpp output of an instrumentation amp such as the LT1167.

Generally speaking, what configuration of active filter produces the least amount of phase shift distortion?

Does phase distortion necessarily increase with the number of stages?

In terms of distortion, is there any advantage in using a part such as the LTC1164-5 that has a clocked cut-off frequency?

Any suggestions most welcome.

Ken Foster

Reply to
Kevin Foster
Loading thread data ...

Bessel, no, no,

Reply to
JM

A Bessel filter has minimal phase error vs frequency, approximating constant delay at all frequencies. As you add stages, the phase:frequency performance gets better, but you add time delay. Bessels are sloppy as regards frequency response.

There are "transitional" filters that behave pretty much like Bessels in the time domain, but roll off faster in the frequency domain.

Switched-capacitor (clocked) filters tend to be noisy, and have aliasing artifacts. They are seldom actually useful.

TI's free FilterPro program is good for active filter design.

The Williams+Taylor book is the must-have if you play with filters.

I have a simple Windows program that designs passive (LC) Bessel filters, if anybody wants it.

formatting link

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

The topology is HOW you implement the filter -- Sallen-Key, state space, etc.. Any one of those topologies would allow you to implement any arbitrary low-pass filter. Perhaps not equally well -- but that's a different discussion.

The filter transfer function is what determines the phase shift properties -- there you're looking at a choice of Bessel, Butterworth, Chebychev, or eliptical (or, if you don't like that ordering, Bessel, Butterworth, eliptical or Tchebychev -- Russian-English spelling is so fun).

What exactly do you need? What are you doing? Bessel has been suggested, but it doesn't minimize the absolute phase shift, rather, it shows the most constant group delay. This is good if you need to maintain pulse shape, but it's bad if you need to have a low-pass filter in a control loop (for instance).

If you want to minimize the phase shift well away from the cutoff frequency, and if you don't care about phase shift around the cutoff frequency, then Butterworth or elliptical is best.

If you're doing a sampled-time control system for the first time and you have this worm in your head that says that if you do sampling you need an anti-alias filter -- don't go there. Anti-alias filters are the wrong choice when the cost of phase shift at baseband is more than the cost of aliasing non-existent noise down to baseband.

When you're playing with filters, keep in mind that a Bessel filter is going to have the worst sensitivity to component variations in the passband, in both amplitude and phase shift. Basically, the more highly curved the amplitude response is at any given frequency, the more the amplitude and phase response at that frequency will be sensitive to component variations. Bessel filters have a frequency response that has significant curvature all the way down to DC, and the sensitivity to component variations follows that.

--
Tim Wescott 
Control systems, embedded software and circuit design 
 Click to see the full signature
Reply to
Tim Wescott

I would suggest you consider a digital approach using a processor or FPGA. A to D, filter, D to A. This can be all in one chip if your frequency range allows it. What is your passband?

Once in the digital domain, you have a lot more you can do with the signal, limited only by your processing capability.

--

Rick C
Reply to
rickman

Check out dis interesting paper (PDF):

file:///home/matt/Downloads/A54_2__Mijat_A_Novel_Third_Order_Leap_Frog_Active_Filter.pdf

Reply to
bitrex

*Cough* Seems to be here at the moment:
formatting link

:-)

I wonder how well the topologies work with respect to op-amp nonideal properties. They've assumed ideal op-amps, which is a silly assumption to make! At least including finite GBW (as an integrator) would be helpful.

If you'd like a calculator instead of tables, check:

formatting link

Tim

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

I'm not sure why you make this claim? The op-amp approximation works as long as the GBW product is adequately large. The other approximations apply as long as those non-ideal properties are adequately small (in the noise so to speak). What is wrong with that?

--

Rick C
Reply to
rickman

The CA3140A for example has a typical gain of 90dB at 35 Hz, input C of

4 pF and an input impedance of 2Tohms...seems fairly close to ideal to me!

The offset is not that great at 5mV but the leapfrog filter topology doesn't seem to need a lot of overall gain

Reply to
bitrex

The other suggestion I would offer is to ask this question in comp.dsp rather than here.

--

Rick C
Reply to
rickman

If distortion matters, the rule is "always invert." Finite (often terrible) AC CMRR gets into the signal.

Sallen-Key filters should be the worst, since the opamps don't invert. But S-K filters have other virtues.

--
John Larkin         Highland Technology, Inc 

lunatic fringe electronics
 Click to see the full signature
Reply to
John Larkin

I would ideally like to BOTH minimize both the phase shift and keep the group delay as constant as possible.

Without going to digital or FIR, what type of filter would be the best compromise?

There is a configuration termed "Besselworth", presumably because it has the properties of both. Is that worth considering? I cannot find any design info on it.

Yes, the frequency band of interest is 2-25Hz, and I was planning to use a 35Hz LPF, mostly to reduce 60Hz. What happens between 25-35Hz is not that critical.

Whatever filter is applied, I was intending to use only two poles to reduce the time delay.

Hopefully, between this and the CMR of the instrumentation amp any residual 60Hz will be negligible.

Kevin Foster

Reply to
Kevin Foster

Here is a schematic of what is termed a Besselworth filter.

formatting link

Kevin Foster

Reply to
Kevin Foster

,
r

The phase shift is determined by the order of your filter - the number of p oles - and the cut-off frequency.

Keeping the group delay constant means going for a Bessel filter, or one of the "transitional" close approximations to the Bessel filter, which allow a certain amount of frequency dependent ripple about the notionally flat de lay that the Bessel offers.

For what?

With only two poles I don't think you can make anything except a classical Bessel filter. If you'd been interested in blocking higher frequency noise, you might have traded of a higher cut-off frequency with more poles to get the same delay, but that won't work here.

Faint hope. Mains hum gets everywhere.

--
Bill Sloman, Sydney
Reply to
bill.sloman

Duh, I read LPF and was thinking high pass. It's not so often 35 Hz is the high end of a band. I would go with a simple MCU with ADC/DAC and take advantage of the wide range of options available digitally. A tiny

8 pin MCU can do this job easily. You might need a crystal if you don't have a good time base already, but then even with the internal RC clock it would be every bit as stable as a typical analog filter.
--

Rick C
Reply to
rickman

In this particular case, nothing.

What about in general?

Even at very low frequencies, a very power sensitive application would be quite concerned about it. You can get op-amps with supply currents less than most battery leakage currents (and GBWs to match!).

If I'm deciding between active RC and passive LC at some arbitrary cutoff frequency, I need to know how much cost and footprint to devote to each. If it takes a $5 op-amp to "optimize out" a pair of $0.10 inductors, it ain't gonna happen. Somewhere between >1cm^3 inductors, at LF, and $5 op-amps at HF, the curves intersect. How can I calculate that if nobody does their theory with /actually real/ op-amp models?

One must resort to suboptimal guard banding strategies, in this case. Use at least as much GBW as the highest Q stage (which is more in high order filters with lots of stages), plus an extra "insurance" factor to account for the amp's phase shift and finite gain skewing the filter curve (say a factor of 2-10).

Suddenly, somehow, a 1MHz filter needs a row of $3 "fast" op-amps!

Sure, you can adjust the values in a simulator, or breadboard, but I can do that without reading any papers in the first place. What's the point of publishing it when there's no value?

Tim

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

On 19.12.16 07:11, Kevin Foster wrote:

The filter seems to be between Bessel and Butterworth, more like plain Butterworth. I was too lazy to compute the poles, but run it on LTSpice instead:

--- clip clip ---

Version 4 SHEET 1 904 680 WIRE -208 -192 -480 -192 WIRE -208 -96 -288 -96 WIRE -480 -32 -480 -192 WIRE -288 -32 -288 -96 WIRE -480 112 -480 48 WIRE -288 112 -288 48 WIRE -288 112 -480 112 WIRE 352 128 320 128 WIRE -480 176 -480 112 WIRE 96 208 -48 208 WIRE 496 208 160 208 WIRE 352 272 352 128 WIRE 320 288 240 288 WIRE 496 304 496 208 WIRE 496 304 384 304 WIRE 624 304 496 304 WIRE 752 304 624 304 WIRE -480 320 -544 320 WIRE -384 320 -480 320 WIRE -240 320 -320 320 WIRE -176 320 -240 320 WIRE -48 320 -48 208 WIRE -48 320 -96 320 WIRE 0 320 -48 320 WIRE 128 320 80 320 WIRE 320 320 128 320 WIRE 496 336 496 304 WIRE 624 336 624 304 WIRE -240 368 -240 320 WIRE -480 416 -480 320 WIRE 128 432 128 320 WIRE 240 448 240 288 WIRE 496 448 496 416 WIRE 496 448 240 448 WIRE 624 448 624 400 WIRE 624 448 496 448 WIRE 496 480 496 448 WIRE 352 496 352 336 WIRE 352 496 320 496 WIRE 496 592 496 560 WIRE -480 608 -480 496 WIRE -240 608 -240 448 WIRE 128 608 128 496 FLAG 128 608 0 FLAG 496 592 0 FLAG -240 608 0 FLAG 752 304 Out IOPIN 752 304 Out FLAG -480 176 0 FLAG -480 608 0 FLAG -544 320 In IOPIN -544 320 In FLAG -208 -192 +2.5V IOPIN -208 -192 Out FLAG -208 -96 -2.5V IOPIN -208 -96 Out FLAG 320 496 -2.5V IOPIN 320 496 In FLAG 320 128 +2.5V IOPIN 320 128 In SYMBOL Opamps/LTC6085 352 304 R0 SYMATTR InstName U1 SYMBOL res 480 320 R0 SYMATTR InstName R1 SYMATTR Value 100k SYMBOL res 480 464 R0 SYMATTR InstName R2 SYMATTR Value 8.2k SYMBOL res 96 304 R90 WINDOW 0 0 56 VBottom 2 WINDOW 3 32 56 VTop 2 SYMATTR InstName R3 SYMATTR Value 15k SYMBOL res -80 304 R90 WINDOW 0 0 56 VBottom 2 WINDOW 3 32 56 VTop 2 SYMATTR InstName R4 SYMATTR Value 10k SYMBOL res -224 464 R180 WINDOW 0 36 76 Left 2 WINDOW 3 36 40 Left 2 SYMATTR InstName R5 SYMATTR Value 1Meg SYMBOL cap 112 432 R0 SYMATTR InstName C1 SYMATTR Value 220n SYMBOL cap 608 336 R0 SYMATTR InstName C2 SYMATTR Value 1n SYMBOL cap 160 192 R90 WINDOW 0 0 32 VBottom 2 WINDOW 3 32 32 VTop 2 SYMATTR InstName C3 SYMATTR Value 33n SYMBOL voltage -480 -48 R0 SYMATTR InstName V1 SYMATTR Value 2.5V SYMBOL voltage -288 64 R180 WINDOW 0 24 96 Left 2 WINDOW 3 24 16 Left 2 SYMATTR InstName V2 SYMATTR Value 2.5V SYMBOL voltage -480 400 R0 SYMATTR InstName V3 SYMATTR Value SINE(0) SYMATTR Value2 AC 0.01 SYMBOL cap -320 304 R90 WINDOW 0 0 32 VBottom 2 WINDOW 3 32 32 VTop 2 SYMATTR InstName C4

TEXT 584 -16 Left 2 !.ac oct 8 0.01 10000 TEXT 128 -192 Left 5 ;Besselworth filter 100 Hz

--- clip clip ---

The input capacitor spoils the group delay at low frequencies (1 Hz and below), and there is a typical group delay peak at the low-pass corner (100 Hz).

--

-TV
Reply to
Tauno Voipio

I was trying to stay within my abilities and avoid micros. Can you refer me to an existing application, including schematic and code, of the type of circuit you describe?

Kevin Foster

Reply to
Kevin Foster

I am having trouble finding any tutorials online for transitional filters.

Does anyone know of any, or a circuit component calculator that would apply this topology?

Thank you,

Kevin Foster

Reply to
Kevin Foster

How about using a 60Hz notch? Should give greater rejection than a 2 pole LPF.

piglet

Reply to
piglet

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.