Use bleeder resistors in audio opamp design?

There you go ;-) ...Jim Thompson

I'm fond of shunting the integrating capacitor with an R... so you get integrators of the form 1/(p+1)

"p" = frequency normalized "s" (Laplace/Heaviside for you amateurs :-)

Tosses any dissipation problems with VLF filters.

Also helps control dynamic range. ...Jim Thompson

=A0 =A0 ...Jim Thompson

=A0 =A0| =A0 =A0mens =A0 =A0 |

=A0 | =A0 =A0 et =A0 =A0 =A0|

=A0|

=A0 =A0 =A0 |

You can control dynamic range by scaling gain paths.

Out-of-band is often problematic... my way has no integrator with gain

Run the math, you'll like it :-)

Drops right into leap frog, AND saves one OpAmp per "lump". ...Jim Thompson

Or for people who design PPM-accurate instruments. I can't afford to add a dozen or two resistors into my gain paths... the TCs would be deadly or the cost would be insane. An S-K has a DC gain of 1, independent of resistor values.

And they are not hyper-sensitive to component values if the Qs are low. We generally use them to front digital filters, so don't need brickwall performance. But we do need DC accuracy.

John

...

That's why I'm using a couple windoze apps to do it for me ;)

Anyway, as I was uncomfortable with series caps I ordered some 10n today for where I was going to series up to get lower value. I might even use the remainder of the 50 caps one day on ADC/DAC stuff.

I did say up front I'm processing audio, though I do like the DC unity gain because I'll possibly use the low pass on data, dunno yet. Direct coupled is good too. Better than back to back electros.

If I was a tad brighter, younger and better at maths I'd look further into other topologies, but you need to try harder to scare me into doing more than taking a brief glance at your leapfrog whatsits :) I never heard of them before. I used to work in industrial electronics, was called on to make a filter card once, multistage low pass thing with lots of trimpots, yuck. Stayed out there so it must've worked.

That was DC accurate stuff for signal conditioners too. So there's my bias showing. If it don't work I'll chalk it up as lesson learned, be here back with a sad story to tell.

Grant.

...

Stuff that, spent most of today learning Eagle :(

So what? I've not heard of them, therefore they have no track record for this particular application, where the topology I'm using does have a track record.

No, I really don't. So I didn't, now.

I like nice symmetrical swap the Rs and Cs to get the complimentary function with S-K.

Grant.

I'm so glad to have dropped away from your target audience :)

Grant.

Relax. The toothless old git can't actually do anybody any harm.

John

No kidding! The S-K filters I worked with at Microdyne had 14 1% components and the bandwidth & roll off had to meet 5% specs. A run of boards usually required at least eight of the 12 filters to be tweaked by changing pairs of resistors or in some cases, swapping the pairs 100 pF capacitors that were used in every filter. I wrote a program to select new pairs to get the desired response. I found out later that the engineer had simply made a spreadsheet and picked the closest values in inventory. He didn't take the capacitance of the traces into account, so as the frequency went up, his calcs were useless.

Another problem was that once we passed 6 MHz we switched from op amps to current amps, due to the impedance problems.

s.

Wow! Have you built one of those ten opamp filters? I've built the 'classic' three opamp two pole S-V filters, could I string three together and get 6 poles with 9 opamps? (Or am I displaying my filter ignorance?)

George H.

ors.

in a

t?

The op amps are cheap and don't play much into the filter response if chosen correctly. It is the Rs and Cs that are troublesome in creating products for volume production. Hence some designs limit the value of Cs used to get volume pricing.

In the SCF days, inversion comes for free with switch phasing, so leapfrog was basically what was done for filters with tight specs.

I can toss in another fact of filter life: generally you should add transmission zeros to your filters rather than all pole designs. Now there are exceptions if you are pulse shaping. What you need to do is philosophize on what you are trying to achieve with a filter. For an all pole filter, eventually the roll off will be such that the signal you are trying to filter will be in the noise of the filter itself. Thus, what you really want is a filter with a defined level of stop band attentuation, rather than a filter that rolls off for ever. Transmission zeros in SCF don't add op amps, so their is no additional noise added. With transmission zeros, you can often get by with less poles of filtering, hence you can make a lower noise filter.

Getting back to leapfrog designs, let's assume the filter is relatively high order, say 9th order. If you compare a cascade of four biquads plus one single pole stage to the leapfrog design, the "localized" Q of the op amp outputs will be lower with leapfrog versus the cascade of biquads.

Localized Q isn't necessarily a standardized term, so I will elaborate. Every op amp node is essentially a filter output. You can use the traditional definition of Q (i.e. relate delta F to center frequency), even though for the ladder the filter order will be higher than second order. [My recollection is every node of a leapfrog design has the same denominator due to Mason's Rule, but don't quote me on that.] If you compared the Qs, the cascade of biquads would have higher Qs than the localized leapfrog Qs. A high Q section is prone to shaping the noise at the center frequency. Much worse, the PSRR of the filter is worse with high Q.

All that said, much of the datacom filtering was done as a cascade of biquads and all passes. It was only after sophisticated software was developed that leapfrog (i.e. ladder) filters were used in high volume designs.

itors.

in

uld

in

s in a

es

it?

Thanks Miso, that's interesting. What sort of frequencies are you talking about? For the two pole (Q=3D.707) S-V filters I've built I see a few percent of Q-enhancement at 100kHz. (8 MHz GBW)

George H.

acitors.

em in

would

on

s in

ros in a

ries

des

ix it?

ere

the

ter

SV as in state variable? At two poles, there really probably isn't much difference in leapfrog versus SV. You would only see the difference in high order filters. You would have to measure all the components just to make sure it isn't some tolerance issue that makes the Q appear higher.

...

Measure to what sort of tolerance, for 2 pole Butterworth response (Q=.707) filters?

Grant.

apacitors.

them in

r would

es on

aps in

ctros in a

series

nodes

fix it?

There

en

be

t the

ilter

see

Yeah SV is state variable. It's nice having all three ouputs (LP, BP and HP).

S. Franco gives a formula for the Q-enhancement in his opamp book. That fits what I measure. Q(actual) ~=3D Q/(1-4*Q*fo/ft) Where fo is the corner frequency and ft the opamp transistion frequency.

George H.

er

gn

to

he

me

see

Ahh how good is my measurment?

So for the S-V filter there are three ouputs. LP, BP and HP. I find the frequency at which the LP and HP outputs are equal and call this the corner frequency, and then measure the band pass response at this same frequency and call that the Q. (That comes from the theory of the S-V filter function.) Here's a comparison for the single point fit with a 'complete' measurment of the one filter. (10kHz low pass.) (The two highest frequency data points were not used in the fit.)

I've got similar fit's to the BP (f0=3D10.05kHz, Q=3D0.705) and HP (fo=3D10.06kHz, Q=3D0.708)

So something a bit better than 1%.

George H.

It would take a lot to get me to use 10 op amps in a filter--like several wild horses. DC and low frequency accuracy is one reason, and weird overload and noise behaviour is another.

If I need anything that fancy, I far prefer to use a gentle passive filter just to preserve dynamic range, followed by digitizing, followed by doing the fancy stuff digitally. I might lose a factor of 2 in SNR that I'd have to make up elsewhere, but OTOH I wouldn't be losing dynamic range to internal peaking in the op amp filter, so it's probably a wash.

If it had to be really fancy, e.g. a lowpass with a high-Q notch, I'd sooner use active tweaking of a two-pole LC than deal with the noise peaks and poor overload behaviour of a fancy active filter. I've also used switched cap filters for moderate dynamic range jobs, and they work fine.

Cheers

Phil Hobbs

The S-K filters we used were four op or current amps. Since the unit had a 68340 CPU and A/D converter, it checked the DC offset and automatically nulled it to less than a mV. it was sampled at the output of the 16/1 analog mux used to select which of the 12 filters, or no filter was selected for the application.