EMRFD Message Archive 11810

Message Date From Subject
11810 2015-10-19 09:09:14 Andy Talbot Classic Multiple Feedback OPA BPF
I think this group is the right place to ask the question - as many members here seem to be well-versed in analogue circuitry.

This Multiple Feedback opamp bandpass filter forms a backbone of many audio filtering tasks, but I recently designed a built a cascade of them to give some anti aliasing on a PIC based 1kHz I/Q sampling circuit (to reduce products below 750HZ and above 1250Hz that would alias into the baseband after downconversion).   However, measurements of the final don't stack-up with my implelemtatino of its transfer function.

I have a cascade of three identical filters, 200Hz bandwidth tuned on the nose at 1kHz with each havign Av = 1.7.  Component values were generated using design equations I've had for decades, (literally),  then I used a formula for the transfer function found on the web.  I've lost the URL for that but it occurs frequently enough in a Google search.   Implemented that in a spreadsheet and checked the attenuation curve did what I wanted .  The plots suggested  at least -50dB at the two frequency points of interest.

The circuit diagram showing the filters and the spreadsheet with the curves can be found in http://www.g4jnt.com/DownLoad/1kHz_IQ_Sampler.zip

However, when I came to measure the response using broadband noise fed to its input, and monitoring the output going to the A/D using Spectran audio analaysis tool, I found the levels at 750 / 1250Hz were only some 25dB down from the peak.

Does anyone have any idea why roll-off is so much poorer than analysis suggests?  I assume the transfer function equation is either wrong, or the design equations are.

For information / interest, I am doing this to be able to coherently receive signals at LF (137kHz and lower) .  I have a QSD receiver for LF with a 48 bit DDS LO, clocked from a 10MHz GPS or Caesium master source.   The output from the QSD appears at exactly 1kHz, passes though a simple quadrature network for sideband cancellation then into the PIC A/D converter.  There is it multiplied by a quadrature squarewave at 1kHz and the outputs sent on a RS485 interface to the PC for analysis.  As a squarewave LO is used, alias products will appear if signals  outside a 500Hz bandwidth, ie +/-250Hz on the centre tone are present.

I first did this back in the late 1990's using a simple PIC of that era, and a Racal 1792 receiver for the Rx. Back then, I could use the receiver's 300Hz CW filter for the alias reduction.  But now,  having got rid of that Rx and not having another suitable one with fully locked LO chain, the Softrock type clone has to serve.

Andy  G4JNT

11812 2015-10-19 10:33:29 victorkoren Re: Classic Multiple Feedback OPA BPF
Hi Andy,
I simulated one BPF stage in LTspice and indeed it seems that your spreadsheet is to blame.
Gain is OK at 4.6dB on the peak frequency of 1 KHz, but attenuation is not as expected, it is -9.9 dB below peak gain on 750 Hz and -7.8 dB on 1250 Hz.
Back to the design table.
Victor - 4Z4ME
11813 2015-10-19 11:53:32 andy_jnt Re: Classic Multiple Feedback OPA BPF

Yes, I feared that might be the case.   I was too lazy to try to derive the transfer function from first principles and just took the function from the web - and other sites tend to agree with the equation.  So it looks as if my spreadsheet conversion is at fault.

To those who replied individually:

 I had already tried lower component values,  each R was originally  1/10 of those quoted, with 100nF caps, but once I realised I couldn't use SMT ceramic caps as they were far too unstable, they were replaced with small 10nF polyester ones that fitted on the SMT pads neatly .   There is no difference between the 1/10 and current values.

I did probe each stage, and the response from each was indeed 1/3 (in dB)  of the total.  And about the 9.something dB down at the edges as mentioned here - giving the 25dB or so for the three.

I do have a copy of Ansoft Designer, but prefer to derive the equations themselves from first principles for simple designs like this rather than just entering components into a simulator package.

Must check the spreadsheet implementation first though.    My calculated values are about twice those in reality, so it looks as if there could a spurious squared term present, perhaps.

Then its Kirchoff and equation manipulation time.

And the worse annoyance, that 25db is not enough attenuati
11815 2015-10-19 22:47:29 victorkoren Re: Classic Multiple Feedback OPA BPF
You have a few free active filter calculators. Try Analog Devices, TI or Microchip.
Victor - 4Z4ME
11818 2015-10-20 09:27:41 wb8yyy_curt Re: Classic Multiple Feedback OPA BPF

the op amp filter design tools at TI are very good and easy to use, and give trustable results.  the new one, on line I think is Webench, and they may have filterpro still there as well.  there you can explore different kinds of filter shapes.  there are some good articles on the web also.  if you email me I can also send you a recent QRP quarterly article (if nothing else the references are good). 

73 CUL

Curt WB8YYY (my email at qrz.com)
11819 2015-10-21 10:02:59 andy_jnt Re: Classic Multiple Feedback OPA BPF
I sat down with pen and paper, manipulated S and jw  and got the transfer equation to my own satisfaction. Scrapped the old spreadsheet, put the new equations in and it gives the right answers.  Which, of course didn't do the job I wanted with the original RC values...

So now, changed the R values for a 100Hz design tuned on teh nose for all three, and I am now indeed getting around -50dB at the  750/1250Hz points.

Still can't work out what was wrong with the old one now, but that was consigned to the bin so its irrelavent now.

Being of the (slightly) older school, I believe that if one can't work out he basic transfer function for a simple lumped element design, you have no right to use complex simulation software.  Leave that for the jobs that are near impossible to do from first principles - like microwave amplifier design.

Andy  G4JNT