Filter Question

[Eric Foster]

Hi All,

Here is something that has puzzled me for some time, and maybe someone can help me find an answer. As many of you know, a PI has to filter the wanted signal out of a lot of noise. This is done at various points in the receiver circuit, including the final dc signal path. I usually include some low pass filtering in this dc path, to iron out higher frequency fluctuations, which cause audio threshold flutter. I use an RC filter on the non-inverting input of an opamp that starts to roll off at about 5Hz. If instead of a simple 1st order filter, I now add a couple of components to make it a 2nd order filter (Butterworth type), the noise is greater.

The 2nd order filter is standard text book stuff and has the input resistor split, with a capacitor half the value of the first, taken from the opamp output to the junction of the two resistors. The circuit has been simulated in Electronics workbench and the Bode plots for each are exactly what one would expect. Gains for the two circuits are the same. Running the dc output of the filters on a chart recorder shows the higher noise level for the 2nd order filter. Having the feedback capacitor smaller prevents any peaking at the corner frequency, so that is not the cause.

I just would have expected that the 2nd order filter would have been better.

Hope there are some filter specialists among the PI forum readers.

Eric.

 

[Randy]

Eric,

Is this lowpass filter unity gain? What is the impedance of the RC network that the input of the opamp sees? What is the noise floor of the opamp?

Randy Seden

 

[wm]

Hi Eric,

could it be the higher impedance connected to the input terminals of the OpAmp, multiplied with his input noise current? What about doubling the capacitor values / half the resistor values?

best regards
Walter

 

[Frank-S]

Hi Eric,
Will you show us what a noise filter looks like by posting a
drawing of a noise filter for a threshold, like the one you talked
about.Use a comen IC,a 062 IC.Threshold comes off pin 6,a feed back ressitor
between pin 6 and 7,a.047 cap across top.ThankS
Frank-S.

 

[Eric Foster]

Here is a link to a webpage, with the schematic I am using.

http://engr.calvin.edu/courses/engr332/lectures/activeFilters/secondOrder.html

It is the second schematic and the values are as follows mR = R = 47K, Ra = 22K, Rb = 47K, C = 0.47 and nC = 0.22. IC is a TL082, with the other half being used as a follow up dc amplifier.

To change to a 1st order I just remove nC.

Low frequency noise filtering is worse when connected as 2nd order filter. This shows on a chart recorder where the inertia of the pen is a big bandwidth limiter anyway.

Eric.

 

[Reg]

Hi Eric,

You might try simulating the circuit with EWB but use a pulse as an input. I tried a 1 hz pulse with a duty cycle of 7% and found the gain increased when using the Butterworth circuit with the cap feedback. This should result in a higher noise level, I would think.

What was interesting was the scope showed the gain while the bode plot didn't show anything obvious.

Reg

 

[Frank-S]

Thank,s Eric.
My PI has noise in 3 places.these filters will give me a
new toy to tinker with.
Question.Do you install filter between incomeing signal to threshold,or out going
of threshold.
If some one has time will they draw a simple threshold circuit with one of these
filters.I am sure that it will help people that wont ask questions.
Thank,s
Frank-S.

 

[Eric Foster]

Hi Frank,

It is installed before the threshold and audio circuit.

Eric.

 

[Frank-S]

Thank,S
Frank.

 

[Carl-NC]

What opamp are you simming with? Is output resistance modeled?

- Carl

 

[wm]

Hi Eric,
I simulated your schematic and found out, that both filters have a DC-gain of 10dB but different bandwiths (-3dB) - 3.5Hz for the 1.order and 10Hz for the 2.order. In case of flat OPV-noise density this would make about 3x more noise power for the 2.order, but according to the TL082 data sheet (from national), this frequencies are in the 1/f noise region, which should relax the 1:3 relation a bit. Disconnecting C instead of nC would result into a 1.order bandwith of 7Hz, making noise power more similar to 2.order case.
If you want to have noise power allways like the 1.order case, then increase C to 1,0 and reduce R to 22k - this results into 3,3Hz bandwith. Disconnecting nC and shortening R will result into a 1.order bandwith of 3,3Hz either.
Hope this helps
regards
Walter

 

[Eric Foster]

Hi Walter,

I think that it could be that simple. I posted on the Geotech forum earlier, as JC, who hasn't got a working password for this forum, and Reg, had posted a response there. As I was doing the post, I was running the simulation again, to make sure my figures were right. Only then did I realise the bandwidth difference was probably significant. Here is a copy of my Geotech post.


Many thanks for the help and info., but I haven't had time yet to look at it further. The EWB simulation uses the Bode plotter and sig. gen., which I presume sweeps through the frequency band you set up. Using the values I have, shows no peak in the response. If I increase the feedback capacitor, then I start to get peaking. I haven't tried a pulse input yet. The input to the filter is the differential integrator output, which is a dc level (near 0V), with superimposed residual random noise, only a few tens of mV at most.

With the values I have on the PI Forum, the gain is just over 9db and flat to 5Hz. -3db point is 21Hz and the response is -60db at approx 600Hz.

If I take out the feedback capacitor, the gain doesn't change much, but the flat bandwidth reduces to 1Hz, -3db at 15Hz, and -60db at 10kHz.

Maybe that is the problem. Out to the -3db point the bandwidth of the 2nd order filter is wider, so letting through more lf noise. Perhaps the next move is to increase the capacitor values on the 2nd order, to give a -3db point of 15Hz.


All the high frequency noise has long gone, due to previous filtering and integrating, so it is just noise from the pulse rep rate downwards, that I am trying to reduce. Some would likely advocate digital filtering, but I am basically an analog man and am not that great with analog filters, as I have just demonstrated.

Thanks to all who have offered thoughts and suggestions here. It is a great help just talking a problem through, and looking at the different angles that have been presented. I'll try a second order filter again with modified bandwidth and post the result.

Eric.