Need advice on Anti-alias filter design

[beej2]

Have E-series DAQ card and I want to add some anti-alias filtering at the inputs (SCB-68). Can't afford to purchase but can afford to build my own.
Have an 8-pole MFB filter design courtesy of TI's FilterPro program for active filter using TL07x op amps with capacitors and resistors. Because my inputs will be either ground referenced (NRSE) or Differential floating (DIFF) I am unsure how the input to my home made filter should be constructed. I am looking at applying the inputs +IN and -IN to a AMP02 instrumentation amp set up at unity gain.
A diagram of the proposed input circuitry is below.

1.0k
+ ------/\/\/\/---------> + IN of AMP02

1.0k
- --*---/\/\/\/---------> -IN of AMPo2

|
| 10.0M
|--/\/\/\/----------> Pwr Supply GND

The AMP02 and TL07x are run from +/-15 VDC. The REF pin of the AMP02 is externally connected directly to Pwr Supply GND. The output of the AMP02 feed the first stage of the filter. The output from the last filter stage is referenced to the same Pwr Supply GND.

This filter works on the bench when all test gear is connected to the same outlet and ground.
In the field, the Common Mode voltage should be less than 1v for almost all measurements. Almost all, but not all.
On the bench, the circuitry works without the AMP02 or any of the input resistors with the -IN connected to Pwr Supply GND. The test signal generator is a BK function generator.

Is the contemplated design with the AMP02 and input circuits a good design? The AMP02 ($9) costs as much as everything else put together, so if there is a good method without the cost of the AMP02, that would be great.

Sorry about the long-winded question. I am attac
hing a bitmap of the entire circuit for anyone willing to assist.
Thanks,
Barry Jensen

[Dr._Imad]

If you are using Labview, you may use a digital filter (look at the Labview examples - Digital filters).
Depending on what you are trying to achieve (eleminate high frequency noise, or 60 Hz noise) etc, and the bandwidth of the signal, simple simulations or runs may acheive your goal. (I always do that in many applications )
I am not sure if anybody in this forum will help you with the electronic design though.

[beej2]

Dr. Imad,
Thank you for the response.
I will digitally filter within Labview in addition to the analog filtering. The environment (paper mills) is notorious for wide band EM noise which I have experienced. (From several hundred HP PWM drives and the like). This is process control work where data is sometimes collected for hours, then analyzed in the time and frequency domain due to the very low frequencies of interest. I beleive I need analog filtering to so as to BE CERTAIN there is no aliasing of larger high frequency noise into the lower areas of the spectrum. I haven't tried really high frequency sampling rates. I am concerned about the amount of data gathered when collecting at high sampling rates.

I was sort of hoping I wasn't the first to not
want to spend thousands on analog LP filtering.
Barry

[Dr._Imad]

Hello Barry,

If you know that the high frequency component will not exceed say 100 Hz, scanning at 1000 Hz or higher (at least 10 times above and then using a LowPassFilter will eleminate all the noise in real time,with no computational burdens, and no aliasing.

If the type of signal your monitoring is DC or low frequency, then you may scan at lower scan rates as long as you are scanning (10 times higher the max signal frequency). Averaging several scans ( or readout where each read out is an average of several scans) will in most cases do it

(Theoretically, miniumm scanning is at the Nyquist frequency, twice the highest frequency component, but 10 times is good enough to eleminate aliasing)
With analog filtering, if you build your own f
ilter, you may not be able to build an exactly similar one (componet tolerances, among other things),
With digital filtering, once the filter is designed and tested, you will always get similar performance from day to day, and you do not have to worry about finding a replcement capacitor or resistor, that may not exactly match

[Conseils]

The basic filter design looks OK, I have built similar before. The TL064 has 4 OP amps in one chip and will reduce chip count. But what I didn't see was values for the filter elements (resistors and capacitors).

Just to clarify, no amount of digital filtering will prevent problems associated with signal frequency components which are higher / outside the system bandwidth.

Anyone sampling should fit antialias filters suitable for thier used bandwidth.

The design is quite complex, from what I remember the SCB68 will accomodate fitting of a simple (1 Pole) RC filter that would probably suffice for your purposes (Page 55). (A couple of resistors and a capactor will fit easily in the component holes provided)

Watch the corner frequency of your filter and mak
e sure that the filter is FLAT in the range of frequencys required!!!!! Are you up to making sure that the filters gain is measured and what about phase matching of multiple channels. Are you happy with the noise that will be introduced, especially if you build it on proto board without screening.

In essence I would suggest that unless you get the construction of the filter to a high standard, you would be better off using your sound card for an analogue input channel or better yet, guess the value rather than using an NI board!

Take care that the maximum POSSIBLE (Vcc - 1.5v approx) ouput voltage from the OP Amp does not exceed the maximum input voltage for the card.

[Dr._Imad]

If you check the NI web site for web events on Demand, you may check on
"
Increase the Accuracy of Your Measurements with Filtering"

http://www.ni.com/webevents/archive/#1

http://digital.ni.com/demo.nsf/websearch/0D166E066989B9EB86256E58007F1282

[beej2]

Thanks for the input!

Your comment about using a sound card or quessing rather than an NI board. Are you saying it is very hard to produce an analog filter of the proper quality, or are you saying something about the quality of NI stuff? Or both?

I assume from your comments that you agree with using the AMP02 with the three input resistors is a satisfactory method for taking the differential voltage input and getting an output referenced to Pwr Supply GND?

SOme great comments to boot. Here is some more info on what I am attempting and how and why.


I will be using TL074s as they have more than sufficient bandwidth for the cuttoffs I need (fastest is 1khz), they are CHEAP, and seem to work OK in testing where others (LM833, for instance) did not.

I did not include the component values as I will build 4 different cuttoffs and I didn't think including them was germain, I trust the TI FilterPro program and have tested a 1k filter on a breadboard and its performance matches expectations.

Your comment about a 1-pole filter being added to the SCB is correct. My thinking is that -20dB/decade is way too slow a rolloff. In process control the signal is essentially small magnitude AC imposed on large mag DC. The AC of interest is almost always .01hz or longer, and having a period of 1 hr is not uncommon. Cross correllation is always of interest and sampling of 10 - 16 signals so as to find potential interactions is commonplace. For a stopband attenuation of 80dB I would have to sample at 4 decades times the cutoff. If cutoff is 10 hz, that means I must sample at 200,000 s/sec (4 decades + Nyquist) which is not not a particularly desirable sampling rate just to evaluate frequencies in the 5 hz range! Maybe 80dB is too much to ask for considering the noise floor of the entire system.

I am not particualarly interested in phase information. Gain accuracy of the filter is not particularly important either and so I opted not include gain adjustment and left at ~unity gain.

The final filter will be implemented on a four layer PCB which, amazingly, I find I can have produced in small quantities, fast and for reasonable prices. I think, but certainly don't know because I am not an analog circiut designer, that the filter will perform acceptably. The protoboard filter scopes at 3-4mV noise (when measuring output in the stopband) on a +/-7.5V input sine wave, and I am betting that noise is independent of input amplitude, something I am going to check. My real world signals will be in the range of 3v and the AC of interest is in the neighborhood of .05v so the signal to noise is not too good,as you say. I'd like better noise performance,and hope to get it on the "production" boards. I am using 1%tolerance resistors and Class I COG caps which are the recommended quality of components. Hey, if I want top performance, I ought dig out my wallet, but then, I am not certain the performance would be that much better.

Thanks for reminding me about not maxing out the DAQ card. With this arrangement, I could do it. I guess I should adjust the gain of the first stage to make sure the output can't swing full rail.

It was a pleasure hearing from you.
You scared me at the end, with the comments about guessing. Am I in water over my head? If so, let me know and I will start saving my pennies and in a year or so I can buy this stuff off the shelf. In the meantime, I will have to fake it.

Barry

[Conseils]

My comment was to indicate (in my opinion) the very high quality of the equipment and the trouble one would have to got too - too do the data aquisition system justice. As an aside you have to use a MAC to get the full benefit as IBM compatable P.C.s PSU spec. is crud comparatively. I have P.Cs 😞

Sounds like your prepared to go the distance on the stuff and all sounds well, although I would watch the op-amps for zero-offset, thermal drift and input offset errors which will cause problems in your application. You also don't have the facility to null the op-amps with the TL074 so you would have to capacitively couple the outputs if you only want A.C. As the final output offset voltage could be more than you want and mean the gain for the A/D channel is not optimal with a D.C component, you will then have issues with the coupling capacitor.

I think you will be lucky to get better than 65dB S/N myself. Have you considerd the PSU type you will employ? What about mains A.C. born noise and how about cross channel interference. Selected your I/O connectors and shielding?

I am not trying to frighten you it's just I have built almost exaclty the same for general instrumentation purposes before. I also had the kit to check it's performance after all my hard work. 😞 B.S.D.

If you want a source for Anti-Alias filters you could do a lot worse than Kemo (http://www.kemo.com/) I have no affiliation with the company. I think you would be ecstatic with the performance!

Were it my A/D card I would want to be sure that there was no way it could exceed the maximum, by design!

Well my control systems experience is all high frequency stuff so I guess that's what I'm 'calibrated to'.

In summary, I believe that to get better than 65dB to 70dB S/N you will have done extremely well. However you may manage it if you screen extremely carefully, decouple the power supplies, shield the channels and select an analogue power supply with a low noise supply. Watch the regulation technique as many have nasty high frequency (1Mhz) components that add to the overall noise floor from many regulators.