op-amps, bias, and integration circuits 1

[keystoneclimber]

Hello all,

I have a signal, of which the integral represents a process variable. (Fig. 1 of references.pdf) I designed a gated integrator (Fig. 2) to isolate the usable portion of the signal. I then sample this output with an ADC and scale appropriately. This works well, however the input signal can be bi-directional (Fig. 3) and the circuit doesn't support this. My first thought was to simply make everything bipolar. As it turns out though, my task of selecting an appropriate and reasonably priced ADC is much simpler if the output remains unipolar. My next thought was to apply both a DC offset to the input signal and an appropriate bias to the integrator. The idea was to get an output similar to the simulated signal in Fig. 4. I'm not sure if this is even possible. Typical non-gated integrators produce a triangle wave output from a square wave input. The gated variety produce a sawtooth wave. Here in lies the problem with my concept. Resetting the gate on the integrator sends the output to the bias point. I've yet to be able to figure out a way to send the output below the bias point. Maybe I have tunnel vision and am too focused on a single approach to the problem. I was wondering if anyone can provide some insight as to alternative solutions?

Brandon

http://home.comcast.net/~keystoneclimber/imagehost/references.pdf

 

[Noway2]

I am not entirely certain I follow what you are trying to do, but.....

A lot of amplifiers will take a reference input. This is the signal that they give if there is zero input. For you application, this means, that you could use one of these amplifiers as a unity gain follower with a 2V reference. This would then offset your signal onto the 2V reference and giving you a +/- 2V input signal range that would run between 0 and 4V to the ADC. If this is too much or too little swing, you could use the amplifier with a gain or attenuation factor in addition to the reference offset.

 

[VEBill]

If the input signal is centered around zero, and you don't really care which side of zero it is on, then perhaps (at least conceptually) you could use a bridge rectifier.

 

[DHambley]

For your process, is a negative signal important and needs to be integrated? or would a negative signal be some kind of noise or transient which you want to ignore.
If you need to integrate both pos and neg, I'd go with the suggestion from Noway2.

 

[OperaHouse]

I too am a little lost. If the negative portion is just as important, use a precision op-amp rectifier and treat the negative portion as a positive. That way you do not loose more than half the bits offsetting the input.

 

[keystoneclimber]

The positive and negative signals are both important. So, this being the case I do care which side of zero it's on since it will indicate the polarity of the process variable. Think positive and negative pressure or flow. This is simply the way the transducer outputs the signal. I'm a little lost on the suggestion of a bridge rectifier. Thinking in terms of pure AC, rather than my signal, I would get out pulsed DC. In this case I wouldn't be able to tell if the original signal was of negative polarity. I understand offsetting, biasing, and various circuit implementations to do so. I guess what I'm trying to figure out is if "negative integration" is possible. I did a simulation (a bunch actually) and have yet to be able to get it to work. The maximum transducer input signal is +- 100mV. I offset this at 2Vdc and set my gain appropriately for a 2Vdc full scale output. The results are displayed in the following...

http://home.comcast.net/~keystoneclimber/imagehost/simcircuit.png

http://home.comcast.net/~keystoneclimber/imagehost/simoutput.png

 

[xray]

Keystone- I do not see where you have put in a 2V offset on your amplifier configuration, figure 2. negative input voltage of 1 division at 100 mV/div, signal 3 purple, results in an integration up to what looks like 2 divisions at 2V/div 4V on the green signal 4 of figure 1.
In figure 3 it looks like the output is at a 0.0V and a positve 100mv from the transducer would produce a -4V which is out of the rail to rail voltage of the AD8626. Negative integration is possible but it looks like you are on the bottom rail with the output in figures 1 and 3.
Again, I am not sure where the 2V offset has been produced in the circuit on the output with a zero voltage signal in.

Best Regards

 

[Noway2]

Keystone, Check out the AD628. I believe that it will do exactly what you require. You can provide a two volt offset reference, set a precision gain of 20 and feed it your +/- signal to get an output that goes between 0 and 4V centered around 2V.

 

[keystoneclimber]

xray,

Sorry to be confusing but fig. 2 is the original circuit I built before attempting any bi-polar stuff. It has a 0 volt bias. The traces in fig. 1 are actual measurements from that circuit, and you are correct with your interpretation.

I can add a negative supply rail to that circuit but I'm not sure it will help because the lower limits of integration have been (in my experience) somehow limited by the bias. I used a negative rail in the simulation. The bias is set via the non-inverting input with a voltage divider, or by simply grounding if no DC bias is required. In the case of the first circuit, I use some feedback from the output for this which allows a linearity correction of sorts for distorted input signals. In the more recent circuit simulation, I've removed this and opted for the voltage divider configuration for simplicity while working out the bi-polar issues.

What I've found is two fold. First I've been unable to drive a signal more than about 0.5V below the bias point. Second, whenever the integrator is reset, the output is driven to the bias point. Both of these conditions can be seen in the simulation output on the red trace.

I'm not sure the AD628 will work in this situation since it is internally configured as a difference amplifier. It seems like it would be difficult to configure as gated integrator.

Again, thanks for the help. Any more ideas?

 

[xray]

keystone,

I think your whole setup is fine and a gated integrator is working. I think that you must have a 2V bias on the output of the ad628 with no input signal or 0v input. Instead of grounding your linerarity 50 pot terminate it into a 2V regulated supply.

Best Regards

 

[Noway2]

keystone,

I am curious, is there any reason you don't do the integration in software?

There are plenty of very good integration algorithms, even for dirt cheap 8 bit micros.

 

[keystoneclimber]

One idea I have is to duplicate the integrator with a unity gain inverting amplifier on the input. I can use a dual analog switch for the resets and a dual channel ADC to monitor both outputs simultaneously. The dual switch and triple (most likely quad) op amp packages don't add too much to the cost, but the dual ADC adds a significant amount. I was thinking that I could probably use a bi-polar ADC to do this using only a single part. Any other thoughts?

 

[Noway2]

I think you are making this too complex. In your original post, you suggested that possibility that you are getting tunnel vision and focusing too heavilly on a single approach to the problem.

Instead of answering my question as to whether or not you had considered performing the integration in software, which personally I think has several advantages, you responded by suggesting an even more complex circuit.

Your latest post has you looking at dual ADCs, and dual analog switches, in addition to a integrator stage. This sounds a lot worse that the bipolar ADC that you were so desperately trying to avoid.

I don't have enough information on exactly what it is you are trying to accomplish to fully understand your application. However, from what you have described, it seems to me that it would be much simpler to physically offset the transducer output onto a bias that represents the midpoint of your ADC and then process it digitally in software. Otherwise, as I suspect you are starting to see, you will be running into a complex circuit full of precision components that will more than likely suffer from parasitic effects and noise loops.

 

[keystoneclimber]

Noway2,

Thanks for your insight. Sorry for not addressing your question. This could be done in software and that is a good suggestion. I would however like to avoid that approach. I don't consider any of the circuitry that I've presented to be complex, and certainly not enough to justify coding calculus algorithms in firmware. Also, their are some timing restraints that make this approach undesirable. In a different scenario though, I would certainly consider this approach. Thanks again.

 

[Noway2]

If you are comfortable with the circuit that you have developed and don't feel that it is too complex, then perhaps it is the way to go.

If you change your mind and decide to look at a software approach, I would recommend reading Jack Crenshaw's book Real Time Math Programming. Once I read it, I was truly amazed at how little processing power (read few instructions) it takes to "code calculus algorithms" to a high degree of accuracy.

 

[itsmoked]

There's a star for an excellent couple of suggestions.

Keith Cress
Flamin Systems, Inc.-

http://www.flaminsystems.com