If your sensor is an appreciable distance from your amplifier, you should definitely use an instrumentation amp. They have superior common mode rejection over anything you can build. The sensor leads will almost certainly pick up common mode noise, which, in your circuit, will be converted to differential noise, which may be higher than the signal amplitude.
The sensor has 5 second response time, so the 1kHz bandwidth of the INA326 is more than adequate. In fact, you might want to add filtering on the output, although you can probably do that in software.

 

Consider the MAX4238/4239 ~ $2.50 at digikey. ~2 microvolt input offset. pay attention to pin 5 strobe pin. It needs to be tied high.

 

OK, that's not what you described.
That circuit will give you 1001*Vsig -1000*Vref.

I'm sorry about the circuit confusion. I just re-read my first post. I think it may not have been very clear. I briefly mentioned the circuit I uploaded. But I spent most of the post describing a virtual circuit I built to test the output of the OpAmp models in the CircuitMaker Student version.

If your sensor is an appreciable distance from your amplifier, you should definitely use an instrumentation amp. They have superior common mode rejection over anything you can build. The sensor leads will almost certainly pick up common mode noise, which, in your circuit, will be converted to differential noise, which may be higher than the signal amplitude.
The sensor has 5 second response time, so the 1kHz bandwidth of the INA326 is more than adequate. In fact, you might want to add filtering on the output, although you can probably do that in software.

With any luck, I'm hoping to put the entire circuit and sensor in a box not much bigger than the components. Hopefully small enough to easily fit in a small pocket. I'm pretty sure I can get the leads between the sensor and the first OpAmp down to less than an inch. I may put an LCD display on box. But I'm planning on having the end product communicate the results to a host, via Bluetooth (which may add so much noise, that I will definitely need to switch to the INA326).

 

Consider the MAX4238/4239 ~ $2.50 at digikey. ~2 microvolt input offset. pay attention to pin 5 strobe pin. It needs to be tied high.

Sounds good, I won't have a chance to look over the spec sheet until this afternoon. But I'm a big fan of the MAX line drivers (I like how easy they are to obtain and how they always seem to work they way I think they will).

I just down loaded the text of the spice model for the MAX4238. I will import it into the CircuitMaker Student version this afternoon and give it a try.

I have a lot of things to try this afternoon...

 

At what output voltage is the clip point?

It looks like it clips around 68mV. I'm including PDFs of the Circuit, a sine wave from the LT1077 (in the circuit), and a sine wave from the LT1677 (in the circuit).

The LT1077's sine wave looks like I expected it to look, almost exactly like the IDEAL OpAmp5's sine wave, when I use it in the circuit.

The LT1677's sine wave looks really good above 75 or so mV. But below 75mV (down to around 68mV) it appears to be clipped off.

It is this clipping effect, that I would like to be able to identify by looking at the spec sheet. And not only be able to tell it will exist by having to put it in a virtual or real circuit and then monitor the results.

 

In post #3, I told you what to look for in the datasheet:

There are NO op amps whose outputs truly swing rail-to-rail, or even all the way to ONE rail. The LT1078 comes pretty dang close to the negative rail (Ground, in your case). Look at "output voltage swing".
On the input side, there are op amps that can handle rail-to-rail signals, or even higher (or lower). Look at "input voltage range" or "input common mode range".

 

There are NO op amps whose outputs truly swing rail-to-rail, or even all the way to ONE rail. The LT1078 comes pretty dang close to the negative rail (Ground, in your case). Look at "output voltage swing".
On the input side, there are op amps that can handle rail-to-rail signals, or even higher (or lower). Look at "input voltage range" or "input common mode range".

I think this just sunk in... By comparing the output sine waves from the LT1077 and the LT1677 (and a couple of others) and looking at the data sheet values you mentioned, the cut offs at the top and bottom ends are making sense.

It looks like the "Output Voltage Swing" aren't values to be applied to the formula, they are literal maximum and minimum voltages the OpAmp can output, based on the load the signal is driving. When the sine wave approaches one of the "Output Voltage Swing" limits, the sine wave flattens out, and the voltage stays within the "Output Voltage Swing" limits.

Based on the "Output Voltage Swing" values, it looks like I am pushing the LTA1077/8/9 right to its "Output Voltage Swing" limits. And that would explain why, this afternoon, when I was attempting to do the amplification before I applied the differencing voltage to bring the low end of the sine wave back to 0, I was experiencing clipping on the top end of the sine wave.

Tomorrow afternoon, I will attempt to take the "Output Voltage Swing" limits into account when I attempt to modify the circuit to do some of the amplification prior to the voltage differencing.

Thanks!

It looks like I just got lucky when I imported the LT1677 OpAmp spice data into the CircuitMaker Student version. I wasn't so lucky with the INA326 or the Max4238. It appears as if the CircuitMaker Student version isn't designed to have additional modules imported into it. So I guess I was only able to import the LT1677 because I got lucky and chose the right spots to add it descriptions. And probably also, since it was an LT device, and most of the other OpAmps also appear to be LT devices (maybe the format of the LT1677 file, just happened to be compatible).

 

If you want help with simulations, switch to Linear Technology's LTspiceIV. Another big advantage is, all of LTC's op amp models are in the library. We can also show you how to add models.
Attached are results of some simulations, along with the .asc files, which will run if you download them to LTspiceIV.
Note that I added a 1k/1Meg attenuator to the input, to make gain=1000*(Vin-Vref), instead of gain=1001*Vin-1000*Vref.
I ran the INA326 sim with a 100Hz source, because the sim runs VERY slowly. With a 4Hz source, it would have taken all day, and possibly used hard drive space for additional memory (although I doubt it). Be aware that, for the INA326 sim, it takes a LONG time for the simulator to find the operating point before the sim actually starts.
Be aware that, in your schematic, you will probably not be able to get 8mV from the output of your voltage follower. If you choose to go that route, you might want to forego the voltage follower. The Thevenin resistance of your voltage divider can be very low, and you can compensate for it if you think you need to.

 

If you want help with simulations, switch to Linear Technology's LTspiceIV. Another big advantage is, all of LTC's op amp models are in the library. We can also show you how to add models.

Last night I played a little bit with LTspiceIV. I found a couple of really simple schematics I had done with it previously. I will give it another try today or tomorrow (unfortunately, this afternoon is looking a little to busy for Me time).

Be aware that, in your schematic, you will probably not be able to get 8mV from the output of your voltage follower. If you choose to go that route, you might want to forego the voltage follower. The Thevenin resistance of your voltage divider can be very low, and you can compensate for it if you think you need to.

Early this morning, I re-worked the circuit. It made a lot more sense after I understood the effect of "Output Voltage Swing" numbers.

I added a 10x preamp OpAmp to the Sensor line. I let the 8mV offset go through the 10x preamp and get multiplied to around 87mV. I upped the 8mV voltage divider reference signal to 87mV and moved it to a second amplifying OpAmp, after the 10x preamp OpAmp. The second amplifying OpAmp uses 1k and 85k resistors, to up the output voltages to pretty close to the entire "Output Voltage Swing" range of the LT1079.

The end result looks pretty clean. But it is definitely not as clean as the INA326 circuit.

Attached are results of some simulations, along with the .asc files, which will run if you download them to LTspiceIV. Note that I added a 1k/1Meg attenuator to the input, to make gain=1000*(Vin-Vref), instead of gain=1001*Vin-1000*Vref.
I ran the INA326 sim with a 100Hz source, because the sim runs VERY slowly. With a 4Hz source, it would have taken all day, and possibly used hard drive space for additional memory (although I doubt it). Be aware that, for the INA326 sim, it takes a LONG time for the simulator to find the operating point before the sim actually starts.

I have to say, I like the looks of the INA326 simulation and circuit. It looks a lot better than any of the LT1077/8/9 circuits. I will break out LTspiceIV and play with it, when I get my next Me time.

 

Don't get too excited about the INA326 results. It appears to have compression on the bottom end. It is just softer than the op amps.
One big advantage on the input side is the ability to adjust the reference without affecting the gain. You can't do this on the input side of the op amp circuit because you can't get an op amp whose output will go down to 8mV.
I think I would use an instrumentation amp (INA326 or MAX4238) and offset the input enough to keep the output out of saturation, then correct for the offset in software.

 

How near zero volt does your output need to be. As mentioned before without a split supply. It would be hard to get zero volt out.

 

How near zero volt does your output need to be. As mentioned before without a split supply. It would be hard to get zero volt out.

It doesn't need to be too close to 0V. I'm trying to get as close to 1000 meaningful samples as possible, I'm attempting to get to around 0.1% accuracy.

I'm going to try to get away with using a PICChip with a 10-bit ADC, that can accept up to 5V. The ADC has a Voltage Reference Input, that overrides the internal 5V Voltage Reference. I am assuming the Voltage Reference Input sets the top end voltage for the ADC and the ADC will return 1024 samples between 0V and the Voltage Reference.

The end application is more concerned with higher than normal oxygen concentrations, than lower than normal. In theory, the gases being sampled should never have less than 20.9% O2 in them. So some clipping on the bottom end is not as bad as clipping on the top end. Though, I would prefer to make it as accurate as possible across the entire range.

The current circuits deliver a maximum voltage of around 4.4V. So at 1024 samples per 4.4V, that works out to about 4.29mV per sample. Since I am looking for 1000 meaningful samples, from 4.4V down, that means I can throw out the lower 24 Samples. 24 Samples works out to about 103mV. If the bottom of the curve falls around .1V, without much clipping, I should be good.

Most all of the circuits so far, look like they will work. The real trick now, is just waiting for the OpAmps to arrive, so I can test the circuits on the bench.

I realize that the real world isn't usually as kind as the simulations. So if the bench tests don't come close enough to the simulations, if adjusting the reference voltages becomes more work than it is worth, or if I'm wrong about the function of the Voltage Reference Input... I do have some PICChips with 12-Bit ADCs that I can switch to. With 4096 samples, 0.1% accuracy becomes a lot easier, but not nearly as challenging, educational, or fun...

For now, everything is looking really good. I have to say, I've learned a lot from the All About Circuits forums, and after this project is done, I am sure I will be back with more questions for some of my next projects...

Funny, I always thought as I grew older and older, I would accomplish more and more and feel like I had less and less that I still wanted to accomplish... I had no idea it worked the other way around...