LM358 Differential Two Stage Amplifier Misbehaving (unanticipated output voltages)

Question

Hello all and thank you in advance for taking a look at my problem.

I have made a circuit which takes a differential voltage from a Wheatstone bridge (pressure transducer) and converts to an analog voltage to be read by an ADC.

The first stage of the amplifier takes the differential input and outputs at unity gain. Pin1 of IC4G1 is outputting ~.2 volts with no pressure applied to the transducer. I think the output of the first stage should equal the voltage measured across the output of the transducer.

On the pressure transducer, IC2, the voltage Vpin4-Vpin2 = -5mV. After cutting the traces going to the op amp, that same measurement yields 0V (as I would expect with zero pressure on the transducer.)

The second stage of the op amp should have a 100x gain but behaves like it has around 50x gain.

Here is a screenshot of our layout.

Hopefully there is something simple we're missing.

If there is any information I've missed let me know and I'll get it posted ASAP. We're in a bit of a crunch to get this solved.

Answer 1

Your differential amplifier is loading the bridge- the differential input impedance is not very high. The effect will depend on your bridge resistors and may have a temperature coefficient.

At least as important, you have a single supply, so the accuracy is going to peter out close to 0V (differential) input. The output of the first op-amp cannot go down to 0V, and especially not when it's trying to sink all that current through R74+R75 (if 2.5V is the nominal CM voltage that will be about 62uA so the 50uA sink inside the LM358 will likely be overwhelmed). Above 1V out it should be accurate. At 200mV, let alone 5mV or zero, nah. This is true of any op-amp including rail-to-rail, by the way. They don't work all the way to the rails. The LM358 is not quite as good as some more modern chips, but it does illustrate the problem well.

You would do better with an instrumentation amplifier (which does not load the bridge significantly) and a bipolar supply (which allows internal nodes to actually reach 0V). If you have some kind of special situation you might be able to use a single supply, but that depends on unstated details.

---

Edit: Here is a rough design that will work with any rail-to-rail or single supply op-amp that can handle 20V and avoids both the above design flaws, even with a single supply. The output is G*Vin(diff) + Vbias. In this case, G = 201 and Vbias = 2.0V

You could also lower the supply voltage, subject to possible op-amp limitations, which might help protect the ADC.

Answer 2

The primary problem with this design is a lack of specs for gain and offset error with accuracy on each and a calibration test.

If you had done this exercise you should observe the following design flaws ;

1. Gain error.

• If your gain error tolerance is <1% then your load should be > 100x the source R. Reading Honeywell specs the source impedance (output) is between 2k and 10k. Your Differential Amplifier bridge input impedance is 2R on each input = 40K and a differential impedance of 2R=40K. This results in an unnacceptable gain error <20%.

• Correct by using Zin >= 1M such as an INA with Rail to Rail output and variable gain =100

2. Offset error

• due to lack of dynamic range of Op Amp Bipolar types are not rail to rail
• You ought to also have a null pressure spec and tolerance error . Determine what the worst case null pressure input offset is and consider if that is acceptable or add a calibrated Offset to the Op Amp. in mV to be null out in software or with a pot.

Recommendation.

• Start with design specs before design with error tolerances and environmental limits

• verify initial design against specs , consider noise and filtering requirements

• if meets specs, then do layout.

• Choose a CMOS Instrument Amp (INA) with design specs for Gain, Offset tolerance over all environmental stress limits (Vdd, T'C, etc)

  • if you wanted to use LM358 , you need to add a Calibrated Offset before like (10mV x100) then subtract later for null pressure.
  • This may also require a pot to null 3mV max input offset and 1V output offset at 1mA for Vo(neg)

Other issue: 50x gain vs 100x ?? schematic is OK with Av=101. Component tolerances undefined. and unverified.

• I presume pressure is unipolar, but that is also undefined in missing design spec