Thermocouple Front End With 741: Offset and Common-Mode Errors
A final-year project uses a K-type thermocouple to measure a furnace over 0–1000 °C, giving about 0–40 mV, and needs 0–4 V at the ADC input, i.e. a gain of 100. The first prototype uses a µA741 in a non-inverting configuration with Rf = 99 kΩ and R1 = 1 kΩ, powered from ±12 V, with a 1 m unshielded twisted pair from the thermocouple whose loop resistance is about 20 Ω. The furnace contactor produces about 1 V of 50 Hz common-mode voltage on both input leads. The 741 data sheet quotes VIO = 2 mV typical, 5 mV maximum, IIB = 80 nA, CMRR = 90 dB, GBW = 1 MHz and slew rate 0.5 V/µs. Test results show the output sitting at 0.6 V with the thermocouple at room temperature and drifting slowly, plus about 0.2 V of 50 Hz ripple. A teammate suggests replacing the 741 with a "faster" op-amp and adding a 100 kΩ resistor from the non-inverting input to ground "to bias it".
1. Refer the offset voltage and input bias current to the output and compare them with the resolution needed for 1 °C. Judge the 741 and the proposed 100 kΩ resistor.
Read the worked answer
Resolution target. A K-type pair gives about 41 µV/°C, so 40 mV spans roughly 975 °C. One degree is 41 µV at the input, which at gain 100 is 4.1 mV at the output.
Offset. VIO = 5 mV maximum × 100 = 0.5 V at the output, which is about 12% of full scale and matches the observed 0.6 V sitting at room temperature. Referred to temperature, 5 mV is equivalent to about 122 °C of error - unusable for a furnace controller.
Bias current, existing circuit. 80 nA × 20 Ω = 1.6 µV, i.e. 0.16 mV at the output: negligible, because the source resistance is small.
The proposed 100 kΩ. It creates a DC path but converts IIB into an error: 80 nA × 100 kΩ = 8 mV at the input, or 0.8 V at the output, making the offset worse. The resistor is the wrong fix and should be removed.
Verdict. The 741's offset, not its speed or its CMRR, is the dominant error. The part is unacceptable unless the offset is nulled with the balance pins, and even then offset drift (typically 15–20 µV/°C) would add several degrees of error over the furnace's ambient range. A low-offset, low-drift amplifier or a chopper-stabilised part is required.
2. Analyse the 50 Hz ripple: is the 741's CMRR or the wiring responsible, and would a faster op-amp help?
Read the worked answer
Predicted CMRR error. 90 dB is a ratio of 31,623. One volt of common-mode input therefore appears as 1/31,623 = 31.6 µV referred to the input, or 3.16 mV at the output.
Observed error. 0.2 V at the output is about 63 times larger than the CMRR prediction, so the op-amp's common-mode rejection is not the limiting factor. The extra error comes from the common-mode voltage being converted into a differential signal before it reaches the amplifier: the two leads of the unshielded pair pick up different amounts, and the thermocouple's small loop resistance is easily unbalanced by contact and wiring resistance. This is a wiring and topology problem, not an op-amp specification problem.
Would speed help? No. At 50 Hz, a 4 V output needs a slew rate of only 2π × 50 × 4 = 1,257 V/s = 0.0013 V/µs, far below 0.5 V/µs, and the closed-loop bandwidth is 1 MHz/100 = 10 kHz. Speed is irrelevant here.
Appropriate fixes. Shielded twisted pair with the shield grounded at one end, short leads, a differential or instrumentation front end with high input impedance, a 50 Hz notch or a low-pass filter after the amplifier, and physical separation from the contactor wiring.
3. The team proposes a three-op-amp instrumentation amplifier built with 1% resistors. Estimate its common-mode rejection and decide whether it meets the requirement.
Read the worked answer
Resistor matching sets CMRR. In a difference stage the worst-case CMRR is approximately 1/(4 × tolerance). With 1% resistors that is 1/0.04 = 25, or about 28 dB - a very ordinary figure that no op-amp specification can rescue, because the mismatch creates a differential error from the common-mode voltage.
Effect at the ADC. A 1 V common-mode signal divided by 25 gives 40 mV of input-referred error, which at gain 100 is 4 V at the output - enough to saturate the ADC or to look like a 1,000 °C reading.
How much is needed. The 0.2 V observed ripple must fall below the 4.1 mV resolution target, so the front end needs at least about 48 dB of CMRR, and preferably 80 dB for margin. One-percent resistors fail this by a wide margin; 0.1% resistors give roughly 48 dB worst case, which is marginal, and 0.01% or trimmed networks are needed for comfort.
Practical recommendation. Use a monolithic instrumentation amplifier such as the INA128 or AD620, whose laser-trimmed internal resistors give CMRR above 100 dB at DC, and place a differential low-pass filter (for example 10 kΩ and 1 µF, cutoff about 16 Hz) ahead of it. Keep the first-stage gain high, add cold-junction compensation, and drive the ADC differentially where possible. The 741-plus-1%-resistors plan should be rejected on the numbers.