Thermocouple Measurement Circuit Noise Problem

[dkaikas]

Hello to all,

I am designing a system for analog measurement of RTD, Thermocouple & 4-20mA with 2 inputs, using Sigma-Delta ADC converter.

The system is powered from 9->28VDC. This voltage is converter to 5VDC using OKI-78SR switching power supply module. The 5VDC is then fed to BQ24070 LiIon battery charger. This battery charging IC provides 4.4V on the output or the LiIon battery voltage when main power supply is off. Finally, a linear LDO is used to provide 3.3V from 4.4V.

I am facing the following problem and behaviour of the circuit:
- RTD measurement noise is on the order of 0.02 degrees Celsius, which is fine and represents 1 LSB of the ADC. The noise remains the same using battery, external linear 12V power supply (supply A), external switching mode power supply (supply B), and another external switching mode power supply (supply C).
- Thermocouple measurement noise is on the order of 1 LSB using battery, external supply A, supply B, BUT NOT with supply C. When the system is connected to supply C, it measures Thermocouple temperature with an offset of about 3 degrees Celsius when battery charging is off and 30 !! degrees Celsius when battery is charging with about 850mA. Also, there is very high noise except from the offset !!

From the above, I assume that the problem is the noise in power supply which passes to the analog domain of the board. I have noticed that between power supply B and C, there is one more small transformer on the primary side of supply B (the "good" one). Both power supply are "budget price" 12V/1A.

Some facts that might help (using supply C):
- When I connect the system to PC through USB or Serial port (actually connecting PC and system GND together), the problem disappears as long as they remain connected, no matter if the battery is charging
- Using second LDO to produce A3V3 from 4V4 did not solve the problem.
- Removing U14 (DG2731/33 analog switch) solves the problem, so I assume that noise passes from 3V3->A3V3->DG2731/33->AINx+ and AINx-.

The weird thing is that RTD measurement is not affected by the power supply in use but Thermocouple measurement does !!!

The design files:
**broken link removed**
**broken link removed**
**broken link removed**
**broken link removed**
**broken link removed**

Any ideas would be very appriciated because I am struggling to find the problem a long time now. Thank you very much.

Dimitris.

 

[FvM]

The schematic is unclear in several regards. How do you connect thermocouples to the differential input, how's a suitable common mode bias achieved?

 

[dkaikas]

The schematic is unclear in several regards. How do you connect thermocouples to the differential input, how's a suitable common mode bias achieved?

Thanks FvM.

The circuit has 2 analog inputs in which we can connect RTD, TC and 4-20mA.

J3 connections:
1. Analog input 1+
2. Analog input 1-
3. Analog input 2+
4. Analog input 2-
5. Analog inputs common connection (used for RTD measurements).
6. System GND (used for 4-20mA measurements)

I have attached drawings for connections based on sensor type.

For TC measurement, U13 AD7792 generates and connects the bias voltage (AVDD / 2) on AIN1(-) or AIN2(-). U13 has internal reference.
For RTD measurement, equal excitation currents are generated by U13 on IOUT1 and IOUT2. These currents are fed to Channel 1 or 2 through U14, which has 2 x SPDT switches. The equal currents cancel wiring resistance on the measurement and return through J3-5, where they produce the reference voltage on R9 for RTD measurements. The 100nF was added because the same noise problem existed also on RTD measurements but canceled by filtering reference voltage also with this capacitor.

SJ1 on U13 is just a jumper made up of pads on the pcb to account for DG2731 and DG2733 differences, so assume that 2-3 terminals are shorted.

For 4-20mA measurement, S4 and/or S5 shorts 2-3 and 5-4 terminals so the current is converted to voltage. Voltage is added through 33 Ohm resistor on the (-) side of the input because U13 needs any absolute input voltage to be at least 100mV above GND for correct offset and gain results.

All resistors on analog paths AIN1(+/-) and AIN2(+/-) and U14 are used for current limiting except the ones together with capacitors just before U13 inputs, which are used as low pass RC filters on signal.


Some more facts about the circuit:
- Regarding any possible ground problems. Internal plane 1, is the GND plane. It is separated from the GND to the rest of the board and connected only on the upper left corner as shown in the picture. Although I have set the minimum clearance from GND plane to 0.7mm, I believe that the resulting area is sufficient for the analog node which draws about 5mA maximum during conversion. Please correct me if I am wrong. But I am not sure if I have joined analog node GND plane to the rest of the PCB GND plane correctly. The connection path remains on the internal plane and it is 5mm wide.
- I noticed that if I switch on a CCFL light, the circuit picks up noise during light startup.
- By removing U14 (analog switches) the problem is canceled and thermocouple measurement becomes also very stable on CCFL light EMI.

We can assume that U14 injects all the noise to the signal path. But then it is connected to the same VCC plane as U13. Is it possible to inject so much noice from power supply on the signal path while U13 is not ?

We can also assume that the noise on VCC exists on all signal paths and is cancelled at U13 because of the differential design of the measurement. But then why analog switch inserts noise ? Does this has to do with phase change issue or any mismatch between the two indepentent analog switches that creates differential errors ?


Dimitris

- - - Updated - - -

All resistors on analog paths AIN1(+/-) and AIN2(+/-) and U14 are used for current limiting except the ones together with capacitors just before U13 inputs, which are used as low pass RC filters on signal.
Dimitris

I mean current limiting for overvoltage protection.

 

[FvM]

TC uses a higher gain than RTD, so it's quite normal that it's more sensitive to interferences.

The observation suggests a problem of common mode interferences, e.g. capacitively coupled to the TC input. Supply C might be floating against ground with high common mode voltage unless the system is earthed through USB connection.

What's the integration time used for TC measurement? Usually a long period like 100 ms that's also an integer multiple of mains frequency period is suitable.