[Help] Multi-channel Thermocouple Meter

I wish to build a multi-channel (say 3) thermocouple meter. Which of the following options should be opted?

1) Thermocouples -> analog multiplexer -> one instrumentation op-amp -> ADC

2) Thermocouples -> each thermocouple connected to one instrumentation op-amp -> analog multiplexer -> ADC

3) Or, other better options ...

Thanks

Why do you change your position of blocks in your system.

Using ADC before analog multiplexer. It is reality.

So, do you mean the following topology:

Thermocouples (TC) -> Each TC connected to an instrumentation amplifier (IA) -> Each IA connected to an ADC -> Digital mulitplexer -> Processing Unit

Both of your designs in the original post are valid and can only display one TC measurement at a time. The slowest component in your design will be the display. ADC response time will not be a bottleneck.

First ask:

1. Which designs perform to specification.

2. Of those, which is cheapest. (Because, if you don't ask it yourself a boss or your client someday will.)

In your case, your option (1) uses three active components. But option (2) uses five (adding two relatively expensive ADCs to the cost). Option 2 might make sense if and only if you need to independently calibrate the thermocouples themselves. But I have never seen that on a commercial TC meter.

Sorry, but "hoangthanhtung" is not making sense. How does valid data flow if the ADC is in front of the analog multiplexer?

The design in your second post is even more wasteful. Now you have seven components for basic conversion instead of three.

You had it right to begin with. Option 1 of your orignal post is good engineering. Simplicity is a beautiful thing.

Extended question ...

If I wish to build a thermocouple meter with 0.5 deg celcius resolution, do you think 2-layer PCB is enough? Do I need to go for higher number of layer, say 4-layer?

I've never built one before, but I really wish to built one (multi-channel) to monitor my system. If I'm not mistaken, the output voltage of J-type or K-type is in the range of uV (micro-volt). So, I don't know whether 2-layer PCB is enough or not for 0.5 degree celcius resolution.

I cannot think of any reason why thermocouple measurement accuracy would be affected by the number of copper layers on your interface board. Go with a 2-layer board if cost is a primary factor!

onemilimeter said:

I wish to build a multi-channel (say 3) thermocouple meter. Which of the following options should be opted?

1) Thermocouples -> analog multiplexer -> one instrumentation op-amp -> ADC

2) Thermocouples -> each thermocouple connected to one instrumentation op-amp -> analog multiplexer -> ADC

3) Or, other better options ...

Thanks

Click to expand...

Hey onemilimeter,

Why not use something simpler - like Dallas Semi DS18S20 sensor - those use 1W bus and you can have an array of hundreds of sensors and read them from Linux - each chip appears as a directory within a virtual file system - then you read the actual files from bash or c++ or whatever - it is super simple. Each sensor has unique lasered serial # and that is the name of each file.

Take a look at http://owfs.sourceforge.net/

Then you can build your array of sensors - each sensor is about $3 bucks



Then with something as simple as MRTG or RRD_Tool you can build beutiful graphs from cgi perl script - they look really cool:

r u going to interface with microcontroller , then u can choose controllers with inbuilt amplifiers with programmable gain .
Check www.silabs.com . they have controllers with 8bit to 24bit ADCs (8051 variant)

onemilimeter said:

Which of the following options should be opted?

Click to expand...

Option 1 is best, with these modifications:

Thermocouples/CJCSensor/VoltageRefs -> analog mux -> one instrumentation amp -> adc

First, you must have a cold junction compensation sensor because the tc "cold" end is not at 0°C. You must measure this sensor, so you may as well share the same measurement circuitry that is used for the tc.

Second, your circuitry will drift with time and temperature, so you must somehow calibrate it. Instead of using trimpots (which are problematic at best), provide precision voltage references which can also be switched through the measurement circuitry. Now the cpu can normalize all tc and cjc measurements against the references and automatically correct for any gain and offset errors. No trimpots!