MCP9801 issues with LSB

[jacquesdejager]

Hello guys --

The LSB returned from a MCP9801 is not correct, to do a repeated start condition is it correct in doing the following --

[temtronic]

This comment.....The LSB returned from a MCP9801 is not correct
does NOT explain what is wrong with the reading. How do you know the value returned is 'not correct'.
You'ld have to show us your code, not just the I2C section as well as expected values and returned values for at least 5 samples.
I can think of 6 different reasons even without seeing more but without more info, I can't say which it is.
The more information you supply the better, faster answers we can give you.

[jacquesdejager]

Hello temtronic --

Thank you for prompt response.

I resolve the problem, still doesn't explain why I need to multiply the value twice with the value 2^-4.

The datasheet indicate that you only need to calculate the temperature like this : Ta = Code x 2^-4 for 12 bit resolution, maybe you can make something out, but the code works now except for negative temperature.

[Douglas Kennedy]

In mathematics notation is used to represent numbers. Since we are familiar with decimal ( base 10 ) notation we often consider it the only true notation.
This is reinforced by hand held calculators that use binary coded decimal ( base 10) . The PIC is strictly a binary device. All numerical ( float integer signed unsigned ) calculations are done in binary using binary notation positive voltage for 1 zero voltage for 0. Since the registers etc are limited a space restriction is made on the binary calculation like 23 bit mantissa or 32 bit integer. These restrictions create upper and lower bounds on the notation that can be employed by the PIC. Only with integers or power series of 2 numbers can binary and decimal notations be exactly aligned. Notation of rational numbers or irrational numbers is different between binary and decimal and only approximations ( often called rounding or precision restrictions) are possible.
Your source data is binary which you manipulate using binary floating point. The source code used decimal notation EX 0.0625 the compiler will translated this notation to binary since this is a whole number power of 2 the binary and the decimal notations will align.
Within the restrictions of this notation the calculation by the PIC is accurate.
This binary notation if transformed to decimal notation will if it involves rational or irrational numbers will always be approximate ( only agreeing to the first few significant digits) .
Now a pic could do everything in BCD but it really needs compiler support and would make the code bulkier and slower. Next best is fixed decimal support in which rational decimal notation is promoted to binary integer notation and the decimal point is tracked separately. Since binary and decimal notation are aligned for all integers it avoids any differences between binary notation and decimal notation.

[jacquesdejager]

Hello Douglas --

I understand the notation systems, but why should I do the calculation twice to get to the right answer, is it better to convert the answer to BCD then ?

Looks like this is what Microchip did in the development kit for MCP9801, but with 7 Segment displays.

Thank you
_________________
The Definition of an Upgrade: Take old bugs out, put new ones in.

[temtronic]

Have another look at the MCP9801 data sheet,especially the Ambient temperature register information.You'll see the lower 4 bits of temperature data in the 'lower half' of the 'temperature reading' are actually the high nibble. The raw reading must be right shifted by 4, then some math done on it to convert to actual temperature, based on resolution that you've chosen.

[jacquesdejager]

Hello temtronic --

You are right, all work out well now !!