ADC conversion simple question :)

Although you are getting meaningful readings from the ADC, I suggest you to have 2 functions: init the ADC only once, at the start of your readings, and then simply get every reading, without calling CloseADC().
 
A reading of 625 for your setup suggests a room temperature of 32°C, is that the room temperature when you get a 625 reading?
 
The conversion from the sensor 'Kelvin' output voltage to Celsius temperature would be:
 
Tcelcius = (Vout - Vzero) / 10mV
 
where Vzero is 2.7315V, Vout is the actual voltage reading, and Tcelsius is the equivalent temperature in celsius degrees. For example, if Vout is 3.051V, Tcelsius would be = (3.051 - 2.7315) / 10mV = 31.95°C
 
Now, the ADC has the following ideal transfer function, for a right-justified ADRES, and 5V VREF:
 
ADRES = Vin / ((Vref+ - Vref-) / 1024) => Vin / (5/1024) => Vin * 1024 / 5 => Vin * 204.8
 
So, our 3.051V would ideally be converted to: 3.051 * 204.8 = 624.84 => 624
 
The conversion of the ADRES code to celsius temperature would similarly be:
 
Tdigital = (ADRES - Tdzero) / Td_1C
 
where Tdzero is the ADC theoretical digital code for 2.7315V (559.41), and Td_1C is (2.048), the theoretical digital code for 1°C.
 
So, our value 624 would evaluate to = 31.54°C. The error (0.41°C) is due to the quantization error of the LSb size of 4.883mV/LSb, meaning that for your 10mV/°C analog range, the ADC has a resolution of 0.4883°C/LSb.
 
The above formula has fractional values for the Tdzero and Td_1C constants, and requires that you process it with floating point math, or with enough fractional length fixed point math, to avoid rounding errors.
 
But you are one lucky guy, because if you do process that formula with integer math only, with the constants rounded to integer, you get the rounding bias to almost cancel each other. The rounded final values will have a 0.5°C non-linearity. That can be bad or not for your application, you have to decide. For example, if we evaluate the above formula entirely in integer it will become:
 
Tdigital = (ADRES - 559) / 2
 
So, our value of 624 would be: = (624 - 559) / 2 => 32°C.
 
Therefore, for your circuit, the above integer formula will get you a linear range of temperatures with an error of ±0.5°C.
If 1°C accuracy is good for you, that is the simplest conversion formula that you could have: just need to get the ADRES value in 16bit, subtract 559 from it, and shift right one position.
 
 

j_doin,
First thank you very much for detailed explanations,

A reading of 625 for your setup suggests a room temperature of 32°C, is that the room temperature when you get a 625 reading?

Although the cheap thermometer I have in hand might not be preious, but it seems the temperature at the point sensor is put was almost 32.

Curently it shows 31/30, and ADC value is 621/622 while my voltmeter reads sensor in-circuit 3.03v.  So I guess ADC is reading almost correctly.

I used this formula before reading your reply:

int t;
signed short  t1;
t1 = (short)(t*(0.48828125)-273.15);

Why I used that formula temporarily was that having 5000/1024 (to have it in milivolts), was making floating point problems.  So tried to have a better number to not loose the important values.

using fprintf(_H_USART, "%f", t1) is where I face  a worst problem, and I guess it is because I am  not yet enough experienced with C18.  I simply can not print a floating point number to serial port.

I hope you also help me with this, as all of type castings etc. didn't help either, the only way I got some results was to use %d for t1, which is obviously not the best way.

The 0.5°C precision is quite enough for my purpose in this case, but being able to do it better helps in later cases ofcourse.

So I guess the current question is:

1- How to calculate the formula and have it in Floating point format (float, etc.),

2- How can I fprintf a floating point in C18?

3- How can I convert a floating point to a string in C18?!

to Artic:

Thanks you too, the Vdd in this pproject is from a 7805, and I guess it can keep itself almost fixed, but I accept that having a external Vref will give much better precision.

You can get some very good repeatable results from..

1. Using fixed floating point maths (depending on the precision of the maths [typcially no more than 24 to 32 bit is required])

I am having problems with floating points at the moment, so my own preferred method is the same,

2. Over-sampling of the ADC result (this technique should easy give you an additional 1 bit resolution)

Another factor that will limit your design is the Sensor Response Time. There is no point in running the ADC at 1000's of samples per second - if your sensor has a dynamic rate of change of 10mV per second...

Can you please explain the Over-sampling?

I only read the sensor each 500ms, so that is ok.

Cheers,

* short is not a floating point type. On C18, short is the same as int (16bit integer).

You are right, I forget to menion I have also tried that with Float, but as printf didn't show anything, I thought there might be a problem with float casting, and changed it to short to see some output at first.

Also thanks for explaining the over sampling, It will be for sure helpful for more precise conversions.  The sensor I am using at the moment has only a 0.5°C precision so the current method seems to be enough.  Now I am getting the voltage with  9 digits after point like this: fp_a=30.560943616  Which is amazing.

You can't print floating point with the printf() function family. There is a discussion on how to print flating point numbers at this thread: http://forum.microchip.com/fb.aspx?m=122316.

Thanks for your great hints also on printing the floating point number, and I didn't know printf can not print floats in C18, so it was the main problem in my case.

Thanks alot,
Cheers,

... or use a precision 5V regulator for VDD, like the LP2950ACZ-5.0, a 0.5% low-dropout precision regulator with the sme pinout as the LM78L05. Alternatively, you can use the LP2950-5.0 as a VREF+ source also. For a 0.5% application, this may be enough to have uncalibrated accuracy.
 
 

ORIGINAL: Paymaan

I checked the LP2950ACZ-5.0 datasheet, typical applications differs from 7805 and it only outputs 100mA, which is not much for the rest of circuit. 

Not much? What is your circuit that draws over 50mA? Is the PIC driving any high-current loads directly from its pins?
 

On the other hand much Voltage Refrences are at 1.2, 2.5 or around those voltages, can you please suggest something above 4V, which is also high precision?

I use the Analog Devices REF198 as VREF+ in my analog measurements applications. It is a 4.096V ±2mV guaranteed. That is a 12bit accuracy voltage reference. The nice thing about it is that you get a 4mV quantization at 10bits, and 1mV quantization at 4x oversampling.

And to have a better output, what about using LM135A/235A which have a much better uncalibrated accuracy?

If you want better precision, you can use the LM35A, a 10mv/°C linear sensor with 0V = 0°C. It is the best sensor from National analog temperature sensors family, with guaranteed min 0.5°C accuracy at 25°C, and a maximum of 1°C error at temperature extremes (-55°|+150°C).
Its typical accuracy and max absolute nonlinearity is 0.25°C, and relative linearity is 0.08°C.
 
The nice thing about the LM35 is that you can expand the sensor voltage range by using a precision OpAmp as a simple non-inverting amplifier, and have an output that is larger than 10mV/°C. With a 6V/V amplification factor you can have 60mV/°C, and have a range from 0°C to 68.2°C (if using a 4.096V VREF) with 10bits base resolution, or 0.066°C/LSb. You can make a high-resolution temperature acquisition system with that.
 
 

I see. In that case (a general-purpose board) I agree with you, use just the bare kelvin scale LM135. If you use the recommended REF198 and do a filtering of the analog input you can achieve 12bits of resolution, or 0.1°C.

You must follow good analog design practices: good filtering for the VREF regulator, good filtering for the VREF line, quiet ground plane for the analog circuitry, good decoupling capacitors for the PIC, very close to every VDD/VSS pin. With a good analog circuit board, you can achieve noise of less than one LSb at 10 bits (noise < 4mV), just enough to have a little dithering.

>>> added:
 
For calibration, please do not use any potentiometers. If you have to calibrate, use software to compute the calibration constant from the ADC values read. Use at least 2 thermal stable points (water fusion and boiling points, for example), and correct the PVT values for your ambient pressure. Remember, these sensors are very slow, so you must allow for temperature stabilization.
 
 

ORIGINAL: Paymaan

What about AD586 series instead of REF198? (I see I can obtain it faster),

The AD586 is a 5V VREF regulator, not a 4.096V. It is equivalent to the REF198 in terms of accuracy and stability, but you must pay attention to the suffix letter. Only the AD586M has a 2mV accuracy.

You may also check other excellent voltage references:

• MAX6341 (0.02%, 0.5ppm/°C)
  
• LT1461 (<0.04%, Low-Dropout)
  
• LT1019 (< 0.05%)
  
• ADR43x (ADR430, ADR431, ADR433, ADR434, ADR435, ADR439)
  
• LM4140

And thanks for nice hints for Analog, I really need them, as I am very unexperienced with Analog design.  The only filtering I use at the moment is using 100nf capacitors at Vdd/Vss pins of ICs.  Can you please also suggest a link on those topics to study?

So you mean potentiometers add noise as well?

Potentiometers are inherently innacurate devices. The rule is: what can be easily changed can easily uncalibrate. If you have software at your service, you should design analog circuitry with the most stable signals possible, and use software to calibrate the fixed offsets and gains of the circuit.
 
For good analog sources of theory and practical advice:

• ADI Grounding techniques collection: http://www.analog.com/UploadedFiles/Associated_Docs/217348021sect7b.pdf
  
• Understanding ADC performance specs: http://ww1.microchip.com/downloads/en/appnotes/00693a.pdf
  
• Thread with design hints and links: http://forum.microchip.com/fb.aspx?m=145543
  
• More circuit design links: http://forum.microchip.com/fb.aspx?m=121355