Sensor Calibration Problem? Help needed

You did not say how you power the Arduino, but imagine this. The power has to dissipate either through a regulator to the processor or to other things as well. If you look at https://en.wikipedia.org/wiki/Voltage_regulator it will mention temperature coefficent which changes based on load. More load less heat near the regulator. Question is the sensor near the regulator? This may affect you. Also temperature sensor needs calibration. See Temperature Sensor Calibration Resources which can be affected by nearby things...

Clem

Ok let me elaborate more.. The board/shield i am taking about  has a buck converter configuration which drops the voltage of the battery from 11.2v to 9v. this 9v will be used for driving motors.(the issue of incorrect readings persists even when i am not driving the motor).This 9v is then dropped by LDO 1117 regulator. This 5v is supplied to two sensors one of them is a LM35 other i a MQ series gas sensor.  I have a lead that goes from the board to the barrel jack on arduino. the arduino Uno has a xbee shield over it. The sensors are too far from the regulator. After taking into account the 10mv/degree rise and  ADC conversion i got this and the values are fairly near to the actual temperature when the Arduino is powerd using USB which is connected to usb of my comuter. so the algo sould be right.

  temp = analogRead(tempPin);

  temp = temp * 0.48828125;

Is there any other way to  calibrate LM35?

deepankarmaithani Hi Deepankar

for clarification:

What range of temperature are you measuring (hence what range of analog voltage are you working with)?

How are you defining your analog reference for the ADC?

Neil

Hi @Neil Kenyon

i am measuring temperature of range +55 -+60 degree. I am using 5v as ADC reference voltage.  here is what i did. to find out the step size ,for a 10 bit ADC with 5v as refferenc. stepsize= 5/1024 =0.0048828 = 4.88mV

from data sheet of LM35 we know for every 1 degree C rise  there is increase of  10mv in the output.

temperature= (analogRead(A0)*4.88mV.)/10mv

and it should finally be  temperature= analogRead(A0) *0.488

Can you post a schematic of your system and ideally a photograph as well. It sounds to me as if you may have a common path for some  supply current mixed up with your temperature signal - but without a detailed and accurate schematic it isn't possible to even guess at where it might be.

MK

deepankarmaithani Hi Deepank

I don't know the answer to your precise problem, but I can try and shed some light on overall accuracy:

If your MAXIMUM temperature is 60 celsius, then your MAXIMUM measured voltage is only 600mV

With the default Analog  Reference Voltage of 5V, you are only using just over 10% of the ADC range.

This link: http://www.arduino.cc/en/Reference/AnalogReference

Explains how you can use the internal Analog Reference Voltage to increase the accuracy of your measurements. I have used the INTERNAL (ie: 1.1V) when working with a UNO - this increases accuracy by a factor of just under 5.

However, note this:

" After changing the analog reference, the first few readings from analogRead() may not be accurate."

I was initially caught by that problem.

Finally - and sticking my neck out here -  is it possible that the Vcc derived from USB power and the Vcc derived from the power jack have slightly different values - but still within spec.? If so, how might that impact on the default 5V ADC reference? Perhaps someone else in the community knows the answer.....

Hope this helps

Neil

As far as the 5V reference for the UNO, yes it is using the 5V supply and therefore inherently unstable and prone to interference from other things including the CPU itself as it varies its load, coupled with the fact that USB is rarely 5V and can be as low as 4 it should be relegated to experimenting only.

Switching to the internal BandGap 1.1V reference will increase the resolution of your measurements but should not be confused with accuracy. Im not sure how accurate the BandGap reference is but that plus the base accuracy of the ADC and then the tolerances of the sensor all together will define your accuracy and linearity.

In most cases ADCs don't need to be accurate, nor does the VRef. But they do need to be stable over temperature, time, load and supply changes. The rest can be compensated for in software.

Neil is correct in suggesting the 1.1V Internal reference though as it will increase your resolution to ~1mV instead of ~5mV and as the band gap is more stable than the supply it will provide more consistent readings over time (It will mostly ignore the changes to running the CPU on 5V vs 4V for instance).

If you want to increase the resolution even further then (As you dont need fast samples per second) you can try over sampling where you take say 16 measurements, add them all together and then shift the bits 2 bits to the right. This will increase your effective resolution to 12Bits instead of 10 (300uV instead of 1mV) resulting in a resolution of 0.03dec C per ADC count.

you can read all about over sampling here:- http://www.atmel.com/images/doc8003.pdf

As far as the 5V reference for the UNO, yes it is using the 5V supply and therefore inherently unstable and prone to interference from other things including the CPU itself as it varies its load, coupled with the fact that USB is rarely 5V and can be as low as 4 it should be relegated to experimenting only.

Switching to the internal BandGap 1.1V reference will increase the resolution of your measurements but should not be confused with accuracy. Im not sure how accurate the BandGap reference is but that plus the base accuracy of the ADC and then the tolerances of the sensor all together will define your accuracy and linearity.

In most cases ADCs don't need to be accurate, nor does the VRef. But they do need to be stable over temperature, time, load and supply changes. The rest can be compensated for in software.

Neil is correct in suggesting the 1.1V Internal reference though as it will increase your resolution to ~1mV instead of ~5mV and as the band gap is more stable than the supply it will provide more consistent readings over time (It will mostly ignore the changes to running the CPU on 5V vs 4V for instance).

If you want to increase the resolution even further then (As you dont need fast samples per second) you can try over sampling where you take say 16 measurements, add them all together and then shift the bits 2 bits to the right. This will increase your effective resolution to 12Bits instead of 10 (300uV instead of 1mV) resulting in a resolution of 0.03dec C per ADC count.

you can read all about over sampling here:- http://www.atmel.com/images/doc8003.pdf

Thanks everyone ,I will incorporate the suggestions tomorrow and will get back to you. I think this should solve my issue