Temp sensor issue

Code 2 can't reliably work (in other words: it is pure coincidence):

abhishek Sabnis said:

ADC10CTL0 |= ENC + ADC10SC; // Sampling and conversion start
temp = ADC10MEM;

Also, after switching the reference on, it requires a certain settling time, so after setting REFON, you'll have to wait (30µs, according to the datasheet). Else your readings are using an instable reference.

Now to yor formula: you are using the typical values for offset and gain. However, the temperature sensor values vary widely for each individual MSP. You may have a gain of 3.55mV/°C, but it can be significantly higher or lower. And the datasgeet doesn't even give a typical value for the offset.
On some MSPs, there are calibration values available for 30°C and 85°C. With these, you can calculate offset and gain for each MSP individually (at runtime). But AFAIK the 2553 doesn't have them (calibraiton takes tiem and tiem is money and the 22553 is cheap). So you'll need to calibrate it yourself. One by one.

Thanks Jens for your reply.

I modified my software according to your suggestion.

Here is the updated code:

#include <msp430g2553.h>
unsigned int T_count;
unsigned int emflag =0;
unsigned int B_count, timeout;
long ActualTemp;
long IntDegC;
void main(void)
{

WDTCTL = WDTPW + WDTHOLD; // Stop WDT
// ADC10CTL1 = INCH_10 + ADC10DIV_3; // Temp Sensor ADC10CLK/4
// ADC10CTL0 = SREF_1 + ADC10SHT_3 + REFON + ADC10ON;
T_count =0;
DCOCTL = 0;
P1DIR |= BIT0+BIT6;
BCSCTL1= CALBC1_8MHZ; // System Clock Settings
DCOCTL = CALDCO_8MHZ; // use calibrated 8MHz settings
P1DIR |= BIT0+BIT6; 
_bis_SR_register(GIE);
TA0CCR1 = 5000;
TA0CCTL1 = CCIE; // CCR0 interrupt enabled
TA0CTL = TASSEL_2 + MC_2 + TAIE; // SMCLK, contmode
}
void VarDelay(volatile unsigned long int time)
{
do (time--);
while(time != 0);
}
#pragma vector=TIMER0_A1_VECTOR
__interrupt void Timer0_A1 (void)
{
switch (__even_in_range(TA0IV, 10)) // Efficient switch-implementation
{
case 0:
break;
case 2:
TA0CCR1 += 50000; 
break;
case 10: 
T_count++;
if(T_count >= 1220)
{
T_count = 0;
P1OUT ^= BIT0;
ADC10CTL1 = INCH_10 + ADC10DIV_3; // Temp Sensor ADC10CLK/4
ADC10CTL0 = SREF_1 + ADC10SHT_3 + REFON + ADC10ON;
VarDelay(40); // Delay 10ms, 1 ms = 400
ADC10CTL0 |= ENC + ADC10SC; // Sampling and conversion start
timeout = 1;
while (!(ADC10CTL0 & ADC10IFG) && ++timeout);
ActualTemp = ADC10MEM;
ADC10CTL0 &= ~(ENC + ADC10SC); // Sampling and conversion stop
IntDegC = ((ActualTemp - 673) * 423) / 1024;
if(ActualTemp >= 697)
P1OUT &= BIT6;
if(ActualTemp < 692)
P1OUT |= BIT6;
__no_operation(); // break point here
}
break;
}
}

Please look the code written in Red. i wait for approximatly 100 micro sec. after ref is turned on. then checking ADC10IFG bit before getting results. I am working on caliberation of MSP430G with know temperatures and adjusting formula.

One question though: you mentioned using 2MHz clock, i am using 8MHz SMCLK in my application. do i really need to change it to 1MHz or 2 MHz?

Debasish,

Cool down man. why are you getting so aggressive. I just want to make sure i am on right path or not.

In my application, i use temeperature sensor just to differetiate 2 conditions, like

if( temp >10C)

{ do this}

if (temp <8C)

{do this}

i can do it just comparing ADC10MEM directly to evqivalent temperatures. so in that case correct my ADC10 register setting, btw i already made clk to run 1MHz

ADC10CTL1 = INCH_10 + ADC10DIV_3; 
ADC10CTL0 = SREF_1 + ADC10SHT_3 + REFON + ADC10ON;
VarDelay(40); // Delay 0.5msec

ADC10CTL0 |= ENC + ADC10SC; // Sampling and conversion start
timeout = 1;
while (!(ADC10CTL0 & ADC10IFG) && ++timeout);

ActualTemp = ADC10MEM;
ADC10CTL0 &= ~(ENC + ADC10SC); // Sampling and conversion stop

Thanks,

Abhishek

Thanks Debasish for your information. Just clarifying your statments:

When i use  ADC10CTL0= ADC10ON + ADC10SHT_3+ADC10IE;  you said reference voltage of the internal temperature sensor is automatically used according to the datasheet. So, i need to caliberate ADC10MEM values for corresponding temperatures right?

Debasish,

I ran experiment with your suggestions. I read 1C as 280, 5C=285, 11C= 292 etc. if i subtract 280 insted 273 i get approximate equivalent to degC   

Debasish,

If you think its nonsense then just forget and don't answer. Its not compulsory to answer everything. But keep in mind that you are on social site.  

abhishek Sabnis said:

I ran experiment with your suggestions. I read 1C as 280, 5C=285, 11C= 292 etc. if i subtract 280 insted 273 i get approximate equivalent to degC

Hi abhishek,

As Jens-Michael said earlier:

Jens-Michael Gross said:

However, the temperature sensor values vary widely for each individual MSP. You may have a gain of 3.55mV/°C, but it can be significantly higher or lower. And the datasgeet doesn't even give a typical value for the offset.
On some MSPs, there are calibration values available for 30°C and 85°C. With these, you can calculate offset and gain for each MSP individually (at runtime).

Although AFAIK this was not always the case, the msp430g2553 does now have temperature sensor calibration values. They're stored in the segment A TLV structure. Unfortunately msp430g2553.h doesn't define identifiers to allow easy access to the ADC calibrations. Also the value of TAG_ADC10_1 specified in the family user guide is incorrect - the value to use is in the 2553 datasheet. Still, the calibration data is available should you need it.

In case anyone's struggling to understand how this ADC10MEM - 273 = tempInCelsius method works (as I was last night) here's an explanation.

As Debasish quoted from the 2553 datasheet earlier:

 VSensor,typ = TCSensor (273 + T [°C] ) + VOffset,sensor [mV]

TCSensor is typically 3.55mV/°C and 273 + T [°C]  is approximately T[°K], so that's:

 VSensor,typ = 3.55 * T [°K] + VOffset,sensor [mV]

Which means that the temperature sensor's output voltage increases by 3.55mV for each 1°K rise in temperature.

The next step is to configure the ADC10 so that a 3.55mV increase in the input voltage corresponds to an increase of 1 in the output value. With that done, each increment of the ADC10 output will correspond to a 1°K rise in temperature. That makes it possible to simply subtract an offset from the ADC10 output to get a temperature in °C. The offset used would need to account for both the 273.15 difference between Kelvin and Celsius temperatures, as well as the offset of the temperature sensor (which might not output exactly 0V at 0°K).

Debasish uses this to set up the ADC:

 ADC10CTL0 = ADC10ON + ADC10SHT_3 + ADC10IE

That implicitly sets the ADC's reference to SREF_0 (VR+ = VCC and VR- = VSS). If VCC happens to be 3.6V, then the ADC increments are 3.6/1024 = 0.00352V. Ideally VCC would be 0.00355 * 1024 = 3.635V, but 3.6V is close enough.

As an aside, it also configures the internal reference-generator voltage to 1.5V but leaves it switched off. Looking at the ADC10 block diagram you can see that the temperature sensor uses Ref_x, which is the reference-generator output. For this reason the reference-generator is automatically enabled when INCHx=0Ah, even if REFON is 0. Far from not needing a reference, the temperature sensor and ADC use two reference voltages (+ground) for this method.

It's a clever technique that allows minimal code for reading the temperature, but the following points need to be considered before using it:

1. VCC must be 3.6V and reference-generator gets enabled during conversion.
2. Factory-supplied temperature sensor calibrations cannot be used, as they are only provided for 1.5V and 2.5V ADC reference voltages.
3. It's possible to calibrate for the sensor offset of a specific chip, but not for variations in TCSensor (which may differ from the typical value).
4. This method restricts output temperature to integer values, which may be desirable for user display but sacrifices some precision.

Robert,

Thanks for thorough response. 

One question though. If i am using 3.3V can i still use this scheme.?

You could try it, but it will be less accurate than at 3.6V. The ADC increments will be equivalent to 3.3/1024= 0.00322V instead of 0.00355V. In that case, to get a 1°C change in your measured value would take only a 0.907°C change in the actual temperature.

This will also affect the offset value quite a bit. Instead of producing an ADC result of 273, 0°C would give 273/0.907 = 301 (typical)

Debasish Banerjee said:

even 2553 datasheet recommends what i did .

Debasish Banerjee said:

open page 40 see 10-Bit ADC, Temperature Sensor and Built-In VMID (MSP430G2x53 Only), can u find:

The sensor current ISENSOR is consumed if (ADC10ON = 1 and REFON = 1) or (ADC10ON = 1 and INCH = 0Ah and sample signal is

high). When REFON = 1, ISENSOR is included in IREF+. When REFON = 0, ISENSOR applies during conversion of the temperature sensor
input (INCH = 0Ah).

That isn't a recommendation to use the method you proposed. It just explains the conditions under which the temperature sensor current (I_SENSOR) needs to be accounted for.

Specifically, the temperature sensor uses the reference-generator, so if the ADC isn't using the reference-generator (ie if it uses VCC as V_R+) then the extra sensor current needs to be accounted for. If the ADC is using SREF_1 then the reference generator is already enabled so there's no need to add I_SENSOR when calculating the total current.

Yes, but it's not just offset that needs to be corrected when VCC is not 3.6 volts.

If VCC is 3.3V then a 0.9°C change in real temperature is measured as a 1°C change. If VCC is 2.2V then a 0.61°C change is measured as 1°C. With that level of error you only have to move 5°C away from the offset calibration point to get a 1°C error in the reading.

I'm not saying this is a bad method that should never be used; as I already stated I think it's a clever idea. That said, people should be aware of all the advantages and disadvantages so they can make an informed decision of whether it works in their particular application.

No, it would require a multiply or divide to correct for the slope error

Debasish Banerjee said:

But voltage Tc sensor is 3.55mV/dc when vcc is 3v...as per the datasheet says, then it is just that VSensor,typ = TCSensor (273 + T [°C] ) + VOffset,sensor [mV], why cant one just deduct the digital values for the offset from (ADC10MEM-273)?

The sensor output changes 3.55mV/°C, but if the ADC's V_R+ is VCC(=2.2V) the ADC increments correspond to 2.2/1024=0.00215V. For each 1°C that the real temperature rises, the temp sensor output increases by 3.55mV (typical). The problem is that the sensor output voltage needs to be read by the ADC, and that counts in 2.15mV steps when V_R+ is VCC(=2.2V).

This mismatch between the sensor output voltage steps and the ADC input voltage steps is what needs to be calibrated out with a multiply. If VCC is 3.6V then the multiply is not needed because the ADC steps are 3.6/1024, which is close enough to 3.55mV to give good results.

Edit: To give a concrete example, say the offset is chosen so that the measured temperature is correct at 0°C. If the temperature rises by 10°C then the sensor output voltage will rise by 35.5mV. The ADC is using VCC (=2.2V) as reference voltage V_R+, so the ADC value increases by 1 for each 2.15mV increase in input voltage (as explained above). ADC10MEM value will rise by 35.5/2.15, which is 16 (ignoring the decimal places). The measurement is correct at 0°C, but at 10°C it measures 16°C.

abhishek Sabnis said:

Please look the code written in Red.

i wait for approximatly 100 micro sec. after ref is turned on.

You don't need a timeout for the conversion. Since the ADC is an independent piece of hardware, it will even then complete its conversion if the CPU had crashed in the meantime :)
And by the seatings you make, you can exactly determine how long it will take (with some tolerance when using the ADC10OSC as clock source).

abhishek Sabnis said:

i am using 8MHz SMCLK in my application. do i really need to change it to 1MHz or 2 MHz?

Debasish Banerjee said:

as datasheet says "When REFON = 0, ISENSOR applies during conversion of the temperature sensor input (INCH = 0Ah)

Debasish Banerjee said:

.the internal temparature sensor of MSP430G2553IN20 is Kelvin calibrated

The current into the sensor isn't calibrated, it is just ensured to be constant  for a constant (reference) supply voltage. So the voltage produced by the temperature sensor has a gain and an offset. It is not ensured that the output voltage will be 0V at 0°K. It may have its zero point at 10°K or 100°K. Also the gain per °K (or °C) is not calibrated. But can be considered the same for the same reference voltage on the same MSP.
So a manual calibration is always required:

Put the circuit into your oven, measure the temperature when the oven is off with a precision thermometer and write down the ADC output. Now turn the oven on on lowest possible temperature (50°C to 80°C). When it has settled, read the precision thermometer and the ADC output again. You now know the ADC value difference and the temperature difference and you can easily calculate the offset and the gain per degree (If you want in °F or °R or °C or °K, it doesn't matter).

Debasish Banerjee said:

" VSensor,typ = TCSensor (273 + T [°C] ) + VOffset,sensor [mV]" which shows the temparature reading is Kelvin calibrated

Robert Cowsill said:

If the ADC is using SREF_1 then the reference generator is already enabled so there's no need to add I_SENSOR when calculating the total current.

Th e5x family, however, has a specific bit in its REF module that disables the temperature sensor and removes I_SENSOR from I_REF if you don't need it. So when migrating to 5x family, this somethign to take in mind for low-power operation.

Debasish Banerjee said:

But voltage Tc sensor is 3.55mV/dc when vcc is 3v...

However, taking a 'typical' value as an exact factor for a calculation, or to use it to reverse-calculating teh offset, is at least naive.