Now is the time to start evaluating how good your results are. The spreadsheet contains a number of features to help you with this.

The most important measure of your design is the resolution. The resolution is the smallest change in temperature the measurement system is able to detect. The resolution graph plots this as a function of temperature.

A typical best case figure for resolution is 0.2°C. Usually the best resolution (low point on the resolution curve) will occur at or near the low end of the temperature range, though invoking the shunt resistor may move the low point to the right. You need to evaluate the resolution predicted by the spreadsheet and decide if it is adequate for your application. If you are not happy with the results you may still be able to get an improvement by trying a different thermistor and repeating the process.

Accuracy is a different matter, and much harder to evaluate. If you look carefully at the Temperature versus Count graph, you will see it consists of two curves. One, drawn with a thick line, is the real raw AnIn reading versus temperature curve. The other, drawn in a thin line, is a “trendline” that Excel has calculated to try and match the real. This trendline represents the outcome of the calculations your SPLat program is going to perform on the raw AnIn readings, and is represented by the polynomial equation displayed at the top of the graph. As long as the two curves lie on top of each other, all is well and the polynomial equation will contribute negligible errors to your measurements.

Try this experiment: Add 200 to Tmax. This will force the polynomial to cover a much larger range of values and you will see it no longer tracks the ‘real’ curve. Remember to reset Tmax!

The polynomial is only one source of errors. Other sources are:

Thermistor accuracy. This was discussed earlier. The best reason for buying premium accuracy thermistors is that they are interchangeable. That means you can calibrate your system with one thermistor, and know it will work to a very close tolerance with any other thermistor of the same type.

Analog measurement errors. The SPLat analog inputs have a finite accuracy specification. This varies from board to board, but is typically in the 1% to 2.5% of full scale (2 to 6 counts). That means the possible error from this source is 2 to 6 times the resolution (remember, resolution is the temperature change represented by a 1-count change in the raw analog reading).

Resistor errors. The Rfeed and optional external shunt resistors contribute potential errors.

Vfeed errors. If you are using an onboard source of Vfeed, the error contribution of Vfeed will be minimal because the analog output used to generate Vfeed uses the same reference voltage as the analog input measurement, and any changes tend to cancel (this is called ratiometric measurement). If you use an external Vfeed (i.e. with the OEM36) then there is no longer any cancellation. That is why we suggest you use an external Vfeed only if it is from a regulated power supply.

Thermistor self-heating. The spreadsheet contains a column headed “mW” to allow you to evaluate this.

What to do about the errors?

If your application is non-critical, say heating the water in a commercial dish washer, then simply ignore the errors. You will get to within 2-3°C without calibration, so don’t worry about it.

In most applications there is only a certain temperature, or a narrow range of temperatures, that are of interest, and accuracy is important. For example, a fermentation vessel may require a controlled temperature of 33°C±1°C, so that’s the temperature that really counts. The trick then is to make a calibration adjustment to the reading at or close to that particular temperature. This can be done via an onscreen menu, using shadow memory or permanent memory to store the calibration offset. A simpler, but less elegant way is to simply add in a calibration offset determined before your program is downloaded to the SPLat board. Naturally, every system you make will need to be individually calibrated, if you need high accuracy.

If you are using a low accuracy thermistor, you will need to subject it to a precisely known temperature, than adjust the system to get the right reading.

If you are using high accuracy, interchangeable thermistors, then you can use a fixed resistor equal to the thermistor resistance at the temperature of interest. That saves having to generate a known temperature.