How to llinearize? sensor with an anti-log response

[swanann2000]

How to llinearize?

Hi all!

I have a sensor with a anti-log response, and i have to linearize it to determine all the points off the plot.
How i can do it?, with a Microcontroler, with a analog driver?...

Thanks in advance.

 

[techie]

linearization

it depends on the number of decades of the log that you want to use. The most practical way is to use the log-amp which is an op-amp with a transistor in the feedback loop. But they are highly temperature dependent. So you have to compensate.

If you can tell me about the application in more detail, maybe i can help.

 

[swanann2000]

Hi Techie, and thanks

It goes by 2 dec.

What do you think about doing a spline interpolation of the plot with a microcontroler to avoid aditional electronics, or it is not possible?

 

[FANT]

Linearize

With two decades i think it is quite easy to linearize it in analog way.
You can find some schematics in the applications notes of the National company; look for the application note AN30 ( Log converters) I have found it on the Linear Applications handbook at page 68; if you can tell me your e-mail or another way to send it, I will scan it and put on the network.
hope to be helpful

Mandi

 

[techie]

linearize

I agree that with only 2 decades, it should be easy to linearize by software. You will lose some resolution at one end of the curve. So probably the AD convertor will have to be of a higher resolution. If the sensor is resistive, then it would be better to use a resistance to pulse-width or resistance to frequency conversion as it gives a much higher resolution. For example, with thermistors, It have found using a simple monostable to be the easiest way to acquire non-linear data and then linearize in software. For some humidity sensors, however the scheme does not work due to the AC excitation requirements.

 

[swanann2000]

Thank you to answer.

The sensor is purely resistive, how can I carry out a monostable to acquire the signal?

And if I try to linearize using software, what do you think about splines interpolation?

 

[RegUser_2]

Yup, you can use 555 timer as a resiatance-to-frequency converter.
Why do you need spline interpolation, if the resolution is high enough for your needs (e.g the quantization error is smaller than the noise of the process you want to measure, or say the intrinstic noise of the sensor) and the linearization could be made by dirrectly solving the equation of the functional dependence of the resistance vs whatever you are measuring?

But than again the analog solution is much more elegant and simple, for 1E2 dynamic range. The thermal dependence of the
compensated log amp gain and the gain error is much better than 1%.

 

[mdamdam]

Hi, You can use a Atmel u-controler and the excellent AVRCO from E-lab.
The pascal compiler include a linearization tool for any kind of curve. You can enter a curve manually.
M

 

[techie]

monostable

A 4098 or 4538 monostable can be used to convert resistance to pulse width. The 4538 is a more precise and gives a linear reading of resistance to pulsewidth for quite a range. It fails however on reisistances less than about 1kohm. The 4098 can be used if the resistance is expected to drop below 1k. Either way you just need a stable capacitor to make the RC combination and a trigger pulse each time you need to measure. This fairly small arrangement can give reisistance measurement of a few decades easily.

 

[swanann2000]

i will try a software solution (low cost solution).

Thanks for all!!!!!!!

 

[johnyaya]

Both techniques (analog and uC) have merit for the info given. I assume that the sensor needs calibration too. I personally hate trim pots, someone is always tweaking them. I lean towards the uC with eprom to store (and update later if required) cal data for your fit.

 

[swanann2000]

i will asume the calibration with a digital potentiometre.

 

[techie]

pot

Even the digital pot has its problems. THe best way is to do a CAL in software and the store the calibration coefficients in EEPROM.