Matlab has a curve fitting toolbox (installed on machines in Hicks, but perhaps not elsewhere on campus - as of Sept. 2007), that can be used to fit any curve to data. This page describes how to do this with data collected (i.e., from an oscilloscope).

We'll do two examples:

- Easy fit (Equation to fit is included in toolbox).
- Hard fit (Equation to fit is not included in toolbox).
- Bad fit
- Generic fit (for exponential).

Consider the data from a square wave input driving an RC lowpass circuit.

```
plot(t,vin,t,vout);
legend('Vin','Vout');
title('RC Circuit with Square Wave Input');
ylabel('Volts');
xlabel('Time, (seconds)');
```

Let's say we want to find the time constant of the circuit from the falling edge data. Immediately we have a problem not only because our data rises and falls, but because none of the falling edges aligns with t=0.

First lets get the falling edge data. By using the magnify option on the plot (or by looking through the data manually) we can find where the drop occurs. The diagram below shows the falling edge that is near t=-4 seconds.

From this diagram it is obvious that the falling edge occurs at about t=-4.34 seconds. By expanding about the next rising edge, we can see that we want the data from -4.34<t<-1.22 seconds.

```
tIndex=find((t>-4.32) & (t<-1.22)); %Get only the chosen values of time
t1=t(tIndex); %Get new time vector
vin1=vin(tIndex); %Get Vin
vout1=y(tIndex); %Get y (Vout)
plot(t1,vin1,t1,vout);
legend('Vin','Vout');
title('RC Circuit with Square Wave Input, Falling only');
ylabel('Volts');
xlabel('Time, (seconds)');
```

Now we need to adjust the data so t=0 corresponds to the first point. This is easily done by just subtracting the first time value from each of the other time values. This is shown below.

t1=t1-t1(1); %Subtract first time value, so t1 begins at 0. plot(t1,vin1,t1,vout1); legend('Vin','Vout'); title('RC Circuit with Square Wave Input, Falling only, start at t=0'); ylabel('Volts'); xlabel('Time, (seconds)');

We would like to fit the data to the equation:

To do this we will use Matlab's curve fitting tool, "cftool." To do this enter "cftool" at the Matlab prompt. You should get a window like the one below.

Let's start by importing the data. Hit the "Data" button, and then choose "t1" as the "X data", and "vout1" as the "Y data." Then hit the "Create data set" button.

Your screen should look like the one below:

Now choose "Fitting." Set your "Fit Name" to "Exponential Fit" and your "Type of fit" to "Exponential." Choose "a*exp(b*x)" as your fit.

When doing non-linear curve fitting, it is helpful to give the program as much information as possible. It is apparent from looking at the data that "a" is about 1 and "b" is about -2. So open the "Fit options" and enter these initial guesses.

Close the fit options, and start the fit, by hitting "Apply." The final fit looks good:

and we get a=0.92 , and b=-2.0 as our best-fit parameters.

In the same way as before, let's get the rising edge data, and make it start at t=0.

This time would like to fit the data to the equation:

Unfortunately this isn't one of Matlab's standard fitting equations. So... Choose "Fit Name" as "Asymptotic Fit." Choose "Type of fit" as "Custom Equations." Hit "New Equation," and select general equations. Now enter the equation we want (with initial guesses for "a" and "b"):

Now, do the fit. Again it looks quite good.

See what happens with aa bad initial guess for "b."

Using the examples above, you would use a different fit for a decaying exponential and an increasing (asymptotic) exponential. However, you know that the generic first order step response given by:

or

can be used for either both cases. This might simplify your work.

You can fit just about any equation with this tool. However, it is more successful if you have somewhat accurate initial guesses for you parameters. Sometimes this is easy, sometimes not so.