How to implement a RNN in MATLAB?

I am currently working on developing a RNN to identify a non-linear function. I am writing my own code for back-propagation learning. My code is able to identify the a simple linear SISO function with random inputs.

• It is working only for y(t)=a*f{x(t)+y(t-1) where 'a' is a constant.
• x= 2*rand-1 ;

As soon as I change even the sign of the term y(t-1) the code fails. I am using a single layer with twice the number the neurons as the number of inputs in the network.

This is the response for y(t)=x(t)+0.5*y(t-1)

This is the response for y(t)= sin(x(t))+ y(t-1)