How can i make a training of square pulse to neural network ? how

when i am training a square wave to a neural network ,network reply response below

</matlabcentral/answers/uploaded_files/24263/111.JPG> figure 1 : response of training of network ; figure 2: square pulse train ; why don't have same network's response and pulse train? code: fs = 100000 t = 0:1/fs:5; x2 = square(2*pi*t); net = newff([-1.5 1.5],[5,1],{'logsig','purelin'}); net= train(net,t,x2); Y=net(x2); subplot(211) plot(t,Y) axis([0 5 -1.2 1.2]);

subplot(212) plot(t,x2)

[Greg Heath]

Reformat so that the program will run when cut and pasted into the command line

Answer 1

[Greg Heath]

% when i am training a square wave to a neural network , network reply response below % /matlabcentral/answers/uploaded_files/24263/111.JPG

 % figure 1 : response of training of network ; 
 % figure 2: square pulse train ; 
 % why don't have same network's response and pulse train? 
 % code: 

 fs = 100000 
 t = 0:1/fs:5;
 N = length(t)           % 500,001
 x2 = square(2*pi*t); % 0 <= 2*pi*t <= 31.416
 net = newff([-1.5 1.5],[5,1],{'logsig','purelin'});

0. I don't have a toolbox with the square function. Therefore, I used x2 = sign(sin(2*pi*t))

1. You are trying to use a half of a million points to model a non-differentiable square wave over 5 periods.

2. For a sine wave

    a. As few as ~12 points per period may be sufficient
    b. At least 2 hidden node sigmoids are needed for each peak

3. You erroneously entered the input range to be [-1.5 1.5] instead of [ 0 5 ].

   Hope this helps.

   *Thank you for formally accepting my answer*

   Greg

[mahmoud shojaee]

Hi Greg,thank you that help me

I implement whatever you said;

I put:

fs = 1000;

x2 = sign(sin(2*pi*t));

net = newff([0 5],[10,1],{'logsig','purelin'});

net= train(net,t,x2);

but the network's response is in contrast phase or Vice Versa of main function *so thanks Mr.Heath

[Greg Heath]

Dr. Heath or Prof. Heath. However, I prefer Greg.

Label your plots Target and Output, respectively. Also increase the vertical range on the lower plot.

Greg

[mahmoud shojaee]

graph with increasing

the vertical range on the lower plot.