training neural network with noise input

Hi,

I have a set of inputs for training a neural net. On these inputs i added white gaussian noise for a certain Signal to noise ratio.

Then i train the network. The outputs for each SNR individual are good but now i have 3 different neural networks for 3 snr ratios.

I was training for -5db 0db and 5db SNR.

I want one network which is working correct for all SNR values.

I was thinking of combining the training inputs with noise for all SNR ratios but the results are very bad.

How is this normally done? Or is it also possible that it is not working for my application?

Edit:

Maybe i can merge different trained neural nets to one? But then i get a really big neural net i think.

thanks in advance

Answer 1

[Anshika Chaurasia]

Hi,

I would suggest you to use k-fold cross-validation to train a neural network with the combination of all three datasets.

You can use cvpartition function.

Hope it helps!

Answer 2

[soepblik]

How can i train it with a range of SNR between -5db and 5db?

so it is working for all SNR values inbetween?

I was trying to train it with just the inputs. add random gaussian noise with random SNR(between -5 and 5) on each input and then train it. But then the tool is not training anything and get zero results.

What is the best way?