matlab - training neural network pattern recognition -

here network setup

have 100000 samples of 1d vector (1440 data points) 1/3 of data original signal + noise

1/3 of data set original signal 1/3 of data set original signal filtered

i have 45 second window sampled @ 32 hz 1440 data points. pulse sequence can start anywhere in 45 second window. generated different pulse sequences @ different points each output.

for given pulse sequence should correspond number in output.

output 1 hot vector mapped set of values.

for example  input --|_|--|_|--|_|--|_|-- output should 0 = [0 0 ......];                --|_|--|_|--|_|----|_|-- output should 1 = [0 1 ......];               --|_|--|_|--|_|------|_|-- output should 2 = [0 0 1......]; 

i have tried train network 20 times 20000 iterations doesn't seem recognize patterns. error percentage seems high around %60

what need data network train better?