|
The whiteness test computes autocorrelation sequence of residual error. The
correlation should be as close to zero as possible for all lags except 0,
where it is 1 by definition. If you look at the power spectrum of such as
sequence, it should be a flat line across all frequencies (essential PSD of
white noise). The level of the signals denotes noise variance.
The correlation test computes correlation between error and input. In
frequency domain, you could equivalently look at transfer function from
input to error which should have low gain in the bandwidth of the system
(provided input had sufficient power in the bandwidth of system).
Rajiv
"pietro " <bracardi82@email.it> wrote in message
news:h63puk$k9v$1@fred.mathworks.com...
> "Rajiv Singh" <Rajiv.SinghNoSpam@mathworks.com> wrote in message
> <h4n0t6$lh0$1@fred.mathworks.com>...
>> System Identification Toolbox handles estimation of transfer function
>> from
>> freqyency response data, since version 6.0. For transfer functions, try
>> OE
>> and IDPROC/PEM.
>>
>> Rajiv
>>
>> "pietro " <bracardi82@email.it> wrote in message
>> news:h4mf1q$t60$1@fred.mathworks.com...
>> >I have only system identification toolbox. I have frequency response
>> >data
>> >and i have to identificate the transfer function. I can do that with
>> >system
>> >identification toolbox or it is necessary to have frequency domain
>> >identification toolbox?
>>
>
> Thank you.
>
> I have another question: how can i test the whitness and the correlation
> between residuals and transfer function? I have idenficated a model and i
> have to validate it. I have plotted the residual but i don't understand
> this graph.
>
> http://www.megaportal.it/imghost/2009/08/14/1535341250256934.jpg
>
> Can you explain me?
> thank you in advance
|
Feed for this Thread