Re: Hilbert transform

[Sergei Steshenko]

----- Original Message -----
> From: Przemek Klosowski <address@hidden>
> To: address@hidden
> Cc: 
> Sent: Monday, July 9, 2012 9:29 PM
> Subject: Re: Hilbert transform
> 
> On 07/09/2012 02:00 PM, Przemek Klosowski wrote:
> 
>>  Curiously, as the vector length increases from two to 20, odd vector
>>  lengths satisfy Sergei's identity u = -Hilb(Hilb(u)), but odd ones have
>>  quite large errors:
> 
> Nevermind---as Ben mentioned, this is due to hilbert() assuming the 
> periodicity 
> of the signal.
> 

Since Hilbert transform is implemented through FFT, periodicity is implied 
automatically.

As I wrote in an earlier Email, for some signals Hilbert transform can have +/- 
infinity values - which can't be reached in discrete numerical implementation.

In that same Email I gave a link to an article measuring Matlab Hilbert 
transform errors for various input signals.

Regards,
  Sergei.