hello forum,

given an input signal x(t)=cos(3t), its FT will be 
X(w)=.5*[delta(w-3)+delta(w+3)]

If we have a filter that passes the positive part of the spectrum
(therefore the term delta(w-3) and completely blocks everything to the
left, w<0,

the output signal will be the complex valued exp(j*3t).

The filter that produced this result is asymmetric: |H(-w)| is different
from H|(w)|, i.e. the transfer function magnitude is not even and the phase
spectrum is not odd.

That implies that the impulse response is a complex value function....
Now the output y(t)=exp(i*3*t) and the input x(t)=cos(3t) but somehow
shifted?
I am trying to physically interpret the output exp(i*3*t) in relation to
the cosine input. Being real-valued, the component exp(-i*3*t) is redundant
but that does not means that is is superfluous. It missing should change
the signal a lot....

thanks for any hint.....
fisico32

On 08/27/2010 03:05 PM, fisico32 wrote:
> hello forum,
>
> given an input signal x(t)=cos(3t), its FT will be
> X(w)=.5*[delta(w-3)+delta(w+3)]
>
> If we have a filter that passes the positive part of the spectrum
> (therefore the term delta(w-3) and completely blocks everything to the
> left, w<0,
>
> the output signal will be the complex valued exp(j*3t).
>
> The filter that produced this result is asymmetric: |H(-w)| is different
> from H|(w)|, i.e. the transfer function magnitude is not even and the phase
> spectrum is not odd.

It would have to be, yes.

> That implies that the impulse response is a complex value function....

That, too, would have to be.

> Now the output y(t)=exp(i*3*t) and the input x(t)=cos(3t)

Yes.

> but somehow shifted?

What do you mean shifted?  The filter has filtered out one component 
(exp(-i*3*t)), and left the other -- that's a perfectly valid linear 
filter sort of thing to do.

> I am trying to physically interpret the output exp(i*3*t) in relation to
> the cosine input.

Well, first you would need to assign a physical meaning to the imaginary 
part of your data.  For baseband work, there's not much point in the 
filter.  If you were feeding the filter I and Q data from a 
downconverter it would give you a response to just one sideband.

> Being real-valued,

What's real valued?

> the component exp(-i*3*t) is redundant
> but that does not means that is is superfluous. It missing should change
> the signal a lot....

I think you're either borrowing trouble with a filter that's more 
complex than necessary, or you've gone and gotten yourself off into the 
mathematical weeds.

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

thanks Tim.
i am doing an optics experiment. It is optically possible to spatially
filter a real-valued image (2D signal) and create a filter like that, say a
step filter that block all negative freq and passes all positive ones....
If the image is a focal length away from a converging lens, the back focal
plane of the lens will have the 2D FT (power spectrum) of the image.
I can physically block regions of the FT plane.
If x is the spatial coordinate and I have an input  signal
cos(x)=.5[exp(-ix)+exp(ix)] and eliminate the exp(-ix) part by filtering, I
will be left with that .5*exp[ix] and I can't exactly understand if looks
like....it will not be a low passed version of the cosine either a bandpass
or high pass. 

I guess .5(cos(x)+j sin(x)) is the original signal .5*cos(x)+ j.5*sin(x).

But physically, if I take the real part, I would seem to simply get
.5cos(x), which is the original signal reduced in amplitude...



>On 08/27/2010 03:05 PM, fisico32 wrote:
>> hello forum,
>>
>> given an input signal x(t)=cos(3t), its FT will be
>> X(w)=.5*[delta(w-3)+delta(w+3)]
>>
>> If we have a filter that passes the positive part of the spectrum
>> (therefore the term delta(w-3) and completely blocks everything to the
>> left, w<0,
>>
>> the output signal will be the complex valued exp(j*3t).
>>
>> The filter that produced this result is asymmetric: |H(-w)| is
different
>> from H|(w)|, i.e. the transfer function magnitude is not even and the
phase
>> spectrum is not odd.
>
>It would have to be, yes.
>
>> That implies that the impulse response is a complex value function....
>
>That, too, would have to be.
>
>> Now the output y(t)=exp(i*3*t) and the input x(t)=cos(3t)
>
>Yes.
>
>> but somehow shifted?
>
>What do you mean shifted?  The filter has filtered out one component 
>(exp(-i*3*t)), and left the other -- that's a perfectly valid linear 
>filter sort of thing to do.
>
>> I am trying to physically interpret the output exp(i*3*t) in relation
to
>> the cosine input.
>
>Well, first you would need to assign a physical meaning to the imaginary 
>part of your data.  For baseband work, there's not much point in the 
>filter.  If you were feeding the filter I and Q data from a 
>downconverter it would give you a response to just one sideband.
>
>> Being real-valued,
>
>What's real valued?
>
>> the component exp(-i*3*t) is redundant
>> but that does not means that is is superfluous. It missing should
change
>> the signal a lot....
>
>I think you're either borrowing trouble with a filter that's more 
>complex than necessary, or you've gone and gotten yourself off into the 
>mathematical weeds.
>
>-- 
>
>Tim Wescott
>Wescott Design Services
>http://www.wescottdesign.com
>
>Do you need to implement control loops in software?
>"Applied Control Theory for Embedded Systems" was written for you.
>See details at http://www.wescottdesign.com/actfes/actfes.html
>