using FFT as a channelizer

Hi, I am learning multi-channel DDC techniques using FFT. My design is
referenced to 
http://www.pentek.com/tutorials/15_4/digdown.cfm
Something in this design I do not understand. 
The FFT will split the input stream into N(FFT size) bins equally spaced
in frequency. Totally 4 FFTs are used so that the output data rate is
fs/N*4. Followed by the FFT a compensation filter is used to flatten the
frequency droop at the passband edge. 
What I do not understand is that after FFT each bin output is a bandpass
signal whose center frequency is fs/N*k (k=0:N-1). Do we need a phase
shifter to translate these bandpass signals down to baseband before
applying the compensation filter?
I am confused on this part, I appreciate for any help.

On Oct 21, 9:55&#4294967295;am, "gobruins" <chunmei.k...@gmail.com> wrote:
> Hi, I am learning multi-channel DDC techniques using FFT. My design is
> referenced tohttp://www.pentek.com/tutorials/15_4/digdown.cfm
> Something in this design I do not understand.
> The FFT will split the input stream into N(FFT size) bins equally spaced
> in frequency. Totally 4 FFTs are used so that the output data rate is
> fs/N*4. Followed by the FFT a compensation filter is used to flatten the
> frequency droop at the passband edge.
> What I do not understand is that after FFT each bin output is a bandpass
> signal whose center frequency is fs/N*k (k=0:N-1). Do we need a phase
> shifter to translate these bandpass signals down to baseband before
> applying the compensation filter?
> I am confused on this part, I appreciate for any help.

What you're trying to build is called a "transmultiplexer". The Pentek
article has oversimplified the structure somewhat, leaving out the
polyphase filter that's usually placed on the input of the FFT. With
the polyphase filter you won't need a compensation filter after the
FFT. Do a bit of googling on the term transmultiplexer to get a better
idea of how to build one. You might also throw in the name 'harris',
since fred harris has written a lot about this structure - even has a
pretty good book out on multi-rate systems which goes into a fair
amount of detail on it.

When properly implemented, a transmultiplexer converts each of the
bandpass channels down to be centered at DC.

Eric

Thanks, I found a paper by fred Harris
Digital Receivers and Transmitters Using Polyphase
Filter Banks for Wireless Communications
I will take a look.

when you say "When properly implemented, a transmultiplexer converts each
of the bandpass channels down to be centered at DC". is this by aliasing
the pass band to baseband?



>On Oct 21, 9:55=A0am, "gobruins" <chunmei.k...@gmail.com> wrote:
>> Hi, I am learning multi-channel DDC techniques using FFT. My design is
>> referenced tohttp://www.pentek.com/tutorials/15_4/digdown.cfm
>> Something in this design I do not understand.
>> The FFT will split the input stream into N(FFT size) bins equally
spaced
>> in frequency. Totally 4 FFTs are used so that the output data rate is
>> fs/N*4. Followed by the FFT a compensation filter is used to flatten
the
>> frequency droop at the passband edge.
>> What I do not understand is that after FFT each bin output is a
bandpass
>> signal whose center frequency is fs/N*k (k=3D0:N-1). Do we need a
phase
>> shifter to translate these bandpass signals down to baseband before
>> applying the compensation filter?
>> I am confused on this part, I appreciate for any help.
>
>What you're trying to build is called a "transmultiplexer". The Pentek
>article has oversimplified the structure somewhat, leaving out the
>polyphase filter that's usually placed on the input of the FFT. With
>the polyphase filter you won't need a compensation filter after the
>FFT. Do a bit of googling on the term transmultiplexer to get a better
>idea of how to build one. You might also throw in the name 'harris',
>since fred harris has written a lot about this structure - even has a
>pretty good book out on multi-rate systems which goes into a fair
>amount of detail on it.
>
>When properly implemented, a transmultiplexer converts each of the
>bandpass channels down to be centered at DC.
>
>Eric
>