[Pafgbt] PAF beamformer size and cost

[Brian Jeffs]

A correction on my computational burden analysis follows.

Brian

On Feb 4, 2010, at 1:14 PM, Brian Jeffs wrote:

>
> On Feb 4, 2010, at 12:23 PM, Rick Fisher wrote:
>
>> Karl,
>>
>> To a first approximation I think the computational burden is
>> independent
>> of beamformer subchannel bandwidth.  The data rate in each
>> subchannel and
>> the number of subchannels are both inversely proportional to
>> subchannel
>> bandwidth.
>>
>> Rick
>
> Yes, I agree with this (almost).  Computations go up with Log(no. of
> freq channels) for the FFT portion of the F engine and beamformer.
> If polyphase filter banks are used instead of FFTs, then the growth is
> linear, and is therefore nearly exactly compensated for by the higher
> decimation rate for narrower channels.
>

Correction:
Yes, I agree with this (almost).  Let N= # of freq. channels.   
Computations
go up with Log(N) for the FFT portion of the F engine and beamformer.
If polyphase filter banks are used instead of FFTs, then the total
computation growth is linear with N.

Brian