Re: [Discuss-gnuradio] isolate channels from wideband

[Tom Rondeau]

On Wed, Jul 22, 2015 at 4:57 AM, Daniele Nicolodi <address@hidden> wrote:

On 21/07/15 21:39, Tom Rondeau wrote:
> Here's my presentation from last GRCon:
>
> http://gnuradio.squarespace.com/grcon14-presentations#tut-rondeau

Hello Tom,

browsing through your presentation I see that on page 58 and 59 you
recommend to use firdes filter design tool and not optfir to build re
reconstruction filter.  However, I don't quite understand why the filter
generated by one tool is better than the other is this case.

Can you please comment on it?

Thanks! Cheers,
Daniele

The shape of this filter matters greatly. The inband, transition, and stop band behavior all determine if the filter can be used for the reconstruction purposes. The image on slide 59 shows the specific transition between the pass band and stop bands. To match that with the PM (i.e., Remez) algorithm, you can't get the same stop band performance for that given transition. Plus the equal response in the stop band is bad when channelizing because all channels will alias at equal powers, whereas the roll off in frequency with the windowed (firdes) filter continues to decrease with f. Remez also produces a pass band ripple, which will also affect things. The ripple with the firdes is not equiripple like Remez promises, but it's much, much smaller.

Tom