Re: Efficiency of PFB symbol sync filtering

[Jeff Long]

True that most of the filters are producing unused results. There isn't a great way to put only the needed input into only the filter that will be used for output on a given cycle (at least using the usual FIRs in GNU Radio).

Are you saying that the staircase version of the output is acceptable for some applications? What kind of speedup do you think it would give?

On Wed, Dec 8, 2021 at 8:28 PM Moses Browne Mwakyanjala <mbkitine@gmail.com> wrote:

Hi everyone,

I've been thinking about the PFB and its operations (match-filtering and interpolation in one go). From the source code of the PFB block Lines 390 and 427, the match/diff filter output are

out[i + d_out_idx] = d_filters[d_filtnum].filter(&in[count + d_out_idx]);

gr_complex diff = d_diff_filters[d_filtnum].filter(&in[count]);

That is, the filtering operations produce a single symbol. Given that filters work on a block of samples, I was wondering how many samples are actually filtered in the case above.  What becomes of the other output samples from the filter? Are they just discarded? Correct me if I'm wrong but shouldn't it be more efficient not to discard the extra samples produced by the filters? That way, we could "fix" the filter index (d_filtnum) for a block of the filter output vector? The transient filter index characteristics will then become a staircase as shown by my simulation results below. 

Regards,

Moses.