[Discuss-gnuradio] Re: When we combine several streams

[Eric Blossom]

On Fri, Jul 15, 2005 at 05:53:08PM -0700, Sachi wrote:
> Hi, Matt and Eric
> 
> I have a basic problem regarding the flow schedueling
> in GNU Radio, especially when we try to combine
> several sub-streams using a multiplier or an adder. 
> Can we make sure the indexes are still aligned?
> 
> For example, suppose I want to implement:
> z[n] = x[n] + c * y[n]
> Source block A is used to genereate x[1], x[2],
> x[3]...
> source block B is used to genereate y[1], y[2],
> y[3]...
> Block C, gr_multiply_const_cc, is used to multiply
> y[n], as the constant c.
> Block D, gr_add_cc, is used to add the two streams,
> the flow graph looks like:
> A->D
> B->C->D
> 
> So my question is, can the scheduling mechanism in GNU
> Radio guarantee the index alignment?  I mean, will the
> block C introduce any delay, or the two sourced
> generating data in different speeds, so that at the
> adder D, the index for the two sequence is not aligned
> any more, but something looks like:
> z[n] = x[n] + c * y[n - 1]??

They will be aligned.  If the sources are producing data at different
rates, the pipeline will block (the adder block) waiting for data to
be available from both input streams.

> Another more complicated example is the FM stereo
> receiver, after fm-demodulating the input signal, we
> need to split the data stream into two, process the
> two sub-streams separately (filtering), and recombine
> them using an adder and a substractor.  Something
> like:
> A->B->C->F
> A->D->E->F
> I am concerned about whether the index is still
> aligned at F.  Can we make sure (BC) and (DE)
> introduce the same delay?  

As long as the group delay through B->C is the same as D->E this will work.
You control the group delay by your choice of filter taps, etc.

In general, if B->C's overall input/output ratio is the same as
D->E's, you'll be OK.  E.g., if they're both 1:1, or 1:N or N:1,
everything should work fine.

> Thank you very much, the answer to this is very
> important to me, because I am going implement some
> complex algorithms. I know the simulink in Matlab
> could guarantee this, but I am not sure whether GNU
> Radio could gurantee this, especially when we split up
> stream and recombine again.

Eric