Re: [Discuss-gnuradio] is there possible to have item drops?

[Marcus Müller]

Hello Mostafa,

filtering is a inherently lossy operation: it's an operation that involves multiplication and addition of floating point numbers.
Matlab internally uses double (float64) values, whereas GNU Radio, usually, uses single precision (float32) numbers for performance reasons.
Now, without knowing what kind of signal you are filtering with what kind of taps you are using with what kind of GNU Radio filters and what Matlab functions, there's nothing to guess on how these differences occur.
What I *can* say though is that I've never seen an item drop in GNU Radio's runtime, and we have strong tests against that. Now, it's entirely possible that any of our filter implementations might have bugs, but I kind of doubt that. It's most probably simply numeric accuracy that kills your samples.

Greetings,
Marcus

On 15.10.2014 18:07, Mostafa Alizadeh wrote:

Hi all,

It seems this is a ridiculous question, however, I didn't find any solution
for my problem.

I have already a bunch of custom blocks connected to each other which play
role as a transmitter (tx). I want to connect tx to a fir filter by using
fir_filter_ccc block. I set the filter taps with specific ones. I also have
this filter in Matlab. The output of the filter is connected to a sink file.

When I see the output (tx) and filtering it by Matlab and comparing it with
the output of the gnuradio filter, I see both of the following:

1) Both matlab and gnuradio filtered output, have little difference in a
range of 0.01. (I'm using gr_complex data).

2) After a while, Matlab output and GNURadio's output will differ as high
as 1000 in magnitude!!

I guess, there is item drops in this filtering. I know, the probability of
this event is so weak, anyway, item drop even can happen in GNURadio?

Where is my problem?

Best,
Mostafa

_______________________________________________
Discuss-gnuradio mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio