Re: [Discuss-gnuradio] Channel model - the identity channel

[Martin Braun]

This is not the first time this has come up, and yes, patches are
welcome. The simplest fix might be to add a factory-time argument that
disables the resampler entirely -- this will guarantee no sample
offsets, ever, but of course will not allow to change the timing offset
at runtime.

M

On 06/28/2016 07:47 PM, Raj Bhattacharjea wrote:
> Thanks for the hints for where to look and why this happens from a code
> perspective. I don't think it makes sense from a semantic perspective
> (if the user says taps=1, she should get the unity gain delay free
> channel) but I'm guessing "patches are welcome" is the appropriate
> response to this minor usability issue.
> 
> On Tuesday, June 28, 2016, Andrej Rode  > wrote:
> 
> Hey Raj,
> 
> > introduces a delay that can be corrected by applying the taps
> [0,0,0,1].
> > See the attached flowgraph that subtracts the signals before and after
> > the channel model; if you let taps = 1, the two signals don't
> cancel. If
> > you use taps = [0,0,0,1], they do.
> 
> The delay you are seeing is introduced by the fractional_resapmler
> inside the channel_model block. The FIR filter inside the fractional
> resampler needs to build up and therefore introduces a initial delay of
> three samples.
> 
> If you want to know more about the insides of the channel model I
> suggest you look at the sources [0][1] yourself.
> 
> Greetings,
> Andrej
> 
> [0] ./gr-channels/lib/channel_model_impl.cc
> [1] ./gr-channels/lib/channel_model_impl.h
> 
> 
> 
> -- 
> Raj Bhattacharjea, PhD
> Georgia Tech Research Institute
> Information and Communications Laboratory
> http://www.prism.gatech.edu/~rb288/ 
> 404.407.6622
> 
> 
> 
> ___
> Discuss-gnuradio mailing list
> Discuss-gnuradio@gnu.org
> https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
> 


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

Re: [Discuss-gnuradio] Channel model - the identity channel

[Raj Bhattacharjea]

Thanks for the hints for where to look and why this happens from a code
perspective. I don't think it makes sense from a semantic perspective (if
the user says taps=1, she should get the unity gain delay free channel) but
I'm guessing "patches are welcome" is the appropriate response to this
minor usability issue.

On Tuesday, June 28, 2016, Andrej Rode  wrote:

> Hey Raj,
>
> > introduces a delay that can be corrected by applying the taps [0,0,0,1].
> > See the attached flowgraph that subtracts the signals before and after
> > the channel model; if you let taps = 1, the two signals don't cancel. If
> > you use taps = [0,0,0,1], they do.
>
> The delay you are seeing is introduced by the fractional_resapmler
> inside the channel_model block. The FIR filter inside the fractional
> resampler needs to build up and therefore introduces a initial delay of
> three samples.
>
> If you want to know more about the insides of the channel model I
> suggest you look at the sources [0][1] yourself.
>
> Greetings,
> Andrej
>
> [0] ./gr-channels/lib/channel_model_impl.cc
> [1] ./gr-channels/lib/channel_model_impl.h
>
>

-- 
Raj Bhattacharjea, PhD
Georgia Tech Research Institute
Information and Communications Laboratory
http://www.prism.gatech.edu/~rb288/
404.407.6622
___
Discuss-gnuradio mailing list
Discuss-gnuradio@gnu.org
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio

Re: [Discuss-gnuradio] Channel model - the identity channel

[Tom Rondeau]

On Tue, Jun 28, 2016 at 6:14 PM, Raj Bhattacharjea  wrote:

> Consider the identity channel that passes through it's input to its
> output. This is an idealization of a very short run of high quality cable
> between synchronized systems. Using GNURadio's channels_channel_model, I
> expect that the following provides the identity channel:
>
> Noise voltage: 0
> Frequency offset: 0
> Epsilon: 1
> Taps: 1
>
> That says add no noise, simulate no timing or carrier offset, and simulate
> a channel impulse response that is a single impulse with amplitude 1.
> However, I find that using such a channel model block introduces a delay
> that can be corrected by applying the taps [0,0,0,1]. See the attached
> flowgraph that subtracts the signals before and after the channel model; if
> you let taps = 1, the two signals don't cancel. If you use taps =
> [0,0,0,1], they do. Anyone know why this occurs? For my purposes, such a
> channel model is a stand-in that is ideal for now, but general enough to
> allow modeling of channel effects when I'm ready.
>
> --
> Raj Bhattacharjea, PhD
> Georgia Tech Research Institute
> Information and Communications Laboratory
> http://www.prism.gatech.edu/~rb288/
> 404.407.6622
>


For some value of "short" compared to the samples, otherwise, your cable is
also introducing delay.

Tom
___
Discuss-gnuradio mailing list
Discuss-gnuradio@gnu.org
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio

Re: [Discuss-gnuradio] Channel model - the identity channel

[Andrej Rode]

Hey Raj,

> introduces a delay that can be corrected by applying the taps [0,0,0,1].
> See the attached flowgraph that subtracts the signals before and after
> the channel model; if you let taps = 1, the two signals don't cancel. If
> you use taps = [0,0,0,1], they do. 

The delay you are seeing is introduced by the fractional_resapmler
inside the channel_model block. The FIR filter inside the fractional
resampler needs to build up and therefore introduces a initial delay of
three samples.

If you want to know more about the insides of the channel model I
suggest you look at the sources [0][1] yourself.

Greetings,
Andrej

[0] ./gr-channels/lib/channel_model_impl.cc
[1] ./gr-channels/lib/channel_model_impl.h



signature.asc
Description: OpenPGP digital signature
___
Discuss-gnuradio mailing list
Discuss-gnuradio@gnu.org
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio