Re: GSoC 2020: gr-dpd - New blogpost for Week 4

[Matthias Brändli]

Hello Alekh,

glad to hear about your progress on this topic that interests me much
even though I don't have time to work on it anymore.

I did work on DPD for Digital Audio Broadcasting some time ago.[1]


Regarding what you mention here:

> - Getting to work upon the synchronisation of signal for making this
> module useful for real-world physicals applications.

This is quite important, and (at least in our application), I found out
that aligning samples wasn't sufficient. Subsample alignment was necessary.


We tried two ways:

1. upsample - find correlation peak - downsample, which was quite costly

2. coarse correlation to sample accuracy using correlation peak, then
subsample delay estimation, inspired by this:
https://www.dsprelated.com/showarticle/26.php
https://www.dsprelated.com/showcode/207.php

The second option worked better.

All this to say that it's worth looking into subsample alignment.


I'm very curious to hear about your findings on this aspect, and of
course on the whole topic of DPD. All the best!
mpb

[1]
Application: 8192 ksps samplerate, 1.5MHz effective signal bandwidth,
transmitted in Band III (around 200MHz)

We went so far as to do a memoryless polynomial predistorter, in C++
inside odr-dabmod, and the alignment and coefficient calculation was
done using SciPy.
https://github.com/Opendigitalradio/ODR-DabMod/tree/master/python