Re: Correlation Estimator Questions

[Andy Walls]

> Date: Sat, 18 Jul 2020 12:06:11 -0400
> From: Nicholas Savka <nicholas.s.savka@gmail.com>
> To: discuss-gnuradio@gnu.org
> Subject: Correlation Estimator Questions
> 
> Hello,
> 
> My ultimate goal is to use the correlation estimator block for
> synchronization using a preamble in order to send packets of digital
> data
> from one sdr to another, both running gnu-radio. In order to
> understand how
> the correlation estimator works, I am looking at the provided example
> "example_corr_est_and_phase_sync." It looks like this example is
> using a
> BPSK modulation scheme and I am trying to modify it to work with QPSK
> instead to make sure I know how it works, but I am running into
> problems. I
> have four questions:
> 
> 1.)
> I do not understand the purpose of the "tag marking delay" entry in
> the
> correlation estimator block. How should this value be determined?

The block will output a tag on the output sample stream when it detects
a correlation.

By default (delay of 0) this tag will be set at the start of your
preamble.

The tag marking delay lets you adjust the position of the output tag in
units of samples.  You should set this empirically to be at the center
of the first symbol in the preamble.


> 2.)
> How does the correlation estimator "talk to" the costas loop and
> Polyphase
> Clock Sync blocks? Is there something going on behind the scenes here
> that
> I cannot see? Is it somehow passing the frequency/phase correction
> information to these blocks?

Using the time_est tag on the output sample stream, to tell the clock
sync blocks, "This is the approximate center of a symbol, reset your
symbol clock timing recovery".

A CPO estimate is also output on the "phase_est", but I don't think any
block uses it.


> 3.)
> I think my main issue is that I do not understand how to format the
> "Data
> Vector" entry in the "Modulate Vector" block. This is my (probably
> incorrect) understanding:
> 
> I want my preamble symbols to be [1+1j, 1+1j, 1+1j, -1-1j, -1-1j,
> 1+1j,
> -1-1j]
> 
> My constellation object has the following parameters:
> Symbol Map: [0,1,2,3]
> Constellation Points: [(-1-1j),(-1+1j),(1-1j),(1+1j)]
> 
> Therefore, my desired preamble would map to: [3,3,3,0,0,3,0]
> So, in the "Data Vector" box in the "Modulate Vector" block, I should
> enter:  [0x3,0x3,0x3,0x0,0x0,0x3,0x0]



> 4.)
> Why is the Modulate Vector block needed at all? Why can I not simply
> enter
> [1+1j, 1+1j, 1+1j, -1-1j, -1-1j, 1+1j, -1-1j] into the field for
> "Symbols"
> in the correlation estimator block?

The corr_est block "Symbols" input wants the correlation/matched filter
taps (*before* time-reversal and conjugation) for the preamble it is
trying to match.

It can't use symbols from the constellation, because it doesn't make
any assumptions about your pulse filter, you modulation, or how you
truncate the transient of pulse filtering your preamble symbols.

I suggest that you do not use the Modulate Vector block, as it doesn't
give you insight or much control into what it is actually producing.

Instead, produce your correlation/matched filter taps yourself using
MatLab or Octave or Python.  See the attached Octave scripts for a QPSK
example with your preamble and symbol mapping.

BTW, the GNURadio Constallation object will silently scale your
constellation so that the constellation points have an average radius
of 1.  So your constealltion points are really 1/sqrt(2) * [(-1-1j),(-
1+1j),(1-1j),(1+1j)], as far as GNURadio is concerned.


Regards,
Andy

qpsk_corr.m
Description: Text Data