Re: [Discuss-gnuradio] Square Wave Pulse Shapping Filter

[Andy Walls]

Hi Jose:

Please keep conversations on the list.

On Sun, 2018-05-20 at 00:54 -0300, Jose Marcelo Lima Duarte wrote:
> Andy,
> 
> Thank you very much for all information.

You're welcome.  But after I thought about it, it may not work in your
situation.

The GNURadio symbol timing synchronization blocks require symbol
centers to be "peaked" to perform their function properly.  Matched
Filtered RRC pulsed QPSK and Matched Filtered Square pulsed baseband
PAM are nicely peaked.

I'm not sure a Matched Filter will do anything to peak Square pulsed
QPSK unless there are spaces in between the pulses or the pulses are
not truly square.  Recovery of symbol centers of Square pulsed QPSK
will be very difficult, given that the QPSK can have an arbitrary phase
rotation in the receiver.

However, I don't believe engineers of an old satellite system would
make reception so difficult; especially if they were content to waste
TX energy in square pulses (which would be fairly broad spectrum).

Does the downlink provide an explicit symbol clock, on a frequency
separate from the data?  Is the satellite downlink specification
publically available?



> I will use the architecture you propose with minor parameters change.
> I am planing to work with a very high oversample rate to make ISI
> insignificant, 32 or higher prior to the MF and 16 after it.

Note that square pulse matched filters are not good antialias filters. 
Don't decimate with the MF.  Do the anti-alias filtering and down-
sampling with a separate Decimating FIR filter.


> Thus, I won't need more than 2 tap at the interpolation filter.

Well the GNURadio 8-tap MMSE interpolation filter used by the symbol
synch blocks is fixed at 8-taps, since the taps are all precomputed
offline and compiled in as constants.

Note that the MMSE interpolation filter is only MMSE (i.e. bounded
error) when the input signal is band-limited to the frequencies in the
interval [-Fs/4, Fs/4].  When using a high number of samples per
symbol, like 16, usually you'll be OK.

 
I would say, for more information, please see my presentation on Symbol
Synchornization here:

https://www.gnuradio.org/grcon-2017/program/grcon17-presentations/

but the web page seems to be broken/gone.

Instead, you can suffer through the video of me talking through my
slides, if you want:

https://www.youtube.com/watch?v=uMEfx_l5Oxk

Regards,
Andy

> Regards
> 
> Marcelo
> 
> 
> 
> On Sat, May 19, 2018 at 6:32 PM, Andy Walls <address@hidden
> net> wrote:
> > > Date: Sat, 19 May 2018 11:29:14 -0300
> > > From: Jose Marcelo Lima Duarte <address@hidden>
> > 
> > > Hello everyone,
> > > 
> > > I was giving the task of quickly develop a receiver for an old
> > > satellite system that use a pulse shaping filter with square wave
> > > impulse response and QPSK constelation.
> > > 
> > > Does GNU Radio has a prompt solution for this? By a prompt
> > solution a
> > > mean a single block in GNU Radio Companion that handles carrier
> > phase
> > > synch, symbol time synch and sampling. From its documentation its
> > > seams to me that it lacks of prompt solution for old fashion
> > square
> > > wave pulse shaped signals. Am I right?
> > 
> > You are correct in that there is no single block in GNURadio that
> > handles all that.
> > 
> > 
> > > I have seen that GNU Radio has a prompt solution for demodulating
> > > QPSK signals with Squared RRC filter. And I known that it can be
> > used
> > > for square wave signals too. However, I want to avoid the
> > performance
> > > degradation from not using a matching filter at the receiver end.
> > 
> > Actually the one in GNURadio specifically can't be used for square
> > pulses, if it uses the Polyphase Clock Sync block in the underlying
> > implementation, as that block can't make a proper derivative pulse
> > filter for a square wave.
> > 
> > > Any suggestion for a quick solution for my problem?
> > 
> > For coarse carrier frequency synchronization, if you don't have
> > high
> > doppler, the "FLL band-edge" block will keep RRC pulse filtered PSK
> > signals roughly on frequency.  You can probably abuse it for PSK
> > with
> > square pulses too.
> > 
> > For matched filtering of a square pulse, just use a "Decimating FIR
> > Filter" block with a decimation of 1 (no decimation) and taps of
> > [1.0/N]*N  where N is the number of samples in th unit pulse width.
> > This will introduce receiver side ISI, since your square pulses are
> > not
> > Nyquist pulses, but you should get nice triangular pulse peaks.
> > 
> > For symbol syncronization, use a "Symbol Sync" block with the
> > signal
> > times slope ML approximation TED.  Also select the 8 tap MMSE
> > interpolation filter. 
> > 
> > For fine carrier frequency and phase synchronization use the
> > "Costas
> > Loop" block.
> > 
> > Regards,
> > Andy 
> > 
> > > 
> 
> 
> 
>