Re: Symbol timing estimator block and CPM modulations
Hi Nick: On Fri, 2020-07-31 at 17:21 -0700, Nick Foster wrote: > Thanks for the feedback, Andy! I'm back at this now. I've filed a bug > on it with my findings so far. Comments welcome. > > https://github.com/gnuradio/gnuradio/issues/3678 I've added some comments, some plots, and an Octave script to the issue on github. TL;DR: The original TED from the paper is correct, if computed when shifted by a 1/4 symbol and the result negated. -Andy
Re: Symbol timing estimator block and CPM modulations
Thanks for the feedback, Andy! I'm back at this now. I've filed a bug on it with my findings so far. Comments welcome. https://github.com/gnuradio/gnuradio/issues/3678 Nick On Mon, Jun 29, 2020 at 1:06 PM Andy Walls wrote: > Hi Nick: > > On Wed, 2020-06-24 at 17:39 -0700, Nick Foster wrote: > > Hi, > > > > I'm finally getting around to doing some more CPM work > > I forgot to mention, my company found the GMSK demoulation technique > found in this paper: > > > https://www.researchgate.net/publication/241629765_Coherent_demodulation_of_AIS-GMSK_signals_in_co-channel_interference > > coupled with the "method 0" CFO correction in this paper: > > > https://www.researchgate.net/publication/290776860_Preprocessing_AIS_Signals_for_Demodulation_in_Co-Channel_Interference > > along with analog RF components in front of an SDR, and precise timing > with a GPSDO w/PPS to aid in looking for preambles, gave us fantastic > results for AIS reception in GNURadio. > > Although it required quite a few custom blocks. > > Anyway, Douglas Nelson seems to have a lot of cool papers available in the > public: > > https://www.researchgate.net/profile/Douglas_Nelson8 > > Regards, > Andy > > > Best, > > Nick > > >
Re: Symbol timing estimator block and CPM modulations
Hi Nick: On Wed, 2020-06-24 at 17:39 -0700, Nick Foster wrote: > Hi, > > I'm finally getting around to doing some more CPM work I forgot to mention, my company found the GMSK demoulation technique found in this paper: https://www.researchgate.net/publication/241629765_Coherent_demodulation_of_AIS-GMSK_signals_in_co-channel_interference coupled with the "method 0" CFO correction in this paper: https://www.researchgate.net/publication/290776860_Preprocessing_AIS_Signals_for_Demodulation_in_Co-Channel_Interference along with analog RF components in front of an SDR, and precise timing with a GPSDO w/PPS to aid in looking for preambles, gave us fantastic results for AIS reception in GNURadio. Although it required quite a few custom blocks. Anyway, Douglas Nelson seems to have a lot of cool papers available in the public: https://www.researchgate.net/profile/Douglas_Nelson8 Regards, Andy > Best, > Nick
Re: Symbol timing estimator block and CPM modulations
On Mon, 2020-06-29 at 14:11 -0500, Alex Roberts wrote: > Copy that. Would a preamble with better properties like a gold code > allow to raise that threshold? I guess I need to dig into what that > threshold means. On the examples I’ve seen it’s always 0.9 or higher. > You could use a Gold code. Although the length 13 Barker code is probably good too, just to play around with. Yeah, so the fixed correlation threshold algorithm is somewhat staright forward, but I caused to to have a bit of a wrinkle. You'll notice in the algorithm, it checks using the square of the correlation's magnitude. That's because long ago, to save CPU cycles, I coded the block to compute the magnitude squared of the correlation and check against that. Taking the square-root of every I^2+Q^2 sample of the correlation output seemed like a waste of CPU to me at the time. I now regret that decision, as it has been a source of confusion for people working with this block, oh well. The block has undergone a few revisions regarding the thresh holding algorithm, since I last touched it myself. I know that the dynamic algorithm only works with preambles with good correlation properties (gold codes, barker codes). The non-adaptive algorithm is sensitive to the incoming signal level compared to the reference level of the correlation filter taps. In hindsight, I probably should have had my correlation filter taps script normalize h_iq to some nominal gain value. Anyway, I highly recommend you do test correlation runs in Matlab/Octave first, duplicating the corr_est block's thresh hold crossing detection algorithm, before jumping into a GNURadio implementation. The analysis in Matlab/Octave will much better inform you on how your GNURadio parameters should be set. Regards,Andy > On Monday, June 29, 2020, Andy Walls > wrote: > > On Sun, 2020-06-28 at 16:31 -0500, Alex Roberts wrote: > > > > > Andy, > > > > > > > > > > I had a copy of an old flowgraph that you posted on the mailing > > list > > > > > a long while back ( > > > > > > > https://lists.gnu.org/archive/html/discuss-gnuradio/2016-03/msg00735.html > > > > > ). This is what initially confused me on what to pass in as gmsk > > > > > symbols to corr_est block as it has strictly real values. > > > > > > > > > > I modified it to mux in the preamble that's in the octave code > > with a > > > > > random source. I also modified the octave code to match the > > samples > > > > > per symbol of the flowgraph (10) and generated the modulated > > symbols > > > > > for the corr_est block. The corr_est block does indeed > > > > > perform conjugation and reversal of the symbols, so I used the > > "h_iq" > > > > > value from the octave code as the modulated symbols since it has > > not > > > > > been conjugated or reversed. I'm not getting corr_est tags. > > > > > > > > On that flowgraph, set your threshold on the corr_est block to > > 0.23. > > > > > > > > BTW, that preamble has horrible correlation properties. > > > > > > > > > > > > > Thanks,Alex. > > > > > > > > -Andy > > > > > > > >
Re: Symbol timing estimator block and CPM modulations
Copy that. Would a preamble with better properties like a gold code allow to raise that threshold? I guess I need to dig into what that threshold means. On the examples I’ve seen it’s always 0.9 or higher. On Monday, June 29, 2020, Andy Walls wrote: > On Sun, 2020-06-28 at 16:31 -0500, Alex Roberts wrote: > > Andy, > > > > I had a copy of an old flowgraph that you posted on the mailing list > > a long while back ( > > https://lists.gnu.org/archive/html/discuss-gnuradio/2016-03/ > msg00735.html > > ). This is what initially confused me on what to pass in as gmsk > > symbols to corr_est block as it has strictly real values. > > > > I modified it to mux in the preamble that's in the octave code with a > > random source. I also modified the octave code to match the samples > > per symbol of the flowgraph (10) and generated the modulated symbols > > for the corr_est block. The corr_est block does indeed > > perform conjugation and reversal of the symbols, so I used the "h_iq" > > value from the octave code as the modulated symbols since it has not > > been conjugated or reversed. I'm not getting corr_est tags. > > On that flowgraph, set your threshold on the corr_est block to 0.23. > > BTW, that preamble has horrible correlation properties. > > > > Thanks,Alex. > > -Andy > >
Re: Symbol timing estimator block and CPM modulations
On Sun, 2020-06-28 at 16:31 -0500, Alex Roberts wrote: > Andy, > > I had a copy of an old flowgraph that you posted on the mailing list > a long while back ( > https://lists.gnu.org/archive/html/discuss-gnuradio/2016-03/msg00735.html > ). This is what initially confused me on what to pass in as gmsk > symbols to corr_est block as it has strictly real values. > > I modified it to mux in the preamble that's in the octave code with a > random source. I also modified the octave code to match the samples > per symbol of the flowgraph (10) and generated the modulated symbols > for the corr_est block. The corr_est block does indeed > perform conjugation and reversal of the symbols, so I used the "h_iq" > value from the octave code as the modulated symbols since it has not > been conjugated or reversed. I'm not getting corr_est tags. On that flowgraph, set your threshold on the corr_est block to 0.23. BTW, that preamble has horrible correlation properties. > Thanks,Alex. -Andy
Re: Symbol timing estimator block and CPM modulations
Andy, Yes, I did mean the corr_est block. Thank you for the matlab code. I'll hopefully get to play with it today. The flowgraph I have write now is actually doing quite well with high AWGN just with switching to the enhanced symbol block and using the Mengali and d'Andrea timing recovery. For what it's worth, I'm stuck on GRC 3.7 so I had to check out from your git from before you removed the symbol sync block (assuming it was removed since it's been moved in-tree in 3.8) Thanks, Alex. On Sun, Jun 28, 2020 at 7:32 AM Andy Walls wrote: > Hi Alex: > > On Sat, 2020-06-27 at 19:42 -0500, Alex Roberts wrote: > > Andy, > > > > I’m not sure how integrate the correlation sync block with gmsk. > > The correlate_and_sync block is an old block that was specific to RRC > filtered PSK. You must mean the corr_est block. > > > It expects modulated symbols and I’m not sure how to generate a > > modulated vector of gmsk symbols. There doesn’t seem to be a gmsk > > class that can be used by the modulate vector block. > > You can create a modulated GMSK preamble with the modulate vector block > in GNURadio. However, I do not recommend it. It provides little > insight or control over the exact correlation filter taps, so trimming > off start and end transients becomes a hassle. > > Instead, generate your correlation filter taps in MatLab or Octave. > Then you can also use Matlab or Octave to assess the performance of > your correlation filter taps against real or simulated data. > > See the attached script for an example of generating GMSK correlation > filter taps. Note that this script generates the conjugated and time > reversed filter taps. You'll have to check what the corr_est block is > expecting, as it may be performing a conjugation and time reversal for > you, under the assumption that you didn't do it. > > Regards, > Andy > > > > Thanks, > > Alex > > > > On Thursday, June 25, 2020, Andy Walls > > wrote: > > > Recommend reading > > > > > > > https://www.gnuradio.org/grcon/grcon17/presentations/symbol_clock_recovery_and_improved_symbol_synchronization_blocks/Andy-Walls-Samples-to-Digital-Symbols.pdf > > > > > > Yeah, the MSK TED selections for the symbol sync block expect the > > > constant envelope FSK waveform on the input. Demodulation from FSK > > > to baseband pulses should happen after the symbol sync block. > > > > >
Re: Symbol timing estimator block and CPM modulations
Hi Alex:
On Sat, 2020-06-27 at 19:42 -0500, Alex Roberts wrote:
> Andy,
>
> I’m not sure how integrate the correlation sync block with gmsk.
The correlate_and_sync block is an old block that was specific to RRC
filtered PSK. You must mean the corr_est block.
> It expects modulated symbols and I’m not sure how to generate a
> modulated vector of gmsk symbols. There doesn’t seem to be a gmsk
> class that can be used by the modulate vector block.
You can create a modulated GMSK preamble with the modulate vector block
in GNURadio. However, I do not recommend it. It provides little
insight or control over the exact correlation filter taps, so trimming
off start and end transients becomes a hassle.
Instead, generate your correlation filter taps in MatLab or Octave.
Then you can also use Matlab or Octave to assess the performance of
your correlation filter taps against real or simulated data.
See the attached script for an example of generating GMSK correlation
filter taps. Note that this script generates the conjugated and time
reversed filter taps. You'll have to check what the corr_est block is
expecting, as it may be performing a conjugation and time reversal for
you, under the assumption that you didn't do it.
Regards,
Andy
> Thanks,
> Alex
>
> On Thursday, June 25, 2020, Andy Walls
> wrote:
> > Recommend reading
> >
> > https://www.gnuradio.org/grcon/grcon17/presentations/symbol_clock_recovery_and_improved_symbol_synchronization_blocks/Andy-Walls-Samples-to-Digital-Symbols.pdf
> >
> > Yeah, the MSK TED selections for the symbol sync block expect the
> > constant envelope FSK waveform on the input. Demodulation from FSK
> > to baseband pulses should happen after the symbol sync block.
>
>
pkg load signal; % Octave needs this
% One of the possible encodings of the AIS preamble symbols
h_bits = [-1,-1, 1, 1, ...
-1,-1, 1, 1, ...
-1,-1, 1, 1, ...
-1,-1, 1, 1, ...
-1,-1, 1, 1, ...
-1,-1, 1, 1, ...
-1,-1,-1,-1,-1,-1,-1, 1].';
% AIS specific GMSK parameters
Rb = 9600;
BT = 0.4;
L = 3;
modulation_index = 1/2;
% Working sample rate
sps = 5;
Fs = Rb * sps;
% Build Gaussian pulse filter
Ls = round(L*sps);
alpha = sqrt(2/log(2)) * pi * BT;
k = [(-Ls/2+1):1:(Ls/2-1)];
taps = (erf(alpha*(k/sps + 0.5)) - erf(alpha*(k/sps - 0.5)))*0.5/sps;
K = length(taps);
if (mod(K,2) == 0)
delay = K/2;
else
delay = (K-1)/2;
end
% Upsample and pulse shape the correlation sequence
x = sps*[upsample(h_bits,sps); zeros(delay, size(h_bits)(2))];
h_ibaseband = filter(taps, [1], x);
% Trim the pulse shaped correlation sequence
h_baseband = h_ibaseband((delay):(end-(sps-1)),:);
% modulate the correlation sequence
fm_gain = pi/(Fs/2) * Rb/2 * modulation_index;
x = h_baseband * fm_gain;
phase = cumsum(x); % phase is integral of frequency
h_iq = exp(1i*mod(phase, 2*pi));
% create the correlation filter
h = conj(h_iq(end:-1:1,:));
figure(1);
t1 = [1:size(h_ibaseband)(1)];
plot(t1, h_ibaseband(:,1), 'x-');
title('Premable Baseband Gaussian Pulses');
figure(2);
t2 = [1:size(h_baseband)(1)];
plot(t2, h_baseband(:,1), 'x-');
title('Trimmed Premable Baseband Gaussian Pulses');
figure(3);
t3 = [1:size(h)(1)];
plot(t3, real(h(:,1)), 'x-', t3, imag(h(:,1)), 'x-');
title('Correlation Filter Taps');
h(:,1)
Re: Symbol timing estimator block and CPM modulations
Hello Alex, Few months ago, I was also having some similar issues, with respect to GFSK modulation. I was using a M TED and I was getting a loss percentage of about 7% mainly due to having a large amount of tx packets (100 pkts/s) . So, it was suggested to me to use a correlator sync block to pre sync the packets, and then do timing recovery. But, since I also couldn't find a M-fsk or MSK modulation class I start doing some research and I found that the timing recovery block works perfectly if there are no noise in the demodulated output, so I created my own version of demodulator which outputs values only of signal is present, and in case of noise I just output some negative value, say like -5. It just a concept if signal power is above a threshold then it is actual tx if not it is a noise I am not sure if this can solve your problem. Thought just in case you may find it useful. Cheers.! Regards, Mohamed Yaaseen On Sun, 28 Jun 2020 at 02:43, Alex Roberts wrote: > Andy, > > I’m not sure how integrate the correlation sync block with gmsk. It > expects modulated symbols and I’m not sure how to generate a modulated > vector of gmsk symbols. There doesn’t seem to be a gmsk class that can be > used by the modulate vector block. > > Thanks, > Alex > > On Thursday, June 25, 2020, Andy Walls > wrote: > >> Recommend reading >> >> https://www.gnuradio.org/grcon/grcon17/presentations/symbol_clock_recovery_and_improved_symbol_synchronization_blocks/Andy-Walls-Samples-to-Digital-Symbols.pdf >> >> Yeah, the MSK TED selections for the symbol sync block expect the >> constant envelope FSK waveform on the input. Demodulation from FSK to >> baseband pulses should happen after the symbol sync block. >> >
Re: Symbol timing estimator block and CPM modulations
Andy, I’m not sure how integrate the correlation sync block with gmsk. It expects modulated symbols and I’m not sure how to generate a modulated vector of gmsk symbols. There doesn’t seem to be a gmsk class that can be used by the modulate vector block. Thanks, Alex On Thursday, June 25, 2020, Andy Walls wrote: > Recommend reading > https://www.gnuradio.org/grcon/grcon17/presentations/ > symbol_clock_recovery_and_improved_symbol_synchronization_blocks/Andy- > Walls-Samples-to-Digital-Symbols.pdf > > Yeah, the MSK TED selections for the symbol sync block expect the constant > envelope FSK waveform on the input. Demodulation from FSK to baseband > pulses should happen after the symbol sync block. >
Re: Symbol timing estimator block and CPM modulations
Recommend reading https://www.gnuradio.org/grcon/grcon17/presentations/symbol_clock_recovery_and_improved_symbol_synchronization_blocks/Andy-Walls-Samples-to-Digital-Symbols.pdf Yeah, the MSK TED selections for the symbol sync block expect the constant envelope FSK waveform on the input. Demodulation from FSK to baseband pulses should happen after the symbol sync block.
