Coherent vs Noncoherent GFSK Demodulation

Hi,

I've been working on demodulation of low-deviation GFSK signals similar
to Bluetooth, with h=0.3 and BT=0.5. I have two simulations, one using
a discriminator plus integrate and dump, and another with a second
order PLL. I am computing BER vs Eb/No for each case. In the PLL case,
I expect to see a few dB of improvement compared to the discriminator,
but I'm not getting that. I'm either misunderstanding something, or I
have made a mistake in the implementation.

The PLL locks onto the average phase of the carrier, leaving a phase
error term that, when differentiated and filtered (with a boxcar),
gives the bit waveform. What I don't get is how this is any better than
skipping the PLL entirely.

Is it possible to use a PLL to obtain better sensitivity for this class
of FSK signals?

Thanks,

John

"john" <johns@xetron.com> writes:

> Hi,
>
> I've been working on demodulation of low-deviation GFSK signals similar
> to Bluetooth, with h=0.3 and BT=0.5. I have two simulations, one using
> a discriminator plus integrate and dump, and another with a second
> order PLL. I am computing BER vs Eb/No for each case. In the PLL case,
> I expect to see a few dB of improvement compared to the discriminator,
> but I'm not getting that. I'm either misunderstanding something, or I
> have made a mistake in the implementation.
>
> The PLL locks onto the average phase of the carrier, leaving a phase
> error term that, when differentiated and filtered (with a boxcar),
> gives the bit waveform. What I don't get is how this is any better than
> skipping the PLL entirely.
>
> Is it possible to use a PLL to obtain better sensitivity for this class
> of FSK signals?

Hi John,

How do you get baud lock with your integrate/dump in the discriminator
model? That's the real issue. 

Also, how do you know your phase lock is good in the PLL model? You might want to try
using the carrier of the transmitter in the demodulator as a test to see
that the PLL is getting locked correctly. If it is their shouldn't be much
difference in the BERs, I would think. Caveat: I'm no expert.
-- 
%  Randy Yates                  % "I met someone who looks alot like you,
%% Fuquay-Varina, NC            %             she does the things you do, 
%%% 919-577-9882                %                     but she is an IBM."
%%%% <yates@ieee.org>           %        'Yours Truly, 2095', *Time*, ELO   
http://home.earthlink.net/~yatescr

Randy Yates wrote:
> "john" <johns@xetron.com> writes:
>
> > Hi,
> >
> > I've been working on demodulation of low-deviation GFSK signals similar
> > to Bluetooth, with h=0.3 and BT=0.5. I have two simulations, one using
> > a discriminator plus integrate and dump, and another with a second
> > order PLL. I am computing BER vs Eb/No for each case. In the PLL case,
> > I expect to see a few dB of improvement compared to the discriminator,
> > but I'm not getting that. I'm either misunderstanding something, or I
> > have made a mistake in the implementation.
> >
> > The PLL locks onto the average phase of the carrier, leaving a phase
> > error term that, when differentiated and filtered (with a boxcar),
> > gives the bit waveform. What I don't get is how this is any better than
> > skipping the PLL entirely.
> >
> > Is it possible to use a PLL to obtain better sensitivity for this class
> > of FSK signals?
>
> Hi John,
>
> How do you get baud lock with your integrate/dump in the discriminator
> model? That's the real issue.
>
> Also, how do you know your phase lock is good in the PLL model? You might want to try
> using the carrier of the transmitter in the demodulator as a test to see
> that the PLL is getting locked correctly. If it is their shouldn't be much
> difference in the BERs, I would think. Caveat: I'm no expert.
> --
> %  Randy Yates                  % "I met someone who looks alot like you,
> %% Fuquay-Varina, NC            %             she does the things you do,
> %%% 919-577-9882                %                     but she is an IBM."
> %%%% <yates@ieee.org>           %        'Yours Truly, 2095', *Time*, ELO
> http://home.earthlink.net/~yatescr

Thanks, Randy. I have a couple methods of obtaining baud lock.
Remember, this is just a simulation at this point. The first method is
that there is no timing error, i.e. perfect sampling. The second method
is to use an S-curve to choose the best phase over a packet. The third
method is to use a transition tracking PLL to track the best sampling
phase even when clocks drift. In no case do I see the improvement I
expect with coherent detection.

Regarding the phase, I check the PLL transient step response with an
unmodulated input, and plot against the theory for a second order loop.
The agreement is excellent. Now I turn on the GFSK modulation, with no
change to the PLL parameters. 

John