# GMSK: GSM initial phase for a burst

Hello everyone.

I am looking at the frequency correction burst of a GSM signal.
According to spec the original data defines 148 bits of 0 that are
differentially encoded and GMSK modulated. After the data goes
through the differential encoder it ends up as [-1,1,1, (142 1s),
1,1,1]. I have taken this burst and plotted I vs Q (which is
attached). I was expecting to see the amplitude ramp up followed by a
negative pi/2 phase shift, followed by 147 positive pi/2 phase shifts,
and a ramp down. I never see the initial negative phase shift though.
Am I missing something in my sanity? Any thoughts or ideas?

Thank you for your time
Jeff

Jeffrey K. wrote:

Am I missing something in my sanity? Any thoughts or ideas?
Its hard to read the graph like that. Plot angle vs time or even
better, derivative of angle vs. time.

Matt

Great suggestion!

I have attached a couple of things:

1. The IQ plot (1e6 sample rate)
2. The burst data (MATLAB array)
3. The da vs time plot. (1 symbol = 3.69us)

Another quick question. I can now see that after the amplitude ramp
up that there is a short section of negative phase movement (.7
radians clockwise) prior to the remaining counterclockwise rotations.
Should I expect a full (-pi/2) to start off the burst? Is my sampling
rate just not high enough to catch the remaining rotation? Any other
suggestions?

Thank you for your time
Jeff

On 6/18/07, Jeffrey K. [email protected] wrote:

Another quick question. I can now see that after the amplitude ramp
up that there is a short section of negative phase movement (.7
radians clockwise) prior to the remaining counterclockwise rotations.
Should I expect a full (-pi/2) to start off the burst? Is my sampling
rate just not high enough to catch the remaining rotation? Any other
suggestions?

GMSK has a certain level of ISI built into the waveform. 0.7 radians
is about a pi/2 shift (0.2228 versus 0.25?) which would be pretty darn
close.

You can look at the inherent ISI within GMSK in a nice pretty graph in
this PDF:
http://www.ieee802.org/16/tg1/phy/contrib/802161pc-00_11.pdf

You may want to filter your samples down to 1 sample per symbol and
see if you get 3 dots close to each of your 4 constellation points (if
you have already compensated for any phase rotations).

Brian

On 6/18/07, Brian P. [email protected] wrote:

GMSK has a certain level of ISI built into the waveform. 0.7 radians
is about a pi/2 shift (0.2228 versus 0.25?) which would be pretty darn
close.

Whoops - looks like I’ve got QPSK on the brain today. Sorry about that.

Not really sure why you couldn’t see at least something close to a pi/2
shift.

Brian

Jeffrey K. wrote:

Should I expect a full (-pi/2) to start off the burst? Is my sampling
rate just not high enough to catch the remaining rotation? Any other
suggestions?

I don’t see anything at all wrong with this data.

Matt

Is there an initial condition to the differential encoder? Seems odd
that
they wouldn’t just send the 148 pi/2 phase shifts. -Clark

differentially encoded and GMSK modulated. After the data goes
through the differential encoder it ends up as [-1,1,1, (142 1s),
1,1,1]. I have taken this burst and plotted I vs Q (which is
attached). I was expecting to see the amplitude ramp up followed by a
negative pi/2 phase shift, followed by 147 positive pi/2 phase shifts,
and a ramp down. I never see the initial negative phase shift though.
Am I missing something in my sanity? Any thoughts or ideas?

Thank you for your time
Jeff

<< fcch2.png >>

Get a preview of Live Earth, the hottest event this summer - only on MSN
http://liveearth.msn.com?source=msntaglineliveearthhm