See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/2575678

GMSK in a nutshell

Article · October 1996

Source: CiteSeer

CITATIONS READS
9 35,914

1 author:

Thierry Turletti
National Institute for Research in Computer Science and Control
234 PUBLICATIONS   8,787 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Direct Code Execution View project

Practical architectures for Information-Centric Networks View project

All content following this page was uploaded by Thierry Turletti on 29 April 2013.

The user has requested enhancement of the downloaded file.

 GMSK in a nutshell
Thierry Turletti
Telemedia Networks and Systems Group
Laboratory for Computer Science
Massachussets Institute of Technology
[email protected]

April, 96

Abstract
This report presents a brief overview of the GMSK modulation scheme.

1 What is GMSK ?
GMSK stands for Gaussian Minimum Shift Keying. This is a modulation scheme in
which the phase of the carrier is instantaneously varied by the \modulating" signal (i.e.
the information to transmit). GMSK diers from MSK (Minimum Shift Keying) in
that a Gaussian Filter of an appropriate bandwidth (dened by the BT product) is used
before the modulation stage. The time-domain impulse response of the lter is described
in Equation 1, where k1 = p2ln 2 and B is the half-power bandwidth.
k1 B ? k12 B2 t2
h(t) = p e (1)
MSK is binary digital FM with a modulation index of 0.5. It has the following important
characteristics: constant envelope, relatively narrow bandwidth and coherent detection
capability. The most important characteristic of MSK is that it is a constant-envelope
variety of modulation. This makes the modulation scheme more immune to noise than
the Amplitude Shift Keying (ASK) scheme. However, MSK does not satisfy the require-
ments with respect to out-of-band radiation for single-channel-per-carrier (SCPC) mo-
bile radio. GMSK uses a pre-modulation Gaussian lter which makes the output power
spectrum more compact. The pre-modulation Gaussian lter has narrow bandwidth and
sharp cuto properties which are required to suppress the high-frequency components.
Moreover, it has a lower overshoot impulse response which allows to protect against
excessive instantaneous deviation.

1
 Figure 1 shows the 16-bit NRZ (Non-Return-to-Zero) sequence (?1; ?1; ?1; +1;
+1; ?1; +1; +1; +1; +1; ?1; +1; ?1; +1; ?1; ?1) and the corresponding phase trajectory
of MSK (left) and GMSK (right) signals. The phase increment per symbol is  2 for
the MSK signal. Figure 2 shows the in phase I (real) and quadrature Q (imaginary)
components of the MSK (left) and GMSK (right) corresponding baseband equivalent
signals. Finally, gure 3 shows the MSK and GMSK modulated signals for two dier-
ent symbols. Notice the slight dierence of frequency between the modulated signal of
symbol (?1) and symbol (1). This shows the FM nature of MSK and GMSK signals.

2 How to implement a GMSK modulator ?

There are many ways to implement a GMSK modulator. A block diagram of such a
modulator is shown in Figure 4.
An algorithm of a GMSK modulator is described below:
 Create the NRZ (?1; 1) sequence from the binary (0,1) input sequence.
 Create N samples per symbols (N depends of FsT).
 Integrate the NRZ sequence.
 Convolute with a Gaussian function.
 Compute the corresponding I and Q components. At this stage, we have the
quadrature components of the baseband GMSK equivalent signal.
 Multiply the I and Q components by the corresponding cos(n w0); and ? sin(n w0)
carriers.
 Add the two resulting ows.

3 How to implement a GMSK demodulator ?

A coherent detector can be used to demodulate GMSK signal, see Figure 5.
However, to avoid the need for the receiver to have its own reference, (i.e. the
necessity of a coherent demodulation) a dierential encoding can be used to create the
NRZ signal at the input of the demodulator. Table 1 shows how to convert the binary
input (0,1) to dierential (?1; 1) symbols. It can be easily computed using the following
operations:
X = x[n]  x[n ? 1]; Y = 1 ? 2 X (2)
 6 6
"symbols.dat" "symbols.dat"
"phase_msk" "phase_gmsk"

4 4

2 2
value

value
0 0

-2 -2

-4 -4

-6 -6
0 100 200 300 400 500 600 0 100 200 300 400 500 600
samples (1 symbol = 36 samples) samples (1 symbol = 36 samples)

1.5
Figure 1: Symbols and phase (in radians) of MSK and GMSK signal vs samples 1.5
"msk_I" "gmsk_I"
"msk_Q" "gmsk_Q"

1 1

0.5 0.5
value

value

0 0

-0.5 -0.5

-1 -1

-1.5 -1.5
0 100 200 300 400 500 600 0 100 200 300 400 500 600
samples (1 symbol = 36 samples) samples (1 symbol = 36 samples)

Figure 2: Baseband (I,Q) MSK and GMSK signals vs samples for fsT=36
"symbols.dat" "symbols.dat"
"msk.dat" "gmsk.dat"

1 1

0.5 0.5
value

value

0 0

-0.5 -0.5

-1 -1

60 70 80 90 100 110 120 130 140 150 60 70 80 90 100 110 120 130 140 150
samples (1 symbol = 36 samples) samples (1 symbol = 36 samples)

Figure 3: MSK and GMSK signal vs samples for fsT=36

 cos (nwo)

I
cos

+
NRZ signal
+1 o
Gaussian s (n) = cos ( nwo+ o )
Integrator
Filter + m
−1
s (n) Q
sin

+
− sin (nwo)

Figure 4: GMSK modulation block diagram

cos (nwo)

Low−Pass I
+

Filter cos (o (n)) NRZ signal

I
+1
cos ( nw + o(n)) Arctan ( ) Derivator
Q
−1
Low−Pass sin (o(n) )
+

Filter Q

sin (nwo)

Figure 5: GMSK Demodulation block diagram

4 A specic GMSK scheme for GSM

GSM uses GMSK with modulation index h = 0:5, BT (lter bandwidth times bit
period) equal to 0.3 and a modulation rate of 271 (270 5/6) kbauds. That makes
B = 81:3 kHz when T is about 3:7 s. The GMSK modulation has been chosen as a
compromise between a fairly high spectrum eciency (of the order of 1 bit/Hz) and a
reasonable demodulation complexity. The constant envelope allows the use of simple
power ampliers and the low out-of-band radiation minimizes the eect of adjacent
channel interference.
Theoretically, a modulating symbol in uences the signal during an innite period.
Luckily, this in uence becomes negligible outside a period during 3 T . This source of in-
terference can be limited by separating geographically the usage of adjacent frequencies.
The protocol for frequency use in GSM is an operator dependent function of frequency
planning activities. The GSM specications leave these issues to manufacturers who can
incorporate clever dynamic channel allocation schemes within their equipment.
This source of interference is not the only one a receiver is confronted to: a GSM
receiver has also to tackle the following problems:
 x[n] x[n-1] X Y
0 0 0 +1
0 1 1 ?1
1 0 1 ?1
1 1 0 +1
Table 1: Logical table for dierential encoding

 variable signal attenuation: because of free space loss and also because of the
attenuation due to the presence of obstacles along the way.
 multipath propagation: incoming RF signals results from re ections of the trans-
mitted signal from the environment (houses, mountains and various obstacles)
where each path yields valid signals with dierent relative time delays.
 adding noise: co-channel interference (interferences from other GSM emitters us-
ing the same frequency band at the same time), adjacent band interference (as
discussed before), thermal noise, etc.
The GSM specications do not impose one particular demodulation algorithm. How-
ever, they impose minimal performance gures measured after correction of errors by
channel decoding. The algorithm used must be able to cope with two multipaths of
equal power received at an interval of up to 16 s (i.e. more than four symbols). With
such a level of intersymbol interference, simple demodulation techniques are ineective
and an equalizer is required. The aim of the equalizer is to separate the dierent signals,
to sharpen and recover the original signal. It is a lter that uses the 26-bit training
sequence included in each burst transmitted1 to clear up the distorted signals. The
longer the possible delays, the fancier the equalizer should be. Viterbi demodulation is a
maximum likelihood technique which nds the most probable emitted sequence, accord-
ing to assumptions on the possible signals and on the adding noise [1]. The algorithm
uses a nite set of possible signal shapes received during one bit period. The received
signal shapes is in uenced by several modulating bits. If the signal depends on n bits,
2n dierent shapes have to be stored in memory. In order to be able to cope with two
multipaths of equal power received at an interval of up to 16 s, 5 bits (i.e. 32 dierent
shapes) are required.

References
[1] Forney, G.D. JR. \The Viterbi Algorithm" Proc. of the IEEE, Vol. 61, No 3, Mar.
1973, pp. 268-278.
1 All the signals in a particular cell share the same training sequence.
 [2] Mouly, M. and Pautet, M-B. \The GSM System for Mobile", ISBN: 2-9507190-0-7,
1992.
[3] Murota, K., Hirade, K. \GMSK Modulation for Digital Mobile Radio Telephony",
IEEE Trans. on Communications, Vol. com-29, No 7, July 1981.
[4] Redl, S.H., Weber, M.K., Oliphant, M.W. \An Introduction to GSM", The Artech
House Mobile Communications Series, ISBN 0-89006-785-6, 1995.

View publication stats