Wireless Pers Commun

DOI 10.1007/s11277-008-9509-y
An All Digital Implementation of Constant Envelope:
Bandwidth Efcient GMSK Modem using Advanced
Digital Signal Processing Techniques
Arjun Ramamurthy fredric j. harris
Springer Science+Business Media, LLC. 2008
Abstract Gaussian Minimum Shift Keying (GMSK) has been the most common modula-
tion format belonging to the class of partial response Continuous Phase Modulation (CPM)
scheme. It is primarily adopted in the GSMstandards (B=0.3) for land mobile radio commu-
nication systems because of its high bandwidth efciency and constant envelope modulation
characteristics. The focus of this paper is the design of the demodulator wherein we demon-
strate an all digital implementation of sub-optimal synchronization techniques for a GMSK
modem based on two Laurent Amplitude modulation pulse (AMP) streams approximation
representing the matched lter. In this all digital implementation, we perform a joint estima-
tion of the symbol timing and carrier offset wherein the symbol timing is performed using
interpolation techniques.
Keywords GMSK synchronization O-QPSK Joint timing and carrier recovery
Polyphase matched lter
1 Introduction
For several years Continuous Phase Modulation (CPM) has been a well studied class of
modulation characterized with high performance metrics, primarily efcient spectrum uti-
lization and power efciency. CPM schemes can be broadly classied into two categories,
namely full response and partial response depending upon whether the modulation frequency
pulse is of single symbol duration of longer. Minimum Shift Keying (MSK) has been the
popular example of a full response spectrally efcient modulation scheme [12]. Within the
class of partial response CPMscheme, Gaussian MinimumShift Keying (GMSK) is the most
A. Ramamurthy (B)
San Diego State University, 6450, Sequence Dr., San Diego, CA 92121, USA
e-mail: arjun.ramamurthy@motorola.com
f. j. harris
Department of Electrical and Computer Engineering, San Diego State University, 5500 Campanile Drive,
San Diego, CA 92182-1309, USA
e-mail: fred.harris@sdsu.edu
1 3
A. Ramamurthy, f. j. harris
{a}
n
p(t-nT)
a g(t-nT )
n b
a (-1,+1)
n
Gaussian
Filter
h(t)
Frequency
Modulator
Data NRZ Pulse Train

f
c
s(t)
Fig. 1 GMSK transmitter (CPM representation) Redrawn from [1]
common example for its high spectral efciency and constant envelope modulation charac-
teristics. It is an h=0.5 partial response continuous phase modulation scheme derived from
MSK with the addition of baseband Gaussian ltering applied to the identically and inde-
pendently distributed (i.i.d) random rectangular pulse shaped input signal prior to frequency
modulation of the carrier. As indicated in [1], it is important to emphasize that although the
acronymGMSKwas assigned to the termGaussian-ltered MSKin [2], the modulation actu-
ally described in this reference applies to Gaussian ltering of rectangular pulses at baseband
as shown in Fig. 1 [1], i.e. prior to modulation onto the carrier, and, hence, it does not destroy
the constant envelope property of the resulting modulation. It is assumed that the frequency
pulse g(t) in Fig. 1 is the result of a convolution (ltering) operation performed between the
rectangular pulse p(t) and the Gaussian lter h(t) as illustrated in Eqs. 13.
Although g(t) of Eq. 1 (or Eq. 4) appears to have a Gaussian-looking shape, we empha-
size that the word Gaussian in the GMSK refers to the impulse response h(t) of the lter
through which the input rectangular pulse train is passed and not the shape of the resulting
frequency pulse g(t).
g(t ) = p(t )

h(t ),

i s convoluti on operati on (1)

p(t ) = 1, 0 t T
s
(2)
h(t ) = 1/(2
2
)
1/2
. exp(t
2
/2
2
),
2
= ln2/(2B)
2
(3)
Alternatively, as illustrated in [1: Eq. 2.8.52], the GMSK frequency pulse is the difference
of the two time displaced (by T
s
seconds) Gaussian probability integrals (Q functions), i.e.
where B is the single sided 3-dB bandwidth of the Gaussian lter h(t ) as shown in Fig. 2.
g(t ) =
1
2T
S
_
Q
_
K
_
t
T
S
1
__
Q
_
K
_
t
T
S
___
K = 2
BT
S

ln(2)
Q(x) =
_

x
1

2
exp
_

y
2
2
_
dy
(4)
It is a common practice to refer to the product of the Gaussian lter 3dB bandwidth and
the coded symbol period as the BT
s
factor, commonly used to vary the spectral (bandwidth)
efciency of the modulated signal. However, in practice it is only the 3dB bandwidth (B) of
the Gaussian lter that controls the spectral occupancy as shown in Fig. 2. Smaller values of B
result in lesser spectral bandwidth occupancy (Fig. 2) but greater ISI (Figs. 35) resulting in
higher BER. Hence, depending upon the application, a particular value of Bis selected result-
ing in a compromise between spectral efciency and BER performance. Viterbi equalization
or trellis demodulation [3] techniques are normally used to compensate for the increased ISI
due to lower B values (or equivalently lower BT
s
product). The Gaussian shape provides an
additional advantage of reduced sidelobe levels. The constant envelope of the GMSK signal
reduces spectral re-growth and signal distortion due to any amplier nonlinearity.
1 3
An All Digital Implementation of Constant Envelope
Fig. 2 Impulse response and Frequency response of Gaussian lter h(t), and Gaussian ltered frequency pulse
g(t) with B=0.5, 0.3, 0.25
Fig. 3 Eye diagram I-Q representation of GMSK signal with B=0.5
1 3
A. Ramamurthy, f. j. harris
Fig. 4 Eye diagram I-Q representation of GMSK signal with B=0.3
Fig. 5 Eye diagram I-Q representation of GMSK signal with B=0.25
1 3
An All Digital Implementation of Constant Envelope
{a }
n
p(t-nT)
a g(t-nT )
n b
a (-1,+1)
n
Gaussian
Filter
h(t)
Integrator
Data NRZ Pulse Train

f
c
s(t)
cos(--)
cos(--)
-
/2
Fig. 6 GMSK transmitter (CPM representation) Redrawn from [1]
2 Modulator
The GMSK modulator can be implemented as a FSK modulator or as an offset QPSK
modulator [3]. It is a common practice to adopt the I-Q implementation to generate the
GMSK signal resulting in a representation equivalent to staggered I-Q modulation format.
Subsequently, in addition to the intentional and controlled ISI introduced in the GMSK
signaling format, there exists correlation between adjacent symbols on the quadrature chan-
nels. As a result, the conventional receiver synchronization techniques, involving the carrier
and symbol timing recovery needs to be modied to accommodate this cross talk between
adjacent symbols on the I and Q channels.
There are various optimal and suboptimal methods used to estimate the carrier offset and
symbol timing to achieve successful signal detection and decoding at the demodulator. We
follow the offset quadrature implementation approach as shown in Fig. 6. From Eq. 4 it
can bee seen that the Q function is doubly innite in extent and its a common practice to
time truncate the GMSK frequency pulse resulting in nite ISI. It has been shown in [46]
that g(t) is truncated to 4 symbol intervals and 3-symbols intervals for B=0.25 and B=0.3
respectively.
The GSM application adopts the B=0.3 considering ISI only from adjacent neighbors
(say, for symbol
n
, only
n1
and
n+1
introduce ISI) as shown in [6]. Hence, in practical
GMSK implementations, an I-Q representation (Fig. 6) with an approximation as shown in
Eq. 5 [1:Eq. 2.8.54] is adopted where L is chosen based on the value of B. Based upon this,
in our computer simulations we have illustrated the modulation-demodulation technique for
B = 0.3.
g(t ) =
1
2T
S
_
Q
_
K
_
t
T
S
1
__
Q
_
K
_
t
T
S
___
where (L 1)
T
S
2
t (L + 1)
T
S
2
0 Elsewhere (5)
3 Channel Impairments
The signal received from the channel arrives with unknown amplitude, time delay, frequency
shift, phase shift and noise. This is shown in Eq. 6
r(t) = Re{A(t) . R(t ) . e
j(t)
. e
jc(t)
. e
jt
+ n(t) . e
jc(t)
} (6)
We will assume that the unknown channel attenuation A(t), varies slowly with respect
to the modulation signal and is removed by an AGC loop. If A(t) introduces frequency
1 3
A. Ramamurthy, f. j. harris
Fig. 7 Signal degradation due to channel impairments
Fig. 8 First and Second Laurent AMP streams with B=0.5
dependent distortion, the received signal will have to be processed by an equalizer. We
assume this is done and can ignore interaction between the various loops (AGC, carrier,
timing and equalizer). Thus the signal delivered to the demodulator can be described as a
known signal, s(t) with unknown values of parameters namely, frequency offset (), phase
offset () and time delay () as shown in Eq. 7 and illustrated in Fig. 7
r(t) = s(t : , , ) (7)
4 Demodulator
The demodulator can be implemented in a number of different optimum and sub-optimum
methods as indicated in [1]. We use the Laurents sub-optimal two pulse approximation
method in our simulations. Laurent [7] described a representation of CPM signal as a super-
position of phase shifted amplitude modulation pulse (AMP) streams. In [1], a detailed
interpretation of the key results of this paper has been conducted for both, the exact and
approximate AMPrepresentationof GMSKsignal. Inour computer simulations of the demod-
ulator we have used only the rst two Laurent AMP streams (matched lter) because the rst
and second AMP component corresponding to the pulse streams C
0
[n] and C
1
[n] contains the
significant fraction of 0.991944 and 0.00803 of the total energy [1]. Figures 810 demonstrate
the rst two pulse stream AMP representation for different values of B.
1 3
An All Digital Implementation of Constant Envelope
Fig. 9 First and Second Laurent AMP streams with B=0.3
Fig. 10 First and Second Laurent AMP streams with B=0.25
In this all digital implementation, we perform a joint estimation of the symbol timing and
carrier offset. The carrier phase synchronization can be performed based on the MAP and/or
ML phase estimation techniques using data aided/non-data aided or decision directed con-
gurations. We perform the carrier phase synchronization using the non-data aided method
as detailed in [8]. The symbol timing is performed using an interpolation technique [9].
1 3
A. Ramamurthy, f. j. harris
Fig. 11 Three samples, early,
punctual, late samples on positive
and negative correlation.
Redrawn from [9]
5 Symbol Timing EstimationRecovery
In modern digital receivers the sampling is performed with a xed clock i.e. the received
signal is not synchronized with the incoming data symbols. Hence, subsequent to the ana-
log-to-digital conversion the timing adjustment is performed using interpolation techniques.
In this method, the phase alignment occurs not by moving the sample instances to the correct
position in the time waveform, but rather by interpolation of the matched lter samples from
the collected sample locations to the desired sample locations. This is achieved by designing
a polyphase matched lter with an increased sample rate using multirate signal processing
techniques whose lter coefcients are aligned with the data samples. This matched lter
operating at an increased sample rate is then re-sampled to attain the lter response at the
original sample rate with successive time offsets of 1/M, 2/M, 3/M etc, to form a bank of M
lters matched to different time offsets between the input sample location and the envelope
of the received waveform. The timing recovery process determines one of the M possible
paths required to align the lter with the signal time offset.
We emphasize that the M paths (or phases) of the polyphase matched lter correspond to
the acceptable time granularity of the re-sampling process. Also, for a xed roll-off factor ,
the width of the eye openings decrease with an increase in the number of levels in the eye
diagram, and for a xed number of levels in the eye diagram, the width of the eye opening
also decreases as the roll-off factor , decreases [9].
We have an intuitive sense that timing information in a modulated signal resides in signal
level transitions. The transition detector seeks the midpoint of the matched lter output as it
transitions between different modulation states. The timing recovery process must ascertain
if the clock sample position is at the correct position or needs to be advanced or retarded
relative to the input time waveform. A derivative operator is used to detect the signal tran-
sitions i.e. the correlation function of the correct matched lter (selected from the bank of
M polyphase matched lters) would have a zero derivative at the peak. This has been moti-
vated from the idea of the early and late gate output values used in the conventional analog
timing recovery systems. However, in addition to the information provided by the derivative
function, the timing recovery process also needs to know a conditional piece of information
to reliably decide the direction of movement with respect to the received time waveform.
This conditional additional information represents the sign of the sample of the correlation
function which is folded into the observable error parameter in the timing recovery process.
If the sample is positioned prior to the peak, a positive sign represents positive slope and a
negative sign represents negative slope as illustrated in Fig. 11 [9]. However, if the sample
is located ahead of the peak the reverse is true.
1 3
An All Digital Implementation of Constant Envelope
MF
PA
DMF
Index
Selection
LF
1:2
Tanh
Z
-1
Tanh
Z
-1
I
Q
Q
I
MF
PA
DMF
Index
Selection
LF
1:2
Tanh
Z
-1
Tanh
Z
-1
I
Q
Q
I
PA LF
C
0
Path
C
1
Path
DDS
2 samples/symbol
*
Fig. 12 Joint symbol timing and carrier recovery block diagram (MF: Matched Filter, DMF: Derivative
Matched Filter, LF: Loop Filter, PA: Phase Accumulator, DDS: Direct Digital Synthesizer)
Fig. 13 Eye diagram at the output of the combined (C
0
[n] and C
1
[n]) matched lter
1 3
A. Ramamurthy, f. j. harris
Fig. 14 Transition diagram and constellation plot at the output of the combined (C
0
[n] and C
1
[n]) matched
lter
Furthermore, zero crossing points can also be used to align the symbol timing during the
timing recovery process. At this point the derivative function would hold a maximum value.
Using a simple sine wave as a model, it is easy to visualize that the midpoint of the transition
corresponds to the location of the maximumvalue of the derivative. Thus the transition detec-
tor often seeks the time location for which the magnitude of the average slope is a maximum.
As shown in Fig. 11 this time corresponds to the midpoint of the symbol interval and also
coincides with the center of mass of the zero-crossings in the eye-diagram.
In sampled data systems, the derivative of the bank of polyphase lters, say, when testing
segment (k) is derived using polyphase segments (k 1) and (k+1) as shown in Eq. 8.
(n + k/M) = y(n) . h
k+1
(n) y(n) . h
k1
(n)
= y(n) . [h
k+1
(n) h
k1
(n)]
= y(n) .

h
k
(n) (8)
Since the input signal is collected at two samples/symbol, one sample, the one with an
even index for instance, is chosen as a data sample, and the other is the non data sample
sometimes called the timing sample. The data sample is the one delivered to the detector. If
the address pointer tries to cross the address boundaries, the address wraps circularly and the
input sample identied as the data sample is switched to the odd index as detailed in [9].
These parameters can be estimated independently in a specied order, or can be estimated
with aid from previous estimators in the sequential chain, or can be estimated concurrently
but independently, or concurrently and cooperatively [11]. The most common practice is
concurrent independent acquisition or concurrent aided acquisition.
Another common practice involves switching between modes, initializing unaided and
then operating in the aided mode to improve noise immunity. A severely distorted wave-
shape may defeat the acquisition of the timing parameters by the timing recovery PLL.
To reliably demodulate a received signal, in addition to estimation of symbol timing,
the carrier phase of the receiver must be synchronized to the carrier phase of the received
signal. Phase lock loops operating in a closed loop mode provides the appropriate control
to the carrier phase recovery process. In this paper we illustrate through computer simula-
tions an all digital implementation of joint symbol and phase estimationrecovery scheme
1 3
An All Digital Implementation of Constant Envelope
Fig. 15 Symbol timing using polyphase matched lter
Fig. 16 PLL error and input-output phase plots
1 3
A. Ramamurthy, f. j. harris
Fig. 17 Comparison of input-output phase prole and GMSK I-Q signal
in the presence of non-ideal channel, carrier offset and noise. Figure 12 demonstrates the
block diagram for the joint estimation of symbol timing and carrier recovery used in our
simulations.
6 Computer Simulations
Figures 1317 illustrate computer simulations using eye diagrams and constellation plots
representing the signal at various stages of the demodulator. Figure 13 presents the eye dia-
grams of the offset in-phase and quadrature components at 2-samples per symbol formed
by the I and Q matched lters. Figure 14 shows the transition diagram and the constellation
diagram of the ordered pairs formed by I and Q matched lters. Figure 15 shows the content
of the timing recovery phase accumulator and the quantized output, the index pointing to the
branch of the polyphase matched lter.
Figure 16 presents the carrier phase error, the input to the carrier PLL, and the input phase
prole of the received signal carrier and the phase prole of the receivers locally gener-
ated direct digital synthesizer. Figure 17 presents the time and phase aligned versions of the
phase and quadrature components of the base band signal formed at the modulator and then
estimated at the receiver.
7 Conclusions
This paper has described a GMSK modem using the sub-optimal Laurents AMP approxi-
mation method for matched lter processing and implemented with the Simon ML receiver
1 3
An All Digital Implementation of Constant Envelope
architecture for carrier offset and symbol timing estimation and recovery. We further have
demonstrated the use of polyphase matched lter in the timing recovery process to support an
all DSP based implementation of a receiver [9]. Afull exposition of DSP based receiver struc-
tures for a larger class of efcient modulation formats related to staggered I-Q modulation
can be found in [10].
References
1. Simon, M. K. (2003). Bandwidth efcient digital modulation with application to deep space communi-
cations, Wiley-Interscience.
2. Murota, K., Kinoshita, K., & Hirade, K. (1981). Spectrum efciency of GMSK land mobile radio Inter-
national Conference on Communications, 2, pp. 23.8.123.8.5, June 1420.
3. Consultative Committee for Space Data Systems (2003). Bandwidth efcient modulations, CCSDS 413.0-
G-1. Green Book.
4. Kaleh, G. K. (1989). Simple coherent receivers for partial response continuous phase modulation IEEE
Journal on Selected Areas in Communications, 7(9), 14271436.
5. Hodges, M. R. L. (1990). The GSM radio interface British Telecom Technological Journal, 8(2).
6. Haspeslagh, J. et al. (1990) A270 Kb/s 35-mWmodulation ICfor GSMcellular radio hand held terminals.
IEEE Journal on Solid State Circuits, 25(12), 14501457.
7. Laurent, P. A. (1986). Exact and approximate construction of digital phase modulations by superposition
of amplitude modulated pulses. IEEE Transactions on Communications, COM-34(2), 150160.
8. Vassallo, E., & Visintin, M. (2002). Carrier phase synchronization for GMSK signals. International
Journal on Satellite Communications, 20, 391415 (2002). doi:10.1002/sat.729.
9. harris, f. j. (2004). Multirate signal processing for communication systems, Prentice Hall PTR.
10. Ramamurthy, A. (2006). Synchronization methods for efcient modulation formats related to staggered
I-Q modulation, Masters Thesis, Spring 2006, San Diego State University, San Diego.
11. harris, f. j. Manuscript on Synchronization for digital communications, available from author, San Diego
State University (unpublished manuscript).
12. Pasupathy, S. (1979). Minimum shift keying: A spectrally efcient modulation. IEEE Communication
Magazine, 17(4), 1422.
Author Biographies
Arjun Ramamurthy received his B.E in Electronics and Communica-
tion from National Institute of Engineering, University of Mysore, India
in 2001. He worked as a DSP Engineer at Deptartment of Embedded
and Product Engineering, Wipro Technologies, Bangalore, India until
2004. He received his M.S in Electrical Engineering from San Diego
State University in 2006 and is currently working as Sr. Software Engi-
neer at the Advanced Technology Group, Motorola Inc, San Diego. His
interests include in Multirate Signal Processing, Modern Wireless Com-
munication, Video Coding algorithms, Embedded software development
for multimedia applications.
1 3
A. Ramamurthy, f. j. harris
fedric j. harris hold the CUBIC Signal Processing Chair of the Com-
munication Systems and Signal Processing Institute at San Diego State
University where since 1967 I have taught courses in Digital Signal Pro-
cessing and Communication Systems. I hold a number of patents on
digital receiver and DSP technology and lecture throughout the world
on DSP applications. I consult for organizations requiring high perfor-
mance, cost effective DSP solutions.
I have written over 140 journal and conference papers, the most well
known being my 1978 paper On the use of Windows for Harmonic
Analysis with the Discrete Fourier Transform. I am the author of the
book Multirate Signal Processing for Communication Systems and
I have contributed to a number of other books on DSP applications
including the Source Coding chapter in Bernard Sklars 1988 book,
Digital Communications and the Multirate FIR Filters for Interpola-
tion and Resampling and the Time Domain Signal Processing with the
DFT chapters in Doug Elliots 1987 book Handbook of Digital Signal Processing.
In 1990 and 1991 I was the Technical and then the General Chair of the Asilomar Conference on Signals,
Systems, and Computers and was Technical Chair of the 2003 Software Dened Radio Conference and of the
2006 Wireless Personal Multimedia Conference. I became a Fellow of the IEEE in 2003, cited for contribu-
tions of DSP to communications systems. In 2006 I received the Software Dened Radio Forums Industry
Achievement Award. I am the Co-Editor-in-Chief of the Elsevier DSP Journal.
The spelling of my name with all lower case letters is a source of distress for typists and spell checkers.
A child at heart, I collect toy trains and old slide-rules.
1 3