REVISTA TELECOMUNICAÇÕES, VOL. 11, NO.

01, MAIO DE 2008

Contributions to the Understanding

of the MSK Modulation
Dayan Adionel Guimarães

Although covered in many papers and good books on

Abstract—This tutorial deals with key aspects of the MSK Digital Communications, some of the concepts of this
(Minimum Shift Keying) modulation, aiming at unveiling some of modulation are hidden or difficult to understand, representing
its hidden concepts. Signal generation and demodulation are opportunities for alternative approaches, like the one adopted
analyzed in detail. Common questions concerning the study of the in this tutorial. This approach is intended to help everyone
MSK modulation are addressed and answered, e.g. the
who wants to have an understanding about the MSK
similarities and differences among MSK, Sunde’s FSK
(Frequency Shift Keying) and SQPSK (Staggered Quaternary
modulation, especially the practicing engineers and the first-
Phase-Shift Keying) or OQPSK (Offset QPSK); the relation level graduate students in Telecommunications. It addresses
among the modulating data stream, its differentially-decoded some key questions about the MSK modulation, such as:
version, the frequency shifts and the phase shifts of the modulated
signal, and the MSK signal-space representation. 1 – To which extent the MSK modulation can be regarded as a
Index Terms—MSK, FSK, SQPSK and OQPSK modulations. special case of the conventional Sunde’s [6] [7, p. 381] FSK
Resumo—Este tutorial trata de aspectos chave sobre a (Frequency Shift Keying) modulation?
modulação MSK, objetivando revelar alguns dos seus conceitos 2 – To which extent the MSK modulation can be detected in
muitas vezes não revelados explicitamente. A geração e a the same way as the Sunde’s FSK modulation?
demodulação do sinal MSK são analisadas em detalhe. Ao longo
3 – To which extent the MSK modulation can be regarded as a
do trabalho procura-se responder a algumas questões intrigantes
relacionadas com, por exemplo, as similaridades e diferenças special case of the SQPSK or OQPSK (Staggered or Offset
entre as modulações MSK, FSK (de Sunde) e SQPSK ou OQPSK QPSK) modulation?
e a relação entre a seqüência moduladora, sua versão decodificada 4 – To which extent the frequency and phase shifts of an MSK
diferencialmente, os desvios de freqüência e de fase do sinal signal are related to the modulating data sequence?
modulado e a representação do sinal MSK no espaço euclidiano. 5 – To which extent the phase shifts of an MSK signal can be
Palavras chave— Modulações MSK, FSK, SQPSK e OQPSK. related to the phase transition diagram on its signal-space
representation?
I. INTRODUCTION
The Minimum Shift Keying (MSK) modulation, also The remaining of this work is organized as follows: Section
known as “fast FSK” [1], was first considered during the early II addresses some fundamental concepts about the signal-space
60s and 70s [2]-[4], and its characteristics have gained the representation, the complex representation of signals and
attention of the scientific community during the subsequent systems, and the minimum separation between tones in an
decades. orthogonal FSK signaling. Section III is devoted to the
MSK modulation has features such as constant envelope, analysis of the signal construction from the signal-space
compact spectrum and good error performance, which are all expansion and the complex representation approaches. The
desirable in many digital communication systems. Its MSK spectral content, receiver structure and system
utilization goes from the Global System for Mobile performance are also analyzed in Section III. Further attributes
Communication (GSM), in which a Gaussian-filtered MSK and uses of the MSK modulation are summarized in Section
(GMSK) modulation is employed, to micro-satellite IV, and Section V addresses the answers to the questions
communications, positioning and navigation systems, hybrid highlighted above, concluding the work.
optical/wireless communication systems, deep space
communications and, more recently, to the Blue Ray disc II. BASIC CONCEPTS
technology [5], only to mention a few examples. In this section the reader are invited to revisit some
Like many recently rediscovered technologies developed fundamental concepts about signal-space representation and
several years, or even decades ago, the MSK modulation complex representation of signals and systems. Although
seems to be one more idea whose time has come. applicable to the study of digital communications in general,
these two concepts are essential for the study at hand, and will
give us insight on different forms of MSK signal generation
Manuscript received on December 26, 2006. Revised in august 21, 2007. and detection. Additionally, the minimum tone separation for
D. A. Guimarães (dayan@[Link]) is with INATEL - Instituto Nacional coherent detection of orthogonal FSK is analyzed, aiming at
de Telecomunicações. Av. João de Camargo, 510 - Santa Rita do Sapucaí -
MG - Brazil - 37540-000.
 GUIMARÃES: CONTRIBUTIONS TO THE UNDERSTANDING OF THE MSK MODULATION

justifying the term minimum in the name of the MSK The set of functions {φj(t)} are also said to be orthonormal
modulation. and to form an orthornormal basis.
Through Figure 1 it can be seen that the value of a
A. Signal-space representation
coefficient is proportional to a measure of the orthogonality
The signal-space representation is constructed on the basis between the analyzed vector and the corresponding base-
of linear combination theory, and it is very analogous to the vector: the greater the orthogonality, the lesser the value of the
vector algebra theory. Let us define an N-dimensional coefficient. By analogy to the vector algebra, we can
Euclidian space spanned by N orthogonal axes. Let us also determine the values of the coefficients in (3) through a
define a set of orthogonal vectors {φj}, j = 1, 2, …, N, measure of orthogonality between the analyzed waveform and
normalized in the sense that they have unit length. These the corresponding base-function, which leads intuitively to
vectors are said to be orthonormal and to form an
orthornormal basis. T ⎧ i = 1, 2,..., M
Any vector vi, i = 1, 2, …, M in the Euclidian space can be sij = ∫ si (t )φ j (t )dt , ⎨ (5)
0
⎩ j = 1, 2,..., N
generated through the linear combination
N In fact (5) has a formal mathematical justification, which
vi = ∑ vij φj (1) can be obtained by operating generically with (3) and (4):
j =1
T N N

∑ x φ (t ) ∑ y φ (t ) dt
T
where the coefficients vij correspond to the projection of the i- ∫ 0
x(t ) y (t )dt = ∫
0
j =1
j j
k =1
k k

th vector on the j-th base vector. Their values can be N N T

determined by the dot product (or inner product) between vi = ∑∑ x j yk ∫ φ j (t )φk (t )dt (6)
and φj, that is j =1 k =1
0

N
vij = v Ti φj (2) = ∑ x j y j = xT y
j =1

where the superscript T denotes matrix transposition,

vi = [vi1 vi2 … viN]T and φj is also an N-dimensional vector with Expression (6) states that the correlation in time domain
a 1 in the j-th position and zeros otherwise, that is has the inner product as its equivalent in the vector domain.
We are now ready to define the signal-space
φj = [0 1 0 … 0]T for j = 2 as an example.
representation: since knowing the set of coefficients and base-
Figure 1 illustrates these concepts for a two-dimensional
functions is as good as to know the waveform signals
(N = 2) Euclidian space and for two vectors (M = 2). The axes
themselves, we can also represent signals in a Euclidian space.
were labeled in a way to resemble the orthonormal base-
In this representation we use points instead of vectors, to avoid
vectors.
polluting unnecessarily the graph. This kind of plot is also
φ2 called signal constellation. Figure 2 shows a two-dimensional
signal-space used to represent the signals s1(t) and s2(t)
v 22 through the corresponding signal vectors s1 and s2.
v1 2 φ2
v2
v1 s2 s 22
φ2
φ1 s1 2 s1
E2
v 21 0 φ1 v11
E1
Figure 1. Vector-space representation for M = 2 and N = 2.
φ1
In a similar way, one can use the Euclidian space to
represent coefficients that, in a linear combination, will give s 21 0 s1 1
rise to signals instead of vectors. Then we have the signals Figure 2. Signal-space representation for M = 2 and N = 2.
N As can be noticed from Figure 2, the norm of a signal
si (t ) = ∑ sijφ j (t ) , i = 1, 2, ..., M (3) vector, that is, the length of this vector can be determined in
j =1
the light of equation (6) by:
where, now, the set {φj(t)} comprises N orthonormal base- T
functions, one function being orthogonal to each other and si21 + si22 = siT si = ∫
0
si2 (t )dt = Ei (7)
having unit energy, that is:
Generally speaking, the distance from any signal vector to
T ⎧1, i = j
∫0
φi (t )φ j (t )dt = ⎨
⎩0, i ≠ j
(4) the origin of the coordinates is equal to the square root of the
corresponding signal energy:
REVISTA TELECOMUNICAÇÕES, VOL. 11, NO. 01, MAIO DE 2008

T
Ei = ∫ si2 (t )dt = sTi si = ∑ j =1 sij2 = si
N 2
(8) ⎧ 2G ( f ), f > 0
0 ⎪
G+ ( f ) = G ( f ) + sgn( f )G ( f ) = ⎨ G (0), f =0 (16)
As a complementary result, the squared Euclidian distance ⎪0, f <0
⎩
between two signal vectors is obtained through
N
Now, consider a band-pass signal g(t) whose bandwidth is
= ∑ ( sij − skj ) = ∫ [ si (t ) − sk (t )] dt
2 T
d = si − s k
2 2 2
(9) essentially confined in 2W Hz and is small compared to the its
ik
j =1
0
carrier frequency fc. According to (16), the analytic spectrum
G+(f) is centered about fc and contains only positive frequency
The concepts just described will be used later for the components. Then, using the frequency-shifting property of the
understanding of a particular form for the MSK signal Fourier transform we can write:
generation and detection.
g + (t ) = g% (t ) exp( j 2π f c t ) (17)
B. Complex representation of signals and systems
We start by reviewing the concept of Hilbert transform. where g% (t ) is called the complex envelope of the signal g(t)
Following [7] and [8], let g(t) be a signal with Fourier and it is clearly a low-pass signal.
transform G(f). The Hilbert transform of g(t) and the Since g+(t) is a band-pass signal, we can determine the low-
corresponding inverse transform are defined respectively by pass signal g% (t ) through a frequency translation of g+(t) back
to about f = 0. Using again the frequency-shifting property of
1 ∞ g (τ ) ∞ 1
π∫
gˆ (t ) = dτ = ∫ g (τ ) dτ (10) the Fourier transform we can write
−∞ t −τ −∞ π (t − τ )
and g% (t ) = g + (t ) exp( − j 2π f c t )
(18)
1 ∞ gˆ (τ ) = [ g (t ) + jgˆ (t )] exp( − j 2π f c t )
g (t ) = −
π ∫−∞ t −τ
dτ (11)
or, equivalently,
In (10) we can identify that the Hilbert transform of g(t) is
the convolution between g(t) and the function 1/πt. g (t ) + jgˆ (t ) = g% (t ) exp( j 2π f c t ) (19)
By recalling that a convolution in the time domain
corresponds to a multiplication in the frequency domain, and Since the signal g(t) is the real part of the left side of the
by using the Fourier transform pair expression above, we can obtain a very useful representation:

1 g (t ) = Re [ g% (t ) exp( j 2π f c t ) ] (20)
− j sgn( f ) , (12)
πt
Generally speaking, g% (t ) can be a complex quantity, which
where sgn(f) is the sign function or signum function defined by can be expressed in the Cartesian form by:

⎧ 1, f > 0 g% (t ) = g I (t ) + jgQ (t ) (21)

⎪
sgn( f ) = ⎨ 0, f = 0 , (13)
⎪ −1, f < 0 where the subscripts I and Q stand for in-phase and
⎩ quadrature. Then, by substituting (21) in (20) we have, after
some simplifications:
then we can write:
g (t ) = g I (t ) cos(2π f c t ) − gQ (t )sin(2π f c t ) (22)
Gˆ ( f ) = − j sgn( f )G ( f ) (14)
Both gI(t) and gQ(t) are low-pass signals and are called the
Analyzing (14) we can see that the Hilbert transform of g(t)
in-phase component and the quadrature component of the
corresponds to a phase shift of –90º for the positive
frequencies of G(f) and +90º for the its negative frequencies. signal g(t), respectively. This is why we call g% (t ) the
Let us now make use of another definition: the analytic equivalent low-pass version of the band-pass signal g(t). This
signal or pre-envelope of g(t): result will be used later on in this tutorial to describe a
particular form for the MSK signal generation and detection.
g + (t ) = g (t ) + jgˆ (t ) (15) Rewriting expression (21) in its polar form we have:

from where, using (14) and the definition of the signum g% (t ) = a (t ) exp[ jθ (t )] , (23)
function given in (13), we can obtain
from where, using (20), we can obtain
 GUIMARÃES: CONTRIBUTIONS TO THE UNDERSTANDING OF THE MSK MODULATION

g (t ) = Re [ g% (t ) exp( j 2π f c t ) ] 1⎡
X ( f ) + X * (− f ) ⎤
2
S( f ) =
= Re {a (t ) exp[ jθ (t )]exp( j 2π f c t )} (24) 4 ⎣⎢ ⎦⎥

= a (t ) cos [2π f c t + θ (t )]
1
{
= ⎡⎣ X ( f ) + X * ( − f )⎤⎦ ⎡⎣ X * ( f ) + X ( − f )⎤⎦
4
}
1 ⎡ X ( f ) X ( f ) + X (− f ) X (− f ) ⎤
* *
In (24), a(t) = | g% (t ) | is the envelope of the band-pass signal (30)
= ⎢ ⎥
g(t), or the amplitude modulated component of g(t), and θ(t) is 4 ⎢⎣ + X ( f ) X ( − f ) + X * ( − f ) X * ( f )⎥⎦
its phase, or the phase-modulated component of g(t). This
1 ⎡⎢ X ( f ) + X ( − f ) ⎤
2 2
result will also be used later as a means for understanding the
MSK signal generation. = ⎥
4 ⎢ + X ( f ) X ( − f ) + X * ( − f ) X * ( f )⎥
Taking the Fourier transform of g(t) we know to obtain its ⎣ ⎦
frequency content. If g(t) is a voltage signal, then the
magnitude of its Fourier transform will result in a, say, By recognizing that X(f) and X(−f) are band-limited, band-
“voltage spectral density”. Then, using (20) we get: pass signals, the products X(f)X(−f) and X*(f)X*(−f) in (30)
vanish to zero. Going back to the normal notation, we get:
{Re ⎡⎣ g% (t ) e }
∞
G ( f ) = ℑ{g (t )} = ∫ j 2π f c t
⎤⎦ e − j 2π ft dt (25) 1⎡ %
G ( f − f c ) + G% ( − f − f c ) ⎤
2 2
−∞
S( f ) =
4⎣⎢ ⎥⎦
Using the identity Re[C] = ½[C + C*] in (25), and applying (31)
1
the Fourier transform properties x*(t) ∩ X*(−f) and = ⎡⎣ S B ( f − f c ) + S B ( − f − f c )⎤⎦
x(t)exp(j2πfct) ∩ X(f − fc), we obtain: 4
Equation (31) states that we can easily obtain the power
1 ∞
2 ∫−∞ ⎣
G( f ) = ⎡ g% (t ) e j 2π fc t + g% * (t ) e − j 2π fc t ⎤⎦ e − j 2π ft dt spectral density S(f) of a band-pass signal by translating the
(26) power spectral density SB(f) of the low-pass equivalent, and its
1
= ⎡⎣G% ( f − f c ) + G% * ( − f − f c )⎤⎦ mirror image, to the frequencies fc and −fc, respectively, and
2 multiplying the result by ¼.
If g(t) is a sample function of an stationary random process C. Minimum frequency separation for coherent detection
G(t), it has infinity energy and, hence, its Fourier transform It may be somewhat obvious for some readers that MSK is
does not exist. In this case the spectral content of G(t) is given a form of orthogonal frequency shift keying modulation, but
by its power spectral density (PSD), which is obtained from our aim in this subsection is to give reasons for the term
the Fourier transform of the auto-correlation function RG(τ) of minimum in the Minimum Shift Keying nomenclature.
the random process, as follows [8, p. 67]: To be coherently orthogonal in the signaling interval T, two
∞
cosine functions with different frequencies must satisfy
S ( f ) = ∫ RG (τ ) exp( − j 2π f τ )dτ (27)
−∞ T
∫ cos ( 2π f1t ) cos ( 2π f 2 t )dt = 0 (32)
0
The PSD for a stationary random process can also be
estimated through [7, p. 51]: Using the identity cosα⋅cosβ = ½[cos(α − β ) + cos(α + β )]
in the expression above we obtain:
1 ⎡
E GΠ ( f ) ⎤
2
S ( f ) = lim (28)
Π→∞ Π ⎣ ⎦ T T
1
2 ∫ cos ⎡⎣ 2π ( f1 − f 2 ) t ⎤⎦ dt + 12 ∫ cos ⎡⎣ 2π ( f1 + f 2 ) t ⎤⎦ dt = 0 (33)
0 0
where GΠ (f) is the Fourier transform obtained from the sample
process gΠ (t), which is g(t) truncated from –Π/2 to Π/2. The from where, after some manipulations, we get:
function |GΠ (f)|2 is called the energy spectral density of the
energy signal gΠ (t). If the signal is deterministic, (28) can also sin ⎡⎣ 2π ( f1 − f 2 ) T ⎤⎦ sin ⎡⎣ 2π ( f1 + f 2 ) T ⎤⎦
+ =0 (34)
be used, without the expectation operation [11, p. 31]. 4π ( f1 − f 2 ) 4π ( f1 + f 2 )
However, if the Fourier transform G(f) exists and is exact,
according to which was stated before equation (25) S(f) can be Since for practical purposes the sum f1 + f2 >> 1, the
simply determined by the squared-modulus of G(f), that is, second term in the left-hand side of (34) is approximately zero,
which results in
1⎡ %
G ( f − f c ) + G% * ( − f − f c ) ⎤
2 2
S ( f ) = G( f ) = (29)
⎢
4⎣ ⎥⎦ sin ⎡⎣ 2π ( f1 − f 2 ) T ⎤⎦ = 0
k (35)
Using a simplified notation, and the fact that |C|2 = CC*, we ⇒ ( f1 − f 2 ) = , k inteiro
can rewrite (31) as follows: 2T
Then, the minimum frequency separation between tones for
an orthogonal FSK with coherent detection is
REVISTA TELECOMUNICAÇÕES, VOL. 11, NO. 01, MAIO DE 2008

Example 1 - Suppose we want to transmit the information

1
( f1 − f 2 ) = , (36) sequence [1 0 1 0 0 0 1 1]. Following (39), with h = 1, we
2T shall have the phase evolution illustrated in Figure 3. Also in
Figure 3 are plotted the waveform b(t) and the resultant FSK
which justifies the name Minimum Shift Keying for the MSK
modulated signal s(t) for fc = 2/Tb Hz. The resultant tones are
modulation.
then at frequencies f1 = 5/(2Tb) Hz and f2 = 3/(2Tb) Hz.
III. MSK SIGNAL GENERATION AND DETECTION A careful look at Figure 3 shows that phase transitions
In this section, the MSK signal generation and detection from one bit to the next lead to the same value, using modulo
are analyzed in detail, based first on a complex representation 2π algebra (a phase transition of +π is equal to a phase
approach, and then, based on a signal-space representation transition of −π, modulo 2π). Then, the receiver is not able to
approach. The MSK power spectral density is also considered. explore any phase information in the conventional Sunde’s
However, first we introduce the basics about the MSK and the FSK modulation.
conventional binary FSK modulation. At the end of the section
these modulations are revisited, aiming at establishing their
1
similarities and differences from the design of the transmitter
b(t) 0
and the receiver perspective.
-1
The receiver structures and performances considered in this
section assume that the system operates on an Additive White
(a)
Gaussian Noise (AWGN) channel. 1π

A. MSK and conventional binary FSK θ(t) 0

−1π
A continuous-phase, frequency-shift keying (CPFSK)
−2π
signal can be described as a phase-modulated signal using
(24), as shown by: (b)

2 Eb
s( t ) = cos [2π f c t + θ (t )] (37) s(t) 0
Tb
t/Tb
where Eb is the average energy per bit and Tb is the bit 0 1 2 3 4 5 6 7 8
(c)
duration.
The time derivative of the phase evolution θ(t) in (37) Figure 3. Information sequence (a), phase evolution (b) and modulated signal
(c) for the Sunde’s FSK modulation.
gives rise to the CPFSK instantaneous angular frequency shift.
Then, in a given bit interval θ(t) increases or decreases Now, let us make h = ½ in (39). In this case we have the
linearly, depending on the desired transmitted tone, as minimum tone separation of 1/(2Tb) Hz, and, through (37), we
described by: shall generate an MSK signal.

πh Example 2 - Suppose again that we want to transmit the

θ (t ) = θ (0) ± t , 0 ≤ t ≤ Tb (38)
Tb information sequence [1 0 1 0 0 0 1 1]. According to (39),
with h = ½, we shall have the phase evolution shown in Figure
where θ(0) accounts for the accumulated phase history until 4. The waveform b(t) and the resultant MSK modulated signal
instant t = 0 and h is a measure of the frequency deviation. If h s(t) for fc = 1/Tb Hz are also plotted. The resultant tones are at
= 1 we have the conventional form of binary FSK modulation, frequencies f1 = 5/(4Tb) Hz and f2 = 3/(4Tb) Hz.
also know as Sunde’s FSK [6] [7, p. 381], in which the tone
separation is obtained from (38) as 1/Tb Hz. As can be noticed from Figure 4, phase transitions from a
Generalizing (38), at any time instant the phase evolution bit to the next one lead to different values, modulo 2π. Then, it
can be determined by is possible to explore some phase information with the MSK
modulation. This is indeed the motivation for the use of MSK:
πh t
θ (t ) = θ (0) +
Tb ∫ b(t )dt
0
(39) the receiver can explore phase transitions in order to benefit
from this additional information to improve performance.

where b(t) ∈ {±1} is the waveform related to the information B. MSK signal generation and detection from the complex
sequence, such that a −1 represents a bit 0 and a +1 represents representation approach
a bit 1. The generation of s(t) through (37) and (39), though
The modulated signal described by (37) and (39) can be straightforward from the implementation point of view, brings
generated by means of a continuous-phase VCO (voltage no or little insight on how the receiver can be constructed in
controlled oscillator) having b(t) as its input, and configured order to explore the phase information in the modulated signal.
with center frequency fc Hz and gain h/(2Tb) Hz/volt. Then we are forced to obtain alternative mathematical models
for representing s(t).
 GUIMARÃES: CONTRIBUTIONS TO THE UNDERSTANDING OF THE MSK MODULATION

Since θ(Tb) = ±π/2, depending on the information bit

1
during the interval (0, Tb], we shall have sin[θ(t)] going
b(t) 0
towards zero during the interval Tb to 2Tb, regardless the
-1
information bit during this interval. Thus, sQ(t) can be viewed
as a half-cycle sine function from the whole interval (0, 2Tb],
(a) the polarity of which depending on the information bit during
0.5π the interval [0, Tb).
θ(t) 0 Using the results (41) and (42) in (40), we obtain:
−0.5π
2 Eb ⎛ π ⎞
t ⎟ cos ( 2π f c t )
−1.0π
s (t ) = ± cos ⎜
(b) Tb ⎝ 2Tb ⎠
(43)
2 Eb ⎛ π ⎞
s(t) 0
m sin ⎜ t ⎟ sin ( 2π f c t )
Tb ⎝ 2Tb ⎠
t/Tb
where the polarity of both terms in a given bit interval are not
0 1 2 3 4 5 6 7 8
(c) necessarily the same.
Figure 4. Information sequence (a), phase evolution (b) and modulated signal
Following [4, p. 18], we can rewrite (43) as:
(c) for the MSK modulation.
2 ⎛ π ⎞
To simplify matters, consider initially only the first bit s( t ) = a I (t ) cos ⎜ t ⎟ cos ( 2π f c t )
Tb ⎝ 2Tb ⎠
interval. Using h = ½ in (38) and the identity cos(α ±β ) = (44)
cosα⋅cosβ m sinα⋅sinβ , we can rewrite (37) as follows: 2 ⎛ π ⎞
− aQ (t ) sin ⎜ t ⎟ sin ( 2π f c t )
Tb ⎝ 2Tb ⎠
2 Eb ⎡ π ⎤
s( t ) = cos ⎢θ (0) ± t ⎥ cos ( 2π f c t )
Tb ⎣ 2Tb ⎦ where we have defined aI(t) and aQ(t) as random sequences of
rectangular pulses with amplitudes ± Eb and duration 2Tb
2 Eb ⎡ π ⎤ (40)
− sin ⎢θ (0) ± t ⎥ sin ( 2π f c t ) , 0 ≤ t ≤ Tb seconds. These sequences are associated to the polarities of the
Tb ⎣ 2 Tb ⎦
half-cycle cosine and sine functions as follows: if aI(t) is
= sI (t ) cos ( 2π f c t ) − sQ (t )sin ( 2π f c t ) positive, sI(t) follows the function cos{[π/(2Tb)]t}; if aI(t) is
negative, sI(t) corresponds to –cos{[π/(2Tb)]t}. The same
Making use of (22) and applying again the identity happens with sQ(t): if aQ(t) is positive, sQ(t) follows the
cos(α ±β ) = cosα⋅cosβ m sinα⋅sinβ to the in-phase function sin{[π/(2Tb)]t}; if aQ(t) is negative, sQ(t) corresponds
component of s(t), and, without loss of generality, assuming to –sin{[π/(2Tb)]t}.
θ(0) = 0, we get From the above discussion we can conclude that,
depending on the information bit to be transmitted, the in-
2 Eb ⎡ π ⎤ phase and quadrature components of s(t) can change their
sI (t ) = cos ⎢ ± t⎥
Tb ⎣ 2Tb ⎦ polarities each 2Tb seconds, and that the half-cycle cosine and
, −Tb ≤ t ≤ Tb (41) sine functions are offset from each other by Tb seconds.
2 Eb ⎛ π ⎞
=± cos ⎜ t⎟ However, we are not still able to easily obtain the information
Tb ⎝ 2Tb ⎠ sequence responsible for generating a given sequence of
polarities. This would demand us to come back to the general
Since θ(0) = 0, before t = 0 the phase evolution was a analysis presented in Section III-A, specifically to equation
positive or negative slope going towards zero, depending on (37), thus making difficult the visualization of the
the previous bit. Then, the result in (41) is an increasing cosine implementation issues for the MSK modem.
function from –Tb to 0. Thus, sI(t) can be interpreted as a half- Then, for the time being we assume a given sequence of
cycle cosine function from the whole interval (–Tb, Tb]. pulses for sI(t) and sQ(t), and later we determine the
Similarly, the quadrature component of s(t) can be written information sequence based on the analysis of this assumption.
as follows: A general rule will arise from this analysis.

2 Eb ⎛ π ⎞ Example 3 – In a 8-bit interval, let sI(t) and sQ(t) assume the

sQ (t ) = ± sin ⎜ t ⎟ , 0 ≤ t ≤ 2Tb (42)
Tb ⎝ 2Tb ⎠ sequence of half-cycle cosine and sine pulses shown in Figure
5. For reference, in this figure the functions cos{[π/(2Tb)]t}
where we have made use of θ(0) = 0 and of the identity and –sin{[π/(2Tb)]t} are also plotted, in dashed lines, and are
sin(α ±β ) = sinα⋅cosβ ± cosα⋅sinβ . We have also made use of given the polarities of the waveforms aI(t) and aQ(t).
the relation cos[θ(0)] = cos[θ(Tb) m π/2] = ± sin[θ(Tb)] = ±1. Combining the waveforms in Figure 5 according to (40), we
get the results in Figure 6. In this figure the waveforms sI(t)
and sQ(t) are also plotted, in dashed lines. The carrier
REVISTA TELECOMUNICAÇÕES, VOL. 11, NO. 01, MAIO DE 2008

frequency in this example is fc = 1/Tb Hz. The resultant tones remaining terms in (44). Figure 7 illustrates the structure of the
are then at frequencies f1 = 5/(4Tb) Hz and f2 = 3/(4Tb) Hz. MSK modulator constructed according to complex
representation approach just described.
The MSK signal just analyzed can also be generated by
sI(t) 0 + + + − − + + + means of a VCO configured with center frequency fc Hz and
gain 1/(4Tb) Hz/volt. However, since the frequency shifts in
the modulated signal do not directly correspond to the
information sequence, the input of the VCO must be the
differentially decoded version of this information sequence,
sQ(t) 0 − − − − − − + + converted to {±1}.

t/Tb
0 1 2 3 4 5 6 7 8

Figure 5. Generation of the MSK signal: base-band in-phase and quadrature

components.

Observing the modulated signal s(t) in Figure 6 we can

notice that, if a bit 1 is associated to the tone of greater
frequency, the corresponding modulating sequence should be
d = [1 1 1 0 0 1 0 0]. Let us now define a new sequence i in
which the exclusive-or (XOR) operation between a given bit
and the previous one results in a bit of the sequence d. This
new sequence is i = [1 0 1 0 0 0 1 1 1]. Sequence d can be
seen as a differentially decoded version of i. Additionally,
Figure 7. MSK modulator constructed according to the complex
suppose that the sequence i is parallelized to form the representation approach.
sequences of odd and even symbols of duration 2Tb, io = [1 1 0
1 1] and ie = [0 0 0 1], respectively. Now, suppose that each Some authors claim that the MSK modulation is a special
symbol of these sequences is converted to ± Eb . The great form of OQPSK (or SQPSK) modulation where the pulse
achievement here is that these new parallel sequences, if they shaping are half-cycle cosine and sine functions, instead of the
are off-set to each other Tb seconds, are exactly the waveforms rectangular shaping functions used in OQPSK. However, in
aI(t) and aQ(t). despite of being true, this statement must be carefully
interpreted. From (43) we can see that, in fact, the shapes of
the pulses that modulate the quadrature carriers are half-cycle
sI(t)cos(2πfct) 0 cosine and sine functions. Nevertheless, they are not a simple
reshaping of the waveforms aI(t) and aQ(t). Before modulating
the quadrature carriers, aI(t) and aQ(t) are modified by the
polarities of the waveforms cos{[π/(2Tb)]t} and sin{[π/(2Tb)]t}
in each 2Tb interval. But we can make a small modification in
sQ(t)sin(2πfct) 0 the above structure to implement the MSK modulation in the
same way we implement an OQPSK modulator, the unique
difference being the shape of the pulses that modulate the
quadrature carriers. We just have to use the modulus
|cos{[π/(2Tb)]t}| and |sin{[π/(2Tb)]t}| in (44). The resultant
s(t) 0
structure is shown in Figure 8. In this figure we have used
additional simplifications to make the MSK modulator
t/Tb
0 1 2 3 4 5 6 7 8 structure closer to a more practical one: the quadrature carriers
Figure 6. Generation of the MSK signal: modulated in-phase and quadrature were generated from a single oscillator and the pulse-shaping
components and the resultant MSK signal. functions were implemented via low-pass filters with identical
We then conclude that the MSK signal can be generated impulse responses given by
through (44), where the waveforms aI(t) and aQ(t) are the ⎧ ⎛ π ⎞
serial-to-parallel (S/P) converted version of the information ⎪sin ⎜ t ⎟ , 0 ≤ t ≤ 2Tb
h(t ) = ⎨ ⎝ 2Tb ⎠ (45)
sequence, with bit 1 converted to + E b and bit 0 converted to
⎪
− E b . Additionally, the sequence aQ(t) has to be offset Tb ⎩ 0, otherwise
seconds from aI(t), before they multiply the corresponding
 GUIMARÃES: CONTRIBUTIONS TO THE UNDERSTANDING OF THE MSK MODULATION

The inputs to these filters are PAM (Pulse Amplitude If, for some reason, it is necessary to represent a bit 1 in the
Modulation) sequences having very short durations sequence d by the tone of lower frequency, the only thing we
(approximating unit impulses) and amplitudes of + E b . have to do is to invert the minus signal in the summation block
We can recall that at the beginning of Section III-B we in Figure 7 or Figure 8, and invert the minus signal in the
have made the assumption that θ(0) = 0. This assumption was bottom multiplier block in Figure 9.
adopted only to facilitate the mathematical description of the
MSK modulation. In fact, from an implementation perspective,
any initial phase is allowed for the quadrature carriers.
However, regardless of this initial phase, the designer must
only guarantee the correct phase alignment among the
quadrature carriers, the pulse shaping functions and the
sequences aI(t) and aQ(t).
In the light of the similarities between the MSK and
OQPSK modulations, we are now able to understand possible
structures for the MSK demodulator. We know that a
conventional QPSK modulator can be interpreted as two
BPSK (Binary Phase-Shift Keying) modulators, each of them
making use of one of the two quadrature carriers. Then, the
QPSK demodulator can be implemented as two independent Figure 9. MSK demodulator constructed according to the complex
BPSK demodulators. The decisions made by each of these representation approach.
demodulators are parallel-to-serial (P/S) converted to form the
estimate of the transmitted bit sequence. The OQPSK C. MSK signal generation and detection from the signal-
demodulator follows the same rule, with the difference that space representation approach
one of the estimated parallel sequences is offset Tb seconds We are now able to determine the orthonormal base-
from the other. Then, before P/S conversion these sequences functions responsible for generating the MSK signal. Recalling
must be aligned in time. that we are talking about a binary orthogonal signaling, the
The presence of the shaping and polarity inversion base-functions can be directly obtained from (44) as follows:
processes caused by the functions cos{[π/(2Tb)]t} and
sin{[π/(2Tb)]t} in the modulator of Figure 7 do not change the 2 ⎛ π ⎞
φ1 (t ) = cos ⎜ t ⎟ cos ( 2π f c t ) (46)
demodulation rule, as compared to the one used for the Tb ⎝ 2Tb ⎠
OQPSK signal. This is also true if we use |cos{[π/(2Tb)]t}| and
|sin{[π/(2Tb)]t}|, according to Figure 8. Then, the receiver 2 ⎛ π ⎞
block diagram shown in Figure 9 can be used indistinctly, φ2 (t ) = sin ⎜ t ⎟ sin ( 2π f c t ) (47)
Tb ⎝ 2Tb ⎠
without any modification.
These base-functions, differently from what is stated in [7,
p. 390], are defined for any interval, not only from 0 to Tb.
Comparing (44) with (46) and (47) we readily see that the
MSK signal vectors are determined by the amplitudes of the
waveforms aI(t) and aQ(t) defined in (44), in each bit interval:

⎡ s ⎤ ⎡ ± Eb ⎤
si = ⎢ i 1 ⎥ = ⎢ ⎥ , i = 1, 2, 3, 4 (48)
⎣ si 2 ⎦ ⎢⎣ ± Eb ⎥⎦

Figure 8. A more practical MSK modulator constructed according to the Then, as shown in Figure 10, the signal-space diagram for
equivalence with the OQPSK modulation. the MSK modulation comprises four signal vectors, despite of
The received signal in Figure 9 is coherently correlated, in MSK be a binary modulation. The mapping between these
one arm of the receiver, with the result of the multiplication vectors and the information bits is determined via the
between the in-phase carrier and the shaping function differentially decoded version of the information bits. The
cos{[π/(2Tb)]t}. In the other arm, the received signal is following example is meant to clarify these statements.
correlated with the result of the multiplication between the
Example 4 – Let the sequence of signal vector polarities be
quadrature carrier and the shaping function –sin{[π/(2Tb)]t}.
[+ –], [+ –], [+ –], [– –], [– –], [+ –], [+ +] and [+ +],
These correlations are made in a 2Tb seconds interval, generated on a bit-by-bit basis. In this sequence, the polarities
reflecting the duration of the half-cycle cosine and sine on the left refer to si1, and those on the right refer to si2. These
functions, and are time-aligned with these functions. The polarities are the same as those considered in Figure 5 and, as
estimated sequences î e and î o are then time-aligned and P/S we already know from Example 3, they are associated to the
converted to form the estimate of the transmitted sequence, î . information sequence i = [1 0 1 0 0 0 1 1 1] and to its
REVISTA TELECOMUNICAÇÕES, VOL. 11, NO. 01, MAIO DE 2008

differentially decoded version d = [1 1 1 0 0 1 0 0]. From this

example it is possible to draw the mapping between the signal 2 ⎛ π ⎞
φ1 (t ) = cos ⎜ t ⎟ cos ( 2π f c t )
vectors and the differentially decoded version of the Tb ⎝ 2Tb ⎠
information bits, as shown in Table I. (49)
1 1
= cos ( 2π f 2t ) + cos ( 2π f1t )
TABLE I – MAPPING BETWEEN THE MSK SIGNAL VECTORS AND THE 2Tb 2Tb
DIFFERENTIALLY DECODED VERSION OF THE INFORMATION BITS

Signal vector coordinates Now, let us expand φ2(t) using the identity sinα⋅sinβ =
i Bits
si1 si2 ½[cos(α − β ) − cos(α + β )]:
1 0 + Eb + Eb ⎛ π ⎞
2
φ2 (t ) = sin ⎜ t ⎟ sin ( 2π f c t )
2 1 + Eb − Eb Tb ⎝ 2Tb ⎠
− Eb − Eb
(50)
3 0 1 1
= cos ( 2π f 2 t ) − cos ( 2π f1t )
4 1 − Eb + Eb 2Tb 2Tb

Since MSK is a continuous phase modulation, no abrupt

+ + + − − + + +
phase transition occurs when a symbol changes. The
circumference in Figure 10 illustrates this smooth phase aI(t)
0
transitions between any pair of symbols. They can be observed φ1(t)

in a x-y plot, with sI(t) applied to the x-axis and sQ(t) applied to
the y-axis (see Figure 5).
− − − − − − + +
φ2 aQ(t)
0
φ2(t)
1 0
+ Eb

− Eb + Eb
φ1
s(t) 0

t/Tb
− Eb
0 1 2 3 4 5 6 7 8
0 1
Figure 11. Generation of the MSK signal: base-functions, coefficients and
the resultant MSK signal.
Figure 10. MSK constellation.
Figure 12 shows the MSK modulator constructed
Observing (46) and (47) we see that the base-functions according to the interpretation of expressions (49) and (50).
φ1(t) and φ2(t) correspond to the modulation of the quadrature The two cosine functions are multiplied to generate the tones
carriers by the waveforms cos{[π/(2Tb)]t} and sin{[π/(2Tb)]t}, with frequencies f1 and f2, according to (49). Each of these
respectively. Comparing (46) and (47) with (44), we see that tones is selected through the band-pass filters shown in this
the base-function φ1(t) are multiplied by aI(t), the base- figure, and the results are combined according to (49) and (50)
function φ2(t) is multiplied by aQ(t), and the results are added to generate the base-functions. Finally, these base-functions
to form the MSK signal s(t). Figure 11 illustrates the are multiplied by the corresponding waveforms aI(t) and aQ(t)
generation of the MSK signal form this signal-space and the results are added-up to form the MSK signal. The
representation approach. The signal vector polarities approach at hand can also consider the demodulator shown in
associated to the waveforms aI(t) and aQ(t) are the same as Figure 9, where we readily identify the use of the base-
those used in Example 4. functions φ1(t) and φ2(t) feeding the correlators.
As we did with the complex representation approach, now We can see that, operating in different ways with the
we shall construct the modulator structure based on the signal- mathematical model of the MSK signal, it is possible to
space representation. As a matter of fact, if we group together construct different, but equivalent structures. More structures
the two upper mixers and group together the two lower mixers would be possible if an alternative mathematical model were
in Figure 7 this job is already done. But we shall manipulate adopted. These comments are also valid to the construction of
the base-function expressions to get an alternative structure. the MSK demodulator. In [11, pp. 299-307] the reader can
First, let us expand φ1(t) using the identity cosα⋅cosβ = find several forms for the implementation of an MSK modem,
½[cos(α − β ) + cos(α + β )]: along with different approaches on its construction.
 GUIMARÃES: CONTRIBUTIONS TO THE UNDERSTANDING OF THE MSK MODULATION

and is 3 dB more energy-efficient than the conventional BFSK

with coherent detection [7, p. 418].
E. MSK signal generation and detection from a
conventional Sunde’s FSK approach
Suppose now that we aim at generating an MSK signal
using the conventional FSK approach, but with the minimum
tone separation (f1 − f2) = 1/(2Tb) Hz. The modulator would
appear like in Figure 13. This form of FSK signal generation
guarantees phase continuity only if the tone separation is a
multiple of 1/Tb and the carrier frequency is a multiple of
Figure 12. MSK modulator constructed according to the signal-space 1/(2Tb). Then, the modulated signal in Figure 13 will show
representation approach. phase discontinuities, which does not correspond to an MSK
signal. MSK and binary FSK signals are the same if they are
D. Bit error probability for the MSK modulation generated according to (37) and (39), using h = ½.
We can see through Figure 8 and equation (44) that the
modulator transmits two independent sequences using two
quadrature carriers, and through Figure 9 we can see that the
demodulator detects these sequences independently.
Consequently, we can state that the modulator can be
interpreted as formed by two independent BPSK-like
modulators and that the demodulator can be interpreted as
formed by two independent BPSK-like demodulators. The
difference to the conventional BPSK modulator and
demodulator is the presence of half-cycle sine and cosine
pulse-shaping functions. The energy per symbol for each of Figure 13. A try for generating an MSK signal from the conventional binary
these two component BPSK modulators is easily found to be FSK implementation approach.

2 Eb 2 Tb ⎛ π ⎞ Now, following [9], suppose that we want to detect an

ξ= ∫ cos 2 ⎜ t ⎟ cos 2 ( 2π f c t ) dt MSK signal using a conventional coherent FSK demodulator.
Tb 0
⎝ 2Tb ⎠
(51) We would be tempted to think that it is just necessary to
2E 2 Tb ⎛ π ⎞ correlate the received signal with base-functions formed by the
= b ∫
2
sin ⎜ t ⎟ sin 2 ( 2π f c t ) dt = Eb cosine tones with frequencies f1 and f2, during Tb seconds
Tb 0
⎝ 2Tb ⎠
intervals, and that the decision would be made in favor of the
where, for simplification purposes, we have adopted the carrier greatest correlator output. However, the phase continuity and
frequency fc as an integer multiple of 1/(2Tb). The energy per phase dependency imposed by the MSK signal construction do
MSK symbol is the sum of the symbol energies in the not permit the use of the above approach. This is illustrated in
quadrature modulated carriers, that is E = 2Eb, a value that can Figure 14, were we have plotted an MSK signal and the cosine
also be obtained from the constellation in Figure 10. base-functions with frequencies f1 and f2 separated by 1/(2Tb)
Confusions may arise here: the duration of one bit is of Hz. Observe that, in several intervals, there are no phase
course Tb seconds, and we must make the bit decisions in a bit- coherence between the modulated signal and the base-
by-bit basis. But the phase information at the MSK receiver is functions with the same frequency, a behavior that would lead
explored in 2Tb seconds intervals, so that the effective energy to detection errors.
collected by this receiver corresponds to observations made Let us elaborate a little bit more on this issue. From Figure
during intervals of 2Tb seconds. 14 we can see that when no phase coherence occurs, the MSK
From the above discussion we can conclude that the bit signal is at 180º out of phase from the corresponding base-
error probability for the MSK modulation on the AWGN function. Then, by comparing the magnitudes of the correlators
channel, considering equally-likely bits, can be determined by outputs we are still able to make correct decisions. But we
the average of the bit error probabilities for the two component cannot forget that, unless the MSK signal is generated directly
BPSK detectors [8, p. 271], which results in: from the realization of (37) and (39) with h = ½, the estimated
bits would correspond to a differentially decoded version of
1 ⎛ Eb ⎞ the information bits. To get the estimates of the information
Pb = erfc ⎜
⎜ N ⎟⎟
, (52) bits we have to apply the inverse operation on the estimated
2 ⎝ 0 ⎠
bits through the exclusive OR (XOR) between a given bit and
the previous XOR result (see Example 3 and the
where N0 is the AWGN power spectral density and erfc(u) is
corresponding comments). However, this operation can lead to
the complimentary error function of the argument. This result
the opposite decisions, since a differentially decoded 1 can
shows that the performance of the MSK modulation is the
result from the information sequence 01 or 10, and a
same as the performance of the BPSK and QPSK modulations,
differentially decoded 0 can result from the information
REVISTA TELECOMUNICAÇÕES, VOL. 11, NO. 01, MAIO DE 2008

sequence 00 or 11. Inserting a differential coder at the the BER was the same as the one obtained with the simulation
transmitter input and a differential decoder at the receiver of the complete system depicted by Figure 15.
output easily solves this ambiguity problem.
Finally, we shall have the transmitter and receiver 0

structures shown in Figure 15. -1

10

-2
10
s(t) 0

Bit Error Rate

cos(2πf1t) -3
10

-4
10
Exact Pe (numerical)
-5
10 BER (simulation result)
s(t) 0 Pe for coherent BFSK
cos(2πf2t) Pe for non-coherent BFSK
-6
10 Pe for MSK
t/Tb
0 1 2 3 4 5 6 7 8 -7
10
Figure 14. MSK signal s(t), the cos(2πf1t) and cos(2πf2t). 0 2 4 6 8 10 12 14

Eb/N0, dB

Figure 16. Performance results for MSK, coherent and non-coherent BFSK
and for the system depicted in Figure 15. The channel is AWGN [9].

F. Power spectral density of the MSK signal

We saw in Section II that in order to obtain the PSD of a
modulated signal, we can determine the PSD of its complex
envelope representation and, using (31), convert the result to
the desired PSD. According to (22), the MSK signal can be
written as:

s( t ) = sI (t ) cos ( 2π f c t ) − sQ (t ) sin ( 2π f c t ) (53)

from where the complex envelope given by (21) is

s% (t ) = sI (t ) + jsQ (t ) (54)
Figure 15. MSK modulator with conventional FSK detection: modified MSK
transmitter (a) and detection via a modified coherent binary FSK receiver (b).
For the MSK modulation, the low-pass in-phase and
Since the receiver in Figure 15 is not exploring any phase quadrature components in (54) are random waveforms in
information, we expect a worse performance as compared to which the pulses with duration 2Tb can assume positive or
the one provided by the appropriate MSK receiver. negative values according to:
Furthermore, although the channel noise is Gaussian, the noise
in the decision variable is not. Then, the analytical process for sI (t ) = ∑ I k p ( t − 2kTb ) , − ∞ ≤ k ≤ ∞
k
obtaining an expression for the bit error probability Pe for the (55)
receiver under investigation is quite involved and is beyond sQ (t ) = ∑ Qk p ( t − 2kTb ) , − ∞ ≤ k ≤ ∞
the scope of this work. Nevertheless, a numerical calculation k
of Pe was made and a simulation of the system in Figure 15 where p(t) is the shaping pulse with half-cycle sine format:
was carried out. Both results agreed and showed that the
performance lies in between a coherently detected and a non- 2 Eb ⎛ π ⎞
coherently detected binary FSK, as shown in Figure 16, and is p(t ) = sin ⎜ t ⎟ , 0 ≤ t ≤ 2Tb , (56)
Tb ⎝ 2Tb ⎠
approximately 3.05 dB worse than the Pe obtained with the
MSK receiver. This is an attractive result, since the Pe curves and {Ik} and {Qk} are random antipodal sequences ∈{±1}
for the coherent and the non-coherent FSK differs associated to the odd and even information bits, respectively
asymptotically in about 1 dB [7 p. 418], and we are using a (see Example 3) or, equivalently, associated to the waveforms
transmitted signal that has the most compact spectrum among aI(t) and aQ(t) in (44).
the coherent and orthogonal CPFSK modulations [9]. It is a well-known result that the power spectral density of
Using a more practical and simplified approach, the MSK a random antipodal sequence can be determined by dividing
modulator in Figure 15-a can be replaced by a VCO, the energy spectral density (ESD) of the shaping pulse by the
eliminating the need for the three differential circuits used by pulse duration [7, p. 48] [8, p. 207]. By recalling that the ESD
the complete system. This alternative was also simulated and of a pulse is the squared-modulus of its Fourier transform, then
 GUIMARÃES: CONTRIBUTIONS TO THE UNDERSTANDING OF THE MSK MODULATION

the PSD of sI(t), which is equal to the PSD of sQ(t), can be spectral efficiency of the MSK modulation, with the penalty of
easily determined. Furthermore, we know that the in-phase and a possibly small reduction in performance due to inter-symbol
quadrature components of the MSK signal are independent to interference introduced by the Gaussian filtering process.
each other. Then, the PSD of (54) can be obtained through As mentioned at the beginning of this paper, the MSK
modulation is also attractive because of its constant envelope,
| P ( f ) |2 a characteristic that can be observed in all FSK-type
SB ( f ) = 2 , (57)
2Tb modulations. Although M-PSK modulations also have constant
envelopes, this is valid only if no filtering is applied to the
and the PSD of the MSK signal can be finally obtained using signal. When the modulated signal is filtered before going
the above result in (31). through some non-linear distortion, such as non-linear
Following the procedure just described, the PSD of the amplification, out-of-band and in-band spurious can be
base-band MSK signal in (54) can be obtained from [8, p. 214] generated due to envelope fluctuations that occur during
and is given by abrupt phase transitions. Non-constant envelopes can also
show high peak-to-average power ratios (PAPR), making it
2
32 E ⎛ cos2π fTb ⎞ difficult the project of high dynamic range and power-efficient
SB ( f ) = 2 b ⎜ ⎟ (58) non-linear amplifiers. The MSK modulation, even after
π ⎝ 1 − 16 f 2Tb2 ⎠
filtering, has low PAPR, becoming attractive in these cases.
Equation (58) is plotted in Figure 17, along with the base- The MSK modulation can also be viewed as a special form
band PSD of the QPSK modulation, for comparison purposes. of coded-modulation scheme in which the phase continuity
To draw this figure, both MSK and QPSK signals were set to restrictions introduce some sort of redundancy and,
the same average power. consequently, error correction capabilities. This attribute is
explored in detail in [10].
0 MSK
In [12], J. K. Omura, et. al apply the MSK modulation to
QPSK achieve code-division multiple access (CDMA) capability in a
-10 spread spectrum system.
Finally, although MSK is usually associated to the binary
-20
case, that is, M = 2, its concepts are generalized to the M-ary
-30 case in [13] and [14]. A multi-amplitude, continuous-phase
modulation approach is considered in [8, pp. 200-203], where
-40
the signal amplitude is allowed to vary, while the phase
-50
trajectory is constrained to be continuous. Generalized MSK is
also considered in [15].
fTb
-1.5 -1.0 -0.5 0 0.5 1.0 1.5
V. CONCLUSIONS
Figure 17. Normalized base-band PSD, in dBm/Hz, for the MSK and the
QPSK modulations with the same average power. We are now armed with enough concepts to give possible
answers (A) to the questions (Q) listed at the end of Section I:
It can be seen from Figure 17 that, although the main lobe
Q: To which extent the MSK modulation can be regarded
of the MSK spectrum is wider that the main lobe of the QPSK
as a special case of the Sunde’s FSK modulation? A: We saw
one, the PSD of the MSK decreases faster with frequency. For
that MSK is in fact a special form of FSK with the minimum
QPSK, approximately 90% of the modulated signal power is
tone separation for orthogonality and coherent detection.
concentrated in the main lobe. For MSK, this quantity
However, the MSK signal construction gives to the receiver
increases to approximately 99%. This is a desired attribute of
the ability to explore phase information for performance
the MSK modulation, which makes it attractive due to easy
improvement, which does not happen with the conventional
filtering and, consequently, low adjacent channel interference.
FSK modulation. As we saw in Section III-E, the conventional
Detailed and more complete considerations about the
binary FSK signal with minimum tone separation does not
power spectral characteristics of continuous-phase modulated
correspond to an MSK signal and does not exhibit phase
signals can be found in [8, pp. 209-219].
continuity for all bit transitions.
Q: To which extent the MSK modulation can be detected
IV. FURTHER ATTRIBUTES AND USES OF THE MSK
as the conventional Sunde’s FSK modulation? A: From the
In this section we summarize some MSK-related topics analysis in Section III-E we conclude that an MSK signal can
concerning additional attributes and applications of this be detected as a conventional binary FSK, but it is necessary to
modulation. We start by revisiting the application of the MSK make modifications at the transmitter and at the receiver,
in the recently-developed Blue-Ray technology [5], and as the according to the block diagram shown in Figure 15. Since this
base for implementing the GMSK modulation used, for modified receiver explores no phase information, the
instance, in the GSM standard [7, pp. 396-400]. In the case of performance will not be the same as that provided by the
the GSM standard, a Gaussian-filtered version of the appropriate MSK receiver.
information sequence is applied to an MSK modulator, Q: To which extent the MSK modulation can be regarded
resulting in the GMSK signal. This is done to increase the as a special case of the SQPSK or OQPSK (Staggered or
REVISTA TELECOMUNICAÇÕES, VOL. 11, NO. 01, MAIO DE 2008

Offset QPSK) modulation? A: The MSK modulation is indeed [5] Blu-Ray Disc Recordable Format – Part 1: Physical Specifications.
Available at [Link]
a special form of OQPSK (or SQPSK) modulation, where the
R_Physical_3rd_edition_0602f1-[Link] (last access: August, 06,
pulse shaping are half-cycle cosine and sine functions instead 2007).
of the rectangular shaping functions used in OQPSK. But this [6] E. D. Sunde, “Ideal binary pulse transmission by AM and FM”, Bell
is not a direct interpretation of the MSK signal construction. Systems Technical Journal, vol. 38, pp. 1357-1426, Nov. 1959.
To shown perfect equivalence with the OQPSK modulation, [7] S. Haykin, Communication Systems, 4th Edition - John Wiley and
Sons, Inc.: New York, USA, 2001.
the MSK transmitter must be implemented according to Figure [8] J. G. Proakis, Digital Communications – 3rd Edition, McGraw Hill,
8. The receiver structure is kept unchanged, according to the Inc.: USA, 1995.
block diagram shown in Figure 9. [9] D. A. Guimarães, A Simple FFSK Modulator and its Coherent
Q: To which extent the frequency and phase shifts of an Demodulator, IEICE Trans. Fundamentals. Vol. E91-A, No. 3, March
2008.
MSK signal are related to the modulating data sequence? A: If [10] H. Leib, S. Pasupathy, “Error Control Properties of Minimum Shift
the modulated signal is generated through the realization of Keying”, IEEE Communications Magazine, vol.31 No.1, pp. 52-61,
(37) and (39), using h = ½, then there will be a direct January 1993.
correspondence, that is, bit 0 will be represented by the tone [11] S. Benedetto, and E. Biglieri, Principles of Digital Transmission With
Wireless Applications. Kluwer Academic and Plenum Publishers: New
with frequency, say, f2 (or vice-versa), and bit 1 will be York, 1999.
represented by the tone with frequency f1 (or vice-versa). [12] J. K. Omura et. al., “MSK spread-spectrum receiver which allows
However, by generating the MSK signal through the other CDMA operations”, United States Patent 5,963,585, October 5, 1999.
ways shown is this tutorial, the frequency shifts will [13] M. K. Simon, “A generalization of minimum shift keying (MSK) type
signaling based upon input data symbol pulse shaping”, IEEE Trans.
correspond to a differentially decoded version of the on Communications, vol. COM-24, pp. 845-856, August 1976.
modulating data sequence. [14] I. Korn, “Generalized MSK”, IEEE Trans. on Information Theory, vol.
Q: To which extent the phase shifts of an MSK signal can IT-26, no. 2, pp. 234–238, March 1980.
be related to the phase transition diagram on its signal-space [15] R. Sadr and J. K. Omura, “Generalized minimum shift-keying
modulation techniques”, IEEE Trans. on Communications, Volume 36,
representation? A: The MSK signal is constructed in a way Issue 1, pp. 32-40, Jan 1988.
that, besides phase continuity, it exhibits phase transitions that
helps the receiver improve the detection performance. This is
done because phase transitions from one bit to the next lead to
different values, modulo 2π (see Figure 4). A bit one increases
the phase in π/2 radians and a bit 0 decreases the phase in π/2
radians. If these bits are or are not the information bits, it
depends on how the MSK signal is generated: directly via (37)
or indirectly (see former question and answer). Concerning the
phase shifts of an MSK signal, they cannot be directly mapped
on the signal-space symbol transitions. Two reasons support
this conclusion: firstly, since a given signal-space diagram can
represent a base-band or a band-pass signaling, it is not always
able to represent phase transitions of a modulated signal,
though it can happen with some modulations, such as M-PSK
and M-QAM. Secondly, discrete points in a signal space
cannot represent continuous-phase signals, because the phase Figure 18. Phase trajectory of an MSK signal. The projections of this
trajectory on all planes are also shown.
of the carrier is time-variant [8, pp. 199-200]. As an example,
two consecutive ones correspond to the same coordinates in
Figure 10, but we know that the carrier phase changes +π/2 Dayan Adionel Guimarães was Born in Carrancas,
radians from its preceding value, in a continuous way. A MG, Brazil, on March 01, 1969. He holds the titles:
Electronics Technician (ETE “FMC”, 1987), Electrical
solution to this is to have a three-dimensional diagram with Engineer (Inatel, 1994), Specialist in Data
axes sI(t), sQ(t) and t, in which the phase trajectory can be Communication Engineering (Inatel, 2003) and in
recorded [8, pp. 194-195]. Figure 18 illustrates this Human Resources Management (FAI, 1996), Master in
representation. Electrical Engineering (Unicamp, 1998) and Doctor in
Electrical Engineering (Unicamp, 2003).
From 1988 to 1993 he developed equipment for Industrial Instrumentation
REFERENCES and Control, and was also the Manufacturing and Product Engineering
[1] R. de Buda, “Coherent demodulation of frequency-shift keying with Supervisor at SENSE Sensores e Instrumentos. Since January 1995 he is
low deviation ratio”, IEEE Trans. on Communications, vol. COM-20, Professor at Inatel where, for eight years, he was responsible for the structure
no. 3, pp. 429-436, June 1972. that supports practical teaching activities for the Electrical Engineering
[2] M. L. Doelz and E. H. Heald, “Minimum shift data communication undergraduate course. His research includes the general aspects on Digital
system”, United States Patent 2,917,417, March 28, 1961. and Mobile Communications, specifically Multi-Carrier CDMA systems, and
[3] S. A. Groaemeyer and A.L. McBride, “MSK and offset QPSK coding for fading channels, specifically Block Turbo Codes.
modulation”, IEEE Trans. on Communications, August 1976. Dr. Dayan is member of the Telecomunicações magazine’s Editorial Board,
[4] S. Pasupathy, “Minimum Shift Keying: A Spectrally Efficient member of the Inatel’s Master Degree Counseling Board and of the IEICE
Modulation”, IEEE Communications Magazine, vol. 17, no. 4, pp. 14- (Institute of Electronics, Information and Communication Engineers), Japan.
22, July 1979.