You are on page 1of 7

1044 l E E E TRANSACTIONS ON COMMUNICATIONS,

V O L . COM-29, NO. 7, J U L Y 1981

L. Kurz.“Nonparametricdetectorsbased on partitiontests,” in mission is currently being studied [ 1] -[ 71 . While digital trans;


Nonparametric Methods in Communications. P. Papantoni-Kazakos mission can surely bring many advantages, some technical
and D.Kazakos.Eds. New York:MarcelDekker., 1977. pp.
73-143. problems must be solved. This paper is concerned with a dig-
S . A..Kasbam. “Conditional tests in nonparametric detection.:: in ital modulation for futuremobile radio communications.
Nonparametric Methods in Communications. P. PapantonirKazakos From the viewpoint of mobile radio use, the out-of-band
and D. Kazakos,Eds..NewYork:MarcelDeliker. 1977.. pp. radiationpower in the adjacentchannelshould be generally
195-201. , I

J. G . Shin and S. A. Kassam. “Multilevel coincidence correlators


suppressed 60-80 dB below that in the desired channel. So
forrandom’signaldetection.” IEEE Trans.Inform.Theqry. vol. as to satisfy this severe requirement, it is necessary to ‘manip-
IT-25.,’pp. 47153; Jan. 1979. ulate the RF output signal spectrum. Such a spectrum manip-
P.Kerstenand L . , Kurz.“Bivariatem-intervalclassifiers with ulation cannot usually be performed at the final RF stage in
application to edge detection.” Inform. Contr.. vol. 34. pp. 152- the multichannel SCPC transceivers because thetransmitted
168.June1977. ’ *

S . A. Kass?m. “Optimum quantization for signal detection.” IEEE RFfrequency is varia.ble. Thekefore, intermediate-frequency
Trans:.Cornmu.n.. vol. COM-25. pp. 479-484. May 1977. (IF)or basebandfilteringwith frequencyup conversion is
H . V: Poor and J. B...Thomas, “Application of Ali-Silvey distapce mostly used. However, wben suchaspectrum-manipulated
measures in’the design of generalized quantizers for binary decision signal is translated up and passed through a nonlinear class€
systems,” ‘Ii5EE Trans.’ Commun., voi. COM-$5, pp. ‘893-900.
power amplifier, the required. sAectrum manipulation should
Sept . 1.977.
H . V.‘Pooiand Y. Rivani. “Input amplitude compression in digital pot be violated by the nonlinearities. In order to mitigate the
signal-detection systems,’‘ IEEE Trans. Commun., vol. COM-29, impahments, somenarrow-band digital modulation schemes
pp. 707-7t0, ,May’1981. with constant or less fluctuated. envelope property have been
T:‘.L. Lim, .
“Noncoherent digital
matchedfilters:
Multibit researched [ 81 -[ l o ] .
quantizatjon.” .\FEE Tians.Commun.. vol.COM-26.pp. 409-
4.19. Apr.1978. In thispaper,premodulation Gaussian filteredminimum
J. ‘Capon,“Ontheasymptoticefficiencyoflocallyoptimum shift keying (GMSK) with coherent detection is proposed as
detectors,” IRE Trans. Inform. Theow. vol. IT-7. pp. 67-71, Apr. an effective digital modulation for the present purpose, and
1961. its fundamental properties areanalyzedwith the aid of ma-
chine computation. The relationship
betweenout-of-band
radiation suppression and bit-error-rate (BER) performance is
made clear. Constitution of the modulator and demodulator
is then discussed. The superiority of this modulation is sup-
GMSK Modulation for Digital Mobile Radio Telephony ported by some experimental test results.
I .

11. GMSK MODULATION


KAZUAKI MUPOTA, MEMBER, IEEE, AND KENKICHI HIRADE,
. I MEMBER,
IEEE A . Spectrum Manipulation of MSK
Minimum shift keying (MSK), which is binary digital FM
Abstract-Tkis paper is concernedwithdigitalmodulationfor with a modulationindex of 0.5, hasthe following good
.futuremobileradiotelephoneservices.First,thespecificrequire- properties:constant envelope, relatively narrowbandwidth,
ments on. the digital modulation for mobile radio use are described. and coherent detection capability [ 1 1 ] -[ 131. However, it
Then, ’ ‘prepiidutation ’ Gaussianfilteredminimum sh{ft keying does not satisfy the severe requirements with respect to out-
(GMSK) dith coherent detection is proposed as an effective digital of-band radiation for SCPC mobile radio. MSK can be gener-
modulation ‘for the present purpose, and its fundamental properties ated by direct FM modulation. As is easily found, the output
are clarihedwith the aid of machine computation. The constitutionof power spectrum of MSK can be manipulated by using a pre-
modulator ‘and demodulator . . i s then discussed from the viewpoints of modulation low-pass filter (LPF), keeping the constant enve-
mobile‘radio‘applications.The‘superiority of thismodulation is
I .lope property, as shown in Fig. I . To make the output power
supported by some experimental test results.
. . spectrumcompact,thepremodulation LPFshould have the
following properties:
I. INTRODUCTION 1) narrow bandwidthand sharp cutoff
It is well known that voice transmission in many VHF and 2) lower overshootimpulse response
UHF mobile radio telephone systems has usually been made 3) preservation of the filter output pulse area which corres-
by using a single-channel-per-carrier (SCPC) analog FM trans- ponds to a phase shift nl2.
mission technique. However, in. order to provide highly secure Condition 1) is needed to suppress the high-frequency com-
voice and/or ‘high-speed data transmission by the use of large- ponents, 2) is to protect against excessive instantaneous fre-
scale integrated (LSI) transceivers, digital mobile radio trans- quencydeviation,and 3)isforcoherentdetectionto be
. -
applicable as simple MSK.
Paper. approved by the Editor for Communication Theory of the Generally, theintroduction of thepremodulation LPF
IEEECommunicationsSocietyforpubljcationafterpresentationat violates themhimumfrequency spacing constraint and the
29thIEEEVehicularTechnologyConference,Chicago,IL;March fixed-phase constraint of MSK. However, the above two con-
1979. Manuscript received May‘28,.1980;reviSed January 5, 1981.. straints are not intrinsic requirements for effective coherent
The authorsarewiththeYokosukaElectricalCommunication
LabQratory, Nippon Telegraph a n d TelephonePublicCorporation, binary FM with modulation index 0.5. Such a premodulation-
Kanagawa-Ken, Japan. filtered MSK signal can be detectedcoherently because its

0090-6778/81/0700-1044$00.75 0 1981 IEEE


IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. COM-29, NO. 7 , JULY 1981 1045

DATA NRZ
LPF
FM MODULATOR
Fig. 1. Premodulationbaseband-filtered MSK.
,@

pattern-averaged phase-transition trajectory does not deviate


from that of simple MSK.

B. Fundamental Propertiesof GMSK


A Gaussian LPF satisfies all theabovedescribed charac-
teristics. Consequently, the modified MSK modulation using a
premodulation Gaussian LPF can be expected to be an excel-
lent digital modulationtechniqueforthe present purpose:
Such a modified MSK is named Gaussian MSK or GMSK in
connection with Gaussian low-pass filtering. Let us nowin-
vestigate the GMSK modulation from various aspects.
Output PowerSpectrum: Fig. 2shows the machinecom-
puted results of theoutput power spectrum of the GMSK
signal versus the normalized frequency difference fromthe
carrier center frequency cf - f,)T where the normalized 3 dB-
down bandwidth of the premodulation Gaussian LPF B b T is
a parameter. The spectrum forGMSK with Bb T = 0.2 is nearly
equal to that of TFM.
The effective variable parameter BbT can be selected by the
system designer considering overall spectrum efficiency of the
cellular zone structure. 0 0.5 1 .o 1.5 2.0 2.5
Fig. 3 shows the machinecomputed results of the fractional
NORMALIZED FREQUENCY : (f-fc)T
power in the desired channel versus the normalized bandwidth
of thepredetection rectangularbandpass filter (BPF) BiT. Fig. 2. Powerspectra of GMSK.
Table I shows the occupied bandwidth for the prescribed per-
centage of power where B b T is also a variable parameter. For
comparison, the occupied bandwidth of TFM is also shown in
Table I.
Fig. 4 shows the machinecomputed results of the ratio of
the out-of-band radiation powerin the adjacentchannel to
the total power in the desired channel where the normalized
channel spacing fsT is taken as the abscissa and both channels
are assumed to have the idealrectangularbandpasscharac-
teristics with BiT = 1 . The situation of fsT = 1.5 and BiT = 1
corresponds to the case of fs 25 kHz and Bi = 1 6 kHz when
f b = 1/T = 16 kbits/s. From Fig. 4, it is found that the, GMSK
with Bb T = 0.28 can be adopted as the digital modulation for
conventional VHF and UHF SCPC mobile radio com,munica-
tions without carrier frequency drift where the ratio of out-of-
band radiationpower inthe adjacent channel# to thetotal
powerin the desired channelmust be lowerthan -60 dB.
When a certain amount of carrier frequency drift (for example
A f = k 1.5 kHz) exists, BbT = 0.2 is needed.
BER Performance: Let us now consider thetheoretical
BER performance of GMSK modulation using- coherent de-
tection in the presence of additive white Gaussian noise.
Since the GMSK modulation of interest is a certain kind of
binary digital modulation, its BER performance bound in the
high SNR condition is approximatelyrepresented as NORMALIZED BANDWIDTH : BiT
Fig. 3. Fractional power ratioof GMSK.

where No is the power spectrum density of the additive white


Gaussian noise and erfc( ) is the complementary error func-
1646 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL.NO.
COM-29, 1981
7, JULY

TABLE I
OCCUPIED BANDWIDTH CONTAINING A GIVEN PERCENTAGE
POWER
-

99 99:9 99.99

0.2 0.52 0.70 0.99 1.22 W

-.
N.

0.25 0.57 0 : 86 1.09 1.37 S

t
._ N
73
'E
1:;;

0.79 0.52 1.37 1.02

0 LPF BANDWIDTH : BbT


Fig. 5. Normalized minimum signal distance of GMSK.

-20
Fig. 5 shows the machinecomputed results for dmin of the
GMSK signal versus BbT where denotesthe signal energy
per bit defined by

+
-40

-60
Eb =
T

:dT
I um(t>1 2 dt = - 1 u,(t> I* dt.
(4)

In the case BbT + 00, which corresponds to the simple MSK


signal, Fig. 5 yields dmin = 2&b, which is that of antipodal
-80 transmission. It is noticedthatthe meaningful observation
time interval for the GMSK signd t2 - t l may be made longer
than 2T, which corresponds to that for the simple MSK signal,
- 1 00 due to the intersymbol interference (ISI) effect on the phase
transitions.
Substitutingthe machine-computed results of d,i, into
(l), the BER performance of the GMSK modulation with
0 0.5 1 .o 1.5 2.0 2.5 coherent detection is obtained. Fig. 6 shows the performance
degradation of GMSK from antipodal transmission due to the
NORMALIZED CHANNEL SEPARATION : fST IS1 effect of the premodulation LPF. This figure shows that
Fig. 4. Adjacent channel interference of GMSK. theperformance degradation is small and thattherequired
Eb/No of GMSK with BbT = 0.25 does not exceed more than
0.7 dB compared to that of antipodal transmission.
tion given by

111. IMPLEMENTATION
A . Modulator
Furthermore, dmin is the minimum
value of the signal distance The Simple and easy method is to modulate the frequency
d between mark and space in Hilbert space observed during the of VCO directly by the use of baseband Gaussian pulse stream,
time interval from t l to t2 and d is defined by as shown in Fig. 1. However, thismodulatorhasthe weak
point that it is difficult to keep the center frequency within
the allowable value under the restriction of maintaining the
linearity and the sensitivity for the r e q ~ e dFM modulation.
Such a weak point can be removed by the use of an elaborate
PLL modulator witha precisely designed transfer character-
where u,(t) and u,(t) are the complex signal waveforms corre- istics oranorthogonalmodulator with digital waveform
sponding to the mark andthe spacetransmissions,respec- generators [ 141. Instead of such a modulator, anl2-shift
tively. binary PSK (BPSK) modulator followed by asuitable PLL
While the BER performance bound given by (1) is attained phase smoother, as shownin Fig. 7, is considered t o be a
only when the ideal maximum likelihood detection is adopted, prominent alternativewhere thetransfer characteristics of
it gives an approximate solution for the ideal BER perform- this PLL are also designed for the output power spectrum to
ance of GMSK modulation with coherent detection. satisfy the required condition.
IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. COM-29, NO. 7 , J U L Y 1981 1047

1---7--==5
1
LPF

FILTER
n

0 0: 2 0.4 0.6 0.8


L P F BANDWIDTH : BbT

Fig. 6 . Theoretical Eb/N, degradation of GMSK.

DATA n / 2 SHI FT OUTPUT


BPSK LPF

t vco

u
Fig. 7. PLL-type GMSK modulator. (b)
Fig. 8. Orthogonalcoherentdetectorfor MSK/GMSK. (a) Analog
B. Demodulator type. (b) Digital type.
Similar to the simple MSK or TFM system, the orthogonal FM
coherentdetectoris also applicable forthe GMSK system. MODULATOR ,
When realizing such an orthogonal coherent detector, one of GAUSSIAN FADING
the most important and difficult problems is how to recover LPF SIMULATOR
the referencecarrierand the timingclock. The mosttypical
method is de Buda's one [ 121. In his method, the reference
camer is recovered by dividing by four the sum of the two
discrete frequencies contained in the frequency doubler out- / A ATTENUATOR
put and the timing clock is directly recovered by their differ- 4
ence, Remembering that the action of the well-known Costas DEM ERROR RATE
loop as a carrier recovery circuit for BPSK systems is equiva- COUNTER
lent to thatof a PLL with a frequency doubler [ 151 , de Buda's Fig. 9. Block diagram of experimental test system.
method is realized by the equivalent one shown in Fig. 8(a).
This modified method can easily be implemented by conven- modulation Gaussian LPF having a variable bandwidth B b , the
tional digital logic circuits and its configuration is also shown PN sequence is put into the synthesized RF signal generator
in Fig. 8(b). In this configuration, two D flip-flops act as the having an external FM modulation capability. The frequency
quadrature product demodulators and both of the Exclusive- deviation of the RF signal generator is set equal to Af, =
Or logic circuits are used for the baseband multipliers. Further- ? 4 kHz, whichcorresponds to the MSK conditionforthe
more, the mutually orthogonal reference carriers are generated 16 kbits/s transmission. Then the GMSK signal of our choice
by the use of two D flip-flops, and the VCO center frequency is obtained as the RF signal generator output, and is trans-
is then set equal to the four timescarrier center frequency. mittedintothe receiver via the Rayleigh fading simulator
This configuration is considered to be especially suitable for [ 161. Predetection bandpassfilteringin the receiver is per-
the mobile radio unit which must be simplified, miniaturized, formed by the precisely designed Gaussian bandpasscrystal
and economized. filter. The bandpass-filtered output is demodulated by the
digital orthogonalcoherentdetectorshown in Fig. 8. The
IV. EXPERIMENTS regenerated output is fed into the error-rate counter for the
BER measurement.
A . Test System
Fig. 9 shows the block diagram of theexperimentaltest B. Power Spectrumand Eye Pattern
system where the carrier frequency and the bit rate are f,= Fig. 10 shows the measured power spectra of the RF signal
70 MHz and f b = 16 kbits/s, respectively. A pseudonoise (PN) generatoroutput when BbT is a variable parameter.It is
pulse sequence with a repetition period of N = (2' - 1) bits clearly seen thatthe measured results agree well with the
is generated by the 15stage feedback shift register (FSR) and machine-computed ones shown in Fig. 2. Moreover, GMSK
is used as a test pattern signal. After passing through a pre- with BbT = 0.25 is shown t o satisfy the severe requirements
1048 IEEE TRANSACTIONS
COMMUNICATIONS.
ON VOL. COM-29, NO. 7 , JULY 1981

BbT = 0. 5

B b T = 0. 2 5

Fig. 11. Instantaneous frequency variation of GMSK.

IS1 effect causes inferior transmission performance. However,


this misgiving is happily unwarrantedbecause the dernodulator
output of GMSK with B b T = 0.25 degrades only slightly from
that of simple MSK. It is easily foundfrom Fig. 12 which
BbT 0.2 shows the respective eye patterns measured by the analog-type
orthogonalcoherentdetectorshown in Fig. 8(a). It is also
certified fromthe BER performancetestresults described
later.

C. Static BER Performance


Fig. 13 shows experimental test results for static BER per-
Fig. 10. Measured power spectra of GMSK ( V : 10 dB/div., H : 10 formance in the nonfading environment where the normalized
kHz/div.). 3 dB-down bandwidth of thepremodulation Gaussian LPF,
BbT, is a variable parameter and the normalized3 dB-down
of the out-of-band radiation of SCPC mobile radio communi- bandwidth of the predetection Gaussian BPF is BiT E 0.63,
cations. The corresponding eye pattern measured at the pre- i.e., Bi = 10 kHz for f b = 1/T = 16 kbits/s. The condition
modulation Gaussian LPF output is shown in Fig. 11. This BtT 0.63 is nearly optimum, as shown in Fig. 14. From Fig.
figure shows that the above satisfactory performance of the 13,performance degradation of GMSK with BbT = 0.25
out-of-band radiation can only be attained by the sacrifice of relative to simple MSK is found to be only 1.0 dB. Moreover,
introducing severe IS1 effects into the baseband waveform of the measured static BER performance of simple MSK degrades
the FM modulator input. It might be feared that such a severe by 0.7 dB from the theoretical one of ideal antipodal binary
IEEE TRANSACTIONS ON COMMUNICATIONS, V U L . LVIVI-LY, I Y U I , J U L Y IY8I 1049

B T = 0.5
b

4 14 6 128 10 16

E b / N O( d B )

Fig. 13. Static BER performance.

BbT = 0 . 2 5

BbT = 0 . 2
I
I 1
0 0.5 1 .o 1.5 2.0
PREDETECTIONBPFBANDWIDTH : BiT

Fig. 14. Degradation of required Eb/No for obtaining BER of 10-3.


Fig. 12. GMSK eye patternsdemodulated by orthogonalcoherent
detector. The aboveabtained results can be estimated by the degrada-
tion of the minimum signal distance shown in Figs. 5 and 6.
transmission system. If y' denotes the received signal energy-
to-noise density ratio, i.e., Eb/No, the measured static BER
D. DynamicBER Performance
performance in thenonfadingenvironment can be approxi-
mated as Inthe practical V/UHF land mobile radioenvironment,
P,(Y) 2 erfc (6) (5) signal transmission between a fixed base station and a moving
vehicle is usually performed via random multiple propagation
where erfc( ) is the complementary error function given by routes. Consequently, fast and deep multipath fading, which
(2) and a is a constant parameter determinedas can generally be treated by the well-known Rayleigh fading
model,appears on the received signals of both stations and
0.68 for GMSK with BbT = 0.25
degrades the signal transmission performance severely.
0.85 for simple MSK (BbT-' -). (6) In particular, when a quasi-stationary slow Rayleigh fading
1050 IEEE TRANSACTIONS ON COMMUNICATIONS, vOL. COM-29, NO. 7, J U L Y 1981

ment, i.e., f D T -+ 0, is also shown by the dashed line in the


same figure.
V. CONCLUSION
As an effective digital modulationfor mobile radio use,
.............................-
L premodulation Gaussian-filtered minimum shift keying (GMSK)
modulation with coherent detection has been proposed. The
fundamental properties have beenanalyzedwith the aid of
machine computation. The constitution of modulator and
fD = 40 Hz demodulator has also been discussed. The superiority of this
-
............ ..-.........-
.I..
- modulation has been supported by experimental results.

PI
w ACKNOWLEDGMENT
m
1 2 Hz The authors wish to thank Dr. K. Miyauchi, S. Ito, K. Izumi,
-c
...............& ....n:..........- and Dr. S. Seki fortheir helpfulguidance.They also are
grateful to Dr. M. Ishizuka and H. Suzukifortheirfruitful
discussions.

4 Hz REFERENCES
L,....
0. . -
Bettinger.
“Digital
speech
transmission for mobile radio
service,’’ Elec. Commun.. vol. 47. pp. 224-230. 1972.
,: J. S . Bitler and C. 0. Stevens, “A UHF mobile telephone system
t. \
using digital modulation: Preliminary study,” IEEE Trans. Vehic.
\ i Technol.. vol.VT-22, pp. 78-81. Aug. 1973.
\ i
\ j N. S . Jayant.R.W.Schafer.andM.R.Karim,“Step-size-
i transmittingdifferentialcodersformobiletelephony,” in Proc.
0 10 20 30 40
60 50 70 IEEE Int. Conf. Commun., June 1975, pp. 3016-30110.
D. L. Duttweiler and D. G . Messerschmitt, “Nearly instantaneous
AVERAGE Eb/NO (dB)
compandingandtimediversityasappliedtomobileradiotrans-
Fig. 15. Dynamic BER performance. mission,” in P r o c . IEEE Int.Conf.Commun.. June 1975,pp.
401 121101 15.
J . C. Feggeler, “A study of digitized speech in mobile telephony,”
presentedattheSymp. on Microwave Mobile Commun., session
model is assumed, dynamic BER performance isgiven by V-3, Boulder, CO, Sept.-Oct. 1976.
H. M.Sachs.“Digitalvoiceconsiderationsforthelandmobile
rm
radio services.’’ in P r o c . IEEE 27th Vehic. Technol. Conf., Mar.
1977. pp. 207-219.
K.HiradeandM.Ishizuka,“Feasibilityofdigitalvoicetrans-
mission in mobileradiocommunications,” PaperTech.Group.
IECE Japan. vol. CS78-2, Apr. 1978.
where I’ is the average Eb/NO and p ( y ) is theprobability F. G. Jenks, P. D. Morgan, and C . S . Warren, “Use of four-level
density function (pdf) of y given by phasemodulation for digitalmobileradio,” IEEE Trans.Elec-
rromagn. Compar., vol. EMC-14, pp. 113-128, Nov.1972.
P. K. Kwan. “The effects of filtering and limiting a double-binary
PSKsignal,” IEEE Trans. Aerosp. Electron. Svst., vol.AES-5.
pp. 589-594,July1969.
S . A.Rhodes,“Effectsofhardlimitingonbandlimitedtrans-
mission with conventional and offset QPSK modulation,” in Proc.
Substituting ( 5 ) and (8) into (7) yields IEEE Nat. Telecommun. Conf., 1972, pp. 20F/I-20F/7.
H. C . van den Elzen and P. van der Wurf, “A simple method of
calculating the characteristics ofFSK signals with modulation index
0.5,” IEEETrans.Commun., vol.COM-20,pp. 139-147, Apr.
1972.
R . deBuda.“Coherentdemodulationoffrequencyshiftkeying
with low deviation ratio.” IEEE Trans. Commun., vol. COM-20,
where a is the constant parametergiven by (6). pp. 46-70, June 1972.
However, the dynamicBER performance in thefast Rayleigh H. Miyakawa e ta l . , “Digitalphasemodulationschemeusing
continuous-phase waveform.” Trans. IECE Japan, vol. 58-A, pp.
fading environment, where thetemporal variation effect of 767-774.Dec.1975.
the fading cannot be neglected, has not yet been theoretically F. de Jager and C. B. Dekker.“Tamedfrequencymodulation,a
estimated because thetrackingperformance of the carrier novelmethodtoachievespectrumeconomy in digitaltransmis-
recoverycircuit in such environmentcannot be analyzed. sion.” IEEE Trans. Commun., vol.COM-20.pp. 534-542. May
Fig. 15 shows the experimental test results of dynamic BER 1978.
R. L.DiddayandW. C. Lindsey.“Subcarriertrackingmethods
performance of the CMSK with BbT = 0.25 in the simulated andcommunication systemdesign,” IEEE Trans. Commun.
fast Rayleigh fading environment where the maximum Doppler Technol.. vol.COM-16.pp.541-550,Aug.1968.
frequency, i.e., thefadingrate f D , is a variable parameter. K.Hirade etal. “Fadingsimulatorforlandmobileradiocom-
For comparison, theoreticallyestimateddynamic BER per- munications,” Trans. IECE Japan, vol. 58-8. pp. 449-459. Sept.
1975.
formance in the quasi-stationary slow Rayleigh fading environ-

You might also like