FSK Demodulation Method Using Short-time DFT Analysis for LEO

8/10/2019 FSK Demodulation Method Using Short-time DFT Analysis for LEO

1/9

IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 46, NO. 3, AUGUST 1997 625

A Novel FSK Demodulation Method UsingShort-Time DFT Analysis for LEO

Satellite Communication SystemsShinsuke Hara, Member, IEEE, Attapol Wannasarnmaytha, Student Member, IEEE,

Yuuji Tsuchida, and Norihiko Morinaga, Senior Member, IEEE

AbstractThis paper proposes a novel frequency-shift keying(FSK) demodulation method using the short-time discrete Fouriertransform (ST-DFT) analysis for low-earth-orbit (LEO) satellitecommunication systems. The ST-DFT-based FSK demodulationmethod is simple and robust to a large and time-variant frequencyoffset because it expands the received signal in a time-frequencyplane and demodulates it only by searching the instantaneousspectral peaks with no complicated carrier-recovery circuit. Twokinds of demodulation strategies are proposed: a bit-by-bit de-

modulation algorithm and an efficient demodulation-algorithmfrequency-sequence estimation (FSE) based on the Viterbi algo-rithm. In addition, in order to carry out an accurate ST-DFTwindow synchronization, a simple DFT-based ST-DFT window-synchronization method is proposed.

Index Terms Discrete Fourier transforms, Doppler effect,frequency-shift keying, satellite communication.

I. INTRODUCTION

PERSONAL communication systems (PCSs) make com-

munication from person-to-person, with a wide range of

services such as voice and data transmission with different

service qualities, whenever they are required, regardless of

where we locate [1].Low-earth-orbit (LEO) satellite systems have the advantages

of the interoperability of terrestrial cellular and mobile systems

as well as shorter transmission delay and lower propagation

path loss as compared with geostationary-earth-orbit (GEO)

satellite systems. The LEO satellite network is a candidate to

provide such truly seamless global personal communications

services because it has all the coverage, capacity, and features

required for the PCS realization. However, the system suffers

from the Doppler frequency offset.

In the LEO satellite system, there exists a large and time-

variant frequency shift due to the Doppler effect, depending

on the carrier frequency, satellite altitude, orbit, and coverage

assigned to each LEO satellite. Fig. 1 shows the Doppler shift

and rate versus the time, where the earth station is located

Manuscript received December 15, 1995; revised August 1, 1996.S. Hara is with the Department of Electronic, Information, and Energy

Engineering, Graduate School of Engineering, Osaka University, Osaka, Japan(e-mail: [email protected]).

A. Wannasarnmaytha and N. Morinaga are with the Department of Com-munication Engineering, Graduate School of Engineering, Osaka University,Osaka, Japan.

Y. Tsuchida is with the Audio Laboratory, Sony Corporation, Tokyo, Japan.Publisher Item Identifier S 0018-9545(97)04630-6.

Fig. 1. Doppler shift and its rate. The earth station is located on thecrosspoint of the equator and footprint of the satellite coverage of a polarcircular orbit with a satellite altitude km and a carrier frequency GHz.

on the crosspoint of the equator and footprint of the satellite

coverage in a polar circular orbit with a satellite altitude of

788 km and a carrier frequency of 2 GHz [2]. In this case, the

Doppler shift ranges from 40 to 40 kHz and the Doppler

rate from 0 to 5.5 kHz/s. For a symbol transmission rate

of 8.0 kb/s as a low-rate service, for instance, the required

bandwidth of the receiver front-end bandpass filter becomes

approximately five times as large as the symbol rate. Therefore,

it is essential to develop modulation or demodulation schemes

to cope with such a large and time-variant frequency offset.

Also, in the PCS, associated with the miniaturization of

personal terminals, the problem of frequency offset is caused

by the frequency instability of the terminal local oscillator.

An efficient automatic frequency control (AFC) loop might

be one of the solutions [3], [4]. However, there exists a

fundamental time-frequency tradeoff: improving the frequency

resolution results in a loss of time resolution and vice versa

[5]. In other words, an accurate frequency estimation requires along preamble and inevitably introduces a loss of transmitted

power efficiency.

Much effort has been devoted to the analysis of the AFC

tracking performance in the presence of frequency offset and

to the proposal of modulation/demodulation schemes robust to

the large and fast frequency offset. For instance, the tracking

performance of the crossproduct AFC in the Costas loop

is discussed in [6]. A double-pilot-assisted QPSK coherent

demodulation method and a Doppler-corrected differential

detection method of MPSK are proposed in [7] and [8],

00189545/97$10.00 1997 IEEE


2/9

626 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 46, NO. 3, AUGUST 1997

Fig. 2. LEO satellite channel model.

Fig. 3. Transmitter model.

respectively, both of which can cope with time-variant fre-

quency offset. A dual-channel PSK demodulator for LEO

satellite DS/CDMA communications is proposed in [9], whichis absolutely insensitive to time-variant Doppler frequency

offset. Also, a simple coarse frequency acquisition method

through fast Fourier transform (FFT) is proposed in [10].

This paper proposes a novel frequency-shift keying (FSK)

demodulation method using the short-time discrete Fourier

transform (ST-DFT) analysis for an LEO satellite commu-

nication channel with a large and time-variant frequency

offset [11]. The ST-DFT-based FSK demodulation method

expands the received signal in a time-frequency plane based

on the ST-DFT analysis and demodulates it by searching

the instantaneous spectral peaks with no complicated carrier-

recovery circuit. Two kinds of demodulation strategies areproposed: a bit-by-bit demodulation algorithm and a novel ef-

ficient demodulation-algorithm frequency-sequence estimation

(FSE) based on the Viterbi algorithm. In addition, in order

to carry out an accurate ST-DFT window synchronization, a

simple DFT-based ST-DFT window-synchronization method

is proposed.

Sections II and III deal with the channel model and trans-

mitter/receiver model, respectively. Sections IV and V explain

the ST-DFT-based demodulation principle and algorithms,

respectively. Section VI explains the DFT-based ST-DFT

window-synchronization method. Section VII shows the com-

puter simulation results on the bit error probability (BEP).

Finally, Section VIII draws the conclusions.

II. CHANNEL MODEL

We model an LEO satellite communication channel as an

additive white Gaussian Noise (AWGN) channel with fre-

quency offset (see Fig. 2). In a burst mode transmission, where

the signal-burst length is small, considering the frequency

variation up to the first-time derivative, we can approximate

the frequency offset introduced in a signal burst as

(1)

where is the time and is the signal-burst length in time.

We call and the (initial) fixed frequency offset and

frequency-offset rate, respectively. Taking this first-order

approximation, we can evaluate the robustness of the proposed

demodulation method against the frequency offset only for

and .

III. TRANSMITTER AND RECEIVER MODELS

A. Transmitter Model

Fig. 3 shows the block diagram of a binary differentially

encoded FSK (BDEFSK) transmitter. The information data

stream ( or ) is dif-

ferentially encoded, passed through the Nyquist filter with

rolloff factor , and then modulated by the FM modulator

with modulation index . The transmitted signal with unit

amplitude is written by

(2)

where and represent the real part of and the center

frequency, respectively. is the modulated phase given by

(3)

where is the symbol duration and ( or ) is

the differentially encoded th symbol

(4)

The impulse response of the Nyquist filter is given by

(5)

In the BDEFSK scheme, the information 1 is transmitted

by shifting the carrier frequency relative to the previous

carrier frequency and information 1 by keeping the same

carrier frequency. We define and as the higher and

lower transmitted frequencies at sampling instant ,

respectively, and as the frequency separation

(6)

(7)

(8)

B. Receiver ModelFigs. 4 and 5 show the block diagram of the ST-DFT-based

differential frequency receiver and the instantaneous energy

distribution of the received signal, respectively. The received

signal through the LEO satellite channel mentioned in Section

II is written as

(9)

where is the amplitude of the received signal and assumed

to be constant and is the complex AWGN. is passed

through the receiver front-end bandpass filter (BPF) with

bandwidth Hz centered at the nominal center frequency


3/9

HARA et al.: NOVEL FSK DEMODULATION METHOD FOR SATELLITE COMMUNICATION SYSTEMS 627

Fig. 4. Receiver model.

Hz and then downconverted by . After analog-to-digital

(A/D) conversion with sampling rate Hz, the output signal

is expanded into a time-frequency plane by the ST-DFT in

order to analyze the instantaneous energy distribution. Finally,

the demodulation is made based on the spectral analysis result.

Defining and as the maximum frequency offset

introduced in the channel and bandwidth of the transmitted

signal (see Figs. 1 and 5) in order to introduce no distortion

in the received signal, the bandwidth of BPF must satisfy the

following condition:

(10)

Also, in order to introduce no aliasing distortion in the

A/D conversion, the sampling rate must satisfy the following

condition:

(11)

IV. ST-DFT-BASED DEMODULATION PRINCIPLE

A. ST-DFT

The time-frequency representation of a signal basedon the ST-DFT, which is often called Spectrogram, is given

by [12]

(12)

(13)

where is the sampling interval and and

represent a finite-time and even-symmetrical window func-

tion and a number of samples in one window, respectively.

Equation (13) represents the spectral component of at

the th time index and th frequency index

. We define as a point (node) atand on the time-frequency plane. Furthermore, we define

as the instantaneous energy spectrum of at

(14)

B. Basic Demodulation Principle

Fig. 5 shows the basic principle of the ST-DFT-based

demodulation method. After the ST-DFT window synchroniza-

tion is established, the differential frequency demodulation is

made by searching the instantaneous spectral peak of

at . Analysis of the

Fig. 5. Instantaneous energy distribution of received FSK signal and basicdemodulation principle.

received signal with the ST-DFT is all the same as observation

through a filter bank with a number of narrowband filters.

Therefore, the demodulation performance depends not on

the front-end BPF output signal-to-noise power ratio (SNR),

but on the narrowband BPF output SNR. Consequently, inprinciple, however wide the front-end BPF may be made,

it introduces no difference in the demodulation performance.

In other words, the ST-DFT-based demodulation method is

insensitive to the SNR degradation caused by the excessively

wide bandwidth of front-end BPF.

Also, when there are frequency-division multiplexed

channels in the received frequency band because of the wide

front-end BPF, the receiver could find distinct peaks in the

instantaneous energy spectrum at every demodulation instance.

When a signal burst is transmitted with a specific preamble

(unique word) in each channel, the receiver can easily identify

the desired channel and carry out demodulation, focusing

attention only on the desired part of the received frequencyband. Therefore, the ST-DFT-based demodulation method can

mask the false spectral peaks in adjacent channels.

C. Maximum-Likelihood Estimation (MLE) Characteristic

The transmitted signal (when an unknown frequency offset

is introduced in the channel) can be considered to be a mono-

tone with an unknown (discrete) frequency .

Assuming that the frequency of the received signal does not

change in one DFT window, the monotone composed of -

time samples in one window is written in a vector form

as

(15)

(16)

where is an unknown phase. Defining

as the received signal vector

composed of -time samples in one window at ,

is written as

(17)

where is a noise vector and

each component is Gaussian distributed. Therefore, the joint


4/9


probability density function (pdf) for conditioned on the

monotone signal can be written as [13]

(18)

where is the power of . Averaging (18) by , the joint pdf

for conditioned on the frequency is written as

(19)

where is the zeroth-order modified Bessel function of

the first kind. Equation (19) shows that although we assume

a rectangular window, the frequency , which maximizes

, is the MLE of the frequency of the transmitted signal.

Therefore, the ST-DFT-based demodulation method is opti-

mum and can minimize the BEP when the frequency-offset rateis negligibly small . However, the following bit-by-

bit demodulation algorithmand frequency-sequence estimation

(FSE) algorithm are suboptimal because the demodulation

principle is modified in order to track the frequency drift due

to the larger frequency-offset rate.

V. ST-DFT-BASED DEMODULATION ALGORITHMS

In order to analyze the instantaneous energy distribution ofthe received signal accurately in the ST-DFT-based demod-

ulation method, the interpolation technique with points is

employed. The frequency resolution is given by

(20)

A. Bit-by-Bit Demodulation Algorithm

The bit-by-bit demodulation is made according to the fol-

lowing algorithm (see Fig. 6).

1) Let .

2) Calculate for the th symbol.

3) Let .

4) Calculate given by

(21)

Fig. 6. Bit-by-bit demodulation algorithm.

where and are the th peak

frequency for and the decided frequency for

, respectively.

5) Make a decision according to

(22)

6) If does not satisfy the condition in Step 5), then

let and go to Step 4).

7) Set to be , then let and go

to Step 2).

The decision criterion in Step 5) ensures the prevention of

the misdetection of the false spectral peak due to background

noise and the tracking of the frequency drift due to the

frequency-offset rate.

B. FSE Algorithm

We propose a novel demodulation algorithm FSE to improve

the demodulation performance for a large and time-variant

frequency offset. The FSE is a kind of Viterbi algorithm [14],

where state and metric in the Viterbi algorithm correspond

to the node on the time-frequency plane and the

amplitude of ST-DFT output , respectively (see Fig. 7).

The FSE algorithm examines all the frequency paths leading

to a given node and chooses the most likely path according

to the accumulated metric. After the procedure is repeated for

all the frequency indexes in a given time period (the data-fieldlength in a signal burst), a frequency-index sequence with the

largest accumulated metric is finally chosen.

The FSE is made according to the following algorithm

(see Fig. 7), where is the accumulated metric at

, is a set of transition frequency indexes, which

allows the previous nodes to transit to , and is

the number of data symbols in a signal burst.

1) Let and set to be for

.

2) Let , examine all the frequency paths leading

to , and choose the most likely path according


5/9


(a)

(b)

Fig. 7. FSE. (a) Metric and node. (b) Frequency-index-sequence estimation.

to for

(23)3) If , then go to Step 2).

4) Find the frequency-index sequence according to the

largest .

We propose the following two sets of the allowable transition

frequency indexes.

1) Maximum Likelihood FSE (MLFSE):

, where the frequency offset is

assumed to be time-invariant and the frequency

transitions associated only with the modulation process

are allowed [see Fig. 8(a)].

2) FSE: , where the

frequency offset is assumed to be time-variant and fre-quency transitions associated with both the modulation

process and the frequency drift due to frequency-offset

rate are allowed [see Fig. 8(b)].

Note that the MLE characteristic can hold only for the

MLFSE, where the frequency variation in one signal burst is

negligibly small .

C. Required Memory, Demodulation Delay,

and Complexity Comparisons

The bit-by-bit algorithm needs to memorize only a previ-

ously decided frequency at every demodulation instant

(a) (b)

Fig. 8. Allowable frequency transition. (a) MLFSE and (b) FSE.

and requires no demodulation delay. This is the simplest

among the three algorithms.

On the other hand, the MLFSE and FSE algorithms need

to store all the frequency paths with a huge memory and

require -symboldemodulation delay similar to conventional

Viterbi algorithms for convolutional codes. Furthermore, the

number of comparisons to choose a most-likely frequency path

leading to each node is two and eight for the MLFSE and FSE,

respectively. In this sense, the FSE is more complicated than

the MLFSE. In order to shorten the demodulation delay, we

have discussed the effect of the frequency-path history length.

A truncated algorithm only with an eight-symbolpath-history

length can achieve almost the same BEP performance as the(nontruncated) FSE algorithm [the associated demodulation

delay is eight (symbols)] [15].

VI. DFT-BASED ST-DFT WINDOWSYNCHRONIZATION

The ST-DFT-based demodulation method requires no car-

rier frequency/phase recovery, but an accurate ST-DFT win-

dow synchronization. Therefore, we propose a DFT-based

ST-DFT window-synchronization method.

Fig. 9 shows a signal burst used in the ST-DFT-based

demodulation method. The preamble is composed of

symbols, where and alternately appear and the tail

symbol is used to identify the end of the preamble.Defining as the number of samples per window, we can

calculate kinds of for the th symbol with different

sets of window timing . The ST-DFT

window-synchronization method finds the best window timing

that can minimize the intersymbol interference due to the

window-timing offset (see Fig. 9).

The ST-DFT window synchronization is made according to

the following algorithm.

1) Let .

2) Calculate for the symbols in the preamble

.


6/9


Fig. 9. DFT-based ST-DFT window-synchronization method.

3) Calculate the cumulative for odd symbols and

for even symbols as

4) Search , which maximizes , and

, which maximizes .

5) Calculate given by

(24)

6) If , then let and go to Step 2).

7) Find the optimum window timing to maximize .

Step 3) ensures the reduction of the effect of backgroundnoise by adding up (averaging) the instantaneous energy

spectra for odd and even symbols, respectively.

VII. NUMERICAL RESULTS

Table I shows the transmission parameters to demonstrate

the BEP performance of the ST-DFT-based demodulation

method. In Figs. 1015, we assume a perfect ST-DFT win-

dow synchronization, and finally, in Figs. 16 and 17, we

show the performance of the DFT-based ST-DFT window-

synchronization method. The theoretical BEP lower bound,

which corresponds to the BEP of the differentially encoded bi-

nary FSK/noncoherent detection scheme in the AWGN channel

with no frequency offset, is given by (see Appendix)

BEP (25)

(26)

where represents the signal-to-noise energy ratio per

bit.

Fig. 10 shows the BEP versus the ST-DFT window width.

It could be impossible to evaluate the performance of all the

window functions because a number of window functions

have been proposed so far. Here, we choose typical three

window functions: the Hamming, Hanning, and rectangular

TABLE ITRANSMISSION PARAMETERS

Fig. 10. BEP versus ST-DFT window width.

window functions [16] (also, see [17] for the performance ofthe Blackman and Kaiser window functions) and try to find

the best window function and width suited to the transmitted

Nyquist pulse.

In general, a shorter window width results in a worse BEP

because of a lack of signal energy, while a longer window also

results in a worse BEP because of being rich in intersymbol

interference. Therefore, there is an optimum value in the

window width to minimize the BEP. It can be seen from thefigure that the Hanning window with the width of two-symbol

duration is the best choice among three window functions.

The rolloff factor and modulation index are im-

portant parameters for determining the required bandwidth of

the transmitted signal. Figs. 11 and 12 show the BEP versus

and , respectively. As increases, the BEP improves

because of less intersymbol interference. On the other hand,

a smaller results in a worse BEP because of narrower

frequency separation, while a larger also results in a worseBEP because of frequent misdetection of the false spectral

peak caused by the Nyquist filter. Therefore, there is an

optimum value in for minimizing the BEP. It can be seen

from the figure that, leaving the required bandwidth out ofconsideration, and are the best choices.

Fig. 13 shows the BEP versus for different values of

interpolation index . As increases, the BEP performance

improves because the instantaneous energy spectrum can be

analyzed in more detail. However, it requires more time for

the calculation. It can be seen from the figure that is a

reasonable choice from the viewpoint of calculation time and

BEP improvement.

Fig. 14 shows the BEP versus the fixed frequency offset

, where the frequency-offset rate is set to be zero.

The ST-DFT-based demodulation method is insensitive to the


7/9


Fig. 11. BEP versus rolloff factor.

Fig. 12. BEP versus modulation index.

Fig. 13. BEP versus

for different values of interpolation index.

fixed frequency offset, as long as the received signal can be

passed through the receiver front-end bandpass filter with no

distortion.

Fig. 14. BEP versus fixed frequency offset.

Fig. 15. BEP versus frequency-offset rate.

Fig. 15 shows the BEP versus the frequency-offset rate .

The bit-by-bit algorithm, which is the simplest among the three

proposed algorithms, can keep a good BEP performance for

[Hz/s]. The MLFSE algorithm can achieve

the best performance for small values of ( [Hz/s]),

which is almost the same as the lower bound. However, the

performance suddenly degrades as the frequency-offset rate

becomes large because the frequency variation

in one signal burst becomes significantly large. The FSE

algorithm, which is the most complicated one, is more robustto the frequency-offset rate than the bit-by-bit algorithm, and

it can keep a better performance for [Hz/s].

Note that the bit-by-bit and FSE algorithms can work well for

the maximum Doppler rate shown in Fig. 1.

Fig. 16 shows the average window offset versus the length

of preamble . is a reasonable choice from the

viewpoint of power efficiency and achievable window-offset

error. Fig. 17 shows the BEP versus for . The

performance of the proposed DFT-based ST-DFT window-

synchronization method is almost the same as that of the

perfect window synchronization.


8/9


Fig. 16. Average window error versus preamble length.

Fig. 17. BEP versus

with ST-DFT-based DFT window-synchro-nization method.

TABLE IIMODULATION/DEMODULATION PARAMETERS

Table II summarizes the best combination of modulationand demodulation parameters obtained in this paper.

VIII. CONCLUSION

This paper has proposed a novel FSK demodulation method

using the ST-DFT analysis for an LEO satellite communication

channel with a large and time-variant frequency offset. A

bit-by-bit demodulation algorithm and a novel efficient de-

modulation algorithm FSE have been introduced. In addition,

this paper has proposed a simple DFT-based ST-DFT window-

synchronization method.

The receiver configuration has shown the simple structure

of the ST-DFT-based FSK demodulation method, and the

numerical results show that it is robust to the time-variant

frequency offset. Also, the ST-DFT principle has revealed that

the performance is insensitive to the front-end signal-to-noise

power ratio.

The ST-DFT-based FSK demodulation method is insensitive

to the fixed frequency offset. The bit-by-bit demodulation algo-

rithm is robust to the frequency-offset rate and can keep goodBEPs for various values of . The maximum likelihood

FSE (MLFSE) can achieve the best BEP performance among

three demodulation methods for small values of the frequency-

offset rate, which is almost the same as the BEP lower bound.

However, when the frequency-offset rate becomes large, the

performance suddenly degrades. The FSE is more robust to

the frequency-offset rate and can keep better BEPs than the

bit-by-bit algorithm in the wider range of the frequency-offset

rate.

Also, the DFT-based ST-DFT window-synchronization

method, when an adequate preamble length is chosen, can

achieve almost the same performance as the perfect windowsynchronization.

APPENDIX

The BEP lower bound is given by

BEP

(27)

where and are the probability of and

the probability of given , respectively, and is the BEP

of binary FSK/noncoherent detection scheme in the AWGN

channel with no frequency offset given by [18]

(28)

REFERENCES

[1] S. Ginn, Personal communication services: Expanding the freedom tocommunicate,IEEE Commun. Mag., vol. 29, no. 2, pp. 3039, 1991.

[2] T. Toyonaga, Research on coherent demodulation scheme suited formobile satellite communication systems (in Japanese), Masters thesis,Osaka University, Osaka, Japan, 1993.

[3] H. Meyr and G. Ascheid, Synchronization in Digital Communications,Vol. I. New York: Wiley, pp. 305329, 1990.

[4] F. D. Natali, AFC tracking algorithm, IEEE Trans. Commun., vol.COM-32, no. 8, pp. 935947, 1984.

[5] Wavelets and signal processing,IEEE Signal Processing Mag.,vol. 8,no. 4, pp. 1438, 1991.

[6] C. R. Cahn, Improving frequency acquisition of Costas loop, IEEETrans. Commun., vol. COM-25, no. 12, pp. 14531459, 1977.

[7] M. K. Simon, Dual-pilot tone calibration technique,IEEE Trans. Veh.Technol., vol. VT-35, no. 2, pp. 6370, 1986.

[8] M. K. Simon and D. Divsalar, Doppler-corrected differential detectionof MPSK, IEEE Trans. Commun., vol. 37, no. 2, pp. 99100, 1989.

[9] A. Kajiwara, Mobile satellite CDMA system robust to Doppler offset,IEEE Trans. Veh. Technol., vol. 44, no. 3, pp. 480486, 1995.

[10] W. K. M. Ahmed and P. J. Mclane, A simple method for coarsefrequency acquisition through FFT, in Proc. IEEE VTC94, June 1994,pp. 297301.

[11] A. Wannasarnmaytha, S. Hara, and N. Morinaga, A novel FSK de-modulation method using short-time DFT analysis for LEO satellitecommunication systems, in Proc. IEEE GLOBECOM95, Nov. 1995,pp. 549553.


9/9


[12] F. Hlawatsch and G. H. Boudreaux-Bartels, Linear and quadratic time-frequency signal representations, IEEE Signal Processing Mag.,vol. 9,no. 2, pp. 2167, 1992.

[13] J. G. Proakis, Digital Communications. New York: McGraw-Hill, pp.340344, 1989.

[14] V. K. Bhargava, D. Haccoun, R. Matyas, and P. P. Nuspl, DigitalCommunications by Satellite. New York: Wiley, pp. 375382, 1981.

[15] A. Wannasarnmaytha, S. Hara, and N. Morinaga, A novel ST-DFTbased M-ary FSK demodulation method, to be published.

[16] L. R. Rabiner and B. Gold, Theory and Application of Digital Signal

Processing. Englewood Cliffs, NJ: Prentice-Hall, pp. 8893, 1975.[17] A. Wannasarnmaytha, S. Hara, and N. Morinaga, A new short-timeDFT FSK demodulation method for LEO satellite communicationssystems, to be published.

[18] M. Schwartz, W. R. Bennett, and S. Stein,Communication Systems andTechniques. New York: McGraw-Hill, pp. 295298, 1966.

Shinsuke Hara (S87M90) received the B.Eng.,M.Eng., and Ph.D. degrees in communication en-gineering from Osaka University, Osaka, Japan, in1985, 1987, and 1990, respectively.

From 1990 to 1996, he was an Assistant Professorin the Department of Communication Engineering,Osaka University. Since April 1996, he has beena Lecturer in the Department of Electronic, Infor-

mation, and Energy Engineering, Graduate Schoolof Engineering, Osaka University. From April 1996to March 1997, he was a Visiting Scientist in the

Telecommunications and Traffic Control Systems Group, Delft University ofTechnology, Delft, The Netherlands. His research interests include satellite,mobile and indoor wireless communications systems, and digital signalprocessing.

Dr. Hara is a Member of the IEICE of Japan.

Attapol Wannasarnmaytha (S94) received theB.Eng. degree in electrical engineering from Chu-lalongkorn University, Bangkok, Thailand, in 1992and the M.Eng. degree in communication engineer-ing from Osaka University, Osaka, Japan, in 1995.

He is currently working toward the Ph.D. degree atOsaka University.His research interests are digital signal processing

and mobile satellite communications.Mr. Wannasarnmaytha is a Student Member of

the IEICE of Japan.

Yuuji Tsuchida received the B.Eng. and M.Eng.degrees in communication engineering from OsakaUniversity, Osaka, Japan, in 1992 and 1994, respec-tively.

Since 1994, he has been with the Audio Labora-tory, Sony Corporation, Tokyo, Japan, working onthe research and development of an advanced digitalaudio system.

Mr. Tsuchida is a Member of the IEICE of Japan.

Norihiko Morinaga(S64M68SM92) receivedthe B.Eng. degree in electrical engineering fromShizuoka University, Shizuoka, Japan, in 1963 andthe M.Eng. and Ph.D. degrees in communicationengineering from Osaka University, Osaka, Japan,in 1965 and 1968, respectively.

He is currently a Professor in the Departmentof Communication Engineering, Graduate School ofEngineering, Osaka University, working in the areasof radio, mobile, satellite and optical communicationsystems and EMC.

Dr. Morinaga is a Member of the IEICE and ITE of Japan.

Documents

FSK Demodulation Method Using Short-time DFT Analysis for LEO