0 valutazioniIl 0% ha trovato utile questo documento (0 voti)

22 visualizzazioni4 pagineSep 09, 2011

© Attribution Non-Commercial (BY-NC)

PDF, TXT o leggi online da Scribd

Attribution Non-Commercial (BY-NC)

0 valutazioniIl 0% ha trovato utile questo documento (0 voti)

22 visualizzazioni4 pagineAttribution Non-Commercial (BY-NC)

Sei sulla pagina 1di 4

Sills Signal Exploitation and Geolocation Division Southwest Research Institute San Antonio, Texas 78238-5166

ABSTRACT This paper addresses automatic modulation classication for PSK and QAM signals under coherent and noncoherent conditions. In particular, the paper extends previous results by treating the classication of higher-state QAM signals. A maximum-likelihood algorithm is presented for coherent classication of PSK and QAM signals. We evaluate the algorithms performance for various PSK and QAM modulation types including 64-state QAM and then compare it with a psuedo maximum-likelihood noncoherent classication technique in terms of error rate, false alarm rate, and computational complexity. The application of these results to the design and performance of an automatic signal recognizer is discussed throughout the paper. 1. INTRODUCTION Automatic modulation recognition is a rapidly evolving area of signal exploitation with applications in DF conrmation, monitoring, spectrum management, interference identication, and electronic surveillance. Generally stated, a signal recognizer is used to identify the modulation type (along with various parameters such as baud rate) of a detected signal for the purpose of signal exploitation. For example, a signal recognizer could be used to extract signal information useful for choosing a suitable counter measure, such as jamming. In recent years interest in modulation recognition algorithms has increased with the emergence of new communication technologies. In particular, there is growing interest in algorithms that treat quadrature amplitude modulated (QAM) signals, which are used in the HF, VHF, and UHF bands for a wide variety of applications including FAX, modem, and digital cellular. Many techniques for modulation identication have been published in the literature. Early work in modulation identication is found in a report by Weaver, Cole, and Krumland [1] in which frequency-domain parameters were used to distinguish between six candidate modulation types. One of the well-known early papers treating digital modulation types was by Liedtke [2] in which he presents results based on a statistical analysis of various signal parameters to discriminate between amplitude shift keying (ASK), FSK, and PSK. Other techniques using signal parameters have been reported in [3], [4], [5], [6], and [7]. A combination of techniques including pattern recognition are used in [8] and [9]. Several authors have applied techniques from higher-order statistics that exploit cyclostationarity to identify modulation [10]. Still others have applied neural networks to the problem [11, 12]. A recent

This work was supported by the Advisory Committee for Research and Development at Southwest Research Institute.

book by Azzouz and Nandi [13] gives more details on these and other recent techniques for modulation identication. Another group of authors have applied techniques from maximum-likelihood (ML) decision theory to modulation identication. Kim et al. use a truncated series approximation of the likelihood ratio function for distinguishing a BPSK from an MPSK ), but these results apply in low SNR only [14]. Extensions (M to high SNR and 16-state QAM are presented in [15]. Sapiano presents a PSK classication technique with improved sensitivity to parametric degradation [16]. Most recently, Boiteau presented a comprehensive review of the literature on signal classication and provided a generalized framework that does not require any restriction on the baseband pulse [17]. In this paper we extend previous results on maximum-likelihood classication for PSK/QAM by developing general solutions for coherent and noncoherent classication of PSK/QAM signals with an arbitrary number of signal states. Performance curves are presented for both the coherent and noncoherent cases for various modulation types including 64-state QAM. The paper begins with an introduction in Section 1. Section 2 presents the signal model for PSK/QAM communications. The coherent classier is presented in Section 3 along with performance curves showing error rates and false alarm rates. The noncoherent case is treated in Section 4. Performance curves are presented and then noncoherent performance is compared with coherent performance. Section 5 contains conclusions and recommendations.

4

st nt, t T where s t We receive a signal r t is a signal emitted from a non-cooperative transmitter and n t is additive white gaussian noise (AWGN) with a two-sided power spectral density (PSD) of N0 . The signal s t is represented by 2

is the carrier frequency, and c is an unknown phase offset. The received signal is one of N candidate modulation types. Let the integers i ; ;:::;N enumerate the candidate ; ;:::;N denotes modulation types, such that mi for i the event that the intercepted signal belongs to the ith modulation type. We will assume equal a priori probabilities P mi . A t ej t denote the complex envelope of s t . Let s t The complex envelope of a PSK/QAM signal can be expressed in terms of

= + 0

st = Atcos!ct + c + t = Re Atej t ej!c t+c where At and t are the modulated amplitude and phase, !c =01

,1

=01

,1

~ =

st = ~

Ea Ep

2Es

X

n

where an is a sequence of symbols taken from a set of M i com1 ; 2 ; : : : ; M i , Rs T1s is the plex numbers I mi symbol rate, and td is an unknown timing offset. The pulse shape p t is any of the standard symmetric types such as a square-root raised cosine or a square pulse. The symbol energy is Es provided that (1) we dene Z

=f

~ = rI + j rQ = 4 r

The PDF governing is given by

r1 3 ~ ~ 7 6 r2 7 6

2

Ep =

and (2) we model an as independent random variables with equally likely assignment from the set I mi such that

,1

p2 tdt

~ r

rN ~

. . .

where E denotes the expected value operator. It is customary to normalize the set n such that Ea . The input signal-toEs noise ratio (SNR) is dened by N0 , Each modulation type is characterized by its symbol conguration in the complex plane, which denes the amplitude and phase values for the set I mi .

M i X

p~ rI ; rQ jm = mi = r

N Y n=1

pr rI;n ; rQ;n jm = mi ~

The coherent maximum likelihood (ML) classier is simply a rule for choosing among the candidate modulation types given . Choose m mk if and only if

~ r

f g

=1

3. COHERENT ML CLASSIFICATION Automatic signal classication is a rather difcult problem in composite hypothesis testing since so many parameters are unknown: symbol rate Rs ; carrier frequency !c ; carrier phase c ; pulse shape p t ; SNR ; and timing offset td . A common approach is to rst estimate the unknown parameters and then attempt to classify the signal according to modulation type. Although estimating these parameters is nontrivial, it is not impractical. There are a wide variety of techniques for estimating the signal parameters some of which are given in [18]. In this section we evaluate the performance of coherent ML classication in which all of the signal parameters are known. In this case, the signal is classied by forming likelihood ratios from the demodulated matched-lter output

is maximum for i k. We investigate the performance of the coherent classier by evaluating its error rate as a function of SNR for the following PSK/QAM modulation types: (m1 ) BPSK; (m2 ) QPSK; (m3 ) 8PSK; (m4 ) QAM-16; (m5 ) QAM-32; and (m6 ) QAM-64. For PSK, the symbol congurations are well known. For QAM however, there are many possiblities including rectangular and circular congurations. We consider the rectangular congurations dened by the V.32 and V.33 standards [19] and shown in Figure 1.

p~ rI ; rQ jm = mi r

(a) BPSK

(b) QPSK.

(c) PSK-8.

rn = ~

Z1 ,1

rte,j!c t+c

components nI;n and nQ;n are independent, zero mean, with variance N0 . 2 Given that the modulation type is mi , the probability density function (PDF) governing the demodulated symbols rn can be expressed in the form

= rI;n + jrQ;n p p It follows that rI;n = Es aI;n + nI;n and rQ;n = Es aQ;n + nQ;n where aI;n = Refan g and aQ;n = Imfan g. The noise ~

(d) QAM-16

(e) QAM-32.

(f) QAM-64.

Figure 1: PSK/QAM symbol congurations. Figure 2 shows the performance in terms of probability of error and false-alarm rate for the coherent classier resulting from 1000 Monte-Carlo simulations. The performance indicates that the coherent ML classier makes less than one error in ten across all six modulation types provided the SNR is greater than or equal to 10 dB. This performance represents the best possible error rate that can be achieved. This level of performance is unlikely in practice due primarily to phase incoherence between the transmitter and receiver; that is, the parameter c is rarely known when the SNR is 10 dB. In fact coherent carrier acquisition for high-state QAM

pr rI;n ; rQ;n jm = mi = ~

N0 . It is where I;k Re k and Q;k Im k and 2 worth noting that rI;n and rQ;n are not necessarily independent for example consider 8-PSK and QAM-32 from the V.32 standard [19].

= f g

f g

requires SNR levels much larger than 10 dB. Nevertheless, Figure 2 provides a benchmark for classication performance from which to compare noncoherent techniques.

n = tan,1 rQ;n : rI;n Given the transmitted symbol an , the amplitude jrn j is a Ricean~

where distributed random variable with PDF 2 ,rn +Es jan j2 rn I0 pjrn j rn an rn e 2 2 ~ 2

0.9

0.8

j = 0

pE j a j
s n

2

0.6

0.5

where I0 is the th -order modied Bessel function and 2 N0 . 2 The exact expression for the PDF of the phase difference n is very complicated, but for sufciently large SNR, it can be approximated by a Gaussian:

0.4

0.3

0.2

0.1

10

12 Es/No (dB)

14

16

18

20

0.2

, tan,1 aQ;n,11 . The approximation given by (1) follows from the aI;n approximation tan,1 for small . ~ For sufciently large values of SNR, jrn j and n are very nearly independent, hence we can approximate their joint PDF by pjrn j; r; j; an ; an,1 pjrn j rjan p jan ; an,1 [20]. ~ ~ We next apply the law of total probability to nd the conditional density function pjrn j;n rn ; n jm = mi . There are ef~

cient ways to perform this calculation, but as a general expression

where

N 2 ,1 aQ;n n = 20 ja1 j + jan1 1 j and = tan aI;n , n ,

0.18

0.16

j j

j

0.12

0.1

0.08

Given N demodulated symbols, the PDF for the sional decision vectors

N , 1-dimen3 7 7 5

0.06

0.04

R=4

4 6 8 10 12 Es/No (dB) 14 16 18 20

. . .

2 6

P=6 4

1 2

. . .

0.02

N ,1

is

N ,1 Y n=1

= 256 symbols.

4. NONCOHERENT PSUEDO-ML CLASSIFICATION In this section we evaluate the performance of noncoherent ML classication in which all of the signal parameters are known except the carrier phase c . In this case the demodulated symbol is rotated by an unknown carrier phase

rn = Es an ejc + nn ~ ~

In this case, the signal is classied by nding the amplitude and the phase difference

jr n j ~

We select the modulation type that corresponds to the largest of the pR;P ; m mi . The performance of the psuedo-maximimum-likelihood modulation classier is illustrated in Figure 3. This gure shows the performance in terms of probability of error and false-alarm rate for the noncoherent classier resulting from 1000 Monte-Carlo simulations. These results indicate that the noncoherent psuedoML classier makes less than one error in ten across the tested modulation types provided the SNR is greater than or equal to 13 dB. Comparing Figures 2 and 3, it is evident that the noncoherent classier exhibits a performance loss of approximately 3 dB. A critical parameter in the design of a PSK/QAM recognizer is the number of symbols that are used to decide between modulation types. Using a large number of symbols in the likelihood-ratio test reduces the probability of error and probability of false alarm; hence, the one-error-in-ten performance can be sustained down to lower SNR. When fewer symbols are used, we require higher SNR for this performance level.

R Pj = g

0.9

0.8

[2] F.F. Liedtke, Computer Simulation of an Automatic Classication Procedure for Digitally Modulated Communication Signals with Unknown Parameters, Signal Processing, vol. 6, pp. 311-323, 1984 [3] P.H. Halpern and P.E. Mallory, A Simple Method for Distinguishing Modulation Types, IEEE Trans. ASSP, vol. 30, pp. 97-99, Feb. 1982. [4] N.F. Krasner, Optimum Detection of Digitally Modulated Signals, IEEE Transactions on Communications, vol. Com30, pp. 885-895, May 1982. [5] J.E. Hipp, Modulation Classication Based on Statistical Moments, Milcom-86, vol. 2, pp. 20.2.1-6, 1986.

0.6

0.5

0.4

0.3

0.2

0.1

10

12 Es/No (dB)

14

16

18

20

[6] J. Aisbett, Automatic Modulation Recognition Using TimeDomain Parameters, Signal Processing 13, pp. 323-328, 1987 [7] Y.T. Chan and L.G. Gadbois, Identication of the Modulation Type of a Signal, Signal Processing, vol. 16, pp. 149-154, 1989.

0.2

0.18

0.16

[8] L.V. Dominguez, J.M.P. Borrallo, J.P. Garcia, and B.R. Mezcua, A General Approach to the Automatic Classication of Radiocommunication Signals, Signal Processing, vol. 22, pp. 239-250, 1991. [9] F. Jondral, Automatic Classication of High Frequency Signals, Signal Processing, vol. 9, pp. 177-190, 1985. [10] W.A. Gardner and C.M. Spooner, Signal Interception: Performance Advantages of Cyclic-Feature Detectors, IEEE Transactions on Communications, vol 40, No. 1, January 1992. [11] A.K. Nandi and E.E. Azzouz, Modulation Recognition Using Articial Neural Networks, Signal Processing, vol. 56, pp. 165-175, 1997.

0.12

0.1

0.08

0.06

0.04

0.02

10

12 Es/No (dB)

14

16

18

20

[12] A. Bernardini and S.D. Fina, Optimal Decision Boundaries for M-QAM Signal Formats Using Neural Classiers, IEEE Trans. on Neural Networks, vol. 9, no. 2, March 1998. [13] E. Azzouz and A.K. Nandi, Automatic Modulation Recognition of Communication Signals, Kluwer, 1997.

[14] K. Kim and A. Polydoros, Digital Modulation Classication: the BPSK v.s. QPSK case, Milcom 1988. [15] C.S. Long, K.M. Chugg, and A. Polydoros, Further Results in Likelihood Classication of QAM Signals, Proceedings of MILCOM-94, pp. 57-61, Fort Momounth, NJ, October 2-5, 1994. [16] P.C. Sapiano and J.D. Martin, Maximum Likelihood PSK Classier, Milcom-96, pp. 1010-1014, 1996. [17] D. Boiteau and C. Le Martret, A Generalized Maximum Likelihood Framework for Modulation Classication, International Conference on Acoustics, Speech, Signal Processing, 1998. [18] J.A. Sills, A QAM Demodulator for Digital Wideband Architectures, Technical Report, Southwest Research Institute, 1998. [19] CCITT Blue Book, Vol. 8, 1988. [20] J.K. Patel and C.B. Read, Handbook of the Normal Distribution, 2nd Ed., Marcel Dekker Inc., New York, 1996.

5. CONCLUSIONS In this paper we investigated automatic modulation classication for PSK and QAM signals. We presented a maximum likelihood framework for both the coherent and noncoherent cases. This general approach treated PSK and QAM up to an arbitrary number of signal states. The performance of both the coherent and noncoherent classier was investigated for various modulation types including 64 state QAM. Noncoherent performance exhibited a 3 dB loss compared to coherent performance. 6. REFERENCES [1] C.S. Weaver, C.A. Cole, R.B. Krumland, and M.L. Miller, The Automatic Classication of Modulation Types by Pattern Recognition, Standford Electronics Laboratories, Technical Report No. 1829-2, April 1969.

## Molto più che documenti.

Scopri tutto ciò che Scribd ha da offrire, inclusi libri e audiolibri dei maggiori editori.

Annulla in qualsiasi momento.