Difference Threshold Test for M -ary Signaling with Coherent Detection Young Gil Kim Sang Wu Kim University of Seoul Korea Advanced Institute of Science and Technology Seoul 130-743, Korea Taejon 305-701, Korea ygkim@ieee.org swkim@san.kaist.ac.kr We present the difference threshold test (DTT) as a method for generating a reliability information for M -ary signaling with coherent detection. The proposed DTT declares an erasure whenever the difference between the shortest Euclidean distance and the second shortest Euclidean distance from the received signal is less than a threshold. We show that the DTT is an approximation of the Bayesian test, and for binary signaling, it is equivalent to the Bayesian test. We also extend the DTT for L-fold diversity systems. The performance of ReedSolomon (RS) codes with errors-and-erasures decoding is analyzed when the DTT and the conventional SNR threshold test are applied. II. S YSTEM M ODEL We consider M -ary signals in slow frequency-nonselective fading channels with additive white Gaussian noise (AWGN). The low-pass equivalent received signal is I. I NTRODUCTION where r = (r1 , r2 , ..., rN ), x ∈ {s1 , s2 , ..., sN }, n = T (n1 , n2 , ..., nN ), and rk = 0 r(t)fk ∗ (t)dt = a · xk + nk , N M where {fk (t)}k=1 is an orthonormal basis for {sm (t)}m=1 . ∗ We assume that E[nk ] = 0 for all k, and E[nk nj ] = N0 δkj , where δkj is the Kronecker delta function. We assume perfect estimation of fading at the receiver. The use of side information permits identification and erasure of symbols that have been impaired by channel effects such as fading, jamming, background noise, etc. Since more erasures can be corrected than errors, it is advantageous to determine the reliability of the received symbols and erase unreliable symbols prior to the decoding process. There are a number of methods for generating side information, and their performances have been analyzed in [1]-[6]. In [6], difference threshold test (DTT) was proposed for binary frequency shift keying (BFSK) with noncoherent detection. In [7], DTT for M -ary orthogonal signaling with noncoherent detection was proposed and compared with the Viterbi’s ratio threshold test (RTT) [8]. In this paper, we present the DTT for M -ary signaling with coherent detection. The proposed DTT declares an erasure if the difference between the shortest Euclidean distance and the second shortest Euclidean distance is less than a threshold. We show that the DTT is an approximation of the Bayesian test, and for binary signaling, it is equivalent to the Bayesian test. Also, we find that the erasure zone, where erasure is declared, of the Bayesian test is very similar to that of the proposed DTT. We also extend the DTT for L-fold diversity systems. Finally, we apply the proposed DTT to Reed-Solomon (RS) codes with binary phase shift keying (BPSK) signaling, and compare with the conventional SNR threshold test. This paper consists of six sections. In Section II, we describe the system model. In Section III, we propose the DTT for M -ary signaling with coherent detection. In Section IV, the DTT for L-fold diversity systems is presented. In Section V, we analyze a RS-coded BPSK signaling with errors-anderasures decoding using the proposed DTT in slow frequencynonselective Rayleigh fading channels. Finally, conclusions are given in Section VI. GLOBECOM 2003 r(t) = a · x(t) + n(t), 0≤t≤T (1) where a is a complex Gaussian fading, x(t) is the transmitted signal, n(t) is a zero-mean complex AWGN, and T is the symbol duration. We assume that each of M signals, s1 (t), s2 (t), ..., sM (t), is transmitted with equal probability. The received signal can be expressed by a vector with N components: r=a·x+n , (2) III. DTT FOR COHERENT DETECTION Derivation of DTT In this section, we derive the DTT from the Bayesian test. The Bayesian test declares an erasure if the a posteriori probability of correct decision is less than a threshold [1]: Pr(x = x̂|r, a) < φ (3) where x is the transmitted signal, φ is a given threshold, and x̂ is the maximum a posteriori (MAP) detection among {s1 , s2 , ..., sM }: x̂ = arg max sm ∈{s1 ,s2 ,...,sM } Pr(x = sm |r, a) . (4) We define LLR Λm for m = 1, 2, ..., M as Λm = ln Pr(x = x̂|r, a) . Pr(x = sm |r, a) (5) Then, - 1999 - e−Λm Pr(x = x̂|r, a) M e−Λm Pr(x = x̂|r, a) = = m=1 Pr(x = x̂|r, a) Pr(x = sm |r, a) M (6) Pr(x = sm |r, a) (7) m=1 M e−Λm = 1. (8) m=1 0-7803-7974-8/03/$17.00 © 2003 IEEE Thus, the Bayesian test declares an erasure if 1 ≤φ . Pr(x = x̂|r, a) = M −Λm m=1 e Erasure Zone (9) Now, we derive LLR Λm . Assuming that each symbol is transmitted with an equal probability, the LLR Λm in (5) can be expressed as Λm = ln Pr(r|x = x̂, a) . Pr(r|x = sm , a) (10) Since nk ’s are zero-mean Gaussian and E[nk nj ∗ ] = N0 δkj , nk ’s are independent and so are rk ’s. Therefore, Pr(r|x = sm , a) = N Pr(rk |smk , a) , (11) k=1 where 2 mk | 1 − |rk −as N0 . e πN0 Substituting (11) in (10), yields Pr(rk |smk , a) = Λm = (12) N |rk − asmk |2 − |rk − ax̂k |2 /N0 (13) k=1 (14) ||r − asm ||2 − ||r − ax̂||2 /N0 , N where, for x = (x1 , x2 , ..., xN ), ||x||2 = k=1 |xk |2 . Now, we approximate the a posteriori probability of correct decision by taking the largest two terms in the denominator of (9), yielding 1 Pr(x = x̂|r, a) = M (15) −Λm m=1 e 1 (16) ≈ 1 + e−Λm∗ where Erasure zone is the region where we declare an erasure based on (9) or (19). For the proposed DTT, we can simplify (19) as ˆ Re{r • (ax̂ − ax̂)}/N 0 <γ , N where x • y = i=1 xi yi∗ and γ is a threshold that depends ˆ Therefore, the boundary of erasure zone with on ax̂ and ax̂. ˆ the DTT is a plane that is orthogonal to the vector from ax̂ to ax̂. In Figure 1(a) and (b), erasure zones for BPSK and QPSK signals are plotted, respectively. Shaded area represents the erasure zone. Since the proposed DTT is equivalent to the Bayesian test for binary signals, the erasure zones for the proposed DTT and the Bayesian test are the same. In Figure 1(b), the erasure zone of the proposed DTT is darker than the Bayesian test. Note that the boundary of erasure zone for the proposed DTT is always a line for two-dimensional signals. The erasure zone of the Bayesian test looks similar to that of the proposed DTT, but includes the area near the origin. = Λ m∗ = = (22) IV. DTT IN L- FOLD D IVERSITY S YSTEM In this section, we consider extending the DTT with L-fold diversity. We consider a multichannel which consists of L independent frequency nonselective, slow fading channels. The received vector rl , l = 1, 2, ..., L, corresponding to the lth diversity branch is rl = al x + nl (23) where al and nl are the complex fading and the noise vector, respectively, at the lth diversity branch. We assume that {rl } are independent and so are {al } and {nl }. If we let min Λm (17) min ||r − asm ||2 − ||r − ax̂||2 /N0 (18) sm =x̂ x̂L = arg max Pr(x = sm |r1 , r2 , ..., rL , a1 , a2 , ..., aL ) sm (24) sm =x̂ is the difference between the shortest Euclidean distance and the second shortest Euclidean distance from r, normalized by N0 . Since the a posteriori probability of correct decision in (16) increases monotonically with increasing Λm∗ , we propose to declare an erasure if ˆ 2 − ||r − ax̂||2 /N0 < γ , (19) ||r − ax̂|| for some constant γ and call it the DTT, where ˆ = arg min ||r − asm ||2 . x̂ sm =x̂ (20) then Pr(x = x̂L |r1 , r2 , ..., rL , a1 , a2 , ..., aL ) = M m=1 sm Λm,L = ln Pr(r1 , r2 , ..., rL |x = x̂L , a1 , a2 , ..., aL ) . (26) Pr(r1 , r2 , ..., rL |x = sm , a1 , a2 , ..., aL ) Since r1 , r2 , ..., rL are independent, Pr(r1 , r2 , ..., rL |x = x̂L , a1 , a2 , ..., aL ) (21) Thus, the reliability of the ML decision (or MAP decision for equally probable source) can be measured approximately by the difference between the shortest Euclidean distance and the second shortest Euclidean distance from the received vector r. For binary signals such as BPSK, the proposed DTT is equivalent to the Bayesian test because M = 2 for binary signals. GLOBECOM 2003 e−Λm,L (25) where Note that for equally probable source, x̂ = arg min ||r − asm ||2 . 1 L Pr(rl |x = x̂L , al ) (27) ||rl − al sm ||2 − ||rl − al x̂L ||2 /N0 . (28) = l=1 and thus - 2000 - Λm,L = L l=1 0-7803-7974-8/03/$17.00 © 2003 IEEE Note that for equally probable source, L 2 x̂L = arg min ||rl − al sm || . sm Probability of Incorrect Decoding (29) l=1 Following the procedure toward (19) yields that the DTT declares an erasure if L ˆ L ||2 − ||rl − al x̂L ||2 /N0 < γL ||rl − al x̂ (30) l=1 for some constant γL , where L 2 ˆ x̂L = arg min . ||rl − al sm || sm =x̂L (31) pe t (1 − pe ) l=1 r =a·x+n 2 + +1 N Per (i), where Per (i) is the probability of i code symbol erasures taking place in a codeword, i.e. Per (i) = (33) = k · log2 Q · Eb . (34) Thus, n . (35) k We assume that E{a2 } = 1 for convenience of analysis. Thus, the average received energy per information bit Ēb is the same as the transmitted energy per information bit Eb : Eb = Es,b (36) n i p (1 − per )n−i , i er per = 1 − (1 − per,b )log2 Q , (41) where per,b is the BPSK symbol erasure probability. In (39), pe denotes the probability of code symbol error when the code symbol is not erased, and is given by pe = 1 − (1 − pe,b )log2 Q , (42) where pe,b is the BPSK symbol error probability when not erased. In the next subsections, we derive per,b and pe,b for SNR threshold test, and the proposed DTT. SNR Threshold Test: For SNR threshold test, we erase a E BPSK symbol if the received SNR a2 Ns,b is less than a thresh0 old, i.e. (43) if a2 < γ0 , for some constant γ0 . Since a2 is exponentially distributed, the BPSK symbol erasure probability per,b is given by per,b = 0 = γ0 e−x dx 1 − e−γ0 . (44) (45) The BPSK symbol error probability when not erased is given by DTT for BPSK 2Es,b From (19), the proposed DTT declares an erasure if γ0 −γ0 e Q pe,b = e γ0 N0 2 2 (r + a Es,b ) − (r − a Es,b ) /N0 = 4 Es,b ·|ar|/N0 < γ . Es,b Es,b + N0 (37) − Q 2 , γ0 Es,b + N0 N0 Thus, the DTT, which is equivalent to the Bayesian test for binary signaling, declares an erasure if |ar| is less than a given threshold. A detailed derivation of (46) is given in Appendix A. GLOBECOM 2003 (40) where per is the code symbol erasure probability given by (32) = log2 Q · Es,b Ēb = E{a2 } · Eb = Eb . n−i−t i=n−k+1 where a is the fading amplitude, x is Es,b or − Es,b with equal probability 1/2, and n is the AWGN with variance N0 /2. Es,b is the energy per BPSK subsymbol, Eb is the energy per information bit, and Es,c is the energy per RS code symbol. They are related by nEs,c Thus, the probability of incorrect decoding PE is given by [9] n−k n−i n−i PE = Per (i) · · (39) t n−k−i i=0 t= V. E RRORS - AND -E RASURES D ECODING FOR RS- CODED BPSK S IGNALS System Model We apply the proposed DTT to a RS-coded BPSK signals in slow frequency-nonselective Rayleigh fading channels. We use (n,k) Q-ary RS code, where n is the number of code symbols in a codeword and k is the number of information symbols in a codeword. Each Q-ary RS code symbol is composed of log2 Q BPSK symbols. We assume that each of log2 Q BPSK symbols in a code symbol experiences an independent fading. At receiver, we declare a code symbol erasure if the code symbol contains at least one BPSK symbol erasure. The low-pass equivalent received signal for BPSK signaling after phase compensation is Es,c We derive the probability of incorrect decoding with errorsand-erasures decoding for the SNR threshold test [9] and the proposed DTT. With (n, k) RS code, we can correct t errors and e erasures if t + 2e ≤ n − k. (38) - 2001 - (46) 0-7803-7974-8/03/$17.00 © 2003 IEEE DTT: For the proposed DTT, we erase the BPSK symbol if |ar| < γ0 (47) Then, the BPSK symbol erasure probability per,b is given by 1 1 1 per,b = 1 − e−αγ0 + 1 − e−βγ0 , β Es,b + N0 α (48) where 2 Es,b + N0 + Es,b α= (49) N0 and 2 β= Es,b + N0 − Es,b . (50) N0 The BPSK symbol error probability when not erased, pe,b , is given by pe,b = Pr(ar < 0|ar < −γ0 or ar > −γ0 ) (51) Pr(ar < −γ0 ) = , (52) 1 − per,b where N0 Pr(ar < −γ0 ) = 2 Es,b + N0 √ √ 2( Es,b +N0 + Es,b )γ0 1 − N 0 · e (53) . Es,b + N0 + Es,b See Appendix A for the derivation of (48) and (53). Numerical Results and Discussions Figure 2 is a plot of the probability of incorrect decoding PE versus threshold γ0 . We find that the probability of incorrect decoding PE of each system is minimized at different thresholds, and the proposed DTT provides a lower PE . Figure 3 is a plot of the optimum code symbol erasure probability p∗er versus Ēb /N0 that minimizes PE . We find that the proposed DTT erases more code symbols than the SNR threshold test. Figure 4 is a plot of the probability of incorrect decoding PE versus Ēb /N0 when the threshold γ0 is optimally chosen to minimize PE . We also plotted PE of the errors-only decoding, which is a special case of γ0 = 0. We find that the proposed DTT provides a power gain of 1 dB over the SNR threshold test for PE of 10−5 and the power gain increases as the required probability of incorrect decoding decreases. VI. C ONCLUSION We presented the DTT as a method for generating a reliability information for M -ary signaling with coherent detection. The proposed DTT declares an erasure if the difference between the shortest Euclidean distance and the second shortest Euclidean distance from the received signal is less than a threshold. It is shown that the DTT is an approximation of the Bayesian test, and for binary signaling, the DTT is equivalent to the Bayesian test. The proposed DTT is applied to a RS-coded BPSK signaling with errors-and-erasures decoding. The proposed DTT provides a power gain of 1 dB over the SNR threshold test for the probability of incorrect decoding of 10−5 . We also derived the DTT for L-fold diversity. GLOBECOM 2003 R EFERENCES [1] C. W. Baum and M. B. Pursley, “Bayesian methods for erasure insertion in frequency-hop communication systems with partialband interference,” IEEE Trans. Commun., vol. 40, No. 7, pp. 1231-1238, Jul. 1992. [2] C. M. Keller and M. B. Pursley, “Diversity combining for channels with fading and partial-band interference,” IEEE J. Select. Areas Commun., vol. SAC-5, No. 2, pp. 248-260, Feb. 1987. [3] L. Yang and L. Hanzo, “Performance analysis of coded M ary orthogonal signaling errors-and-erasures decoding over frequency-selective fading channels,” IEEE J. Select. Areas Commun., vol. 19, No. 2, pp. 211-221, Feb. 2001. [4] L. Yang, K. Yen and L. Hanzo, “A Reed-Solomon coded DSCDMA system using noncoherent M -ary orthogonal modulation over multipath fading channels,” IEEE J. Select. Areas Commun., vol. 18, No. 11, pp. 2240-2251, Nov. 2000. [5] L. Yang and L. Hanzo, “Low-complexity erasure insertion in frequency-hopping spread-spectrum communications subject to fading and partial-band interference,” IEEE Globecom, Nov. 2529, 2001, San Antonio, Texas, USA, pp. 796-800. [6] I. Su and J. Wu, “Difference threshold test for asynchronous BFSK frequency-hopped multiple access systems over Rician channels,” Electronic Letters, vol. 35, No. 18, pp. 1512-1513, Sept. 1999. [7] Y. G. Kim and S. W. Kim, “Difference threshold test for M-ary orthogonal FSK signaling in Rayleigh fading channels,” Proc. of IEEE ICC, Anchorage, USA, May 2003. [8] A. J. Viterbi, “A robust ratio-threshold technique to mitigate tone and partial band jamming in coded MFSK systems,” IEEE MILCOM, pp. 22.4.1-22.4.5, Oct. 1982. [9] J. Hagenauer and E. Lutz, “Forward error correction coding for fading compensation in mobile satelite channels,” IEEE J. Select. Areas Commun., vol. SAC-5, pp. 215-225, Feb. 1987. [10] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series and Products. Academic Press, corrected and enlarged edition, p. 307, 1980. A PPENDIX A In this Appendix, we derive (46), (48), and (53). Derivation of (46) The BPSK symbol error probability when not erased, pe,b in (46), is ∞ 2Es,b Q pe,b = x e−(x−γ0 ) dx (54) N0 γ0 ∞ ∞ t2 1 = √ · eγ0 e− 2 e−x dtdx (55) 2Es,b 2π γ0 x N 1 =√ 2π ∞ · eγ0 2Es,b N0 0 γ0 N0 2 2Es,b t γ0 t2 e− 2 e−x dxdt (56) ∞ N0 t2 1 − t2 = √ · eγ0 2E e− 2 e−γ0 − e 2Es,b dt s,b 2π N0 γ0 ∞ 2 1 − t2 γ0 −γ0 = √ ·e e e dt 2Es,b 2π N0 γ0 ∞ 2 N − t2 1+ E 0 s,b − 2E e (57) dt . s,b γ0 N0 (46) can be derived from (57) by using the definition of Qfunction. - 2002 - 0-7803-7974-8/03/$17.00 © 2003 IEEE X as 2 Symbol erasure probability using optimum threshold 0.2025 X as 1 O (a) X X as3 as2 0.135 0.09 0.06 0.04 10 DTT 12 - 14 16 18 20 Eb/N0 Fig. 3. Optimum symbol erasure probability vs. E¯b /N0 for (31,16) 32-ary RS code. Bayesian X as 1 O DTT SNR 1 Errors-only SNR DTT 0.1 as4 X 0.01 PE (b) Fig. 1. Erasure zones of the Bayesian test and the proposed DTT for (a) BPSK signals and (b) QPSK signals 0.001 0.0001 0.24 1e-05 10 11 12 14 - 15 16 17 18 19 Fig. 4. Probability of incorrect decoding PE vs. E¯b /N0 for (31,16) 32-ary RS code. SNR DTT PE 13 Eb/N0 0.12 0.06 where we use the equality [10] ∞ q 1 π √ exp(−px2 − 2 )dx = exp(−2 pq) . x 2 p 0 0.03 (59) The BPSK symbol erasure probability per,b in (48) is 0.015 Threshold γ0 Fig. 2. Probability of incorrect decoding PE vs. threshold γ0 for E¯b /N0 = 12 dB and (31,16) 32-ary RS code. Derivation of (48) and (53) 0 For the derivation of the BPSK symbol erasure probability (48) for the proposed DTT, the pdf of ar should be derived first. The pdf of ar, far (y), is ∞ 1 −(y−a2 √ Es,b )2 2 e 2ae−a da1 2πa2 N0 /2 0 √ √ 1 = e2y Es,b /N0 e−2|y| N0 +Es,b /N0 (, 58) N0 + Es,b far (y) = GLOBECOM 2003 per,b = Pr(−γ0 < ar < γ0 ) = √ √ 2y N0 +Es,b Es,b 2y 0 1 N N0 0 e dy e N0 + Es,b −γ0 γ0 2y√Es,b 2y√N0 +Es,b N0 e N0 e− + dy a2 N0 (60) (61) 1 1 , (1 − e−αγ0 ) + (1 − e−βγ0 ) β α Es,b ) and β = where α = N20 ( N0 + Es,b + 2 E ). Also, (53) can be derived from N + E − ( s,b 0 s,b N0 the following integration: = 1 N0 + Es,b (62) Pr(ar < −γ0 ) = −γ0 √ √ 1 e2y Es,b /N0 e2y N0 +Es,b /N0 dy. N0 + Es,b −∞ - 2003 - 0-7803-7974-8/03/$17.00 © 2003 IEEE