Optimum Receive Antenna Selection Minimizing Error Probability Sang Wu Kim∗ Eun Yong Kim† Abstract— The optimum receive antenna selection combining rule that minimizes the bit error probability is presented. It is derived from a general relationship between the bit error probability and the log-likelihood ratio (LLR), and selects the receive antenna providing the largest LLR magnitude. The optimum selection combining rule is applied to single transmit (Tx) antenna and space-time block code (STBC) systems, with NR receive (Rx) antennas, and the bit error probability is derived for BPSK signaling in Rayleigh flat fading channels. A suboptimum selection combining rule based on noncoherent envelope detection is also presented and analyzed. For STBC systems, the optimum generalized selection combining in which M (≤ NR ) Rx antennas providing the largest LLR magnitude are selected and combined is presented. Index Terms— Selection combining, log-likelihood ratio, diversity, space-time block code, multiple antenna, Rayleigh fading channel. I. I NTRODUCTION The radio link is characterized by multi-path fading which causes the link quality to vary with time, frequency, and space. Diversity techniques are known to be very effective in mitigating multi-path fading, and multiple-transmit multiplereceive antennas are known to provide an increased diversity order [1],[2]. However, the implementation of such multipletransmit multiple-receive antenna systems requires multiple RF chains and A/D converters, where much of the hardware and power consumption lies. A promising approach for reducing hardware complexity and power consumption, while retaining a reasonably high performance, is to employ some form of selection combining [3],[4],[5]. In this paper, we present the optimum receive antenna selection combining rule that minimizes the bit error probability in interference-limited systems. The optimum selection combining rule, derived from a general relationship between the bit error probability and the log-likelihood ratio (LLR), is to select the antenna providing the largest LLR magnitude. We apply this rule to a single transmit (Tx) antenna system and space-time block code (STBC) system [6],[7], with NR receive (Rx) antennas, and derive the bit error probability for BPSK signaling in Rayleigh flat fading channels. Also, we present a suboptimum selection combining rule based on a noncoherent envelope detection. We compare the bit error probabilities of optimum and suboptimum selection combining rules with those of traditional combining rules such as signal-to-interference ratio (SIR)-based selection combining and maximum ratio combining (MRC) [3],[8]. We find that ∗ Sang † Eun Wu Kim sangkim@ieee.org Yong Kim eykim@bada.kaist.ac.kr 0-7803-7700-1/03/$17.00 (C) 2003 IEEE the optimum selection combining provides a gain of 3-3.5dB over the SIR-based selection combining, and is only 0.30.6dB away from MRC that requires multiple RF chains. The envelope-based suboptimum selection combining provides a gain of 1.8-2.1dB over SIR-based selection combining, when selecting one out of four Rx antennas. For STBC systems, we present the optimum and suboptimum generalized selection combining rules in which M (≤ NR ) Rx antennas providing the largest LLR magnitude are selected and combined. The remainder of this paper is organized as follows. In Section II, we present the system model. In Section III, we present a general relationship between the bit error probability and LLR, and present the optimum selection combining rule that minimizes the bit error probability. In Section IV, we consider a wireless channel with one Tx antenna and NR Rx antennas, and derive the bit error probability with optimum and suboptimum antenna selection combining rules. In Section V, we consider a STBC system with two Tx antennas and NR Rx antennas and derive the bit error probability with optimum and suboptimum antenna selection rules. In Section VI, we consider a STBC system with two Tx antennas and NR Rx antennas and present optimum and suboptimum generalized selection combining rules. In Section VII, numerical results are provided along with discussions. In Section VIII, conclusion is made. II. S YSTEM M ODEL In this paper, we consider the reverse link (mobile to base) of a digital mobile radio communication system. The system is assumed to be interference-limited. The base station consists of an NR element antenna array, and the antenna element is selected on a symbol-by-symbol basis. The modulation is chosen to be binary phase-shift-keying (BPSK) with coherent detection in two channel models. In Section IV, we consider a wireless communication system with one Tx antenna and NR Rx antennas (Fig.1). The received low-pass equivalent signal yi , given the ith Rx antenna selected, is y i = hi x + K hk,i xk (1) k=1 where hi and hk,i are channel gains between the Tx antenna and the ith Rx antenna for the desired and the kth interfering users, respectively, and K is the number of interfering users. We assume that |hi | and |hk,i | are Rayleigh distributed with 2 2 E{|hi | } = E{|hi,k | } = 1 and that the phases of hi and hk,i are uniformly distributed over [0,2π]. x and xk , representing the desired and the kth interfering signals, respectively, are 441 √ √ + Es or − Es with probability 1/2, where Es is the transmit energy per symbol. We assume that the interference term in (1), denoted hereafter as ni , has a Gaussian distribution with mean zero and variance I0 /2 per dimension. The Gaussian assumption follows from the Cramer’s central limit theorem [10]. In Sections V and VI, we consider a space-time block code (STBC) system with two transmit antennas and NR receive antennas (Fig.2). The received low-pass equivalent signals r1i and r2i from the ith antenna at time t and t + T , respectively, are given by [6] r1i = h1i x1 + h2i x2 + n1i (2) r2i = (3) −h1i x∗2 + h2i x∗1 + n2i where hji is the channel gain between jth Tx antenna and ith Rx antenna, and n1i and n2i , i = 1, 2, ..., N are i.i.d. Gaussian random variables with mean I0 /2 zero and variance per dimension. x1 and x2 are + Es /2 or − Es /2 with probability 1/2. Then, the combiner outputs y1i and y2i are given by y1i y2i ∗ = h∗1i r1i + h2i r2i = (|h1i |2 + |h2i |2 )x1 + h∗1i n1i + h2i n∗2i ∗ = h∗2i r1i − h1i r2i = (|h1i |2 + |h2i |2 )x2 − h1i n∗2i + h∗2i n1i . (4) (5) (6) (7) We assume that the channel is quasi-static flat fading, i.e. the channel gains {hi } and {hji } remain constant over several symbols. We also assume that the channel gains are i.i.d., and are known at the receiver. selection combining rule that minimizes the bit error probability is to select the antenna providing the largest |Λi |. Then, the minimum average bit error probability is 1 Pe,opt = EΛmax (12) 1 + eΛmax where, by definition, Λmax = max |Λi |. 1≤i≤NR (13) It should be noted that averaging Λi in (10) over ni and taking the absolute value yields 4|hi |2 Es /I0 , which is the SIR at diversity branch i. Therefore, selecting the antenna providing the largest SIR or equivalently the largest |hi |2 (assuming that the average interference power is a constant across all antennas) takes the average interference power into account. However, the optimum selection combining rule exploits the interference term Re{h∗i ni } in (10) by selecting the antenna for which signs of x and Re{h∗i ni } are identical and |hi | is large, thereby providing the largest LLR magnitude. As a result, the performance is governed by the peak, as opposed to the mean, channel condition. The bit error probability resulting from the optimum selection combining provides a lower bound on the bit error probability of any selection combining rule. In contrast NRwith the maximum ratio combining (MRC) that Λi , the optimum selection combining selects the yields i=1 best component among NR components used in MRC. In what follows, we apply the optimum selection combining rule to two wireless communication systems. IV. S INGLE T X A NTENNA We consider a wireless communication system with one Tx antenna and NR Rx antennas (Fig.1). The channel output given the ith antenna is selected is given by III. O PTIMUM S ELECTION C OMBINING yi = hi x + ni . Consider a channel where the channel output yi , given the ith Rx antenna is selected, is given by Then, it follows from (10) that the LLR, given the ith Rx antenna selected, is √ 4 Es Λi = [|hi |2 x + Re{h∗i ni }]. (15) I0 If we let √ Es Zi = ||hi |2 x + Re{h∗i ni }| (16) I0 then the pdf of Zi can be shown to be (derived in Appendix II) √ √ 1 −1 −1 fZi (z) = e−2( 1+γ +1)z + e−2( 1+γ −1)z γ2 + γ (17) where yi = hi x + ni , (8) gain√between the Tx antenna and ith where hi is the channel √ Rx antenna, x is + Es or − Es with probability 1/2, and ni is a white Gaussian noise with mean zero and variance I0 /2 per dimension. Then, the log-likelihood ratio (LLR) Λi for x is given by √ P (x = + Es |hi , yi ) √ Λi = ln (9) P (x = − Es |hi , yi ) √ 4 Es = Re{h∗i yi } . (10) I0 γ =|hi |2 x+Re{h∗ n } i i The sign of Λi is the hard decision value, and the magnitude of Λi represents the reliability of hard decision. In general, the bit error probability Pe,i and the LLR Λi is related by Pe,i = 1 1 + e|Λi | (11) A derivation of (11) is provided in Appendix I. Since Pe,i decreases with increasing |Λi |, the optimum receive antenna = E[|hi |2 ]Es /I0 = Ēs /I0 (14) (18) (19) is the average received signal-to-interference ratio (SIR) per diversity branch. The cdf of Zi is then given by z fZi (u)du (20) FZi (z)= 0 √ √ −1 −1 1 − e−2( 1+γ +1)z 1 − e−2( 1+γ −1)z = + .(21) 2(γ + 1 + γ 2 + γ) 2(γ + 1 − γ 2 + γ) 442 If we let Zmax = max {Zi } 1≤i≤NR (22) and assume that Z1 , Z2 , ..., ZNR are i.i.d., then the pdf of Zmax is given by fZmax (z) = NR [FZi (z)]NR −1 fZi (z). (23) Therefore, it follows from (12) and (23) and the relationship Λmax = 4Zmax that the bit error probability with the optimum antenna selection is given by ∞ 1 Pe,opt = fZ (z)dz. (24) 1 + e4z max 0 Envelope-based Selection In this subsection we present a simple suboptimum selection combining rule based on noncoherent envelope detection. First, we note that |Re{h∗i yi }| ≤ |h∗i yi | = |hi ||yi |. (26) SIR-based Selection Since the SIR, given that the ith antenna is selected, is |hi |2 Es /I0 , the antenna providing the largest SIR is the one providing the largest |hi |2 . If we let 1≤i≤NR (29) then the bit error probability, when the antenna providing the largest SIR is selected, is given by ∞ Pe,snr = Q( 2gEs /I0 )fGmax (g)dg (30) 0 where fGmax (g) is the pdf of Gmax given by fGmax (g) = NR (1 − e−g )NR −1 e−g , and γ is the average received SIR per diversity branch. V. S PACE -T IME B LOCK C ODE In this section, we consider a space-time block code (STBC) with two Tx antennas and NR Rx antennas (Fig.2). The baseband combiner output y1i for data x1 , given the ith Rx antenna is selected, is (25) Since noncoherent envelope detection of the received RF signal yields |yi |, we propose a suboptimum selection combining rule that selects the antenna element providing the maximum of {|h1 ||y1 |, |h2 ||y2 |, ... , |hNR ||yNR |}. It will be called envelope-based selection combining. If we let {|h(1) ||y(1) |, |h(2) ||y(2) |, ... , |h(NR ) ||y(NR ) |} be an ordered set such that |h(1) ||y(1) | ≥ |h(2) ||y(2) | ≥ ... ≥ |h(NR ) ||y(NR ) |, then the log-likelihood ratio Λenv for deciding x with envelope-based selection combining is √ 4 Es Λenv = Re{h∗(1) y(1) }. (27) I0 √ We decide that E√ s was transmitted if Λenv > 0, and otherwise, decide − Es was transmitted. Then, it follows from (11) and (27) that the average bit error probability with the envelope-based selection combining is given by 1 Pe,env = EΛenv . (28) 1 + e|Λenv | Gmax = max {|hi |2 } Maximum Ratio Combining The bit error probability with maximum ratio combining of NR independent paths is given by [8] N R −1 NR − 1 + k Pe,mrc = [(1 − µ)/2]NR [(1 + µ)/2]k k k=0 (32) where, by definition, γ µ= (33) 1+γ (31) y1i = (|h1i |2 + |h2i |2 )x1 + η1i (34) η1i = h∗1i n1i + h2i n∗2i (35) where is a conditional Gaussian random variable with mean zero and variance (|h1i |2 + |h2i |2 )I0 /2 per dimension. Then, for a given {hji }, y1i is a Gaussian random variable with mean (|h1i |2 + |h2i |2 )x1 and variance (|h1i |2 + |h2i |2 )I0 /2 per dimension. Therefore, the log-likelihood ratio (LLR) Λi for data x1 1 , given {hji } and y1i , is P (x1 = + Es /2|{hji }, y1i ) Λi = ln (36) P (x1 = − Es /2|{hji }, y1i ) 4 Es /2 = Re{y1i } (37) I0 =Ai x1 +Re{η1i } where, by definition, Ai = 2 j=1 |hji |2 . (38) If we let √ 2Es Zi = |(|h1i |2 + |h2i |2 )x1 + Re{h∗1i n1i + h2i n∗2i }| (39) I0 then the pdf of Zi can be shown to be (derived in Appendix III) 1 fZi (z) = (1 + z 1 + 2/γ) 3 γ(γ + 2) √ √ −z( 1+2/γ+1) [e + e−z( 1+2/γ−1) ]. (40) If we assume that Z1 , Z2 , ..., ZNR are i.i.d., then the pdf of Zmax = max1≤i≤NR {Zi } is given by fZmax (z) = NR [FZi (z)]NR −1 fZi (z), (41) where FZi (z) is the cdf of Zi . Therefore, it follows from (12) and (41) and the relationship Λmax = 2Zmax that the bit error assuming that {|hi |2 } are i.i.d. with pdf f|hi |2 (g) = e−g . 1 The 443 LLR for other information symbols is exactly the same. probability with the optimum antenna selection rule is given by ∞ 1 Pe,opt = fZ (z)dz. (42) 1 + e2z max 0 Envelope-based Selection In this subsection we present a simple suboptimum selection combining rule based on noncoherent envelope detection. We note that 4 Es /2 |Λi | = |Re{y1i }| (43) I 0 4 Es /2 ≤ |y1i |. (44) I0 Since combining the received RF signals and detecting the envelope of the combined signal yields |y1i | and |y2i |, we propose a suboptimum selection combining rule that selects the antenna element providing the maximum of {|y1i |}, i = 1, 2, ..., NR . It will be called envelope-based selection combining. Then, the log-likelihood ratio Λenv for deciding x with envelope-based selection combining is 4 Es /2 Λenv = Re{y(1) } (45) I0 where y(1) is the baseband combiner output associated with the antenna element providing √ the maximum of {|y1i |}, i = 1, 2, ..., NR . We decide that Es was transmitted if Λenv > 0, √ and otherwise, decide − Es was transmitted. Then, it follows from (11) and (45) that the average bit error probability with envelope-based selection combining is 1 Pe,env = EΛenv . (46) 1 + e|Λenv | SIR-based Selection The SIR, given the ith Rx antenna selected, is Ai Es /(2I0 ). Therefore, the antenna providing the largest SIR is the one providing the largest Ai . If we let Amax = max {Ai }, 1≤i≤NR (47) then the bit error probability with SIR-based selection is given by ∞ Pe,snr = Q( 2aEs /(2I0 ))fAmax (a)da, (48) 0 where fAmax (a) is the pdf of Amax given by fAmax (a) = NR [1 − e−a 1 l=0 a1−l ]NR −1 · ae−a (49) assuming that {Ai }’s are i.i.d. with fAi (a) = ae−a , a ≥ 0. Maximum Ratio Combining STBC with two Tx antennas and NR Rx antennas provides a diversity order of 2NR . Therefore, the bit error probability with MRC at the receiver is given by (32) with NR replaced by 2NR and γ replaced by γ/2 because Es /2 is the energy per transmit antenna. VI. O PTIMUM G ENERALIZED S ELECTION C OMBINING In this section we consider the optimum generalized selection combining (GSC) for STBC systems. The optimum GSC for single Tx antenna case is analyzed in [11]. We consider selecting M out of NR Rx antennas and combining those M signals. The baseband combiner output y1i1 ,i2 ,···,iM for data x1 , given Rx antennas i1 , i2 , · · · , iM are selected, is y1i1 ,i2 ,···,iM = = M m=1 M m=1 where y1im (|h1im |2 + |h2im |2 )x1 + η1im (50) η1im = h∗1im n1im + h2im n∗2im . (51) Then, for a given {hji }, y1i1 ,i2 ,···,iM is a Gaussian random M variable with mean m=1 (|h1im |2 + |h2im |2 )x1 and variance M 2 2 m=1 (|h1im | + |h2im | )I0 /2 per dimension. Therefore, the LLR Λi1 ,i2 ,···,iM for x1 , given {hji }, y1i1 ,i2 ,···,iM , is P (x1 = + Es /2|{hji }, y1i1 ,i2 ,···,iM ) Λi1 ,i2 ,···,iM = ln (52) P (x1 = − Es /2|{hji }, y1i1 ,i2 ,···,iM ) √ 4 Es = Re{y1i1 ,i2 ,···,iM } (53) I √0 4 Es = [A1i1 ,i2 ,···,iM x1 I0 + Re{η1i1 + η1i2 + . . . + η1iM }] (54) where, by definition, A1i1 ,i2 ,···,iM = M [|h1im |2 + |h2im |2 ]. (55) m=1 It follows from (11) that the optimum selection combining rule that minimizes the bit error probability is to select those antennas that provide the largest |Λi1 ,i2 ,···,iM |. It can be shown that the bit error probability with the optimum GSC of M = NR 2 antennas is identical to that of MRC, since the sign of the NR 2 optimum GSC output is the same as that of MRC. Envelope-based GSC In this subsection we present a simple suboptimum generalized selection combining rule based on noncoherent envelope detection. We note that √ 4 Es |Λi1 ,i2 ,···,iM | = |Re{y1i1 ,i2 ,···,iM }| (56) I0 √ M 4 Es ≤ |Re{y1im }| (57) I0 m=1 √ M 4 Es ≤ |y1im |. (58) I0 m=1 Since combining the received RF signals and detecting the envelope of the combined signal yields |y1im |, we propose a suboptimum generalized selection M combining rule that selects M Rx antennas maximizing m=1 |y1im |. It will be called envelope-based GSC. 444 SIR-based GSC Given the Rx. antenna i1 , i2 , · · · , iM are selected, the SIR is M proportional to A1i1 ,i2 ,···,iM = m=1 [|h1im |2 + |h2im |2 ]. The conventional GSC rule selects M Rx antennas providing the largest A1i1 ,i2 ,···,iM . It will be called SIR-based GSC. VII. N UMERICAL R ESULTS AND D ISCUSSION In this section, we present numerical results and discussions. Figure 3 is a plot of average bit error probability versus Ēs /I0 with single transmit antenna and NR Rx antennas, where Ēs is the average received energy per diversity branch and is equal to Es since we have assumed E |hi |2 = 1. The performance curves are computed assuming an independent identically distributed (i.i.d.) slow Rayleigh fading model. We find that the diversity order (slope) depends on the number of receive antennas. For NR =4, the optimum selection combining (OSC) provides a gain of 3dB over the SIR-based selection combining (SIR-SC), and is only 0.3dB away from MRC that requires multiple (four) RF chains. The envelope-based selection provides a gain of 1.8dB over SIR-SC. For NR =2, the error probability of MRC can be achieved by OSC. This is because the sign of MRC output is identical to that of LLR-based selection combiner output. We can also notice that MRC of four channels (requiring four RF chains, NR = 4) is outperformed by envelope-based combining of one channel (requiring one RF chain) out of eight antennas (NR = 8). Figure 4 is a plot of average bit error probability versus Ēs /I0 with space-time block code employing two Tx antennas and NR Rx antennas. For NR =4, the OSC provides a gain of 3.5dB over the SIR-SC, and is 0.6dB away from MRC that requires four RF chains. The envelope-based selection combining provides a gain of 2.1dB over SIR-SC. For NR =2, the error probability of MRC can be achieved by the optimum selection combining, as in single Tx antenna case. Figure 5 and 6 are plots of average bit error probability versus Es /I0 with generalized selection combining of M Rx antennas and STBC employing two Tx antennas and NR Rx antennas. We find that the optimum GSC and Env-GSC provide power gains of 1.6 ∼ 3.1dB and 0.9 ∼ 1.6dB over SIR-GSC, respectively, for M = 2 and NR = 4 ∼ 8. We also find that the diversity order (slope) depends on NR , and the SIR gain increases with increasing M . and suboptimum selection combining rules with those of traditional combining rules such as SIR-based selection combining and maximum ratio combining (MRC). For NR = 4, the optimum selection combining provides a gain of 3-3.5dB over the SIR-based selection combining, and is only 0.3-0.6dB away from MRC that requires multiple (four) RF chains. The envelope-based selection combining provides a gain of 1.82.1dB over the SIR-based selection combining. The power gain increases with increasing NR . For STBC systems, we present the optimum and suboptimum generalized selection combining rules in which M (≤ NR ) Rx antennas providing the largest LLR magnitude are selected and combined. A PPENDIX I In this appendix, we show that the bit error probability, Pe (R), with MAP (optimum) detection for a received observation R can be expressed 1 Pe (R) = (59) 1 + e|Λ(R)| where P (x = +1|R) Λ(R) = ln (60) P (x = −1|R) is the log-likelihood ratio (LLR). Moreover, the relationship in (59) is true for any binary signals in any channel. Proof: It follows from (60) and P (x = +1|R) + P (x = −1|R) = 1 that 1 P (x = +1|R) = (61) 1 + e−Λ(R) and 1 P (x = −1|R) = . (62) 1 + eΛ(R) By definition, Pe (R) = P (x̂ = x|R) (63) = P (x̂ = 1, x = −1|R)+P (x̂ = −1, x = 1|R)(64) where x̂ is the detector output. If Λ(R) > 0, i.e. x̂ = 1, then Pe (R) ≤ P (x = −1|R) + P (x̂ = −1|R) =P (Λ(R)<0)=0 = VIII. C ONCLUSION In this paper we presented the optimum receive antenna selection combining rule that minimizes the bit error probability. The optimum selection combining rule, derived from a general relationship between the bit error probability and the log-likelihood ratio (LLR), is to select the antenna providing the largest LLR magnitude. We applied the optimum selection combining rule to single transmit (Tx) antenna and spacetime block code (STBC) systems with NR Rx antennas, and derived the average bit error probability for BPSK signaling in a Rayleigh flat fading channel. Also, we presented a suboptimum selection combining rule based on noncoherent envelope detection and derived its average bit error probability. We compared the average bit error probabilities of optimum (65) 1 . 1 + eΛ(R) (66) Also, Pe (R) = 1 − P (x̂ = x|R) = (67) 1 − P (x̂ = 1, x = 1|R) −P (x̂ = −1, x = −1|R) ≥ 1 − P (x = 1|R) − P (x̂ = −1|R) (68) (69) =P (Λ(R)<0)=0 1 = . 1 + eΛ(R) Therefore, if Λ(R) > 0, then 445 Pe (R) = 1 . 1 + eΛ(R) (70) (71) If Λ(R) < 0, we can similarly show that Pe (R) = 1 . 1 + e−Λ(R) (72) Pe (R) = 1 . 1 + e|Λ(R)| (73) As a result, A PPENDIX II In this Appendix, we derive (17). Let √ Es Yi = (|hi |2 x + Re{h∗i ni }). I0 (74) √ Since the pdf √ of Zi = |Yi | is the same√ whether x = Es or x = − Es , we will assume x = Es without loss of generality. Then, given hi , Yi is a Gaussian random variable with mean |hi |2 γ and variance |hi |2 γ/2, where γ = Es /I0 . 2 Since the pdf of |hi | is Rayleigh, i.e. f|hi | (a) = 2ae−a , the pdf of Yi is ∞ 2 2 2 2 1 fYi (y) = e−(y−a γ) /(a γ) · 2ae−a da(75) 2 πa γ 0 √ 1 −1 = e2y−2|y| 1+γ (76) γ(1 + γ) where we used the equality [9] ∞ 1 π −2√ab −ax2 −b/x2 e dx = e . 2 a 0 The cdf of Zi is FZi (z) = = z 1 2y−2|y| √ 1+γ −1 e γ(1 + γ) √ −1 (1 − e−2( 1+γ +1)z ) 2(γ + 1 + γ 2 + γ) √ −1 (1 − e−2( 1+γ −1)z ) + . 2(γ + 1 − γ 2 + γ) (77) = where we use the equality [9] ∞ √ √ 2 2 1 π x2 e−a/x −bx dx = (1 + 2 ab)e−2 ab . 3 4 b 0 The cdf of Zi , FZi (z), is z FZi (z) = fYi (y)dy −z (85) (86) (87) (88) √ 1 + 2/γ)ey 1+2/γ+y = dy γ(γ + 2)3 −z √ z (1 + y 1 + 2/γ)e−y 1+2/γ+y + dy (89) γ(γ + 2)3 0 2c + 1 − [2c + 1 + c(c + 1)z]e−(c+1)z = γ(γ + 2)3 [(c + 1)]2 2c − 1 − [2c − 1 + c(c − 1)z]e−(c−1)z + (90) γ(γ + 2)3 [(c − 1)]2 where c = 1 + 2/γ. Therefore, the pdf of Zi is given by 0 (1 − y dFZi (z) (91) dz (1 + z 1 + 2/γ) = γ(γ + 2)3 √ √ [e−z( 1+2/γ+1) + e−z( 1+2/γ−1) ]. (92) fZi (z) = dy (78) −z R EFERENCES (79) Therefore, the pdf of Zi is fZi (z) = √ where γ = Es /I0 . If we let b = a then √ y ∞ y2 1 2 2e fYi (y) = √ b2 e− 2γ b2 −(1+γ/2)b db πγ 0 1 + |y| 1 + 2/γ −|y|√1+2/γ+y = e γ(γ + 2)3 dFZi (z) (80) dz √ √ −1 −1 e−2( 1+γ +1)z + e−2( 1+γ −1)z .(81) γ2 + γ A PPENDIX III In this Appendix, we derive (40). Let √ 2Es Yi = [(|h1i |2 + |h2i |2 )x1 + Re{h∗1i n1i + h2i n∗2i }] (82) I0 Since the pdf 1 = + Es /2 of Zi = |Yi | is the same whether x or x1 = − Es /2, we will assume x1 = + Es /2. Then, given h1i and h2i , Yi is a Gaussian random variable with mean Ai Es /I0 and variance Ai Es /I0 , where Ai = |h1i |2 + |h2i |2 . Since the pdf of Ai , fAi (a), is ae−a , the pdf of Yi is ∞ 2 1 √ fYi (y) = e−(y−γa) /(2γa) fAi (a)da (83) 2πγa 0 ∞ √ − y2 1 −(γ/2+1)a ey = √ ae 2γ a da (84) 2πγ 0 [1] W.C.Jakes, Microwave Mobile Communications, IEEE Press, 1974. [2] G.J.Foschini and M.J.Gans, “On the limits of wireless communications in a fading environment when using multiple antennas ,” Wireless Personal Communications, vol.6, pp.311-335, 1998. [3] D.Gore and A.Paulraj, “Space-time block coding with optimal antenna selection,” Proc. of IEEE Inter. Conf. on Acoustics, Speech, and Signal Processing pp.2441-2444, 2001. [4] R.W.Heath, S.Sandhu, and A.Paulraj, “Antenna selection for spatial multiplexing systems with linear receivers,” IEEE Communications Letters pp.142-144, April 2001. [5] A.F.Molisch, M.Z.Win, and J.H.Winters, “Capacity of MIMO systems with antenna selection,” Proc. of IEEE ICC, pp.570-574, 2001. [6] S. M. Alamouti, “A simple transmit diversity technique for wireless communications,” IEEE Jounal on Selected Areas in Communications, Vol. 16, No. 8, pp.1451-1458, October 1998. [7] V.Tarokh, H.Jafarkhani, and A.R.Calderbank, “Space-time block code from orthogonal designs,” IEEE Tr. on Information Theory, Vol. 45, pp.1456-1467, JUly 1999. [8] J.Proakis, Digital Communications, pp.781, 3rd Ed., McGraw-Hill, 1983. [9] I.S.Gradshteyn and I.M.Ryzhik, Tables of Integrals, Series, and Products, pp.341, Eq. 3.472, Academic Press, 1980. [10] H.Cramer, Random Variables and Probability Distributions, Cambridge Univ. Press, 1970. [11] S.W.Kim, Y.G.Kim, and M.Simon, “Generalized selection combining basesd on the log-likelihood ratio,” submitted to IEEE Tr. on Communications. 446 Fig. 1. Channel model for one Tx and NR Rx antennas. Fig. 4. Bit error probability versus Ēs /I0 with 2 Tx and NR Rx: Space-time block code. Fig. 2. Space-time block code systems with two Tx antennas and NR Rx antennas. Fig. 5. Bit error probability versus Ēs /I0 with 2 Tx and NR Rx: Space-time block code, GSC, M=2 Fig. 3. Bit error probability versus Ēs /I0 with 1 Tx and NR Rx. Fig. 6. Bit error probability versus Ēs /I0 with 2 Tx and NR Rx: Space-time block code, Opt.-GSC 447