New Space-Time Code Design for Prime Number of Transmitter Antennas Terasan Niyomsataya Ali Miri Monica Nevins School of Information Technology and Engineering University of Ottawa Ottawa, Canada K1N 6N5 School of Information Technology and Engineering University of Ottawa Ottawa, Canada K1N 6N5 Department of Mathematics and Statistics University of Ottawa Ottawa, Canada K1N 6N5 tniyomsa@site.uottawa.ca samiri@site.uottawa.ca mnevins@uottawa.ca Abstract— A full diversity constellation, that is, a set of unitary matrices whose differences have nonzero determinant, is a design criterion for codes with good performance using differential unitary space-time modulation. In this paper, we present a new Bruhat decomposition design for constructing full diversity unitary constellations for a prime number of transmitter antennas. We give examples of our proposed constellations for 3 and 5 transmitter antennas. Simulations show that these proposed constellations perform well in unknown channel, and also outperform the best abelian group designs. Index Terms— Differential unitary space-time modulation, full diversity, Bruhat decomposition, permutation representation I. I NTRODUCTION channel matrix. The fading coefficients of a channel matrix and the additive noise on receiver antenna are independent complex Gaussian variables with mean zero and variance one, CN (0, 1). Here ρ is the SNR at each receiver antenna. Let Vzτ be an M × M unitary message matrix at block τ which is chosen from an alphabet zτ ∈ {0, 1, . . . , L − 1}. The date rate is R = log2 L/M . The transmitted signal matrix is Sτ = Vzτ Sτ −1 with S0 = IM where IM is an M × M identity matrix. We assume that the channel matrix is constant over two consecutive time periods, that is Hτ ≈ Hτ −1 . Then the received signal matrix of (2) can be written as √ Xτ = Vzτ Xτ −1 + 2Wτ0 (3) The design problem of differential unitary space-time modulation [1], [2] is to maximize the minimum value of the distance (1) d(l, l0 ) = | det(Vl − Vl0 )| ≥ 0, where Wτ0 is also independent CN (0, 1). The ML decoder is used for decode a message ẑτ to be for all distinct Vl , Vl0 ∈ V, a set of M × M unitary matrices where M is the number of transmitter antennas. A constellation V for which d(l, l 0 ) > 0 for all l 6= l0 is said to have full diversity. The application of the Bruhat decomposition to the design of a full diversity unitary subgroup constellations was first introduced in [3] where it was developed only for 2 and 4 transmitter antennas, and used a method not applicable to an odd number of transmitter antennas. In this paper, we present a new Bruhat decomposition design for constructing full diversity unitary space-time constellations for a prime number of transmitter antennas. Simulations show that our proposed constellations perform well in unknown fading channel, and also outperform the best abelian group design, a cyclic group [1], [2]. III. M ATHEMATICAL P RELIMINARIES II. D IFFERENTIAL U NITARY S PACE -T IME M ODULATION Consider a multiple-antenna system in a Rayleigh flat fading channel with M transmitter and N receiver antennas. The M × N received signal matrix Xτ is √ (2) Xτ = ρSτ Hτ + Wτ , τ = 0, 1, . . . where Sτ is the M × M transmitted signal matrix, Wτ is the M × N additive noise matrix and Hτ is the M × N ẑτ = arg min l=0,1,...,L−1 kXτ − Vl Xτ −1 k. (4) Let Sn denote the symmetric group of order n, that is, the group consisting of all permutations of n elements. The symmetric group Sn has a natural representation on C n , called the permutation representation, in which each element ω ∈ Sn is realized as the n × n matrix of the linear transformation which permutes the standard basis vectors e1 , e2 , . . . , en of C n according to ω. For example, in S3 , the permutation ω = (123) sends e1 to e2 , e2 to e3 and e3 to e1 . Hence under the natural representation, we identify ω with the unitary matrix 0 0 1 ω = 1 0 0 . 0 1 0 The symmetric group Sn has a special relationship with the general linear group GL(n) (that is, the group of all invertible n × n matrices) defined by the Bruhat decomposition. The Bruhat decomposition of GL(n) is the disjoint union of double cosets [ GL(n) = BωB (5) ω∈Sn where B denotes the subgroup of all invertible upper triangular matrices, and ω is written in its natural representation. Whereas the group B is not unitary, it contains the subgroup T of diagonal unitary matrices. Restricting to T in equation (5) defines a unitary subgroup of GL(n). A quick calculation reveals that the double coset T ωT is equal to the left coset ωT for any ω ∈ Sn . This defines an infinite unitary subgroup of GL(n), Sn n T , a semi-direct product as: [ [ Sn n T = T ωT = ωT. (6) ω∈Sn OF F ULL D IVERSITY C ONSTELLATION U NITARY In this section, we present the design of a full diversity unitary constellation using Bruhat decomposition as explained in Section III, equation (6). Let M be the number of transmitter antennas and ω be an M × M matrix representing the full permutation (1 2 3 · · · M ) in the natural representation. This cycle has the property that neither ω, nor any power of ω (up to ω M = IM ), has any fixed points. Consequently, the positions of the nonzero entries in ω i , or equivalently, in ω i T , are entirely disjoint from those for ω j for any i 6= j.1 Choose a finite full diversity subgroup D of diagonal unitary matrices 2 having determinant equal to 1. Set λ = e2πj/M . Our proposed unitary constellation V is a subset of Sn n T , given by the disjoint union of sets of unitary matrices defined by V= M −1 [ i=0 ddc (l, l0 ) = | det(D − λi ω i D0 )|. i i λ ω D. (7) Write D ∈ D as D = diag(d1 , d2 , . . . , dM ), with |dj | = 1 and d1 d2 · · · dM = 1. Hence det(λi ω i D) = ±e2πji/M for all i = 0, 1, . . . , M − 1. In order to construct a full diversity constellation, we need to have d(l, l 0 ) > 0 (see (1)). This problem can be broken to two cases: i) Same Coset (sc) Case: Suppose Vl , Vl0 ∈ V lie in the same coset λi ω i D. Then we can write Vl = λi ω i D and Vl0 = λi ω i D0 for some D, D0 ∈ D. We compute dsc (l, l0 ) = | det(Vl − Vl0 )| = | det(λi ω i )|| det(D − D0 )| = | det(D − D0 )|. (8) Thus the minimum value of dsc (l, l0 ) equals the minimum distance between two elements of D. Hence, in order to have dsc (l, l0 ) > 0, it suffices to require the full diversity of the subgroup D. ii) Distinct Coset (dc) Case: Suppose now that Vl , Vl0 ∈ V lie in distinct cosets; says Vl = λi ω i D and Vl0 = λj ω j D0 for some D, D0 ∈ D. Then as above, we may restrict ourselves to the case where Vl ∈ D and Vl0 ∈ λi ω i D, 1 This property makes the computation of the distance between distant cosets more accessible to theoretical analysis. (9) When the number of transmitter antennas is prime, any power ω i of the full permutation ω is a single cycle. It follows that the distance of this case can be simply computed as ddc (l, l0 ) = |λiM − 1|. ω∈Sn However this subgroup does not have full diversity, since T itself does not. The idea of construct a full diversity subgroup is to restrict to a full diversity subgroup of T and constraint a constellation from its cosets. IV. D ESIGN since | det(λk ω k D − λj ω j D0 )| = | det(λk ω k )|| det(D − λj−k ω j−k D0 )| = | det(D − λj−k ω j−k D0 )|. This gives (10) This value is equal to the Euclidean distance between two points on a unit circle which is always greater than 0 (when 2 λiM 6= 1). It follows that the value of λ = e2πj/M maximizes the minimum value of ddc (l, l0 ) by ensuring the M points λiM are equally spaced along the unit circle. Design Method Summary: We now summarize the design method of a full diversity unitary constellation as follows: 2 1) Construct a constellation by (7) which has λ = e2πj/M for a given M . 2) Choose D to be a unitary cyclic diagonal subgroup D −1 {Dl }L l=0 , such that Dl = diag(d1 , d2 , . . . , dM ) = j2πu1 l/LD j2πu2 l/LD , . . . , ej2πuM l/LD ) where ,e diag(e M the values of {ui }i=1 ∈ {1, 2, . . . , LD − 1} must be chosen so that Dl satisfies det Dl = 1, and also so that the distance dsc (l, l0 ) = | det(Dl − Dl0 )| is maximized. 3) The order and the rate of our proposed constellation are LD M and log2 (LD M )/2 respectively. V. E XAMPLES & P ERFORMANCE We give the examples of constructing our proposed full diversity unitary constellations for M = 3 and 5 transmitter antennas using the Bruhat decomposition as discussed in Section IV, and also show their performance in unknown Rayleigh flat fading channels. We use a differential modulation to transmit signals (as explained in Section II). The ML decoder of (4) is used for decoding. A. For M = 3 Our proposed constellation for three transmitter antennas is V3 = D ∪ λωD ∪ λ2 ω 2 D (11) with λ = ej2π/9 . The distance ddc (l, l0 ) may be computed as, for i = 1, ω = (123), we have ddc = |λ3 − 1| = 1.7321 > 0, and for i = 2, ω = (132), we have ddc = |λ6 −1| = 1.7321 > 0. The choice of Dl = diag(d1 , d2 , d3 ) can be made by setting −1 d3 = d−1 1 d2 . Consequently the constellation, of order 3L D , is V3 = 0 0 1 0 1 0 D ∪ ej2π/9 0 0 1 D ∪ ej4π/9 1 0 0 D (12) 0 1 0 1 0 0 −2 where each Dl ∈ D has the form Dl = diag(d1 , d2 , d−1 1 d2 ) with d1 = ej2πu1 l/LD and d2 = ej2πu2 l/LD . The values of u1 , u2 = {1, 2, . . . , L − 1} are also chosen to maximize dsc (l, l0 ) for a given LD . Table I shows some of our proposed constellations with their best diversity products (as defined 0 10 −1 10 bler in [4]) compared with different 3 × 3 unitary constellation designs in the literature. We can see that our proposed constellations have diversity products higher than cyclic groups and some constellations obtained from fixed-point free groups. Figure 1 displays the block error rate (bler) performance of our proposed 3 × 3 constellation u1 = 1, u2 = 4 at R = 2.04 compared with the cyclic group u = (1, 11, 27) at R = 2.00 as given in Table I with N = 1. We can see that our constellation outperforms the cyclic group with ≈ 2 dB gain improvement. B. For M = 5 Our proposed constellation for five transmitter antennas is V5 = D ∪ λωD ∪ λ2 ω 2 D ∪ λ3 ω 3 D ∪ λ4 ω 4 D −2 10 (13) our proposed code cyclic j2π/25 where λ = e . The choice of a subgroup of unitary diagonal matrices D = {Dl }L−1 l=0 is 10 0 2 4 6 8 (14) where d1 = ej2πu1 l/LD , d2 = ej2πu2 l/LD , d3 = ej2πu3 l/LD and d4 = ej2πu4 l/LD . The values of u1 , u2 , u3 , u4 = {1, 2, . . . , LD − 1} are chosen to maximize dsc (l, l0 ) for a given LD . The order of this constellation is 5LD . Table I also compares our proposed 5 × 5 constellation with different 5 × 5 unitary constellation designs at the data rate R ≈ 1.00. Figure 2 shows the block error rate (bler) performance of our proposed 5 × 5 constellation as given in Table I with N = 1. We can see that our proposed constellation outperforms the cyclic group and compares well with the nongroup constellation. 10 SNR in db 12 14 16 18 20 Fig. 1. Block error rate performance for M = 3, N = 1 at R ≈ 2.00 of our proposed constellation and cyclic group 0 10 −1 10 −2 10 bler Dl = −1 −1 −1 diag(d1 , d2 , d3 , d4 , d−1 1 d2 d3 d4 ) −3 −3 M 3 3 3 3 3 3 3 3 3 3 3 3 3 5 5 5 LD M 9 9 57 57 63 63 63 64 69 513 513 4095 4095 32 33 35 R 1.06 1.06 1.94 1.94 1.99 1.99 1.99 2.00 2.04 3.00 3.00 4.00 4.00 1.00 1.01 1.03 ζV 0.6004 0.6004 0.4845 0.4845 0.3301 0.3851 0.3851 0.2765 0.3548 0.1353 0.1436 0.0361 0.0471 0.4095 0.5580 0.5164 Designs fixed-point free group G9,1 [4] proposed design u1 = 1, u2 = 1 nongroup S19,3 , u = (1, 7, 11) [4] proposed design u1 = 1, u2 = 7 cyclic group u = (1, 17, 26) [2] fixed-point free group G21,4 [4] proposed design u1 = 1, u2 = 4 cyclic group u = (1, 11, 27) [2] proposed design u1 = 1, u2 = 4 fixed-point free group G171,64 [4] proposed design u1 = 1, u2 = 22 fixed-point free group G1365,16 [4] proposed design u1 = 1, u2 = 121 cyclic group u = (1, 5, 7, 9, 11) [2] nongroup S11,3 u = (1, 3, 4, 5, 9) [4] proposed design u = (1, 1, 4, 5) TABLE I C OMPARISON OF DIFFERENT UNITARY CONSTELLATION DESIGNS , (O UR PROPOSED CONSTELLATION IS HIGHLIGHTED IN GREY ) 10 −4 10 our proposed code cyclic nongroup −5 10 0 2 4 6 8 SNR in db 10 12 14 16 Fig. 2. Block error rate performance for M = 5, N = 1 at R ≈ 1.00 of our proposed constellation, cyclic group and nongroup ACKNOWLEDGMENT This work was partially supported by grants from Natural Sciences and Engineering Research Council of Canada (NSERC), and Communications and Information Technology Ontario (CITO). R EFERENCES VI. C ONCLUSION We have proposed the design of a full diversity unitary space-time constellation for a prime number of transmitter antennas using Bruhat decomposition. Simulations show that our proposed constellations perform well without the knowledge of the information of the channel. The evaluation of the full diversity condition of ddc (l, l0 ) in (9) can be extended to a general number of transmitter antennas. [1] B.M. Hochwald and W. Sweldens, “Differential unitary space-time modulation,”IEEE Trans. Info. Theory, vol. 48, pp. 2041-2052, Dec 2000. [2] B.L. Hughes, “Differential space-time modulation,” IEEE Trans. Info. Theory, vol. 46, pp. 2567-2578, Nov 2000. [3] T. Konishi, “Unitary subgroup space-time codes using Bruhat decomposition and Weyl groups,” IEEE Trans. Info. Theory, vol. 49, pp. 2713-2717, Oct 2003. [4] A. Shokrollahi, B. Hassibi, B.M. Hochwald and W. Sweldens, “Representation theory for high rate multiple-antenna code designs,” IEEE Trans. Info. Theory, vol. 47, pp. 2335-2367, Sept 2001.