1

On the MIMO Channel Capacity for the Nakagami-m Channel

Gustavo Fraidenraich, Student Member, IEEE, Olivier L v que Member, IEEE, and John M. Ciof, e e Fellow, IEEE

Abstract This paper presents the MIMO channel capacity over the Nakagami-m fading channel. The joint eigenvalue density function of W = HH , where H is the channel matrix, is derived in closed form for H (22) and any integer values of m, as well as for H (23) with m = 2 and m = 3. The marginal eigenvalue distribution of W is also derived in a closed-form expression. All the results are validated by numerical Monte Carlo simulations and are in excellent agreement. Index Terms Fading distributions, Rayleigh distribution, Nakagami-m distribution, Eigenvalue distribution, MIMO channels.

I. I NTRODUCTION It has been acknowledged in recent years that the use of multiple inputs and multiple outputs (MIMO) can potentially provide large spectral efciency for wireless communications in the presence of multipath fading environments. In the papers presented by Winters [1], Foschini [2] and Telatar [3], the capacity was in particular shown to scale linearly with the number of antennas. In most previous research on MIMO capacity, the channel fading is assumed to be Rayleigh distributed. Of course, the Rayleigh fading model is known to be a reasonable assumption for the fading encountered in many wireless communications systems. Nevertheless, many measurements campaigns [4], [5] show that the Nakagami-m distribution provides a much better tting for the fading channel distribution. In fact, since the Nakagami-m distribution has one more free parameter, it allows for more exibility. It moreover contains the Rayleigh distribution (m = 1), the one-sided Gaussian distribution (m 0.5), and the uniform distribution on the unit circle1 (m ) as special (extreme) cases.
Gustavo Fraidenraich and John M. Ciof are with the Electrical Engineering Department, Stanford University, Stanford, CA 94305 USA. The work of Gustavo Fraidenraich was supported by Cnpq (Brazil) Grant Nr 200869/2005-1. Olivier L v que is now with the e e Ecole Polytechnique F d rale de Lausanne, Switzerland, but part e e of his work was supported by Swiss NSF grant Nr PA002-108987, when he was with the Electrical Engineering Department at Stanford University. 1 When the uniform phase distribution is considered

The Nakagami-m distribution is a general, but approximate solution to the random phase problem [6]. The exact solution to this problem involves the knowledge of the distribution and the correlations of all of the partial waves composing the total signal and becomes infeasible due to its complexity [7]. This has been circumvented by Nakagami [6] who, through empirical methods based on eld measurements followed by a curve-tting process, obtained the approximate distribution. Since the publication of [3], other distributions have been considered, as in [8] for the Ricean case, and recently [9] addressed the Hoyt distribution case. A more realistic MIMO Rayleigh channel including correlation has been addressed in [10]. The Nakagami-m distribution has also been addressed for some special cases as in [11] for the SIMO (single input multiple output) and MISO (multiple input single output) cases, and in [12] for the keyhole channel. The MIMO channel capacity can be computed by means of the joint eigenvalue density function (JEDF) of the matrix W = HH , where denotes the complex conjugate transpose. In the classical Rayleigh fading model, the entries of H are assumed to be i.i.d. zero mean complex Gaussian, resulting in a matrix W which is Wishart distributed. This model takes advantages of the numerous results provided by the literature [13], [14]. Unfortunately, as one departs from this model, there is not too much that can be said. The result presented here follows the sequence of decompositions H = LQ, W = LL and W = SS . The decomposition through the unitary matrices Q provides a way to integrate over the volume dQ, which W = HH does not provide. Denoting dH as dH = J dL dQ where J is the Jacobian of the transformation, the distribution of p(H) does not depend on Q for the Rayleigh case. Hence, the integration on the Stiefel manifold can be computed in closed form [15]. Unfortunately, for the Nakagami-m case, p(H) depends on Q, so a larger number of variables has to be integrated. To be more accurate, 2rt r2 integrations over real variables are necessary (the matrix H has 2rt real variables and L has r2 real variables, so Q has 2rt r2 variables).

2 p(r,) r ,

For either m = 1 or m = 0.5, the resulting integral is simply the volume of the Stiefel manifold. This paper presents the MIMO channel capacity over the Nakagami-m fading channel for H with dimensions 2 2 and 2 3. Channel state information is assumed at the receiver side only. In this case, it is shown that the uniform power distribution across the transmitting antennas achieves the channel capacity. Assuming that the entries of H are i.i.d. with Nakagami-m distributed envelopes and uniform phases, an elegant and simple expression for the JEDF of W is derived. The paper is organized as follows: Sec. II introduces the Nakagami distribution in more detail. Sec. III denes channel capacity, and Sec. IV presents the main result for H 22 and any integer m. Sec. V provides the result for H 2 3, and m = 2 or m = 3. Sec. VI presents the asymptotic result. Sec. VII presents some numerical simulations and Sec. VIII draws the conclusion.

Z is given by p(x, y) =

thus
m1

mm x2 + y 2 p(x, y) = m (m)

m(x2 +y 2 )

(5)

III. MIMO C HANNEL C APACITY The following MIMO single-user Gaussian channel is considered, with t antennas at the transmitter and r antennas at the receiver:
y = Hx + n

(6)

H is the r t channel matrix with i.i.d. entries hij , each distributed as the random variable Z dened in (1). The vector y C r , x C t , and n is zero-mean complex Gaussian noise with E n n = I. In addition, a total transmit power constraint E[x x] P is assumed. For a given input covariance matrix , the MIMO instantaneous capacity is given by C() = log2 det I + HH

(7)

II. T HE NAKAGAMI -m DISTRIBUTION The entries of the r t channel matrix H are assumed to be i.i.d. and distributed as
Z = R exp(j)

and the MIMO channel capacity in the absence of channel knowledge at the transmitter is given by [3]
C= sup
0:tr[]P

E [C()]

(8)

(1)

where the phase is uniformly distributed and independent of the envelope R. R is in turn given by
m

R2 =
i=1

Xi2 + Yi2

(2)

Since the entries hij are i.i.d. and the distribution of hij is the same as that of hij for all i, j , one obtains from [16, Corollary 1b] that the uniform power allocation over the t transmit antennas achieves the channel capacity. The capacity is therefore given by
C = E log2 det I + P W t

where Xi and Yi are i.i.d. zero mean Gaussian distributed with variance /2m. The distribution of R is therefore the Nakagami-m distribution [6] given by
2mm r2m1 mr2 p(r) = exp m (m)

(9)

where
W=

HH r t H H r > t
RESULT FOR

(10)

(3) IV. M AIN Theorem 1: The JEDF of the 22 matrix W = HH , where the entries of the 2 2 matrix H are i.i.d. with Nakagami-m envelope and uniform phase, is given by
p(1 , 2 ) = K22 e
m(1 +2 )

where = E R2 , m = E[R2 ]2 /Var[R2 ], and () denotes the Euler Gamma function. In addition, p(r, ) = 1 p(r) 2 , yielding
p(r, ) = mr2 mm r2m1 exp m (m)

H22

(4)

(1 2 )2 F (1 , 2 ) (11)
ij

Note that p(r, ) reduces to the Rayleigh distribution for m = 1 and to the uniform distribution on the circle of radius for m . The above family of distributions therefore allows to interpolate between the classical Rayleigh distribution and the pure random phase distribution. Using the standard polar-rectangular transformation, the joint distribution of the real and imaginary parts of

where
Kij =

mm m (m)

(12)

and F (1 , 2 ) is given by (13), and f (i1 , i2 , k1 , k2 ) is given by (14). In order to gain some intuition on the above result, note that the function F (1 , 2 ) is a homogeneous polynomial of order 4(m 1). In the Rayleigh case

4 F (1 , 2 ) = (4m 2)

m1 m1

i1

i2

i1 =0 i2 =0 k1 =0 k2 =0

f (i1 , i2 , k1 , k2 ) (1 2 ) 2 (i1 +i2 2(k1 +k2 ))+m1 (1 2 )i1 i2 +2(k1 +k2 +m1)

(13)
m1 i2

f (i1 , i2 , k1 , k2 ) =

i1 k1

i2 k2

m1 i1

1 (1)i1 2i2 +3m+1 22(k1 +k2 +1) i2

i2 2

+m

1 2

1 2 1 2 i2 i2 + m i2 i2 + k1 + k2 + m + 1 2 i1 i2 i1 i2 i1 i2 + k1 + k2 + m + + k1 k2 + 2m 1 k1 + k2 + 2m 1 2 2 2 2 2 2

(14)

0.6
Simulated Theoretical

0.4

p(

m m=10
0.2

complex lower triangular matrix with real positive diagonals and Q is a complex unitary matrix QQ = I ; 3) therefore, W = LL ; 4) nally, performing the eigenvalue decomposition W = SS , one obtains the JEDF of W. Since the entries of the channel matrix H are independent, their joint distribution is given by
p (H) = K22 exp m tr HH
2 i,j=1

|hij |2(m1)

m=1
0.0 0 2

m=2

(16) Using the LQ decomposition, the matrix H can be written as H = LQ, where the matrix Q is given as [17]
Q= ej2 sin() ej1 cos() ej(3 2 ) sin() ej(3 1 ) cos()

Fig. 1. The Nakagami eigenvalue distribution function for H (22) and m = 1, 2, 10, 100 ( = 1).

(17)

(m = 1), F is a constant and one recovers the classical JEDF of a Wishart matrix [13]: 1 (1 +2 ) p(1 , 2 ) = e (1 2 )2 (15) 24 The effect of the Nakagami-m envelope distribution on the JEDF is therefore expressed by the polynomial F (1 , 2 ). To illustrate the effect of the parameter m, Fig. 1 shows the resulting marginal eigenvalue distribution p() for m = 1, 2, 10, (see also Corollary 2). It can be observed that as the parameter m increases, the eigenvalues concentrate on the interval [0, 4], with higher probability on the boundary of the interval. It can be also seen in Fig. 1 that the result is in perfect agreement with Monte-Carlo simulations. Proof: In order to get the JEDF of W, the following steps need to be performed: 1) the joint distribution of H can be easily found, since its entries are i.i.d.; 2) the matrix H is decomposed as H = LQ, where L is a

and the variables are dened in the following range 0 1 , 2 , 3 2 , 0 /2. The matrix L is given by l11 0 (18) L= l21R + j l21I l22 The Jacobian of this transformation is given by J = 3 l11 l22 sin() cos(), so the joint probability density function (PDF) of L and Q is given by
m tr(L L ) K22 4m1 p (L, Q) = 2m1 e l11 l22 sin2m1 (2) 2 m1 2 l22 cos2 () + T1L sin2 () T2L sin (2)

2 T1L cos2 () + l22 sin2 () + T2L sin(2)

m1

(19) (20)

where
2 2 T1L = l21I + l21R

and
T2L = l22 (l21I sin (1 + 2 3 ) l21R cos (1 + 2 3 )) (21) Since m is integer, it is possible to use the classical

binomial expansion in (19). Therefore

K22 p (L, Q) = 2m1 e 2
m1 i1 m1 i2

m = 2, the JEDF (11) specializes to 4e (1 2 )2 p(1 , 2 ) = 225 3 8 4 + 2 3 + 262 2 + 3 1 + 4 1 1 2 1 2 2 6561e (1 2 )2 p(1 , 2 ) = 1254400 3 12 8 7 1 + 2 1 + 372 6 + 373 5 + 4784 4 2 1 2 1 2 1 +375 3 + 376 2 + 7 1 + 8 2 1 2 1 2 2
3(1 +2 ) 2(1 +2 )

m tr(L L )

4m1 l11 l22

i1 =0 k1 =0 i2 =0 k2 =0

m1 i1

i1 k1

m1 i2
2(i k2 +i1 k1 )

(29)

In the same way for m = 3, the following is obtained

i2 k2

2m2i1 i2 k1 l22 2 (1)m1i2 T1L+k2 T2L

cos2(k1 +i2 k2 ) () sin2(i1 k1 +k2 ) () sin(2)4m3i1 i2 (22)

(30)

Integrating now over , 1 , 2 and 3 , the distribution of p(L) is obtained as (23). The next transformation is given by
W = LL = w1 w3 jw4 w3 + jw4 w2

(24)

where the Jacobian of this change is given by J = 3 4l11 l22 . Using this transformation, the distribution of p(W) can be easily obtained. The next step it to apply the eigenvalue decomposition W = SS , where the matrix S is given by
S= cos() ei sin() j sin() cos() e

The case m requires a slightly different analysis, which leads to the result below. Theorem 3: The marginal eigenvalue distribution of the 2 2 matrix W = HH , where H (2 2) has i.i.d. entries with Nakagami- envelope and uniform phase, is given by
1 p() = 10<<4 4 2

(31)

(25)

0 2 , 0 /2, and is the eigenvalue matrix given by = 1 0 0 2

Proof: As m , the joint distribution of (R, ) given in (4) tends to the uniform distribution on the circle of radius , therefore ej11 ej12 H= (32) ej21 ej22 where ij are uniformly distributed between 0 and 2 . In this case, the matrix W is given by
W= 2 w12 w12 2

(26)

(33)

The Jacobian of this last transformation can be obtained as |J| = 1 (1 2 )2 sin(2). Finally, applying 2 this transformation and integrating over and , the JEDF (11) is obtained. The following corollary is a direct consequence of Theorem 1. Corollary 2: The marginal distribution p() is given by
p() = K22 4 (4m 2)
m1 m1 i1 i2 2(k1 +k2 +m)i1 i2 i3 =0
m

where w12 = ej(11 21 ) + ej(12 22 ) and w12 denotes its conjugate. Using the decomposition W = SS , it is possible to write 1 + 2 = tr (W) = 4 as well as 1 2 = 2 4 |w12 |2 , so
1 = (2 |w12 |)

(34)

i1 =0 i2 =0 k1 =0 k2 =0

f1 (i1 , i2 , k1 , k2 )f2 (i1 , i2 , k1 , k2 , i3 ) e

1

2 (i1 +i2 +2(m+i3 k1 k2 1)) (27)

where f2 (i1 , i2 , k1 , k2 , i3 ) is given by (28).

A. Special Cases: m = 1, 2, 3 and As already mentioned, the JEDF (11) specializes to the classical Wishart distribution (15) in the case m = 1. For

and 2 = 4 1 . Note therefore that in this case, the correlation between the eigenvalues 1 and 2 is much stronger than in the nite m case. In particular, the joint eigenvalue density p(1 , 2 ) does not exist. The modulus of w12 can be written as |w12 | = 2 (1 + cos(11 21 12 + 22 )). Since the distribution of cos(ij ) is the same as the distribution of y = cos(11 21 12 + 22 ), the distribution of this 1 term is p(y) = 1y2 1|y|<1 . Making the transformation of variable and computing the distribution of |w12 |, the following distribution is obtained
p (|w12 |) = 1 1
|w12 |2 4

1|w12 |<1

(35)

p (L) =

K22 8 5/2 m tr(LL ) 4m1 e l11 22m1

m1

i1

m1

i2

i1 =0 k1 =0 i2 =0 k2 =0

m1 i1

i1 k1

m1 i2

i2 k2

2 (1)m1i2 l21I +

1 (2m i1 i2 1) 2 24m4i1 i2 1 (4m + 2k1 + i2 2k2 2 i1 ) 1 (i1 2k1 + 2k2 + 4m 2 i2 ) 2 2 (4 (m 1)) 1 (2m i1 i2 ) 2
1 k1 +k2 + 2 (2mi1 i2 2) i2 2k2 +i1 2k1 +2m1 2 l22 l21R

(23)

f2 (i1 , i2 , k1 , k2 , i3 ) =
1

i1 i2 + 2 (m + k1 + k2 ) i3

(1)2mi1 i2 i3 +2k1 +2k2

i1 i2

m 2 (6m+i1 +i2 +2i3 2k1 2k2 ) 3m 2 2 i3 +k1 +k2 3m

i1 i2 i3 + k1 + k2 2 2

(28)

From (34) and (35), (31) follows directly. In Fig. 1, the expression (31) is compared with simulations for m = 100, and a nearly perfect match is observable. V. 2 3 CHANNEL
MATRIX

p(1 , 2 ) =

H, WITH m = 2 AND

1 2 (1 2 )2 3307512 6 5 31 + 32 1 + 1012 4 + 1013 3 + 1014 2 2 1 2 1 2 1 +35 1 + 36 2 2

32e

2(1 +2 )

(38)

m=3 1594323e 1 2 (1 2 )2 p(1 , 2 ) = 1795585792000018 12 4501 + 4502 11 + 241102 10 + 241103 9 1 2 1 2 1 +4535394 8 + 4535395 7 + 10670096 6 2 1 2 1 2 1 +45011 + 45012 2 2 +4535397 5 + 4535398 4 + 241109 3 + 2411010 2 2 1 2 1 2 1 2 1
3(1 +2 )

We now address the case where H is a 2 3 matrix. Unfortunately, because the complexity of the computation, only the cases m = 2 and m = 3 will be presented 2 , but the approach is valid for any integer m. Since the entries of the channel matrix H are independent, their joint distribution is given by
p (H) = K23 exp m tr HH
2 3

i=1 j=1

|hij |2(m1)

(39)

(36) where K23 is given in (12). Using the LQ decomposition, the matrix H can be written as H = LQ, where the matrix Q, in this case, will be given by (37). This unitary matrix was generated using [17] for the 3 3 case, and then two out of three columns were chosen. The work in [17] provides a way to generate a nn unitary matrix starting from a 22 unitary matrix. Unfortunately, this method produces a null jacobian J in some cases. For this reason, the matrix in (17) was used as the seed for the generation of the 3 3 matrix. Using this method, the Jacobian of this transformation is given 5 3 by |J| = l114l22 cos(2 )3 sin(2 ) sin(21 ) sin(23 ). Following the same steps, i.e., H = LQ, W = LL , and then W = SS , the following distributions are obtained for m = 2 and m = 3, respectively
In fact, the result can be found for any m, but the number of summations is extremely large.
2

In order to validate these expressions, the theoretical and simulated marginal eigenvalue distributions are plotted in Fig. 2. Note the perfect agreement between them.

VI. A SYMPTOTIC C ASE In the case where the number of antennas grows to innity, the result given in [18] can be used, since the entries of H are i.i.d. The limiting eigenvalue distribution 1 of the matrix n W is given by
p() = 1 2 (b ) ( a) a b (40)

2 2 where a = 1 , b = 1 + , = r/t. Using (40) in (9), the following result is obtained for the asymptotic Nakagami channel capacity C n
b r,t

log2 (1 + P ) p() d

(41)

Q=

ej(1 +5 ) cos(1 ) cos(3 ) ej(2 +4 ) sin(1 ) sin(2 ) sin(3 ) ej(5 2 ) cos(3 ) sin(1 ) ej(4 1 ) cos(1 ) sin(2 ) sin(3 ) ej3 cos(2 ) sin(3 )

ej(2 +3 ) cos(2 ) sin(1 ) ej(3 1 ) cos(1 ) cos(2 ) ej4 sin(2 )

(37)

25
0.30
Theoretical Simulated

Simulated Theoretical

0.25

20

0.20

m=20
m=3

m=1

p(

0.15

m=2
0.10

C (bps/Hz)

15

10

0.05

5
0.00

m=0.5

10

12

0
1

10

15

20

25

30

SNR (dB)

Fig. 2. The Nakagami eigenvalue distribution function for H (23) and m = 2, 3 ( = 1).

Fig. 3. The Nakagami channel capacity for the 2 2 case ( = 1).

80

Using the result presented in [19], one obtains

C n 1 ( ln (1 + P P v (, P )) r,t ln 2 + ln (1 + P P v(, P )) v(, P )) (42) 1 2 1++ 1 P (1 + + P 1 )2 4
60

Asymptotic Simulated

C (bps/Hz)

(10 x10) m=6

40
(6 x 6) m=5

where
v(, P ) =

(4 x 4) m=5

20
(3 x 3) m=10

(43) Although this formula is asymptotic, it is shown below by simulation that it is quite accurate, even for a small number of antennas. VII. N UMERICAL R ESULTS As already presented, Fig. 1 validates with MonteCarlo simulations the expression given in (11) for the JEDF for the cases m = 1, 2, 10, . Note that there is an excellent agreement between the simulations and the theoretical results. Fig. 3 compares the simulated channel capacity to the theoretical result (9) for the 22 case and m = 0.5, 1, 20. As can be seen in the gure, when m increases, the channel capacity also increases. The difference between the case m = 0.5 and m = 20 is small for low values of the power P , and increases as the power increases. The effect of the Nakagami parameter m depends on the power P , but in the worst case, it degrades up to 12.5% of the capacity for a 2 2 channel matrix. In Fig. 4, for
Fig. 4.

0 0 5 10 15 20 25 30

SNR (dB)

The Asymptotic Nakagami channel capacity ( = 1)

the 3 3, 4 4, 6 6, and 10 10 cases for different values of m, simulations are performed and compared with the asymptotic result given in (42). VIII. C ONCLUSION In this paper, a general JEDF for the Nakagamim channel is presented in a closed-form expression considering a 2 2 channel matrix H and any integer m. The marginal distribution is also derived in a closedform expression, and for the m case, a simple and elegant expression is obtained. The JEDF was also found for the case where H is a 2 3 matrix, with

m = 2 or m = 3. The ergodic MIMO channel capacity is computed for the Nakagami-m channel, and for the case H (22), the effect of the parameter m is shown to degrade up to 12.5% of the channel capacity. In all cases, the results are validated by Monte-Carlo simulations.

[19] S. Verd and S. Shamai, Spectral efciency of CDMA with u random spreading, IEEE Trans. Inform. Theory, vol. 45, no. 2, pp. 622640, Mar. 1999.

R EFERENCES
[1] J. H. Winters, On the capacity of radio communications systems with diversity in Rayleigh fading environments, IEEE J. Sel. Areas Commun., vol. 5, no. 5, pp. 871 878, Jun. 1987. [2] G. J. Foschini and M. J. Gans, On limits of wireless communications in a fading environment when using multiple antennas, Wireless Pers. Commun., vol. 6, no. 3, pp. 311 335, Feb. 1998. [3] I. E. Telatar, Capacity of multi-antenna Gaussian channels, Europ. Trans. Telecommun., vol. 10, no. 6, pp. 585595, Nov. 1999. [4] H. Suzuki, A statistical model for urban radio propagation, IEEE Trans. Commun., vol. 25, no. 7, pp. 673679, Jul. 1977. [5] T. Aulin, Characteristics of a digital mobile channel type, IEEE Trans. Veh. Technol., vol. 30, no. 1, pp. 4553, Feb. 1981. [6] M. Nakagami, The m-distribution - A General formula of intensity distribution of rapid fading, ser. Statistical Methods in Radio Wave Propagation. W. C. Hoffman, Ed. Elmsford, NY: Pergamon, 1960. [7] M. D. Yacoub, J. E. V. Bautista, and L. G. de Rezende Guedes, On higher order statistics of the Nakagami-m distribution, IEEE Trans. Veh. Technol., vol. 48, no. 3, pp. 790794, May 1999. [8] S. K. Jayaweera and H. V. Poor, On the capacity of multipleantenna systems in Rician fading, IEEE Trans. Wireless Commun., vol. 4, no. 3, pp. 11021111, May 2005. [9] G. Fraidenraich, O. L v que, and J. M. Ciof, On the MIMO e e channel capacity for the dual and asymptotic cases over Hoyt channels, IEEE Commun. Lett., vol. 11, no. 1, pp. 3133, Jan. 2007. [10] H. Shin and J. H. Lee, Capacity of multiple-antenna fading channels: Spatial fading correlation, double scattering, and keyhole, IEEE Trans. Inform. Theory, vol. 49, no. 10, pp. 26362647, Oct. 2003. [11] F. Zheng and T. Kaiser, On the channel capacity of multiantenna systems with Nakagami fading, EURASIP Journal on Applied Signal Processing, vol. 2006, pp. 111, 2006. [12] A. M ller and J. Speidel, Ergodic capacity and information u outage probability of mimo nakagami-m keyhole channels with general branch parameters, in Wireless Communications and Networking conference WCNC 2007, Mar. 2007, pp. 2184 2189. [13] A. T. James, Distributions of matrix variates and latents roots derived from normal samples, Ann. Math. Stat., vol. 35, no. 2, pp. 475501, Jun. 1964. [14] M. L. Mehta, Random Matrices. Academic Press, 1990. [15] A. Edelman, Eigenvlaues and conditions numbers of random matrices, Ph.D. dissertation, Massachusetts Institute of Technology MIT, May 1989. [16] E. Abbe, I. E. Telatar, and L. Zheng, The algebra of MIMO channels, in Allerton Annual Conference on Communication, Control and Computing, Oct. 2005. [17] K. Zyczkowski and M. Kus, Random unitary matrices, J. Phys, A.: Math Gen., vol. 27, 1994. [18] Z. D. Bai, Methodologies in spectral analysis of large dimensional random matrices, a review, Statistica Sinica, vol. 9, pp. 611677, 1999.

Gustavo Fraidenraich received the Electrical Engineering degree from the Federal University of Pernambuco, UFPE, Brazil, in 1997. He received his M.Sc. and Ph.D degrees from the School of Electrical and Computer Engineering of the State University of Campinas, UNICAMP, Brazil, in 2002 and 2006, respectively. From 2006 to 2008, he held a post-doctoral research position in the StarLab Laboratory at Stanford University. His research interests lie in the elds of fading channels, diversity-combining systems, MIMO systems, and wireless communications in general. Dr. Fraidenraich is a reviewer of many IEEE journals and has been invited as a TPC member to many IEEE conferences.

Olivier L v que was born in e e Switzerland in 1971. He received the engineer physicist diploma from the Swiss Federal Insititute of Technology-Lausanne (EPFL), Switzerland, in 1995, and completed the Ph.D. degree at the Mathematics Department of EPFL in 2001. Since then, he has been with the Laboratory of Information Theory at EPFL. He spent the academical year 2005-2006 at the Electrical Engineering Department of Stanford University, where he was appointed as a lecturer. His research interests include information theory, wireless communications, random matrices and stochastic calculus.

J. M. Ciof (S77-M78-SM90F96) received the B.S. degree from the University of Illinois, UrbanaChampaign, in 1978, and the Ph.D. degree from Stanford University, Stanford, CA, in 1984, both in electrical engineering. He was with Bell Laboratories from 1978 to 1984, and IBM Research from 1984 to 1986. He has been a Professor of Electrical Engineering with Stanford University since 1986. He founded Amati Communications Corporation in 1991 (purchased by Texas Instruments in 1997), and was Ofcer/Director from 1991 to 1997. He is currently Chairman and founder of ASSIA, Inc., a company responsible for the introduction

and use of Dynamic Spectrum Management by several large telephone companies. He has served on the boards of directors of public companies Amati, Marvell, and Integrated Telecom Express. He currently is on the boards of directors of Teknovus, Teranetics, ClariPhy, and ASSIA. He is on the advisory boards ofWavion and Amicus. His specic interests are in the area of highperformance digital transmission. He has published over 280 papers and holds over 80 patents, most of which are widely licensed, including basic patents on DMT, VDSL, and Vectored transmission. Dr. Ciof has been the recipient of various awards: International Marconi Prize (2006), Holder of Hitachi America Professorship in Electrical Engineering at Stanford (2002), Member of the National Academy of Enginereering (2001), IEEE Kobayashi Medal (2001), IEEE Third Millennium Medal (2000), IEE JJ Tomson Medal (2000), University of Illinois Outstanding Alumnus (1999), IEEE Fellow (1996 Committee T1 Outstanding Achievement Award of the ANSI (1995), Outstanding Achievement award from the American National Standards Institute for contributions to ADSL (10/95), NSF Presidential Investigator (1987 1992), IEEE Communications Magazine Best Paper Award (1991), and IEEE ISSLS Best Paper Award (2004), Faculty Development Award from IBM Research (1986 1988).