6

Multiple-Input Multiple-Output
Wireless Communications

6.1 Introduction to MIMO Wireless Communications

Multiple-Input Multiple-Output (MIMO) wireless communications employ multiple antennas

at the transmitter and the receiver. A schematic of a MIMO system with multiple antennas
is shown in Figure 6.1. As MIMO systems have multiple antennas, they can be employed to
increase the reliability of the signal through diversity combining as described in the previous
chapters. This leads to diversity gain and a net decrease in the bit-error rate of the wireless
communication system. In addition, a unique aspect of MIMO wireless systems is that they
enable a several-fold increase in the data rate of the wireless communication system by trans-
mitting several information streams in parallel. This is termed spatial multiplexing. This can be
thought of as transmitting multiple parallel streams in space through different spatial modes,
i.e., multiplexing information streams in the space dimension as illustrated in Figure 6.1.

Figure 6.1 MIMO system schematic

166 Principles of Modern Wireless Communication Systems

6.2 MIMO System Model

Consider a MIMO wireless system with t transmit antennas and r receive antennas. Such
a MIMO system is also termed an r × t system. Let x1 , x2 , . . . , xt denote the t symbols
transmitted from the t transmit antennas in the MIMO system, i.e., xi denotes the symbol
transmitted from the ith transmit antenna 1 ≤ i ≤ t. These transmit symbols can be stacked to
form the t-dimensional vector, also termed the transmit vector,
⎡ ⎤
x1
⎢ ⎥
⎢ x ⎥
⎢ 2 ⎥
x=⎢ . ⎥
⎢ .. ⎥
⎣ ⎦
xt

Corresponding to this transmission, let y1 , y2 , . . . , yt denote the r received symbols across the
r receive antennas in the MIMO systems, which can be stacked as the r -dimensional receive
symbol vector,
⎡ ⎤
y1
⎢ ⎥
⎢ y ⎥
⎢ 2 ⎥
y=⎢ . ⎥
⎢ .. ⎥
⎣ ⎦
yr

Transmit antenna 1 x1 y1 Receive antenna 1

Transmit antenna 2 x2 MIMO y2 Receive antenna 2
Channel

Transmit antenna t xt yr Receive antenna r

X = Transmit vector
y = Receive vector

Figure 6.2 MIMO system input-output schematic

This is shown schematically in Figure 6.2. Let the complex coefficient hij represent the fading
channel coefficient between the ith receive antenna and the j th transmit antenna. Thus, there
are a net of rt channel coefficients in this wireless scenario corresponding to all possible
 Multiple-Input Multiple-Output Wireless Communications 167

combinations of the r receive antennas and t transmit antennas. These can be arranged in a
matrix form as
⎡ ⎤
h11 h12 ... h1t
⎢ ⎥
⎢ h h22 ... h2t ⎥
⎢ 21 ⎥
H=⎢ . .. .. .. ⎥
⎢ .. . . . ⎥
⎣ ⎦
hr1 hr2 ... hrt

where the r × t dimensional matrix H is termed the MIMO channel matrix. Let the additive
noise at the receive antenna i be denoted by ni , i.e., n1 , n2 , . . . , nr denote the additive noise
at the r receive antennas. Thus, the net MIMO input output system model can be represented
in vector form as
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
y1 h11 h12 ... h1t x1 n1
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ y ⎥ ⎢ h h22 ... h2t ⎥⎢ x ⎥ ⎢ n ⎥
⎢ 2 ⎥ ⎢ 21 ⎥⎢ 2 ⎥ ⎢ 2 ⎥
⎢ . ⎥=⎢ . .. .. .. ⎥⎢ . ⎥+⎢ . ⎥
⎢ .. ⎥ ⎢ .. . . . ⎥ ⎢ .. ⎥ ⎢ .. ⎥
⎣ ⎦ ⎣ ⎦⎣ ⎦ ⎣ ⎦
yr hr1 hr2 ... hrt xt nr
y H x n

This is succinctly represented using matrix notation as

y = Hx + n

Observe that the receive symbol y1 is given as

y1 = h11 x1 + h12 x2 + . . . + h1txt + n1

from which it can be seen that all the symbols x1, x2 , . . . , xt interfere at y1 received at the
receive antenna 1. Similarly, the receive symbol y2 is given as

y2 = h21 x2 + h22 x2 + . . . + h2txt + n2

from which it can be once again seen that x1 , x2, . . . , xt interfere at y2 received at the receive
antenna 2. This is, in general, true for all the receive antennas, i.e., at each receive antenna i,
the receive symbol yi is a linear of all the transmit symbols x1 , x2, . . . , xt from the t transmit
168 Principles of Modern Wireless Communication Systems

antennas, observed in additive noise ni . For the special case of t = 1, i.e., single transmit
antenna and multiple receive antennas, this is termed as Single-Input Multiple-Output (SIMO)
system or the receive diversity system as seen earlier. This can be modelled as
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
y1 h1 n1
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ y ⎥ ⎢ h ⎥ ⎢ n ⎥
⎢ 2 ⎥ ⎢ 2 ⎥ ⎢ 2 ⎥
⎢ . ⎥=⎢ . ⎥x + ⎢ . ⎥
⎢ .. ⎥ ⎢ .. ⎥ ⎢ .. ⎥
⎣ ⎦ ⎣ ⎦ ⎣ ⎦
yr hr nr
y h n

Similarly, for the case of one receive antenna, i.e., r = 1 and multiple transmit antennas, it is
termed a Multiple-Input Single-Output (MISO) system model or a transmit diversity system.
Its system model is given as
⎡ ⎤
x1
⎢ ⎥
⎢ x ⎥
⎢ 2 ⎥
y= h1 h1 ... ht ⎢ . ⎥ +n
⎢ .. ⎥
⎣ ⎦
hT
xt
x

Finally, for r = t = 1, i.e., a single receive and transmit antenna, it reduces to the single-input
single-output (SISO) system, modelled as

y = hx + n

As the reader might recall, this was the first system model that was introduced to model the
Rayleigh fading wireless -channel-based wireless communication. The covariance matrix of
the noise Rn of the noise vector n defined as

Rn = E nnH
⎧⎡ ⎤ ⎫
⎪
⎪
⎪ n1 ⎪
⎪
⎪
⎪
⎪⎢ ⎥ ⎪
⎪
⎨⎢ n ⎥ ⎬
⎢ 2 ⎥
=E ⎢ . ⎥ n∗1 n∗2 ... n∗L
⎪
⎪⎢ . ⎥ ⎪
⎪
⎪⎣ . ⎦ ⎪
⎪
⎪
⎪
⎩ ⎪
⎭
nL
 Multiple-Input Multiple-Output Wireless Communications 169

⎡ ⎤
E |n1 |2 E {n1 n∗2 } ... E {n1 n∗r }
⎢ ⎥
⎢ 2 ⎥
⎢ E {n2 n∗1 } E |n2 | ... E {n2 n∗r } ⎥
=⎢ .. .. ..
⎥
⎢ .. ⎥
⎢ . . . . ⎥
⎣ ⎦
E {nr n∗1 } E {nr n∗2 } ... E |nr |2
⎡ ⎤
σn2 0 0 ... 0
⎢ ⎥
⎢
⎢ 0 σn2 0 ... 0 ⎥
⎥
=⎢ .. .. .. .. .. ⎥
⎢
⎣ . . . . . ⎥
⎦
0 0 0 ... σn2

= σn2 Ir

The noise vector n with the covariance structure above is termed spatially uncorrelated additive
noise, since the noise samples at the different antennas i, j are independent, i.e., E ni n∗j = 0
if i = j . Finally, to denote the transmission and reception across different time instants, one
can add the time index k to the MIMO system model to frame the net model as
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
y1 (k) h11 h12 ... h1t x1 (k) n1 (k)
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ y (k) ⎥ ⎢ h h22 ... h2t ⎥ ⎢ x (k) ⎥ ⎢ n (k) ⎥
⎢ 2 ⎥ ⎢ 21 ⎥⎢ 2 ⎥ ⎢ 2 ⎥
⎢ .. ⎥=⎢ . .. .. .. ⎥⎢ .. ⎥+⎢ .. ⎥
⎢ . ⎥ ⎢ .. . . . ⎥⎢ . ⎥ ⎢ . ⎥
⎣ ⎦ ⎣ ⎦⎣ ⎦ ⎣ ⎦
yr (k) hr1 hr2 ... hrt xt (k) nr (k)
y(k) H x(k) n(k)

Thus, the vectors y (k) , x (k) , n (k) define the receive, transmit, and noise vectors of the
MIMO wireless communication system at the time instant k. Notice that above we have
assumed the channel matrix H to be constant or, in other words, not dependent on the time
instant k. This is also termed a slow fading or quasi-static channel matrix, indicating that the
channel coefficients are constant over the block of MIMO vectors that are transmitted. Lastly,
we also assume that any two noise samples across two different time instants are uncorrelated,
i.e., E ni (k) n∗j (l) = 0 if k = l . Hence, the noise covariance matrix is given as

E n (k) n (l)H = σ 2 δ (k − l) Ir

where the delta function δ (k − l) = 1 if k = l and 0 otherwise.

170 Principles of Modern Wireless Communication Systems

This noise process, which is uncorrelated across different antennas and time instants is termed
spatio-temporally uncorrelated noise.

6.3 MIMO Zero-Forcing (ZF) Receiver

We will now describe the process to recover the transmitted signal vector x from the received
vector y at the MIMO receiver. This can be considered as solving the system of linear
equations,

y = Hx

where x1 , x2 , . . . , xt are the t unknowns and there are r equations corresponding to the r
observations y1 , y2 , . . . , yr . Consider a simplistic scenario, where r = t, i.e., the number of
receive antennas is equal to the number of transmit antennas. In this case, the matrix H is
square. Further, if the matrix H is now invertible, the estimate x̂ of the transmit vector x is still
given as

x̂ = H−1 ŷ

However, frequently, one has more receive antennas than transmit antennas, i.e., r > t. In this
scenario, the system y = Hx is given as

⎡ ⎤ ⎡ ⎤
y1 (k) h11 h12 ... h1t
⎢ ⎥ ⎢ ⎥⎡ ⎤
⎢ y (k) ⎥ ⎢ h h22 ... h2t ⎥
⎢ 2 ⎥ ⎢ 21 ⎥ x1 (k)
⎢ .. ⎥ ⎢ .
⎥ ⎢ .. .. .. .. ⎥⎢ ⎥
⎢
⎢ . ⎥ ⎢ . . . ⎥ ⎢ x (k) ⎥
⎥⎢ 2 ⎥
⎢ .. ⎥=⎢ . .. .. .. ⎥⎢ . ⎥
⎢ . ⎥ ⎢ .. . . . ⎥⎢ .. ⎥
⎢ ⎥ ⎢ ⎥⎣ ⎦
⎢ .. ⎥ ⎢ .
⎥ ⎢ .. .. .. .. ⎥
⎢
⎣ . ⎦ ⎣ . . . ⎥
⎦ xt (k)
yr (k) hr1 hr2 ... hrt x(k)

y(k) H

from which it can be seen that the matrix H has more rows than columns. Such a matrix is
popularly known as a tall matrix due to its structure. In this situation, one cannot exactly solve
for x since there are more equations r than unknowns t. Hence, one can resort to choosing the
 Multiple-Input Multiple-Output Wireless Communications 171

vector x which minimizes the estimation error error f (x̂),

2
f (x) = y − Hx

The above error function is also termed the least-squares error function and the resulting
estimator is termed the least-squares estimator. To simplify the analysis going forward, we
consider real vectors/matrices y, x, H. The case for complex quantities will be dealt later. The
above error function can be expanded as
2
f (x) = y − Hx ,

= (y − Hx)T (y − Hx)

= y T − xT HT (y − Hx)

= y T y T − xT HT y − y T Hx + xT HT Hx

= y T y − 2xT HT y + xT HT Hx (6.1)
T
where we have used the relation y T Hx = y T Hx = x̂T HT y in the above simplification.
This is due to the fact that y T Hx is a scalar and, hence, is equal to its transpose. To find
the minimum of the error function f (x) with respect to x, we have to set the derivative with
respect to x equal to 0. For this purpose, the concept of a vector derivative is briefly described
below. Consider a multidimensional function g (x). The vector derivative of g (x) with respect
to x is defined as
⎡ ⎤
∂g(x)
∂x1
⎢ ⎥
∂g(x)
∂g (x) ⎢
⎢ ∂x2
⎥
⎥
=⎢ .. ⎥
∂x ⎢ . ⎥
⎣ ⎦
∂g(x)
∂xt

which is basically a t-dimensional vector with the ith component equal to the derivative of
f (x) with respect to xi . This can be better understood with the aid of an example. Consider
any vector c = [c1 , c2 , . . . , ct ]T . Let the function g (x) be defined as

g (x) = cT x = c1 x1 + c2 x2 + . . . + ct xt
172 Principles of Modern Wireless Communication Systems

∂g(x)
Hence, it can be seen that, ∂xi = ci. Therefore, it can be readily deduced that in this case

⎡ ⎤ ⎡ ⎤
∂g(x)
∂x1 c1
⎢ ⎥ ⎢ ⎥
∂g(x)
∂g (x) ⎢
⎢ ∂x2
⎥ ⎢ c
⎥ ⎢ 2
⎥
⎥
=⎢ .. ⎥=⎢ . ⎥=c
∂x ⎢ . ⎥ ⎢ .. ⎥
⎣ ⎦ ⎣ ⎦
∂g(x)
∂xt ct

In fact, it can also be seen that since cT x = xT c, we have

∂cT x ∂xT c
= =c
∂x ∂x

Going back to the expansion of the error function f (x) from Eq. (6.1), it can be seen that
the derivative of each component with respect to x can be computed as follows. Consider the
T
quantity y T y = y 2 . Observe that this does not depend on x. Hence, we have ∂y∂xy = 0.
Consider the component 2xT HT y . This is in the form of xT c, where c = 2HT y . Hence, the
∂(2xT HT y)
derivative of this component with respect to x is given as ∂x = 2HT y . Now, consider
the last component xT HT Hx. This can be differentiated employing the product rule as
∂ xT HT Hx T
= HT Hx + xT HT H
∂x
= 2HT Hx

Hence, employing the above results, the derivative for the error function f (x) can be simplified
as
∂f (x)
= −2HT y + 2HT Hx
x
At the optimal estimate of the transmit vector x̂ where the above error is minimized, we must
have the derivative equal to 0. Using this condition,
∂f (x)
=0
x x=x̂

−2HT y + 2HT Hx̂ = 0

HT Hx̂ = HT y

−1
x̂ = HT H HT y
 Multiple-Input Multiple-Output Wireless Communications 173

Finally, for the case of complex vectors/matrices x, y, H, the transpose in the above expression
can be replaced by the Hermitian operator to yield

−1
x̂ = HH H HH y

The above decoder for the MIMO wireless system to decode the transmitted symbol vector
x from the received symbol vector y is termed the zero-forcing receiver or simply the ZF
receiver. Hence, the zero-forcing decoder can be expressed as

−1
x̂ZF = HH H HH y (6.2)
FZF

−1
The matrix FZF = HH H HH is also termed the zero-forcing receiver matrix and the
estimate x̂ZF is, therefore, given as

x̂ZF = FZF y

6.3.1 Properties of the Zero-Forcing Receiver Matrix FZF

Consider the matrix product FZFH, which can be simplified as
−1
FZFH = HH H HH H
−1
= HH H HH H = I t (6.3)

where It denotes the identity matrix of dimension t. Thus, multiplying the matrix FZF with
the channel matrix indeed produces the identity matrix. In this sense, the matrix FZF acts
as an inverse of the channel matrix H. However, observe that for r > t, the matrix H is
rectangular and strictly speaking does NOT have a matrix inverse. Hence, this matrix FZF is
termed the pseudo-inverse of H. Observe that it is, however, not the inverse of the matrix H in
the usual sense. For instance, if FZF to be the inverse of H, it must also satisfy the property that
HFZF = I. However, we have,
−1
HFZF = H HH H HH = I t

Thus, the matrix product HFZF is not equal to the identity matrix in general. Further, the
matrix inverse is a unique matrix. However, it can be shown that the left or pseudo-inverse
of the matrix is H when r > t is not unique. Thus, in general, FZF is not the inverse of the
174 Principles of Modern Wireless Communication Systems

channel matrix H. However, if r = t and the matrix H is invertible then the pseudo-inverse is
actually equal to the inverse. This can be seen as follows.
−1
FZF = H HH H HH
−1
= H− 1 HH HH
= H− 1

Thus, for this particular case, the pseudo-inverse reduces to the matrix inverse and is unique.

6.3.2 Principle of Orthogonality Interpretation of ZF Receiver

In this section, we present an intuitive-reasoning-based approach to derive the zero-forcing
decoder above. Consider once again the system of equations at the receiver for r > t.
⎡ ⎤
⎡ ⎤⎢ ⎥
y1 (k) ⎢ h h12 ... h1t ⎥
⎢ 11 ⎥
⎢ ⎥ ⎢ ⎥⎡ ⎤
⎢
⎢ y2 (k) ⎥ ⎢
⎥
⎢ h21 h22 ... h2t ⎥ ⎥ x1 (k)
⎢ .. ⎥ ⎢ .
⎥ ⎢ .. .. .. .. ⎥ ⎢ ⎥
⎢
⎢ . ⎥ ⎢ . . . ⎥ ⎥ ⎢ x2 (k) ⎥
⎢ ⎥
⎢ .. ⎥=⎢ . .. .. .. ⎥ ⎢ .. ⎥
⎢ . ⎥ ⎢ .. . . . ⎥ ⎢ . ⎥
⎢ ⎥ ⎢ ⎥⎣ ⎦
⎢ .. ⎥ ⎢ .
⎥ ⎢ .. .. .. .. ⎥
⎥
⎢
⎣ . ⎦ ⎢ . . . ⎥ xt (k)
⎢ ⎥
yr (k) ⎢ h
⎣ r1 hr2 ... hrt ⎥ ⎦ x(k)

y(k) h1 h2 ht

Denoting the column hi = [h1i , h2i , . . . , hri ]T we denote the ith column of the channel
matrix H. Then, the system of equations above can be succinctly represented as

y = x1 h1 + x2 h2 + . . . + xtht
ŷ

where ŷ is the approximation of y and we are interested in minimizing y − ŷ 2 . Observe that

there are t columns of the channel matrix H, which are h1 , h2, . . . , ht . Hence, they represent
a t-dimensional subspace. However, the vector y can lie anywhere in the r -dimensional space,
and is unlikely to lie exclusively in the t-dimensional subspace represented by the columns of
the channel. This is shown schematically in Figure 6.3, where approximation error e = y − ŷ
 Multiple-Input Multiple-Output Wireless Communications 175

the error is shown for different choices of ŷ . It can be clearly seen that the approximation is
minimum, when it is orthogonal to the space spanned by the columns of H. This is termed the
principle of orthogonality, which is extremely helpful in understanding the intuition behind
complex estimation problems. Observe that the error vector e is orthogonal to hi if hH i e = 0.
Therefore, since e is orthogonal to the subspace spanned by the columns of H, it follows that
it is orthogonal to each of the columns of H.
Vector y belonging to a Min. error vector e,
general r-dimensional orthogonal to the
space column space of H

Approximation of
y in the column
t-dimensional space of H
space of columns
of the matrix
. H
Figure 6.3 Zero-forcing: principle of orthogonality

Hence, we have

hH
1 e=0

hH
2 e=0
..
.
hH
t e = 0,

from which it follows naturally that

HH e = 0t×1

We now employ the above principle of orthogonality to derive the expression for the zero-
forcing estimate x̂ZF of the transmitted symbol vector x. This can be derived by substituting
the expression e = y − Hx̂ZF in the above result as

HH (y − Hx̂ZF) = 0
e

HH y − HH Hx̂ZF = 0
176 Principles of Modern Wireless Communication Systems

HH y = HH Hx̂ZF

−1
x̂ZF = HH H HH y

which is exactly identical to the expression for the zero-forcing MIMO decoder derived earlier
in Eq. (6.2). Thus, the principle of orthogonality can be conveniently employed to deduce
the expression for the optimal ZF MIMO decoder while also giving valuable insights into the
decoding procedure.

EXAMPLE 6.1

Compute the MIMO zero-forcing receiver for the channel matrix H given as
⎡ ⎤
2 3
⎢ ⎥
H=⎢
⎣ 1 3 ⎥
⎦ (6.4)
4 2

Solution: It can be seen that the MIMO channel matrix above is of size 3 × 2, implying that
the number of receive antennas is r = 3, while the number of transmit antennas is t = 2. Thus,
the MIMO system model can be described as
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
y1 2 3 ⎡ ⎤ n1
⎣ x1 ⎦ + ⎢ n ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ y ⎥=⎢ 1 3 ⎥
⎣ 2 ⎦ ⎣ ⎦ ⎣ 2 ⎦
x2
y3 4 2 n3
x
y H n

Thus, as can be seen from the model above, the transmit vector x is of dimension 2 × 1, while
the receive and noise vectors y, n respectively are of dimension 3 × 1. The above MIMO
system model can also be explicitly written to describe the signal received at each receive
antenna as

y1 = 2x1 + 3x2 + n1

y2 = x1 + 3x2 + n2
 Multiple-Input Multiple-Output Wireless Communications 177

y3 = 4x1 + 2x2 + n3

which basically represents a system of r = 3 equations for t = 2 unknowns x1 , x2. To compute

−1
the zero-forcing decoder FZF , we first compute the matrix HH H HH . Observe that
HH H can be simplified as
⎡ ⎤
⎡ ⎤ 2 3
2 1 4 ⎦⎢ ⎥
HH H = ⎣ ⎢ 1
⎣ 3 ⎥
⎦
3 3 2
4 2
⎡ ⎤
21 17 ⎦
=⎣
17 22
−1
Hence, the expression for HH H can be simplified as
⎡ ⎤
−1 1 22 −17 ⎦
HH H = ⎣
21 × 22 − 17 × 17 −17 21
⎡ ⎤
1 ⎣ 22 −17 ⎦
=
173 −17 21
−1
Thus, the zero-forcing receiver matrix FZF = HH H HH can be expressed as
−1
FZF = HH H HH
⎡ ⎤⎡ ⎤
1 ⎣ 22 −17 ⎦ ⎣ 2 1 4 ⎦
=
173 −17 21 3 3 2
⎡ ⎤
1 ⎣ −7 −29 54 ⎦
=
173 29 46 −26
⎡ ⎤
−0.04 −0.17 0.31 ⎦
=⎣ (6.5)
0.17 0.27 −0.15
178 Principles of Modern Wireless Communication Systems

Thus, the zero-forcing estimate of the transmit vector is given as

x̂ZF = FZF y
−1
= HH H HH y
⎡ ⎤
⎡ ⎤ y1
−0.04 −0.17 0.31 ⎢ ⎥
=⎣ ⎦⎢ y ⎥
⎣ 2 ⎦
0.17 0.27 −0.15
y3
These expressions can now be explicitly written for the estimate of each of the transmit
symbols x̂1, x̂2 as

x̂1 = −0.04y1 − 0.17y2 + 0.31y3

x̂2 = +0.17y1 + 0.27y2 − 0.15y3

It can also be confirmed that the zero-forcing receiver matrix matrix FZF is of dimension
2 × 3, i.e., t × r .
One of the chief disadvantages of the zero-forcing receivers is noise amplification. This
can be understood as follows. Consider the SISO wireless system for r = t = 1. We have the
system model given as

y = hx + n

Hence, the zero-forcing receiver is given as fZF = (h∗ h)−1 h∗ = h−1 = h1 . Hence, the zero-
forcing estimate of the transmitted symbol x is given as

x̂ZF = fZF × y

1
= (hx + n)
h
n
=x+ (6.6)
h
Observe that if the fading coefficient h is close to zero then the factor nh is very high. This is
termed noise amplification, which significantly distorts the performance of the receiver. The
MMSE receiver which is proposed next addresses this problem.
 Multiple-Input Multiple-Output Wireless Communications 179

6.4 MIMO MMSE Receiver

In this section, we develop the Minimum Mean-Squared Error (MMSE) receiver for the
MIMO wireless communication system. The MMSE receiver is based on a Bayesian approach,
meaning that the transmit vector x is assumed to be random in nature. Thus, if x̂MMSE denotes
the estimated symbol vector, the MMSE receiver minimizes the average or mean of the squared
error

2
E x̂MMSE − x

Thus, it is aptly named minimum mean-squared error estimator. To illustrate the development
of the MMSE receiver, we consider a single-input multiple-output (SIMO) wireless system,
i.e., t = 1, and generalize the result to the case of a MIMO system. Hence, consider the SIMO
system model given as

y = hx + n

where x is now a scalar transmitted symbol. Thus, the basic problem can be interpreted as
estimating the symbol x given the vector y = [y1 , y2 , . . . , yr ]T . Let c = [c1 , c2, . . . , cr ]T .
One can now define a linear estimator of x as

x̂ = cT y (6.7)
⎡ ⎤
y1
⎢ ⎥
⎢ y ⎥
⎢ 2 ⎥
= c1 c2 ... cr ⎢ . ⎥
⎢ .. ⎥
⎣ ⎦
yr

= c1 y1 + c2 y2 + . . . + cr yr

Such an estimator as above is termed as a linear estimator, since the estimate is a linear function
of y . We now wish to find the best or optimal linear estimator x̂MMSE , which minimizes the
mean squared error and hence is also termed the Linear Minimum Mean Squared Estimator
(LMMSE). This is frequently also referred to simply as the MMSE, although strictly speaking
belongs to the specific class of linear estimators. The average mean squared error is defined as

2
E x̂ − x
180 Principles of Modern Wireless Communication Systems

Employing the form in Eq. (6.7), the resulting equation can be simplified as

2 T
E x̂ − x =E cT y − x cT y − x

=E cT y − x y T c − xT

= E cT yy T c − xy T c − cT yx + xxT

= cT E yy T c − E xy T c − cT E {yx} +E x2 Rxx
Ryy Rxy Ryx

= cT Ryy c − 2cT Ryx + Rxx

where the covariance matrix Ryy is defined as Ryy = E yy H . Similarly, Ryx = E {yx}
T
= RTxy and Rxx = E x2 . Also note that we have used the fact cT Ryx = cT Ryx =
Rxy c. Hence, the average MSE as a function of the receive beamformer c, denoted by
MSE (c), is given as

MSE (c) = cT Ryy c − 2cT Ryx + Rxx

Thus, the optimal beamformer c which minimizes the average or mean squared error can be
obtained by differentiating MSE (c) with respect to c and setting equal to zero as

∂ MSE (c)
=0
∂c
∂
cT Ryy c − 2cT Ryx + Rxx = 0
∂c

2Ryy c − 2Ryx = 0

1
c = R−
yy Ryx

1
Thus, the optimal LMMSE beamforming vector c is given as c = R− yy Ryx . This is also termed
in signal processing as the optimal Wiener filter. The above can be generalized in the case of
complex vectors by replacing the transpose by the Hermitian operation.
 Multiple-Input Multiple-Output Wireless Communications 181

Hence, the MMSE estimate of x is given as

x̂MMSE = cH y

1 H
= R−
yy Ryx y

1
= Rxy R−
yy y

We now compute the MMSE receiver for the MIMO wireless system. Consider again the
MIMO system model given as

y = Hx + n

Let the transmit symbols xi , 1 ≤ i ≤ t be such that each is of power Pd , i.e., E |xi |2 =
Pd , with elements on different transmit antennas being uncorrelated, i.e., E xi x∗j = 0 when
i = j . Hence, the covariance Rxx of the transmit symbols is given as

Rxx = E xxH
⎧⎡ ⎤ ⎫
⎪
⎪
⎪ x1 ⎪
⎪
⎪
⎪⎢
⎪ ⎥ ⎪
⎪
⎨⎢ x ⎥ ⎬
⎢ 2 ⎥
=E ⎢ . ⎥ x∗1 x∗2 ... x∗t
⎪
⎪⎢ .. ⎥ ⎪
⎪
⎪
⎪⎣ ⎦ ⎪
⎪
⎪
⎩ ⎪
⎭
xt
⎡ ⎤
E |x1 |2 E {x1 x∗2 } ... E {x1 x∗t }
⎢ ⎥
E |x2 |2
⎢ ⎥
⎢ E {x2 x∗1 } ... E {x2 x∗t } ⎥
=⎢ .. .. ..
⎥
⎢ .. ⎥
⎢
⎣ . . . . ⎥
⎦
E {xt x∗1 } E {xt x∗2 } ... E |xt |2
⎡ ⎤
Pd 0 0 ... 0
⎢ ⎥
⎢ 0 Pd 0 ... 0 ⎥
⎢ ⎥
=⎢ .. .. .. .. .. ⎥
⎢
⎣ . . . . . ⎥
⎦
0 0 0 ... Pd

= Pd It
182 Principles of Modern Wireless Communication Systems

Hence, Ryy , the covariance of the receive vector y can be simplified as

Ryy = E yy H

= E (Hx + n) (Hx + n)H

= E HxxH H + nxH HH + HxnH + nnH

= H E xxH H + E nxH HH + H E xnH + E nnH

Rxx 0 0 Rnn

= Pd HHH + σn2 Ir
Ryy

where we have used the fact that E nxH = E xnH = 0, since the noise at the
receiver and transmit symbols are uncorrelated, in the above simplification. Further, the
cross-covariance matrix Ryx can be simplified as

Ryx = E yxH

= E (Hx + n) xH

= E HxxH + nxH

= H E xxH + E nxH
Rxx 0

= Pd H

Thus, the optimal MMSE receiver is given as

1
C = R−
yy Ryx

−1
= Pd HHH + σn2 I Pd H

−1
= Pd Pd HHH + σn2 I H

Hence, the MMSE estimate x̂MMSE of the transmit vector x is given as

x̂MMSE = CH y

−1
= Pd HH Pd HHH + σn2 I y (6.8)
 Multiple-Input Multiple-Output Wireless Communications 185

at HIGH MIMO ZF
SNR receiver

MIMO LMMSE
receiver

at LOW MIMO
SNR Matched Filter
(MF) receiver

Figure 6.4 MIMO MMSE receiver: asymptotic behaviour

6.5 Singular Value Decomposition (SVD) of the MIMO Channel

In this section, we will begin to explore the Singular Value Decomposition (SVD) of the MIMO
channel matrix H, which is a very important tool to understand the behaviour of a MIMO
wireless communication system. Consider an r × t MIMO channel H with r ≥ t, i.e., number
of receive antennas greater than or equal to the number of transmit antennas. The SVD of the
channel matrix H is given as
⎡ ⎤
⎢ ⎥⎡ ⎤
⎢ u11 u 12 . . . u 1t ⎥ σ 1 0 . . . 0
⎢ ⎥⎢ ⎥
⎢ ⎥⎢ ⎥
⎢ u21 u 22 . . . u 2t ⎥⎢ 0 σ 2 . . . 0 ⎥
H=⎢ . ⎢
.. ..
⎥
.. ⎥ ⎢ .. .. .. .. ⎥
⎢ . . . . . ⎥ ⎢ . . ⎥
⎢ ⎥⎣ . . ⎦
⎢ ⎥
⎣ ur1 ur2 ... urt ⎦ 0 0 ... σt
u1 u2 ut Σ

⎡ ⎤
⎢
∗
v11 ∗
v21 ... ∗
vt1 v1H ⎥
⎢ ⎥
⎢
⎢
∗
v12 ∗
v22 ... ∗
vt2 v2H ⎥
⎥
⎢ .. .. .. ⎥,
⎢ .. ⎥
⎢ . . . . ⎥
⎣ ⎦
∗
v1t ∗
v2t ... ∗
vtt vtH

VH

= UΣV H (6.10)

where the matrices U, Σ, V , which are r × t, t × t and t × t dimensional respectively, satisfy

important properties, which we examine next. The columns of the matrix U and V are unit-
186 Principles of Modern Wireless Communication Systems

norm, i.e., we have

2 2
ui = vi = 1, 1 ≤ i ≤ t

Further, the columns of matrix U and V are orthogonal, i.e.,

uH H
i uj = vi vj = 0, i = j, 1 ≤ i, j ≤ t

Thus, the columns of matrices U and V are orthonormal. As a result, the t × t dimensional
square matrix V is unitary, i.e.,

V H V = VV H = It
Further, since if r = t, the matrix U is also a unitary matrix. Otherwise, U simply satisfies the
relation UH U = It. Further, the quantities σ1 , σ2 , . . . , σt are known as the singular values of
the matrix Σ. These singular values are non-negative and ordered, i.e., each σi ≥ 0 and

σ1 ≥ σ2 ≥ . . . ≥ σt
Finally, an important property of the singular values is that the number of nonzero singular
values is equal to the rank of the matrix H. Below, we illustrate some examples of singular
value decomposition to give insights into the structure of the SVD.

6.5.1 Examples of Singular Value Decomposition

EXAMPLE 6.2

Consider a 2 × 1 SIMO wireless system with channel matrix H given as

⎡ ⎤
1
H=⎣ ⎦ (6.11)
1

This can be simplified as

⎤⎡ ⎡ ⎤
1 √1 √
H=⎣ ⎦,= ⎣ 2 ⎦ 2
1
1 √
2
⎡ ⎤
√1 √
=⎣ 2 ⎦
1
2 1
√
2
Σ VH
U
 Multiple-Input Multiple-Output Wireless Communications 187

This is a simple example of the SVD and yet illustrates several important properties. For
T
instance, U = √1 , √1 , which is of dimension r × t, i.e., 2 × 1. Further, U only has a single
2 2
T
column u1 = √12 , √12 , which is unit-norm, i.e., u1 2 = 1. Further, the singular value
√
σ1 = 2, which is greater than 0. Also, the number of nonzero singular values is 1, which
is equal to the rank of the matrix. The rank of the matrix is easily seen to be equal to 1 in this
case since it is simply a column vector. Also observe that V = [1]. Hence, we trivially have

V H V = VV H = 1 = I1

Thus, the above decomposition satisfies all the properties of the SVD. Next, we define a slightly
more nuanced example to illustrate the concept of SVD.

EXAMPLE 6.3

Let the 2 × 2 channel matrix H be given as

⎡ ⎤
1 0
H=⎣ √ ⎦
0 5

In this example, it is very intuitive to simplify H simply as

⎡ ⎤⎡ ⎤⎡ ⎤
1 0 1 0 1 0 ⎦
H=⎣ ⎦⎣ √ ⎦⎣
0 1 0 5 0 1
U Σ VH

√
and claim that σ1 = 1, σ2 = 5 are the singular values of H. However, this is incorrect
since σ1 < σ2, meaning that the above is not a valid SVD. However, one can recast H as
follows.
⎡ ⎤⎡ ⎤⎡ ⎤
√
0 1 ⎦⎣ 5 0 ⎦⎣ 0 1 ⎦
H=⎣
1 0 0 1 1 0
U Σ VH
188 Principles of Modern Wireless Communication Systems

where the matrices U, V are 2 × 2 permutation matrices which flip the rows and columns of
the inner diagonal matrix respectively. It can also be observed that VV H = V H V = It . Also,
√
in this case, since r = t, we also have UUH = UH U = It. Further, σ1 = 5 > σ2 = 1 > 0.
Thus, the singular values are positive and ordered. Hence, this is a valid SVD of the channel
matrix H. This example demonstrates how an otherwise seemingly simple diagonal matrix can
have an SVD that is not straightforward to compute.

EXAMPLE 6.4

We now look at another example of a channel matrix H and compute its SVD. Let H be
given as
⎡ ⎤
1 2
H=⎣ ⎦
1 −2

Observe that in this case, the columns of H, i.e., h1 = [1, 1]T and h2 = [2, −2]T are
orthogonal since hH
1 h2 = 1 × 2 + 1 × (−2) = 0. This fact can be employed to compute the
SVD. Thus, the matrix H can be decomposed in steps as
⎡ ⎤
1 2
H=⎣ ⎦
1 −2
⎡ ⎤ ⎡ ⎤
1 1 ⎦×⎣ 1 0 ⎦
=⎣
1 −1 0 2
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
1 1 ⎦ ⎣ 2 0 ⎦ ⎣ 0 1 ⎦
=⎣ × ×
−1 1 0 1 1 0
The above decomposition has a structure that looks very close to the SVD. It now remains to
normalize the columns of the left and right matrices to generate orthonormal vectors. This can
be done as follows.
⎡ ⎤⎡ √ ⎤⎡ ⎤
√1 √1
H=⎣ 2 2 ⎦⎣ 2 2 0 0 1 ⎦
1 1
√ ⎦⎣
− √2 √
2
0 2 1 0
U Σ VH
 Multiple-Input Multiple-Output Wireless Communications 189

√ √
It can be seen that the singular values are σ1 = 2 2 > σ2 = 2 > 0. Further, the number of
nonzero singular values and, hence, the rank is 2. Since r = t = 2, the matrices U, V are both
square 2 × 2, unitary and contain orthonormal columns.

6.6 Singular Value Decomposition and MIMO Capacity

The singular value decomposition is central to understanding the spatial multiplexing

properties of the MIMO channel and deriving the fundamental limit on the capacity of the
MIMO channel. Consider the r × t MIMO wireless system,

y = Hx + n

Let the SVD of the channel matrix H be given as H = UΣV H . Thus, replacing H with its
SVD, the above MIMO system model is given as

y = UΣV H x + n
H

At the receiver, multiplying by UH , we have

UH y = UH UΣV H x + n
ỹ

ỹ = UH U ΣV H x + UH n
It ñ

= ΣV H x + ñ

This operation of multiplying by UH at the receiver is a part of the signal processing at the
receiver or receive processing. Further, prior to transmission, let the transmit vector x be
generated as x = Vx̃, where the vector x̃ contains the transmit symbols. This operation is
termed as transmit precoding. Thus, substituting this expression for x above, we have

ỹ = ΣV H x + ñ
190 Principles of Modern Wireless Communication Systems

= ΣV H Vx̃ + ñ

= Σx̃ + ñ

The above equivalent system model of the MIMO system after the receive and transmit
processing operations can be explicity written as
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
ỹ1 σ1 0 ... 0 x̃1 ñ1
⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥
⎢ ỹ ⎥ ⎢ 0 σ2 ... 0 ⎥ ⎢ x̃ ⎥ ⎢ ñ ⎥
⎢ 2 ⎥ ⎢ ⎥⎢ 2 ⎥ ⎢ 2 ⎥
⎢ . ⎥=⎢ .. .. .. .. ⎥⎢ . ⎥+⎢ . ⎥
⎢ .. ⎥ ⎢ . . . . ⎥ ⎢ .. ⎥ ⎢ .. ⎥
⎣ ⎦ ⎣ ⎦⎣ ⎦ ⎣ ⎦
ỹt 0 0 ... σt x̃t ñt

The above system can, in fact, be written in a much simpler decoupled form as

ỹ1 = σ1 x̃1 + ñ1

ỹ2 = σ2 x̃2 + ñ2

..
.
ỹt = σt x̃t + ñt (6.12)

Thus, the above equivalent system represents the parallelization of the MIMO channel with
t information streams being transmitted in parallel. As can be seen, there is no interference
between the t information streams carrying symbols x̃1 , x̃2 , . . . , x̃t. This is termed spatial
multiplexing, where the t independent information streams are being multiplexed over the
multiple spatial dimensions, arising due to the presence of multiple transmit and receive
antennas in the system. Also, the covariance of the modified noise ñ can be derived as

Rñ = E ññH

H
= E UH n UH n

= E UH nnH U

= UH E nnH U
2I
σn
 Multiple-Input Multiple-Output Wireless Communications 191

= UH σn2 IU = σn2 UH U
It

= σn2 It (6.13)

Thus, once again, the noise ñ has a covariance proportional to the identity matrix, indicating
equal variance, uncorrelated noise components. Further, the variance of each noise component
ñi , 1 ≤ i ≤ t is equal to σn2 . Consider now the ith parallel MIMO channel above. This is
given as

ỹi = σi x̃i + ñi .

E{|xi |2 }
Hence, the SNR of the system is given as σi2 σ2 = σi2 σP2i , where Pi is the power of the ith
n n

data stream xi . Thus, the MIMO system can be viewed as a collection of t parallel channels
each with noise power σn2 and power gain σi2. This is schematically shown in Figure 6.5. From
the above expression for the SNR of the ith channel, the Shannon capacity Ci of the channel

s1 n1

x1 y1

s2 n2

x2 y2

st nt

xt yt

Figure 6.5 MIMO SVD parallel channels

can be derived as

Pi σi2
Ci = log2 1 +
σn2
192 Principles of Modern Wireless Communication Systems

Thus, the MIMO system can be thought of as a collection of t parallel data pipes, with
capacities
P1 σ12
C1 = log2 1 +
σn2

P2 σ22
C2 = log2 1 +
σn2

..
.
Pt σt2
Ct = log2 1 +
σn2
Thus, the net MIMO C capacity is given as the sum of the individual capacities

t
Pi σi2
C= log2 1 +
σi2
i=1

The total power P at the transmitter can be allocated to the individual streams to maximize the
net capacity. Thus, one can maximize the above sum capacity subject to the power constraint

P1 + P2 + . . . + Pt ≤ P

Next, we describe computation of the optimal MIMO capacity.

6.6.1 Optimal MIMO Capacity

Thus, the optimal MIMO power allocation problem can be formulated as
t
Pi σi2
max . log2 1 +
i=1
σi2

t
s.t. Pi ≤ P
i=1

where s.t. in the above problem statement stands for subject to and denotes the optimization
constraint. We employ the standard Lagrange-multiplier-based technique for the above
constrained optimization problem. Denoting the Lagrange multiplier by λ, the Lagrangian cost
 Multiple-Input Multiple-Output Wireless Communications 193

function f P , λ for the above optimization problem can be formulated as

t t
Pi σi2
f P, λ = log2 1 + +λ P − Pi
i=1
σi2 i=1

where P = [P1 , P2 , . . . , Pt ]T . Differentiating f P , λ with respect to Pi and setting equal to

0, we obtain
∂
f P, λ = 0
∂Pi
σi2
2
σn
P σ2
−λ=0
1 + σi 2 i
n

Solving the above equation yields

σi2 1 Pi σi2
= 1 +
σn2 λ σn2
+
1 σn2
Pi = −
λ σi2
where the function x+ = x if x ≥ 0 and 0 otherwise. This is because of the fact that each
power Pi ≥ 0, i.e., power cannot be negative. It now remains to find the Lagrange multiplier
λ, which can be found from the constraint equation as
t
Pi = P
i=1

t +
1 σn2
− =P (6.14)
λ σi2
i=1

The above optimal power allocation is also termed water filling. This can be seen as follows.
2
Consider a vessel with t bars and the height of the ith bar equal to σσn2 . It water is poured into
! i 2
"+
this vessel to the level λ then the level of water at the i bar is λ1 − σσn2 . This is shown
1 th
i

schematically in Figure 6.6. Observe that the power allocated is proportional to the singular
value, i.e., larger the σi , larger is the power allocated. Further, also observe that due to the
nature of water filling, weak channels with low σi are not allocated any power.
194 Principles of Modern Wireless Communication Systems

Figure 6.6 MIMO water-filling capacity

Observe that the water-filling equation in Eq. (6.14) above is nonlinear due to the x+
2
function. It can be solved iteratively as follows. Set N = t initially. Assume λ1 − σσn2 for
i

1 ≤ i ≤ N . Solve the equation

N
1 σ2
− n2 =P
λ σi
i=1

2
1 σn
Now, check if λ − 2
σN ≥ 0. If this is the case then the λ computed yields the desired power
2
allocation. However, if λ1 − σσ2n < 0, then set PN = 0 and N = t − 1 and repeat the process as
N
above. Example clarifies this concept of MIMO capacity and optimal power allocation.

EXAMPLE 6.5

Consider the MIMO channel matrix H given as

⎡ ⎤
2 −6 0
⎢ ⎥
H=⎢
⎣ 3 4 0 ⎥
⎦
0 0 2

Considering a transmit power of P = −1.25 dB and noise power σn2 = 3 dB, compute the
MIMO capacity and optimal power allocation.
 Multiple-Input Multiple-Output Wireless Communications 195

Solution: The above MIMO channel matrix is 3 × 3, i.e., a MIMO system with r = t = 3
antennas. We also have, σn2 = 3 dB = 2 and P = −1.25 dB = 0.75. Observe that the columns
of the channel matrix H are orthogonal. For instance, if c1 , c2 denote the first two columns of
H, we have

cH
1 c2 = 2 × (−6) + 3 × (4)

=0

Hence, the SVD of the channel matrix H is given as

⎡ ⎤
2 −6 0
⎢ ⎥
H = ⎢ 3
⎣ 4 0 ⎥
⎦
0 0 2
⎡ ⎤
√2 − √652 0 ⎥
⎢ 13
⎢ ⎥
= ⎢ √3 √4 0 ⎥
⎣ 13 52 ⎦
0 0 1
⎡ ⎤⎡
√ ⎤⎡ ⎤
− √652 √2
13
0 52 0 0 0 1 0
⎢ ⎥⎢ √ ⎥⎢ ⎥
⎢ ⎥⎢
= √4 3
0 ⎥⎣ 0 13 0 ⎥ ⎢ 0 ⎥
⎢
⎣ 52
√
13 ⎦ ⎦⎣ 1 0 ⎦
0 0 1 0 0 2 0 0 1
U Σ VH

Thus, the singular values σ1 , σ2, σ3 for the above channel matrix are given as

√
σ1 = 52 ⇒ σ12 = 52
√
σ2 = 13 ⇒ σ22 = 13

σ3 = 2 ⇒ σ32 = 4

It can be seen that the channel matrix H has 3 nonzero singular values. Hence, the rank of the
channel matrix is 3. The channel capacity C in terms of the powers P1 , P2 , P3 assigned to the
196 Principles of Modern Wireless Communication Systems

different channel singular modes is given as

P1 σ12 P2 σ22 P3 σ32

C = log2 1 + + log2 1 + + log2 1 +
σn2 σn2 σn2

P1 × 52 P2 × 13 P3 × 4
= log2 1 + + log2 1 + + log2 1 +
2 2 2

The above expression for capacity has to be maximized for total power P1 + P2 + P3 = 0.75.
We set N = t = 3 and solve the Lagrangian in Eq. (6.14) as

1 1 1 2 1 1
− + − + − = 0.75
λ 26 λ 13 λ 2
1 2 1
1 0.75 + 26 + 13 + 2
=
λ 3
1
= 0.48
λ

Let us now compute the power allocation to the channel 3, i.e., P3, which is given as,
P3 = λ1 − 12 = 0.48 − 0.5 = −0.02 ≤ 0. Thus, since the power P3 is coming out to be
negative, this is not a possible allocation. This implies that the power bar corresponding to
the channel 3 lies above the water level λ1 , as per the schematic shown in Figure 6.6. Thus, the
singular mode 3 is allotted 0 power in the optimal allocation. Hence, we now set N = 2 and
resolve the equation for the Lagrangian variable λ as

1 1 1 2
− + − = 0.75
λ 26 λ 13
1 2
1 0.75 + 26 + 13
=
λ 2
= 0.4712

Recomputing the powers P1 , P2 , we have

1
P1 = 0.4712 − = 0.4327 > 0
26
2
P2 = 0.4712 − = 0.3174 > 0
13
 Multiple-Input Multiple-Output Wireless Communications 197

Thus, since P1 , P2 > 0, this is a feasible power allocation. Hence, we have, P1 = 0.4327
= −3.63 dB and P2 = 0.3174 = −4.98 dB. As already described, P3 = 0. Therefore, the
capacity C is given as

52 × 0.4327 13 × 0.3174
Cmax = log2 1 + + log2 1 +
2 2

= 5.23 b/s/Hz

where the units b/s/Hz is read as bits per second per hertz. The optimal MIMO transmission
scheme can now be derived as follows. Observe from the SVD of H = UΣV H computed
above that the matrix V is given as
⎡ ⎤
⎢ ⎥
⎢ 0 1 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 1 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎣ 0 0 1 ⎦
v1 v2 v3

As described in Section 6.6, the transmit vector x is given as

⎡ ⎤
x̃1
⎢ ⎥
x = v1 v2 v3 ⎢ x̃ ⎥
⎣ 2 ⎦
x̃3

= v1 x̃1 + v2 x̃2

√ √
where x̃3 = 0 since P3 = 0. Also, we have P1 = 0.66 and P2 = 0.56. Therefore, the net
transmit vector x is given as
⎡ ⎤ ⎡ ⎤
0 1
⎢ ⎥ ⎢ ⎥
x = ⎢ 1 ⎥ x̃1 + ⎢ 0 ⎥ x̃2
⎣ ⎦ ⎣ ⎦
√ √
0 P1 b1 0 P2 b2
198 Principles of Modern Wireless Communication Systems

⎡ ⎤ ⎡ ⎤
0 1
⎢ ⎥ ⎢ ⎥
= ⎢ 1 ⎥ 0.66 b1 + ⎢ 0 ⎥ 0.56 b2
⎣ ⎦ ⎣ ⎦
0 0

where b1, b2 are unit power-transmit symbols belonging to an appropriate transmit

constellation such as BPSK, QPSK, etc. Thus, one can compute the optimal power allocation
corresponding to the capacity of the MIMO channel to derive the optimal precoded transmit
vectors for the MIMO system.

6.7 Asymptotic MIMO Capacity

We now describe a simple scenario to illustrate the asymptotic capacity of the MIMO channel.
Let the transmit covariance matrix Rx = E x (k) xH (k) . It can then be demonstrated that
the capacity of the MIMO channel, for a general transmit covariance matrix Rx is given as

1
C = log2 I + HRx HH
σn2

where the notation |A| above denotes the determinant of the matrix A. Consider now equal
power allocation, i.e., transmit covariance given as Rx = Ptt I, where the transmit power Pt
is equally allocated to all the t transmit antennas. Also, let us look at a special scenario of
t >> r , corresponding to a much larger number of transmit than receive antennas. Substituting
the value of Rx = Ptt I, the above MIMO capacity C reduces to

Pt
C = log2 I + HHH
tσn2

Let us now examine the structure of the matrix HHH . Let the r columns of the matrix HH be
denoted by h1 , h2, . . . , hr .
 Multiple-Input Multiple-Output Wireless Communications 199

Hence, HHH can be simplified as

⎡ ⎤
hH
1
⎢ ⎥
⎢ hH ⎥
2
HHH
⎢ ⎥
= ⎢ . ⎥ h1 h2 ... hr
⎢ .. ⎥
⎣ ⎦
hH
r

⎡ ⎤
hH
1 h1 hH
1 h2 ... hH
1 hr
⎢ ⎥
⎢ hH h hH ... hH ⎥
⎢ 2 1 2 h2 2 hr ⎥
= ⎢ .. .. .. .. ⎥
⎢
⎣ . . . . ⎥
⎦
hH
r h1 hH
r h2 ... hH
r hr

Observe now that all the vectors hi , 1 ≤ i ≤ r are t-dimensional. Further, considering a rich
scattering environment, they consist of uncorrelated random flat-fading channel coefficients.
Hence, we have

t
hH
i hi = |hik |2
k=1

→ tE |hik |2

≈ t

where we have assumed that all channel coefficients hik and independent identically distributed
with E |hik |2 = 1. Further, the quantities hH
i hj , for i = j can be approximated as

t
hH
i hj = h∗ik hjk
k=1

→ 0

≈ 0
200 Principles of Modern Wireless Communication Systems

since the elements hik and hjk are uncorrelated. Hence, the matrix HHH for t >> r , can be
approximated as
⎡ ⎤
t 0 ... 0
⎢ ⎥
⎢ 0 t ... 0 ⎥
H ⎢ ⎥
HH = ⎢ . .. .. .. ⎥
⎢ .. . . . ⎥
⎣ ⎦
0 0 ... t

= tIr

Therefore, the asymptotic MIMO capacity Ca is given as

Pt Pt
log2 I + × tIr = log2 I + Ir
tσn2 σn2

Pt
= r log2 1 +
σn2

Pt
The above simplification can be better seen as follows. Consider Ir + 2 Ir ,
σn which can be
written as
⎡ ⎤
Pt
1+ 2
σn 0 ... 0
⎢ ⎥
⎢ ⎥
Pt ⎢ 0 1 + σP2t ... 0 ⎥
Ir + 2 Ir = ⎢ .. ..
n
.. ..
⎥ (6.15)
σn ⎢
⎢ . . . .
⎥
⎥
⎣ ⎦
Pt
0 0 ... 1+ 2
σn

! "r
It can be readily seen that the determinant of the above matrix is 1 + σP2t . Hence, the
n
! "r ! "
Pt Pt
capacity is simplified as, log2 1 + σ2 = r log2 1 + σ2 . Also, it can be seen that in this
n n

case, since t >> r , we have min {r, t} = r . The asymptotic capacity can, therefore, also be
written as

Pt
Ca = (min {r, t}) log2 1 +
σn2

from which it can be seen that the capacity increases linearly with the minimum of the number
of transmit or receive antennas, which also indicates the number of degrees of freedom of
 Multiple-Input Multiple-Output Wireless Communications 201

the MIMO system. Further, also observe that this gain is achieved with a constant transmit
power Pt , i.e., one does not need to increase the net transmit power to achieve a higher
throughput using MIMO. This indicates the power of employing MIMO in broadband wireless
communication systems. The throughput can be increased significantly without any need for a
higher transmit power.

6.8 Alamouti and Space-Time Codes

In this section, we describe a power set of codes which are termed space-time codes for
error protection coding in MIMO communication systems. In a traditional error-control coding
framework, the block code is applied only over the time dimension or basically over a block
of concatenated symbols. However, due to the nature of the MIMO system, one can exploit
the spatial dimension as well. That is to say, the one can additionally encode the symbols
over the spatial dimension or across the multiple antennas, in addition to coding over the
time dimension. This gives rise to the paradigm of space-time encoding, which leads to a
significantly superior performance in MIMO systems and multiple-antenna systems in general.
We begin with a basic introduction to the Alamouti code which is described for a 1 × 2
system, i.e., for a system with r = 1 receive antenna and t = 2 transmit antennas. This is
an example of a MISO system, which is a special case of the MIMO system, and has been
described in the beginning of this chapter. Let the 1 × 2 channel matrix be denoted by [h1 , h2 ],
where h1 , h2 denote the channel coefficients between transmit antennas 1, 2 and the single
receive antenna respectively. Hence, the system model can be represented as
⎡ ⎤
x
⎣ 1 ⎦+n
y= h1 h2 (6.16)
x2

where x1 , x2 are the symbols transmitted from the two transmit antennas and n denotes the
additive white Gaussian noise at the receiver. Before we discuss the Alamouti code, let us
discuss another possible mode of transmission, which is termed beamforming. Consider now
∗
a symbol x which is transmitted as follows. Let x1 be generated as hh1 x and similarly, x2 be
h∗2
generated as x2 = h x. Here, h denotes the norm of vector h and is defined as
202 Principles of Modern Wireless Communication Systems

#
h = |h1 |2 + |h2 |2. Thus, the transmit vector is now given as

⎡ ⎤ ⎡ ⎤
h∗1
⎣ x1 ⎦ = ⎣ h ⎦x
h∗2
x2 h

This is termed transmit beamforming, i.e., transmitting the symbol x in the direction given by
the vector
⎡ ⎤
h∗1
⎣ h ⎦.
h∗2
h

Substituting the above expression for the beamformed symbol in Eq. (6.16), the output y is
given as
⎡ ⎤
h∗1
h
y = h1 h2 ⎣ ⎦ x + n,
h∗2
h

|h1|2 |h2 |2
= + x + n,
h h

= h x+n

Observe that the SNR for the above system is, therefore, given as

h 2P
SNR = (6.17)
σn2

which is identical to that of the receive diversity with Maximum Ratio Combining (MRC).
Thus, the above example seems to indicate that the performance that can be achieved through
multiple antennas at the transmitter is equivalent to that achieved with multiple antennas at the
receiver. However, a close examination reveals the following point. Consider again the transmit
vector given as
⎡ ⎤ ⎡ ⎤
h∗1
⎣ x1 ⎦ = ⎣ h ⎦x
h∗2
x2 h
 Multiple-Input Multiple-Output Wireless Communications 203

Observe that to perform this at the transmitter requires knowledge of the channel coefficients
h1 , h2 at the transmitter, which is termed Channel State Information (CSI) in the context
of wireless communications. However, the channel coefficients h1 , h2 are estimated at the
receiver. To implement beamforming at the transmitter, information about these channel
coefficients has to be fed back to the transmitter. This is a challenging task. Therefore, one
cannot always count on possessing the knowledge of the channel at the transmitter. The
Alamouti space-time code is an ingenuous scheme which overcomes this constraint through a
novel transmission procedure and is described next.

6.8.1 Alamouti Code: Procedure

The Alamouti code is a space-time code proposed for a 1 × 2 MISO system. The interesting
aspect of the Alamouti-code is that it achieves a diversity order of 2 without CSI at the
transmitter. Consider two symbols x1 , x2 . In an Alamouti-coded system, in the first transmit
instant, the symbol x1 is transmitted from the transmit antenna 1, while x2 is transmitted from
the transmit antenna 2. Therefore, the transmit symbol vector in the first time instant is given
as
⎡ ⎤
x
⎣ 1 ⎦
x2

Further, the received symbol y(1) at the receiver corresponding to this transmission is given as
⎡ ⎤
x
⎣ 1 ⎦ + n(1)
y(1) = h1 h2 (6.18)
x2

In the second time instant, the symbol −x∗2 is transmitted from the first transmit antenna, while
x∗1 is transmitted from the second transmit antenna. As we will see later, this is the unique
aspect of the Alamouti code which enables it to achieve diversity gain of the order 2 at the
receiver. Therefore, the transmit symbol vector in the second time instant is given as
⎡ ⎤
∗
⎣ − x 2 ⎦
x∗1
204 Principles of Modern Wireless Communication Systems

Further, the received symbol y(2) can be expressed as

⎡ ⎤
−x∗2
y(2) = h1 h2 ⎣ ⎦ + n(2) (6.19)
x∗1

Consider the conjugate of y(2) at the receiver, the above equation can be simplified as
⎡ ⎤
− x2 ⎦
y ∗ (2) = h∗1 h∗2 ⎣ + n∗ (2)
x1
⎡ ⎤
x
⎣ 2 ⎦ + n∗ (2)
= −h∗1 h∗2
x1
⎡ ⎤

= h∗2 −h∗1 ⎣ x1 ⎦ + n∗ (2) (6.20)

x2

Now, the received symbols y(1) and y ∗ (2) from equations (6.18) and (6.20) above can be
stacked to write the combined system model for the first and second time instants in the
Alamouti code as
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
y(1) ⎦ = ⎣ h1 h2 ⎦ ⎣ x1 ⎦ + ⎣ n(1) ⎦
⎣ (6.21)
y ∗ (2) h∗2 −h∗1 x2 n∗ (2)
y H n

Thus, both the symbols have been stacked which effectively converts the Alamouti coded
system into a 2 × 2 MIMO system, with the channel matrix
⎡ ⎤
⎣ h1 h2 ⎦ (6.22)
h∗2 −h∗1

Further, the noise n∗ (2) is statistically identical to n(2), i.e., n∗ (2) is zero mean circularly
symmetric Gaussian noise with variance σn2 . Moreover, observe a very important property of
 Multiple-Input Multiple-Output Wireless Communications 205

the Alamouti channel matrix. Consider the columns c1 , c2 of the channel matrix given as
⎡ ⎤ ⎡ ⎤
h1 ⎦ h2
c1 = ⎣ , c2 = ⎣ ⎦
∗
h2 −h∗1

Then, we have, cH
1 c2 given as
⎡ ⎤
h2
cH
1 c2 = h∗1 h2 ⎣ ⎦
−h∗1

= h∗1 h2 − h2 h∗1
=0

It can, therefore, be seen that the columns c1, c2 are orthogonal. This tremendously simplifies
the receive processing of the Alamouti code. Consider now beamforming using the vector w1
defined in terms of c1 as
1
w1 = c1
c1
⎡ ⎤
1 ⎣ h1 ⎦
=
h h∗2
One can now employ this as a receive beamformer to derive the processed symbol as
⎡ ⎤⎡ ⎤

w1H y = h∗1 h2 ⎣ h1 h2 ⎦ ⎣ x1 ⎦
+ w1H n
h h ∗ ∗
h2 −h1 x2
⎡ ⎤

= h 0 ⎣ x1 ⎦ + ñ1
x2

= h x1 + ñ1
206 Principles of Modern Wireless Communication Systems

Further, since w1 is a unit-norm vector, ñ1 = w1H n is Gaussian noise with variance σn2 .
Therefore, the SNR at the receiver is given as
h 2
SNR = E |x1 |2
σn2
$ 2$
$h $
= P1 ,
σn2
where h 2 = |h1 |2 + |h2 |2 . Therefore, the diversity gain or diversity order of BER at the
receiver is 2, since it is analogous to the maximum ratio combiner with 2 antennas shown in
Eq. (6.17). Similarly, to decode x2 , the beamformer w2 is given as
⎡ ⎤
c2 1 ⎣ h2 ⎦
w2 = =
c2 h −h∗1

2 2
Thus, the SNR of the decoded streams of the Alamouti code is σh2 P1, σh2 P2 , where P1 , P2
n n

are the power allocated to x1, x2 respectively. However, it may be noted that the total transmit
power P is fixed. This has to be allocated to the individual streams. Therefore, we have P1 =
P2 = P2 . Hence, the net output SNR of each stream is
$ $ $ $
P $h2 $ 1 $h2 $
SNR = = P
2 σn2 2 σn2

or in other words, equal to half that with CSI at the transmitter as can be seen from Eq. (6.17).
Thus, the absence of CSI results in a loss of 3 dB in output SNR corresponding to this factor
of 12 . Also, the orthogonality of the columns c1 , c2 of the effective channel matrix is a key
property of the Alamouti code. Hence, the Alamouti code is also termed an Orthogonal Space-
Time Block Code (OSTBC). The term space-time refers to the fact that the Alamouti code
involves two symbols x1, x2 which are transmitted over two antennas over two instants of
time. Therefore, the symbols are coded across both the space and time dimensions, leading to
the name "space-time" code. This is schematically shown in Figure 6.7. Further, also note that
the Alamouti code transmits a net of two symbols x1 , x2 in time instants. Therefore, on an
average it transmits one symbol per time instant. Hence, the net rate of the code is 1 symbol
per time instant, i.e., the rate R = 1. Such a code is termed a full rate code. Therefore, the
Alamouti code is a full-rate code. Example 6.6 given below clarifies the various operations in
the Alamouti code.
 Multiple-Input Multiple-Output Wireless Communications 207

Figure 6.7 Alamouti orthogonal space-time block code

EXAMPLE 6.6

Consider the 1 × 2 wireless system given as

⎡ ⎤
x
⎣ 1 ⎦+n
y= 1+j 3 + 4j
x2

Clearly, indicate the processing at the transmitter and the receiver for the above system with
the Alamouti code.
Solution: Firstly, it can be readily seen that the channel coefficients h1 , h2 are 1 + j, 3 + 4j ,
corresponding to the channels of transmit antennas 1, 2 respectively. Therefore, corresponding
to the transmission of symbols x1 , x2 from the first and second transmit antennas in the first
time instant, we have the received symbol y(1) given as
⎡ ⎤
x
⎣ 1 ⎦ + n(1)
y(1) = 1+j 3 + 4j
x2
208 Principles of Modern Wireless Communication Systems

as described in Eq. (6.18). Further, corresponding to the transmission of −x∗2, x∗1 from the first
and second transmit antennas in the second time instant, we have
⎡ ⎤
∗
− x 2 ⎦
y(2) = 1+j 3 + 4j ⎣ + n(2)
∗
x1

which can be simplified by considering the conjugate y ∗ (2) of y(2) as,

⎡ ⎤

y ∗ (2) = 1−j 3 − 4j ⎣ −x2 ⎦ + n∗ (2)

x1
⎡ ⎤

= −1 + j 3 − 4j ⎣ x2 ⎦ + n∗ (2)
x1
⎡ ⎤

= 3 − 4j −1 + j ⎣ x1 ⎦ + n∗ (2)
x2

as given in Eq. (6.20). Therefore, stacking now y(1), y ∗ (2), we have

⎡ ⎤ ⎡ ⎤⎡ ⎤
y(1) ⎦ = ⎣ 1+j 3 + 4j ⎦ ⎣ x1 ⎦
⎣ +n
y ∗ (2) 3 − 4j −1 + j x2
y H

where H above indicates the effective Alamouti channel matrix and y is the effective received
vector. Therefore, the columns c1 , c2 are given as
⎡ ⎤ ⎡ ⎤
1 + j 3 + 4j
c1 = ⎣ ⎦ , c2 = ⎣ ⎦
3 − 4j −1 + j
 Multiple-Input Multiple-Output Wireless Communications 209

It can now be easily verified that the columns c1 , c2 are orthogonal. Consider cH
1 c2 , which
can be simplified as
∗ ∗
cH
1 c2 = (1 + j) (3 + 4j) + (3 − 4j) (−1 + j)

= (1 − j) (3 + 4j) + (3 + 4j) (− (1 − j))

= (1 − j) (3 + 4j) − (3 + 4j) (1 − j)

=0

Beamformer to detect x1 is given as

⎡ ⎤
c1 1 1 + j
w1 = =√ ⎣ ⎦
c1 27 3 − 4j

To detect x1, one has to perform the receive beamforming operation w1H y . Similarly, w2, the
beamformer to detect x2 is given as
⎡ ⎤
c2 1 3 + 4j ⎦
w2 = =√ ⎣
c2 27 −1 + j

Therefore, x2 can be detected by performing the operation w2H y at the receiver.

6.9 Another OSTBC Example

We now describe another example of an OSTBC for a 1 × 3, i.e., a system with 1 receive
and 3 transmit antennas. Consider the channel matrix h1 h2 h3 corresponding
to this 1 × 3 system. Consider the transmission of 4 symbols x1 , x2, x3 , x4 over the wireless
channel. The corresponding coded block for the OSTBC is given as
⎡ ⎤
x1 − x2 − x3 −x4 x∗1 −x∗2 −x∗3 −x∗4
⎢ ⎥
⎢ x
⎣ 2 x1 x4 −x3 x∗2 x∗1 x∗4 −x∗3 ⎥
⎦ (6.23)
x3 −x4 x1 x2 x∗3 −x∗4 x∗1 ∗
x2
210 Principles of Modern Wireless Communication Systems

Each column in the above code block gives the three symbols transmitted over each of the 3
transmit antennas. Observe, that there are 8 time instants in total in the above block. Therefore,
the four symbols, i.e., x1, x2 , x3, x4 are being transmitted over a total of 8 time instants.
Therefore, the net rate of the code is 84 = 12 . Consider the first received symbols y (1). This can
be expressed as
⎡ ⎤
x1
⎢ ⎥
y (1) = h1 h2 h3 ⎢ x ⎥
⎣ 2 ⎦
x3

= h1 x1 + h2 x2 + h3 x3 + 0x4
⎡ ⎤
x1
⎢ ⎥
⎢ x ⎥
⎢ 2 ⎥
= h1 h2 h3 0 ⎢ ⎥
⎢ x ⎥
⎣ 3 ⎦
x4
Further, the second received symbol y(2) is given as
⎡ ⎤
−x2
⎢ ⎥
y(2) = h1 h2 h3 ⎢
⎣ x1 ⎥
⎦
−x4

= −h1 x2 + h2 x1 − h3 x4 + 0x3
⎡ ⎤
x1
⎢ ⎥
⎢ x ⎥
⎢ 2 ⎥
= h2 −h1 0 −h3 ⎢ ⎥ (6.24)
⎢ x ⎥
⎣ 3 ⎦
x4
Thus, proceeding similarly, the received symbols y(1), y(2), y(3), y(4) and the complex
conjgates of the symbols y(5), y(6), y(7), y(8) can be stacked to obtain the effective system
 Multiple-Input Multiple-Output Wireless Communications 211

model
⎡ ⎤ ⎡ ⎤
y(1) h1 h2 h3 0
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥
⎢ y(2) ⎥ ⎢ h2 −h1 0 −h3 ⎥
⎢ ⎥ ⎢ ⎥⎡ ⎤
⎢ ⎥ ⎢ ⎥
⎢ y(3) ⎥ ⎢ h3 0 −h1 h2 ⎥ x1
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ y(4) ⎥ ⎢ 0 h3 −h2 −h1 ⎥ ⎢ x2 ⎥
⎢ ⎥ ⎢ ⎥⎢ ⎥
⎢ ⎥=⎢ ⎥⎢ ⎥. (6.25)
⎢ y ∗ (5) ⎥ ⎢
⎢ ⎥ ⎢ h∗1 h∗2 h∗3 0 ⎥⎥⎣
⎢ x3 ⎥
⎦
⎢ ⎥ ⎢ ⎥
⎢ y ∗ (6) ⎥ ⎢
⎢ ⎥ ⎢ h∗2 −h∗1 0 −h∗3 ⎥
⎥ x4
⎢ ⎥ ⎢ ⎥
⎢ y ∗ (7) ⎥ ⎢
⎣ ⎦ ⎣ h∗3 0 −h∗1 ∗ ⎥
h2 ⎦
y ∗ (8) 0 h∗3 −h∗2 −h∗1
y H

Therefore, it can be seen that the effective channel matrix H represents an equivalent 8 × 4
MIMO system. Further, the different columns of H are orthogonal. For instance, consider the
cH
1 c2 corresponding to columns c1 , c2 . This can be simplified as
⎡ ⎤
h2
⎢ ⎥
⎢ ⎥
⎢ −h1 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 ⎥
⎢ ⎥
⎢ h ⎥
3
cH
⎢ ⎥
c
1 2 = h1∗
h2∗ ∗
h3 0 h1 h2 h3 0 ⎢ ⎥
⎢ h∗ ⎥
⎢ 2 ⎥
⎢ ⎥
⎢ h∗ ⎥
⎢ 1 ⎥
⎢ ⎥
⎢
⎣ 0 ⎥
⎦
∗
h3

= h∗1 h2 − h∗2 h1 + 0 + 0 + h1 h∗2 − h2 h∗1 + 0 + 0

=0
1
Hence, the columns are orthogonal. Therefore, this is an example of an R = 2 OSTBC, i.e.,
orthogonal space-time block code.
212 Principles of Modern Wireless Communication Systems

6.10 Nonlinear MIMO Receiver: V-BLAST

Previously, we have seen the zero-forcing (ZF) and minimum mean squared error (MMSE)
receivers, which are linear MIMO receivers. We now look at the first nonlinear MIMO
receiver, termed V-BLAST, short for Vertical Bell Labs Layered Space-Time receiver.
V-BLAST employs Successive Interference Cancellation (SIC) in which the impact of each
estimated symbol is cancelled prior to the detection of the next symbol. This SIC principle
on which V-BLAST is based leads to its nonlinear nature. Consider the MIMO system model
given as

y = Hx + n
⎡ ⎤
x1
⎢ ⎥
⎢ x ⎥
⎢ 2 ⎥
= h1 h2 ... ht ⎢ .. ⎥ +n (6.26)
⎢ . ⎥
⎣ ⎦
xt
The vectors h1 , h2 , . . . , ht correspond to the t columns of the channel matrix H. Consider
now the left-inverse or pseudo-inverse of the channel matrix Q described in Eq. (6.3). Let this
matrix be denoted by the t × r matrix Q, i.e., QH = Ir . Further, let the matrix Q be written as
⎡ ⎤
qH
1
⎢ ⎥
⎢ qH ⎥
⎢ 2 ⎥
Q=⎢ .. ⎥
⎢ . ⎥
⎣ ⎦
qH
t

where qH H H
1 , q2 , . . . , qt denote the t rows of the matrix Q. Therefore, QH = It can be
written as
⎡ ⎤
qH1
⎢ ⎥
⎢ qH ⎥
⎢ 2 ⎥
⎢ . ⎥ h1 h2 ... ht = It
⎢ .. ⎥
⎣ ⎦
qHt
 Multiple-Input Multiple-Output Wireless Communications 213

⎡ ⎤ ⎡ ⎤
qH
1 h1 qH
1 h2 ... qH
1 ht 1 0 ... 0
⎢ ⎥ ⎢ ⎥
⎢ qH h qH ... qH ⎥ ⎢ 0 1 ... 0 ⎥
⎢ 2 1 2 h2 2 ht ⎥ ⎢ ⎥
⇒⎢ .. .. .. .. ⎥=⎢ . .. .. .. ⎥ (6.27)
⎢ . . . . ⎥ ⎢ .. . . . ⎥
⎣ ⎦ ⎣ ⎦
qH
t h1 qH
t h2 ... qH
t ht 0 0 ... 1
Therefore, we have,

qH H H
1 h1 = q2 h2 = . . . = qt ht = 0

qH H H
1 h2 = q1 h3 = . . . = qt−1 ht = 0

The above property can be basically summarized as

⎧
⎨ 1 if i = j
qH
i hj = (6.28)
⎩ 0 if i = j

From Eq. (6.26), observe that the system model can be explicitly given in terms of the columns
of the channel matrix H as

y = h1 x1 + h2 x2 + . . . + ht xt + n

Employing now the fact that qH H H

1 h1 = 1 and q1 h2 , q1 h3 , etc., is zero, one can use q1 as a
receive beamformer. Therefore, performing qH
1 y at the receiver, we have

ỹ1 = qH
1 y

= qH
1 (h1 x1 + h2 x2 + . . . + ht xt + n)

= qH H H H
1 h1 x1 + q1 h2 x2 + . . . + q1 ht xt + q1 n
1 0 0 ñ1

= x1 + ñ1
214 Principles of Modern Wireless Communication Systems

Thus, ỹ1 can now be employed to decode x1. Now, the interference caused by x1 is removed
from y to form y̌2 as

y̌2 = y − h1 x1

= h1 x1 + h2 x2 + . . . + htxt + nh1 x1

= h2 x2 + . . . + htxt + n
⎡ ⎤
x2
⎢ ⎥
⎢ x ⎥
⎢ 3 ⎥
= h2 h3 ... ht ⎢ .. ⎥ +n
⎢ . ⎥
⎣ ⎦
H(2)
xt
x(2)

= H(2) x(2) + n

The above system model can now be seen to correspond to a reduced MIMO system with
channel matrix H(2) of r rows and t − 1 columns. Thus, it represents an r × (t − 1) MIMO
system, with x2 , x3, . . . , xt denoting the t − 1 transmit symbols. Now, consider Q(2) as
the zero-forcing receiver for H(2) and repeat the above process by decoding x2 , and so on.
The advantage of this scheme is that the diversity order and the associated diversity gain
progressively increases as we proceed through the scheme for decoding the different transmit
symbols x1 through xt . In fact, in the last stage, when only one symbol xt is left for decoding,
the effective system model is given as

y̌t = H(t)xt + n

Notice that the effective channel H(t) is the column vector ht . This can be decoded by receive
beamforming along ht , in other words, maximum ratio combining. In V-BLAST, streams that
are decoded later experience progressively higher diversity. Example 6.7 clearly illustrates the
working of the V-BLAST MIMO receiver.
 Multiple-Input Multiple-Output Wireless Communications 215

EXAMPLE 6.7

Consider the 2 × 2 MIMO system given below and describe the various stages of the
V-BLAST receiver.
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
⎣ y1 ⎦ = ⎣ 1 2 ⎦ ⎣ x1 ⎦ ⎣ n1 ⎦
+ (6.29)
y2 1 3 x2 n2
y H x n

Solution: The channel matrix H is given as

⎡ ⎤
1 2 ⎦
H=⎣
1 3

Therefore, the left inverse of H, i.e., Q = H† is given as Q = H−1 since the matrix H is
square and invertible. Therefore, we have
⎡ ⎤
3 −2 ⎦
Q = H− 1 = ⎣
−1 1

It can be clearly seen that the rows of the matrix Q are qH

1 = 3 −2 ,
qH
2 = −1 1 . Further, it can be readily seen that qH
1 h1 is given as

⎡ ⎤

qH
1 h1 = 3 −2 ⎣ 1 ⎦=1
1

and qH
1 h2 can be simplified as

⎡ ⎤
2
qH
1 h2 = 3 −2 ⎣ ⎦=0
3
216 Principles of Modern Wireless Communication Systems

Therefore, the first row qH

1 is orthogonal to h2 , i.e., the second column of the channel matrix
H. Therefore, decoding in qH1 in the first stage of V-BLAST, we have

ỹ1 = qH
1 y
⎛⎡ ⎤⎡ ⎤ ⎡ ⎤⎞

= 3 −2 ⎝⎣ 1 2 ⎦ ⎣ x1 ⎦ ⎣ n1 ⎦⎠
+ .
1 3 x2 n2
⎡ ⎤

= 1 0 ⎣ x1 ⎦ + q H
1 n
x2
ñ1

= x1 + ñ1

Therefore, the symbol x1 can be detected from ỹ1 . The interference caused by this can now be
canceled from the received signal y to form y̌2 as

y̌2 = y − h1 x1
⎡ ⎤⎡ ⎤ ⎡ ⎤ ⎡ ⎤
1 2 ⎦ ⎣ x1 ⎦ ⎣ n1 ⎦ ⎣ 1 ⎦
=⎣ + − x1
1 3 x2 n2 1
⎡ ⎤
2 ⎦
=⎣ x2 + n
3
It can now be seen that the above system corresponds effectively to a receive diversity system
with 2 receive antennas and a single transmit antenna. In fact, the optimum receiver scheme is
T
to perform maximum ratio combining with the receive beamformer w = 2 3 . The
diversity order of decoding this stream is, therefore, equal to 2.

6.11 MIMO Beamforming

In this section, we describe beamforming in the context of a MIMO wireless system.

Beamforming basically implies that only spatial dimension is used for transmission amongst