Unit 3

The goal of space-time coding is to achieve the maximum diversity, the maximum coding gain,
and the highest possible throughput. To improve the battery life of mobile wireless
communication, low complexity encoding and decoding is very crucial. On the other hand, the
base station is not as restricted in terms of power and physical size. One can put multiple
independent antennas in a base station. Therefore, in many practical situations, a very low
complexity system with multiple transmit antennas is desirable. Space-time block coding is a
scheme to provide these properties.

Space–time block coding is a technique used in wireless communications to transmit multiple

copies of a data stream across a number of antennas and to exploit the various received versions
of the data to improve the reliability of data transfer.

Real orthogonal designs

To design space-time codes that provide the properties of the Alamouti code for more than two
transmit antennas, first we should understand why these codes behave in this way. For better
understanding, consider the matrix:

(1)

where, (2)

Multiplying both sides of (1) by ΩH gives

(3)

where, N is a Gaussian noise.

Equation (3) consists of two separate equations for decoding the two transmitted symbols. This
separation is the main reason for the first property of simple ML decoding. The second property
comes from the factor |α1|2 + |α2|2 in the right-hand side of Equation (3).

When there is only one transmit antenna available, the power of the signal is affected by a
factor of |α|2 due to the path gain. In a deep faded environment |α|2 is very small and the noise

dominates the signal. On the other hand, (3) shows that using the Alamouti code |α1|2 + |α2|2

Dr.HMV
should be small for a noise dominated channel. To have a small |α1|2 + |α2|2, both |α1|2 and |α2|2
must be small, which is less likely.

Derivation of (3) shows that the following equation is the main reason for both properties:

(4)

where I2 is a 2 × 2 identity matrix. If we show the structure of the Alamouti code by the
following generator matrix

(5)

(4) is a result of the orthogonality of the columns of and the following property:

(6)

So far we have defined the generator matrix and showed that since the generator matrix
satisfies (6) for every possible indeterminant pair (x1, x2), it provides full diversity and simple
ML decoding. For real numbers, matrices satisfying (6) are called orthogonal designs.

Let us understand Radon–Hurwitz theory:

Definition: A set of N × N real matrices {B1, B2,..., BL} is a size L Hurwitz–Radon family of
matrices if

𝐵𝑙𝑇 𝐵𝑙 = 𝐼𝑁 , 𝑙 = 1,2, … , 𝐿

𝐵𝑙𝑇 = −𝐵𝑙 , 𝑙 = 1,2, … , 𝐿

𝐵𝑙 𝐵𝑙 = −𝐵𝑙` 𝐵𝑙 , 𝑙 ≤ 𝑙` ≤ 𝐿

Theorem: Radon Theorem: We write any positive integer number N as N = 2ab, where a = 4c
+ d, c ≥ 0, 0 ≤ d < 4 and b is an odd number. Any N × N Hurwitz–Radon family contains less
than ρ(N) = 8c + 2d ≤ N matrices.

Definition 1: An integer matrix is defined as a matrix that all of its elements are −1, 0, or 1.

Definition 2: A real orthogonal design of size N is an N × N orthogonal matrix GN with real

entries x1,−x1, x2,−x2, . . . , xN ,−xN such that

(a)

Dr.HMV
Using Radon Theorem, we prove the following theorem on the existence of real orthogonal
designs.

Theorem: Real orthogonal design exists if and only if N = 2, 4, 8.

Proof: Let us rewrite , where An is an integer matrix. Then,

(b)

To have both the equations (a) and (b) in simultaneously and for all possible indeterminate
variables x1, x2, . . . , xN , we should have

(c)

By defining for n = 1, 2, . . . , N, we have B1 = IN and the following Hurwitz–

Radon family of size N − 1 is needed:

Using Radon Theorem 4.4.1, we have ρ(N) = N − 1 and therefore N = 2, 4, 8.

Complex orthogonal designs

The previous designs only work for real constellations. Naturally, it is interesting to extend
these schemes to complex signal constellations.

Definition: A complex orthogonal design of size N is an N × N orthogonal matrix GN with

complex entries x1,−x1, x2,−x2, . . . , xN ,−xN , their conjugates 𝑥1∗ , −𝑥1∗ , 𝑥2∗ , −𝑥2∗ , … 𝑥𝑁∗ , −𝑥𝑁∗ , and
multiples of these indeterminate variables by 𝑗 = √−1 𝑜𝑟 − 𝑗 such that

The following construction shows the relationship between real and complex orthogonal
designs.

Construction: Given a complex orthogonal design of size N, we replace each complex variable
xn = R{xn} + I{xn}j, 1 ≤ n ≤ N by the 2 × 2 real matrix

Dr.HMV
 𝑅{𝑥𝑛 } 𝐼{𝑥𝑛 }
( ) (d)
−𝐼{𝑥𝑛 } 𝑅{𝑥𝑛 }

The resulting 2N × 2N matrix is a real orthogonal design of size 2N.

The representation in (d) results in representing 𝑥𝑁∗ by

𝑅{𝑥𝑛 } −𝐼{𝑥𝑛 }
( )
𝐼{𝑥𝑛 } 𝑅{𝑥𝑛 }

Since real orthogonal designs only exist for N = 2, 4, 8, the above construction shows that a
complex orthogonal design can only exist for N = 2 or N = 4. For N = 2, the following matrix
is a complex orthogonal design

Generalized complex orthogonal designs

Definition: A generalized complex orthogonal design is a T × N matrix G with entries that are
linear combinations of the indeterminate variables x1, x2, . . . , xK and their conjugates such that

where IN is the N × N identity matrix and κ is a constant.

Note that κ = 1 is possible by an appropriate normalization of G elements. Also, multiplying a

generalized complex orthogonal design by a unitary matrix results in another generalized
complex orthogonal design. In otherwords, if G is a generalized complex orthogonal design
and U is unitary, that is UH · U = I, then G' = U · G is also a generalized complex orthogonal
design. This can be shown using the fact that G'H · G' = GH · UH · U · G = GH · G. Similarly,
G' = G · U is a generalized complex orthogonal design as well since

Dr.HMV
STBC for any complex constellation:

A space-time block code (STBC) for any complex constellation can be constructed using a
generalized complex orthogonal design.

- The number of transmission antennas is N.

- We assume that transmission at the baseband employs a signal constellation with 2 b
elements.
- At time slot 1, Kb bits arrive at the encoder and select constellation signals s1, s2, .., sK.
- Setting xk = sk for k = 1, 2, . . . , K in G, we arrive at a matrix C = G(s1, s2,..., sK ) whose
entries are linear combinations of s1, s2, .., sK and their conjugates.
- So, while G contains indeterminate variables x1, x2,... xK, matrix C contains specific
constellation symbols that are transmitted from N antennas for each Kb bits as follows.
If Ct,n represents the element in the tth row and the nth column of C, the entries Ct,n, n =
1, 2, . . . , N are transmitted simultaneously from transmit antennas 1, 2, . . . , N at each
time slot t = 1, 2, . . . , T .
- So, the nth column of C represents the transmitted symbols from the nth antenna and the
tth row of C represents the transmitted symbols at time slot t.
- Note that C is basically defined using G. Therefore, the orthogonality of the columns
of G allows a simple ML decoding scheme, which will be explained in what follows.
Since T time slots are used to transmit K symbols, we define the rate of the code to be
R = K/T. The block diagram of the encoder is shown in below figure.

Figure. Encoder block diagram for orthogonal space-time block codes.

The STBCs designed from generalized complex orthogonal designs provide full diversity and
separate decoding:

Theorem: A complex space-time block code designed from a T × N generalized complex

orthogonal design provides a diversity of NM for M receive antennas and a separate maximum-
likelihood decoding of its symbols.

Dr.HMV
Proof:

(i) Full diversity;

We need to show that the difference matrix G(s1, s2,...sk) - G(s'1, s'2,…., s'k) is full rank (non-
singular) for any two distinct set of inputs (s1, s2,...sk) ≠ (s'1, s'2,…., s'k). Since G(s1, s2,...sk)
elements are linear combinations of s1, s2,...sk, we have

G(s1, s2,...sk) - G(s'1, s'2,…., s'k) = G(s1 - s'1, s2 - s'2,,…. sk - s'k,)

Therefore, it is adequate to show that G(s1 - s'1, s2 - s'2,,…. sk - s'k,) is non-singular.

Equivalently, we can show that

Det[GH(s1 - s'1, s2 - s'2,,…. sk - s'k,). G(s1 - s'1, s2 - s'2,,…. sk - s'k,)]

Therefore, the above determinant is zero if and only if (s1, s2,...sk) = (s'1, s'2,…., s'k), which
means the difference matrix is full rank.

(ii) Decode separately;

This is achieved due to the orthogonality of the columns of G. The maximum-likelihood

decoding amounts to finding the codeword Ĉ that maximizes the density function or
equivalently minimizes the cost function.

Therefore, the most likely transmitted signals are

The argument of the minimization in the above equation is a linear combination of sk and |sk|2
and can be written as

where fk(sk), k = 1,2,..,,K.

As a result, the minimization is equivalent to K independent minimizations minsk fk (sk ), k =

1, 2, . . . , K. Therefore, transmitted symbols can be decoded separately.

Dr.HMV
Quasi-orthogonal STBC

The main properties of an orthogonal design are simple separate decoding and full diversity.
Here, we consider codes for which decoding pairs of symbols independently is possible. We
call this class of codes quasi-orthogonal space-time block codes (QOSTBCs).

These codes exhibit partial orthogonality and provide only part of the diversity gain.

First, let us consider the following QOSTBC

where a matrix G∗ is the complex conjugate of G, for example

We denote the ith column of G by Vi. For any indeterminate variables x1, x2, x3, x4, we have

< V1, V2 > = < V1, V3 > = < V2, V4 > = < V3, V4 > = 0,

Where < Vi, Vj > is the inner product of vectors Vi and Vj.

Therefore, the subspace created by V1 and V4 is orthogonal to the subspace created by V2 and
V3. This is the rationale behind the name “quasi-orthogonal” for the code.

The encoding for QOSTBCs is very similar to the encoding of orthogonal STBCs. To transmit
b bits per time slot, we use constellations containing 2b points. Using 4b bits, constellation
symbols s1, s2, s3, s4 are selected. Setting xk = sk for k = 1, 2, 3, 4 in the generator matrix G, we
arrive at a codeword matrix C = G(s1, s2, s3, s4). Then, at time t, the four elements in the tth
row of C are transmitted from the four transmit antennas.

Rotated QOSTBCs

To provide full diversity, we use different constellations for different transmitted symbols. For
example, we may rotate symbols before transmission. The resulting code is very powerful since
it provides full diversity, rate one, and simple pairwise decoding with good performance.

Dr.HMV
Here, we derive the general conditions to achieve full diversity for these codes. First, calculate
the coding gain distance (CGD) for different cases and show that they behave similarly. We
use G to denote the generator matrix with indeterminate variables x1, x2, x3, x4 and G(s1, s2, s3,
s4) to denote the same matrix when xk is replaced by the kth argument in parentheses.

Lemma: The CGD between a pair of codewords C = G(s1, s2, 𝑠̃ 3, 𝑠̃ 4) and C' = G(s'1, s'2, 𝑠̃ '3, 𝑠̃ '4)
of the QOSTBC is given by

(e)

Proof: For QOSTBC, we have

where 𝑎 = ∑4𝑘=1|𝑥𝑘 |2 and b = 2R(x1 x*4 - x2 x*3)

Using the determinant equality:

we have det(GH . G) = (a2 – b2)2. With simple algebraic manipulation, we get

Then, replacing x1 with (s1 - s'1), x2 with (s2 - s'2), x3 with (𝑠̃ 3 - 𝑠̃ '3), and x4 with (𝑠̃ 4 - 𝑠̃ '3) results
in equation (e). Hence proved.

Reference:

Chapter 4 and 5 from Hamid Jafarkhani, “Space-Time Coding Theory and Practise”

STBC - Space-Time Block Code

QOSTBC – Quasi-Orthogonal Space-Time Block Code
CGD - Coding Gain Distance

Dr.HMV