BEC605: Wireless Communication

Module 3

Reference:
Wireless Communications
Andrea Goldsmith
Contents

3.1 Multiple Antennas: Narrowband MIMO Model

3. 2 Parallel Decomposition of the MIMO Channel
3.3 MIMO Channel Capacity
3.4 MIMO Diversity Gain: Beamforming
3.5 Diversity-Multiplexing Trade-offs
3.6 Frequency-Selective MIMO Channels, Smart Antennas
3.1 Multiple Antennas: Narrowband MIMO Model

What is a MIMO system?

A MIMO system uses:
• Multiple antennas at the transmitter (𝑀𝑡 )
• Multiple antennas at the receiver (𝑀𝑟 )
Why use multiple antennas?
Two main benefits:
(a) Diversity gain:
Diversity improves the reliability of communication. Since wireless channels suffer from fading,
transmitting the same information over multiple antennas creates multiple independent paths.
Even if one path is weak, others may be strong, thereby reducing the probability of error.
(b) Multiplexing gain:
Multiplexing increases the data rate. Instead of sending the same signal, different independent data
streams are transmitted simultaneously from different antennas. If the channel supports it, these
streams can be separated at the receiver, resulting in higher spectral efficiency.
Multiple Antennas: Narrowband MIMO Model
• The performance of MIMO systems depends heavily on the
knowledge of the channel matrix 𝐻:
• CSIR (Channel State Information at Receiver):
Usually assumed available. The receiver can estimate the channel
using pilot signals.
• CSIT (Channel State Information at Transmitter):
May be available through feedback or channel reciprocity. This allows
better transmission strategies.
Multiple Antennas: Narrowband MIMO Model
• Consider a narrowband MIMO channel,
where: Signal bandwidth is much
smaller than the coherence bandwidth.
• Therefore:
• Inter-symbol interference (ISI) can be
ignored
• The channel experiences flat fading
• This means that each transmit–receive
antenna pair can be modeled by a single
complex gain.
• Consider a system with: 𝑀𝑡 transmit
antennas , 𝑀𝑟 receive antennas
• The discrete-time baseband model is:
𝑦 = 𝐻𝑥 + 𝑛
Multiple Antennas: Narrowband MIMO Model
Components of the Model
1. Transmit Vector 𝑥
𝑥1
𝑥2
𝑥 = ⋮ This represents the symbols transmitted
𝑥𝑀𝑡 3. Channel Matrix 𝐻
from each antenna.
ℎ11 ⋯ ℎ1𝑀𝑡
2. Receive Vector 𝑦 𝑦 𝐻= ⋮ ⋱ ⋮
1
𝑦2 ℎ𝑀𝑟 1 ⋯ ℎ𝑀𝑟 𝑀𝑡
𝑦= ⋮ Each element ℎ𝑖𝑗 represents the complex gain from transmit
antenna 𝑗to receive antenna 𝑖.
𝑦𝑀𝑟
This represents the signals received at each antenna.
Multiple Antennas: Narrowband MIMO Model
• Noise𝑛 Vector 𝑛
1
𝑛 = ⋮ ; Modeled as complex Gaussian noise
𝑛𝑀𝑟
• Mean = 0 (It does not shift the signal up or down) and Covariance matrix
𝑅𝑛 = 𝜎 2 𝐼𝑀𝑟

Covariance measures how two random variables vary together. Variance gives How much a
signal varies (spreads) around its mean
For two random variables 𝑥and 𝑦:
Cov 𝑥 𝑦 = 𝐸 𝑥 − 𝐸 𝑥 𝑦 − 𝐸 𝑦 ∗
Special case (zero mean)
If 𝐸 𝑥 = 0 ,𝐸 𝑦 = 0 ,then:
Cov 𝑥 𝑦 = 𝐸 𝑥𝑦 ∗
Multiple Antennas: Narrowband MIMO Model

•If covariance = 0 → variables are uncorrelated

Structure
•If covariance ≠ 0 → variables are related
𝐸 𝑛1 𝑛1∗ 𝐸 𝑛1 𝑛2∗ ⋯
• For a vector: 𝑅𝑛 = 𝐸 𝑛2 𝑛1∗ 𝐸 𝑛2 𝑛2∗ ⋯
𝑛1 ⋮ ⋮ ⋱
𝑛2
𝑛= ⋮ Meaning of elements
Diagonal elements:
𝑛𝑀𝑟 𝐸 𝑛𝑖 𝑛𝑖∗ = 𝐸 ∣ 𝑛𝑖 ∣2
Variance (noise power at antenna 𝑖)
• The covariance matrix is:
𝑅𝑛 = 𝐸 𝑛𝑛𝐻
Off-diagonal elements:
𝐸 𝑛𝑖 𝑛𝑗∗ , 𝑖 ≠ 𝑗
Correlation between noise at different antennas
Multiple Antennas: Narrowband MIMO Model
Covariance Matrix of Noise 𝑅𝑛 = 𝐸 𝑛𝑛𝐻
We assume:
• Noise is zero mean Expand 𝑛𝑛𝐻
𝑛1
• Noise at different antennas 𝑛𝑛𝐻 = 𝑛2 𝑛1∗ 𝑛2∗ ⋯
⋮
is independent 𝑛1 𝑛1∗ 𝑛1 𝑛2∗ ⋯
Rn= 𝑛2 𝑛1∗ 𝑛2 𝑛2∗ ⋯
• Each noise has same ⋮ ⋮ ⋱
variance 𝜎 2 𝜎2 0 0
𝑅𝑛 = 0 𝜎2 0
0 0 𝜎2
Multiple Antennas: Narrowband MIMO Model
• 𝐸 𝑛𝑖 𝑛𝑖∗ = 𝜎 2
• Variance tells you: How much a signal varies (spreads) around its mean
• For noise: Mean = 0 . Variance = 𝜎 2
• So:
𝜎 2 = 𝐸 ∣ 𝑛𝑖 ∣2 This is the average power of the noise
Each received signal is a linear combination of all transmitted
Independence: signals:
𝑴𝒕
•𝐸 𝑛𝑖 𝑛𝑗∗ =0 𝑖≠𝑗 𝒚𝒊 = ෍ 𝒉𝒊𝒋 𝒙𝒋 + 𝒏𝒊
2 𝒋=𝟏
• 𝑅𝑛 = 𝜎 𝐼𝑀𝑟 Thus, signals from different antennas are coupled together at
the receiver.
Multiple Antennas: Narrowband MIMO Model
What is 𝑅𝑥 ?
𝑅𝑥 = 𝐸 𝑥𝑥 𝐻 It is the covariance matrix of the transmitted signal
Structure of 𝑅𝑥
𝐸 ∣ 𝑥1 ∣2 𝐸 𝑥1 𝑥2∗ ⋯
𝑅𝑥 = 𝐸 𝑥2 𝑥1∗ 𝐸 ∣ 𝑥2 ∣2 ⋯
⋮ ⋮ ⋱

What is Tr 𝑅𝑥 ?
• Tr 𝑅𝑥 = 𝐸 ∣ 𝑥1 ∣2 + 𝐸 ∣ 𝑥2 ∣2 + ⋯ + 𝐸 ∣ 𝑥𝑀𝑡 ∣2
• It represents: Total transmitted power across all antennas
Multiple Antennas: Narrowband MIMO Model
Power Constraint
• The transmitter has a total power constraint:
𝑀𝑡

෍ 𝔼 ∣ 𝑥𝑖 ∣2 = 𝜌
𝑖=1
Equivalently, in matrix form:
Tr 𝑅𝑥 = 𝜌
where:
• 𝑅𝑥 = 𝔼 𝑥𝑥 𝐻 is the covariance matrix of the transmitted signal
𝑃
• 𝜌represents the average SNR (𝜌 = )
𝜎2
• Assume that Each antenna has noise power = 1 (after normalization)
• Average signal power reaching each antenna is proportional to 𝜌
So each antenna sees:
SNR = 𝜌
3.2
Parallel Decomposition of the MIMO Channel
When both transmitter and receiver have multiple antennas:
• We are not limited to one data stream
• The channel can support multiple independent streams simultaneously
• This is called multiplexing gain
• It increases data rate, not just reliability
MIMO Channel Model
We start with:
𝑦 = 𝐻𝑥 + 𝑛
Where:
• 𝐻: 𝑀𝑟 × 𝑀𝑡 channel matrix
• 𝑥: transmitted signal vector
• 𝑦: received signal vector
• 𝑛: noise
Parallel Decomposition of the MIMO Channel
𝑅𝐻 ​=rank(H)
𝑅𝐻 ≤ min 𝑀𝑡 𝑀𝑟
Interpretation:
• 𝑅𝐻 =number of independent spatial channels
If:
• Rich scattering → full rank → maximum parallel channels, Low correlation
• High correlation → low rank → fewer channels

• In MIMO systems:
Rank = number of parallel data streams you can send
(About rank—Maths)
• What does “linearly independent” mean? Example 1 (Full Rank)
• A set of vectors is linearly independent if: 1 0
𝐻=
No vector can be written as a combination of the others. 0 1
•Rows are independent
Example: •Columns are independent
• Independent: Rank = 2
1 0
• ,
0 1
• Dependent: Example 2 (Rank = 1)
1 2 1 2
• , 𝐻=
2 4 2 4
• Second = 2 × first → not independent) •Second row = 2 × first row
• Rank tells you: Not independent
“How many independent directions exist in the matrix” Rank = 1
Parallel Decomposition of the MIMO Channel
• Any matrix 𝐻can be written as:
𝐻 = 𝑈Σ𝑉 𝐻
Where:
𝑈(Receiver side)
• Size: 𝑀𝑟 × 𝑀𝑟
• Unitary: 𝑈𝐻 𝑈 = 𝐼
𝑉(Transmitter side)
• Size: 𝑀𝑡 × 𝑀𝑡
• Unitary: 𝑉 𝐻 𝑉 = 𝐼
Σ (Diagonal matrix)
• Contains singular values: 𝜎1 , 𝜎2 , . . . , 𝜎𝑅𝐻
• Only 𝑅𝐻 are non-zero
Also:
𝜎𝑖 = 𝜆𝑖
where 𝜆𝑖 are eigenvalues of 𝐻𝐻 𝐻
Parallel Decomposition of the MIMO Channel
Example
Suppose: 𝐻 is 3 × 3
Rank = 2
Then:
𝜎1 0 0
Σ= 0 𝜎2 0
0 0 0
Here:
• 𝜎1 , 𝜎2 ≠ 0
• 𝜎3 = 0
• Each non-zero singular value = one independent channel
• Zero singular value = no channel (dead path)
Parallel Decomposition of the MIMO Channel
In MIMO Interpretation
After decomposition:
𝑦෤𝑖 = 𝜎𝑖 𝑥෤𝑖 + 𝑛෤ 𝑖
Case 1: 𝜎𝑖 ≠ 0
• Signal passes through
• Useful channel
Case 2: 𝜎𝑖 = 0
• Output becomes:
𝑦෤𝑖 = 𝑛෤ 𝑖
• Only noise → no information transmitted
Parallel Decomposition of the MIMO Channel

y=Hx+n Receiver Shaping

𝑦= 𝑈Σ𝑉 𝐻 𝑉𝑥෤ + 𝑛 𝑦෤ = 𝑈 𝐻 𝑦
Now:
𝑉𝐻𝑉 = 𝐼 Final Result:
So: 𝑦෤ = Σ𝑥෤ + 𝑛෤
𝑦 = 𝑈Σ𝑥෤ + 𝑛.
Multiply by 𝑈𝐻 on both sides
Parallel Decomposition of the MIMO Channel

Noise Transformation
•After receiver shaping:
𝑛෤ = 𝑈 𝐻 𝑛
•Since 𝑈is unitary:
𝑈𝐻 𝑈 = 𝐼
• Important result:
• Noise statistics do not change
• 𝑛and 𝑛are
෤ identically distributed

For the 𝑖-th channel:

•Input → 𝑥෤𝑖
•Output → 𝑦෤𝑖
•Noise → 𝑛෤ 𝑖
•Gain → 𝜎𝑖
Parallel Decomposition of the MIMO Channel
Independence of Channels
• Channels do not interfere with each other
• Each behaves like a separate SISO channel
• Gains 𝜎𝑖 are related (all come from same 𝐻)
• Channels are independent in transmission , But linked through
channel matrix & total power
• Can send 𝑅𝐻 independent data streams
• Data rate increases by factor:
Multiplexing Gain = 𝑅𝐻
Parallel Decomposition of the MIMO Channel
Example :
Find the equivalent parallel channel model for a MIMO channel with channel gain matrix:
2×2 MIMO Channel
3 1
𝐻= ;
1 3

𝐻 = 𝑈Σ𝑉 𝐻
Compute 𝐻𝐻 𝐻
Since real matrix → 𝐻 𝐻 = 𝐻 𝑇
3 1 3 1 10 6
𝐻𝐻 𝑇 = =
1 3 1 3 6 10
• Find Eigenvalues
• Solve:
10 − 𝜆 6
det =0
6 2 10 − 𝜆
10 − 𝜆 − 36 = 0
• 𝜆2 − 20𝜆 + 64 = 0
𝜆 = 16, 4
Parallel Decomposition of the MIMO Channel
Singular Values
• 𝜎𝑖 = 𝜆𝑖
• 𝜎1 = 4, 𝜎2 = 2

Form Σ
4 0
• Σ=
0 2
For each eigenvalue:
𝐴 − 𝜆𝐼 𝐱 = 0

Eigenvector for 𝜆1 = 16

10 − 16 6 −6 6
𝐴 − 16𝐼 = =
6 10 − 16 6 −6
Write equations
−6𝑥 + 6𝑦 = 0
6𝑥 − 6𝑦 = 0
Parallel Decomposition of the MIMO Channel
• Solve
−6𝑥 + 6𝑦 = 0 ⇒ 𝑦 = 𝑥
Choose value
• Let: 𝑥 = 1 ⇒ 𝑦 = 1
Eigenvector:
1
1
• Normalize
Length:
12 + 12 = 2
Normalized vector:
1 1
𝐮1 =
2 1
Parallel Decomposition of the MIMO Channel
• Eigenvector for 𝜆2 = 4
10 − 4 6 6 6
𝐴 − 4𝐼 = =
6 10 − 4 6 6

6𝑥 + 6𝑦 = 0
6𝑥 + 6𝑦 = 0
Solve
• 6𝑥 + 6𝑦 = 0 ⇒ 𝑦 = −𝑥
Choose value
𝑥 = 1 ⇒ 𝑦 = −1
Eigenvector:
1
−1
Parallel Decomposition of the MIMO Channel
Normalize
• Length:
12 + −1 2 = 2
Normalized:
1 1
𝐮2 =
2 −1
Final Matrix 𝑈
Put eigenvectors as columns:
1 1 1
𝑈=
2 1 −1
Parallel Decomposition of the MIMO Channel
To find V
𝑉 = 𝐻𝐻 𝑈 Σ −1
3 1
𝐻=
1 3
Compute 𝐻𝐻
3 1
𝐻𝐻 = 𝐻𝑇 =
1 3
1 1 1
𝑈=
2 1 −1
Compute Σ−1
4 0
Σ=
01 2
0
Σ −1 = 4
1
0
2
Parallel Decomposition of the MIMO Channel
Compute 𝐻𝐻 𝑈
𝐻 3 1 1 1 1
•𝐻 𝑈= ⋅
1 3 2 1 −1
1 3+1 3−1 1 4 2
= =
2 1+3 1−3 2 4 −2
𝑉 = 𝐻𝐻 𝑈 Σ −1
1
1 4 2 4 0
=
2 4 −2 0 1
2
Parallel Decomposition of the MIMO Channel
1 1 1
𝑉=
2 1 −1

𝐻
1 1 1
𝑉 =
2 1 −1
Final SVD Expression
𝐻 = 𝑈Σ𝑉 𝐻
3 1 1 1 1 4 0 1 1 1
=
1 3 2 1 −1 0 2 2 1 −1
MIMO Channel Capacity
Case 1: (CSI Known at the receiver)
Static Channels (CSI Known at the receiver)
• For a static MIMO channel, the channel matrix 𝐻 does not change
with time (or changes very slowly), which makes it relatively easy for
the receiver to estimate it accurately using known pilot signals.
Because of this, we assume that the receiver has perfect knowledge
of the channel, a condition referred to as CSIR (Channel State
Information at the Receiver).
• Under this assumption, the capacity of the channel is defined as the
maximum mutual information between the transmitted signal
vector 𝒙 and the received signal vector 𝒚, where the maximization is
taken over all possible input distributions 𝑝 𝑥 .
MIMO Channel Capacity (CSI Known at the
receiver)
• The mutual information 𝐼 𝑋 𝑌 can be written as the difference between two
entropy terms: 𝑯 𝒀 , the entropy of the received signal, and 𝑯 𝒀 ∣ 𝑿 ,the
conditional entropy of the received signal given the transmitted signal. The
term 𝐻 𝑌 represents the total uncertainty or randomness in the received signal,
while 𝐻 𝑌 ∣ 𝑋 represents the remaining uncertainty in the received signal when
the transmitted signal is already know.
I(X;Y)=H(Y)−H(Y∣X)
• From the system model 𝑦 = 𝐻𝑥 + 𝑛, if the transmitted signal 𝑥 is known, then
the only source of randomness in 𝒚 is the noise 𝒏. This means that the
conditional entropy 𝐻 𝑌 ∣ 𝑋 is equal to the entropy of the noise, i.e.,
𝑯 𝒀∣𝑿 =𝑯 𝒏 .
• Since the noise is independent of the transmitted signal and its statistical
properties (such as variance) are fixed, its entropy remains constant regardless of
how we choose the input distribution 𝑝 𝑥 .
MIMO Channel Capacity (CSI Known at the
receiver)
• Starting from the MIMO system model
𝑦 = 𝐻𝑥 + 𝑛
• we assume that the input vector 𝑥has covariance 𝑅𝑥 = 𝐸 𝑥𝑥 𝐻 ,and the noise
𝑛 is zero-mean with covariance 𝐼. Using these, the output covariance is
computed as:
𝑅𝑦 = 𝐸 𝑦𝑦 𝐻 = 𝐸 𝐻𝑥 + 𝑛 𝐻𝑥 + 𝑛 𝐻
Expand the expression
𝑦𝑦 𝐻 = 𝐻𝑥 + 𝑛 𝐻𝑥 + 𝑛 𝐻
Take conjugate transpose:
𝐻𝑥 + 𝑛 𝐻 = 𝑥 𝐻 𝐻 𝐻 + 𝑛𝐻
So: 𝑦𝑦 𝐻 = 𝐻𝑥 + 𝑛 𝑥 𝐻 𝐻 𝐻 + 𝑛𝐻
y𝑦 𝐻 = 𝐻𝑥𝑥 𝐻 𝐻 𝐻 + 𝐻𝑥𝑛𝐻 + 𝑛𝑥 𝐻 𝐻 𝐻 + 𝑛𝑛𝐻 s
MIMO Channel Capacity (CSI Known at the
receiver)
Take expectation 𝐸 ⋅
𝑅𝑦 = 𝐸 𝑦𝑦 𝐻
Apply expectation to each term:
𝑅𝑦 = 𝐸 𝐻𝑥𝑥 𝐻 𝐻𝐻 + 𝐸 𝐻𝑥𝑛𝐻 + 𝐸 𝑛𝑥 𝐻 𝐻𝐻 + 𝐸 𝑛𝑛𝐻
Simplify each term
First term:
• 𝐸 𝐻𝑥𝑥 𝐻 𝐻𝐻 = 𝐻𝐸 𝑥𝑥 𝐻 𝐻𝐻 = 𝐻𝑅𝑥 𝐻𝐻
Second term:
• 𝐸 𝐻𝑥𝑛𝐻 = 𝐻𝐸 𝑥𝑛𝐻 ; but Signal 𝑥 and noise 𝑛 are independent
• 𝐸 𝑥𝑛𝐻 = 0
Third term:
• 𝐸 𝑛𝑥 𝐻 𝐻𝐻 = 𝐸 𝑛𝑥 𝐻 𝐻𝐻 = 0
MIMO Channel Capacity (CSI Known at the
receiver)
Fourth term:
• 𝐸 𝑛𝑛𝐻 = 𝑅𝑛
For normalized noise:
𝑅𝑛 = 𝐼
Final Result:
𝑅𝑦 = 𝐻𝑅𝑥 𝐻 𝐻 + 𝐼
Chanel capacity C=I(X;Y)=H(Y)−H(n).
• Now, to maximize this, we choose 𝑥 to be complex Gaussian, which makes
𝑦 also Gaussian. For a complex Gaussian vector, the entropy is
• 𝐻 𝑌 = 𝐵 log 2
det 𝜋𝑒𝑅𝑦 , 𝐻 𝑛 = 𝐵 log 2
det 𝜋𝑒𝐼 ; e≈2.71828
MIMO Channel Capacity (CSI Known at the
receiver)
• Substituting into the mutual information:
𝐼 𝑋 𝑌 = 𝐵 log 2 det 𝜋𝑒𝑅𝑦 − 𝐵 log 2
det 𝜋𝑒𝐼 .
Using the logarithm property:
log det 𝐴 − log det 𝐵 = log det 𝐴𝐵−1 ,
• 𝐼 𝑋 𝑌 = 𝐵 log 2
det 𝑅𝑦 .
Now substitute the expression for output covariance:
𝑅𝑦 = 𝐻𝑅𝑥 𝐻𝐻 + 𝐼,
so we get:
𝐼 𝑋 𝑌 = 𝐵 log 2 det( 𝐻𝑅𝑥 𝐻𝐻 + 𝐼).
MIMO Channel Capacity (CSI Known at the
receiver)
• Rewriting in standard form:
𝐻
𝐼 𝑋 𝑌 = 𝐵 log 2
det( 𝐼 + 𝐻𝑅𝑥 𝐻 ).
• Finally, capacity is obtained by maximizing over all input covariance
matrices 𝑅𝑥 under the power constraint Tr 𝑅𝑥 = 𝜌

𝑯
𝑪 = 𝒎𝒂𝒙 𝑩 𝒍𝒐𝒈 𝟐
𝒅𝒆𝒕( 𝑰 + 𝑯𝑹 𝒙 𝑯 ).
𝑹𝒙 : Tr 𝑹𝒙 =𝝆
Case 2: MIMO Channel Capacity (CSI Known at
the transmitter and receiver)
• When the channel matrix 𝐻is known at both the transmitter and the
receiver (CSIT + CSIR), we can do something smarter than just sending
signals blindly—we can adapt how we transmit based on the
channel. The key idea is to use the Singular Value Decomposition
(SVD) of the channel:
𝐻 = 𝑈Σ𝑉 𝐻
• where Σcontains the singular values 𝜎𝑖 .By applying appropriate
precoding (using 𝑉) at the transmitter and decoding (using 𝑈 𝐻 ) at the
receiver, the MIMO channel is transformed into a set of independent
parallel channels.
MIMO Channel Capacity (CSI Known at the
transmitter and receiver)
After this transformation, the system behaves like:
𝑦𝑖 = 𝜎𝑖 𝑥𝑖 + 𝑛𝑖
• which means each subchannel is like a simple SISO channel with gain 𝜎𝑖 .If
there are 𝑅𝐻 nonzero singular values, then we effectively have 𝑅𝐻 parallel
channels. This number is called the degrees of freedom of the MIMO
system.
• Now, instead of sending equal power on all antennas, we can distribute
total power optimally across these parallel channels. Let 𝜌𝑖 be the power
allocated to the 𝑖-th subchannel. Then the total power constraint is:
෍ 𝜌𝑖 ≤ 𝜌
𝑖
MIMO Channel Capacity (CSI Known at the
transmitter and receiver)
• We start from the general capacity expression (with CSIR and now
also CSIT):
𝐶 = max 𝐵 log 2 det( 𝐼 + 𝐻𝑅𝑥 𝐻𝐻 )
𝑅𝑥 : Tr 𝑅𝑥 =𝜌
Apply SVD of the channel
• Write the channel matrix as:
𝐻 = 𝑈Σ𝑉 𝐻
Substitute into the capacity equation:
𝐶 = max 𝐵 log 2 det( 𝐼 + 𝑈Σ𝑉 𝐻 𝑅𝑥 𝑉Σ 𝐻 𝑈 𝐻 )
𝑅𝑥
MIMO Channel Capacity (CSI Known at the
transmitter and receiver)
Use unitary matrix property
• Determinant does not change under unitary transformations

det 𝐼 + 𝑈𝐴𝑈 𝐻 = det 𝐼 + 𝐴

• Thus we remove 𝑈and 𝑈 𝐻 :
𝐶 = max 𝐵 log det( 𝐼 + Σ𝑉 𝐻 𝑅 𝑉Σ 𝐻 )
2 𝑥
𝑅𝑥
Define a new variable
Let:
𝑅෨ 𝑥 = 𝑉 𝐻 𝑅𝑥 𝑉
• Since 𝑉is unitary: Trace constraint remains same: Tr 𝑅𝑥 = Tr 𝑅෨ 𝑥

𝐶= max 𝐵 log det( 𝐼 + Σ ෨𝑥 Σ𝐻 )

𝑅
𝑅𝑥 : Tr 𝑅෨ 𝑥
෨ =𝜌 2
MIMO Channel Capacity (CSI Known at the
transmitter and receiver)
Σ is diagonal:
𝜎1 0 ⋯
Σ= 0 𝜎2 ⋯
⋮ ⋱
The optimal 𝑅෨𝑥 is diagonal

𝜌1 0 ⋯
𝑅෨𝑥 = 0 𝜌2 ⋯
⋮ ⋱
with:
෍ 𝜌𝑖 ≤ 𝜌
𝑖
MIMO Channel Capacity (CSI Known at the
transmitter and receiver)
𝜎12 𝜌1 0 ⋯
• Σ𝑅෨ 𝑥 Σ𝐻 = 0 𝜎22 𝜌2 ⋯
⋮ ⋱
1 + 𝜎12 𝜌1 0 ⋯
𝐼 + Σ𝑅෨ 𝑥 Σ𝐻 = 0 1 + 𝜎22 𝜌2 ⋯
⋮ ⋱

𝑅𝐻

𝐶 = max ෍ 𝐵 log 2
1 + 𝜎𝑖2 𝜌𝑖
𝜌𝑖 :෌𝜌𝑖 ≤𝜌
𝑖=1
MIMO Channel Capacity (CSI Known at the
transmitter and receiver
𝑃
• We now express this in terms of actual transmit power. Recall that 𝜌 = 𝜎2 ,
where 𝑃is total transmit power and 𝜎 2 is noise variance. If we let 𝑃𝑖 be the
𝑃𝑖
power allocated to the 𝑖-th channel, then 𝜌𝑖 = 𝜎2 .Substituting this into the
expression gives:
𝜎𝑖2 𝑃𝑖
• 𝐶 = max σ𝑅𝑖=1
𝐻
𝐵 log 1+ ,
𝑃𝑖 2 𝜎2
• with the constraint σ𝑖 𝑃𝑖 ≤ 𝑃.
• To simplify the expression further, define
𝜎𝑖2 𝑃
𝛾𝑖 = 𝜎2 , which represents the SNR of the 𝑖-th channel if the entire power
𝑃 were allocated to that channel.
MIMO Channel Capacity (CSI Known at the
transmitter and receiver
• Using this definition, we rewrite:
𝜎𝑖2 𝑃𝑖 𝑃𝑖
2
= 𝛾𝑖 ,
𝜎 𝑃
and the capacity becomes:
𝑅𝐻
𝑃𝑖
𝐶 = max ෍ 𝐵 log 2
1 + 𝛾𝑖 .
𝑃𝑖 𝑃
𝑖=1
• This expression clearly shows how capacity depends on how we distribute
power across channels. At high SNR, all 𝛾𝑖 are large, so every channel
contributes significantly, and capacity increases roughly linearly with the
number of channels (degrees of freedom). At low SNR, however, only the
channel with the largest 𝛾𝑖 contributes meaningfully, so it is optimal to
allocate almost all power to that single best channel.
MIMO Channel Capacity (CSI Known at the
transmitter and receiver
• To find the optimal 𝑃𝑖 ,we solve this constrained optimization problem, which leads to the water-filling
solution:
1 1
𝑃𝑖 − , 𝛾𝑖 ≥ 𝛾0
= ൞𝛾0 𝛾𝑖
𝑃
0, 𝛾𝑖 < 𝛾0
where 𝛾0 is a cutoff level chosen so that the total power constraint is satisfied.
• This result means that power is allocated only to channels whose SNR is above a certain threshold. Strong
channels (large 𝛾𝑖 )get more power, while weak channels (small 𝛾𝑖 )may get none at all.
• Substituting this optimal allocation back into the capacity expression yields:
𝛾𝑖
𝐶 = ෍ 𝐵 log 2 ,
𝛾0
𝑖:𝛾𝑖≥𝛾0
• which is the final capacity under water-filling. This expression shows that capacity depends only on the ratio
between each channel’s strength and the cutoff level, reinforcing the idea that only sufficiently strong
channels contribute to the total capacity.
Water filling Algorithm- numerical
Problem:
Find the channel capacity and the optimal power allocation for a MIMO
system whose singular values are 𝜎1 = 1.3333 ,𝜎2 = 0.5129 ,and 𝜎3
𝑃
= 0.0965. Assume the total SNR 𝜌 = 2 = 10 (linear scale) and
𝜎
bandwidth 𝐵 = 1Hz. Use the water-filling algorithm to determine how
power should be distributed among the parallel channels and compute
the resulting capacity.
Water filling Algorithm- numerical
First, compute the effective SNR of each parallel channel:
𝛾𝑖 = 𝜎𝑖2 ⋅ 𝜌
𝛾1 = 17.7769, 𝛾2 = 2.6307, 𝛾3 = 0.0931
Now assume all three channels are active and apply the water-filling constraint:
3
1 1
෎ − =1
𝛾0 𝛾𝑖
𝑖=1
which simplifies to:
3 1 1 1
=1+ + + = 12.1749
𝛾0 17.7769 2.6307 0.0931
𝛾0 = 0.2685
Water filling Algorithm- numerical
This value is invalid because 𝛾3 = 0.0931 < 𝛾0 ,so the third channel cannot be used. We discard it and consider only the first two
channels.
Now apply the constraint for two channels: 2
1 1
෎ − =1
𝛾0 𝛾𝑖
𝑖=1
2 1 1
= 1+ + = 1.4364
𝛾0 17.7769 2.6307
𝛾0 = 1.392
This is valid since both 𝛾1and 𝛾2 are greater than 𝛾0.
Now compute power allocation:
𝑃𝑖 1 1
= −
𝑃 𝛾0 𝛾𝑖
1 1
𝑃1 /𝑃 = − = 0.662
1.392 17.7769
1 1
𝑃2 /𝑃 = − = 0.338
1.392 2.6307
Water filling Algorithm- numerical
Finally, compute capacity:
𝛾1 𝛾2
𝐶 = log 2 + log 2
𝛾0 𝛾0
𝐶 = log 2
12.77 + log 2 1.89 ≈ 4.59 bits/s/Hz

Final Answer:
Optimal power allocation:
𝑃1 = 66.2%𝑃, 𝑃2 = 33.8%𝑃, 𝑃3 = 0
• Channel capacity:
𝐶 ≈ 4.59 bits/s/Hz
Case 3: Channel capacity with beamforming:
MIMO Diversity Gain
• In MIMO beamforming, instead of transmitting different data
streams from different antennas (spatial multiplexing), the same
symbol is transmitted from all antennas with carefully chosen
complex weights.
The purpose is to maximize:
• Array gain → stronger received signal (higher SNR)
• Diversity gain → robustness against fading rather than maximizing
channel capacity.
Channel capacity with beamforming:
MIMO Diversity Gain
• These weights are chosen so that the signals combine coherently at the receiver, thereby
improving the received signal strength and reliability.
• The vector 𝑢 appears because the receiver has multiple antennas, and the signals
received at all antennas must be combined into a single output signal. Before combining,
the received signal vector of an 𝑀𝑟 -antenna MIMO system is:

𝑟 = 𝐻𝑣𝑥 + 𝑛
where:
• 𝐻 is the channel matrix,
• 𝑣 is the transmit beamforming vector,
• 𝑥 is the transmitted symbol,
• 𝑛 is the noise vector,
• 𝑟 is the vector of received signals at all receive antennas.
Channel capacity with beamforming:
MIMO Diversity Gain
• Since the receiver obtains multiple signal components, it applies a
linear combiner using the vector 𝑢. The combiner multiplies the
received vector 𝑟by the Hermitian transpose 𝑢𝐻 to produce a single
scalar output:
𝑦 = 𝑢𝐻 𝑟
Substituting 𝑟 = 𝐻𝑣𝑥 + 𝑛:
𝑦 = 𝑢𝐻 𝐻𝑣𝑥 + 𝑛
Expanding:
𝑦 = 𝑢𝐻 𝐻𝑣𝑥 + 𝑢𝐻 𝑛
Channel capacity with beamforming:
MIMO Diversity Gain
• Each transmit antenna sends a weighted version of the same symbol, and
the receiver combines the incoming signals using the receive weights. This
process provides array gain and diversity gain through coherent addition of
the signal components arriving through multiple propagation paths.

• When the channel matrix 𝐻 is known at both the transmitter and receiver,
the received SNR is maximized by choosing the transmit and receive
beamforming vectors as the principal right and left singular vectors of 𝑯,
respectively, obtained from the singular value decomposition 𝑯
= 𝑼𝚺𝑽𝑯 .If 𝜎max is the largest singular value of 𝐻, the resulting
received SNR becomes 𝜎max 2 𝜌 ,where 𝜌 is the transmit SNR.
Channel capacity with beamforming:
MIMO Diversity Gain
• The corresponding channel capacity is 𝑪 = 𝑩 𝐥𝐨𝐠 𝟐 𝟏 + 𝝈𝐦𝐚𝐱 𝟐 𝝆 ,
which is equivalent to the capacity of a SISO channel having an
effective channel gain of 𝜎max 2 .
• Although this capacity is lower than the full capacity achievable
through spatial multiplexing, beamforming significantly reduces
receiver complexity because only a single data stream needs to be
detected.
Channel capacity with beamforming:
MIMO Diversity Gain: Numerical
Consider a MIMO channel with gain matrix
0.7 0.9 0.8
𝐻 = 0.3 0.8 0.2
0.1 0.3 0.9
Find the capacity of this channel under beamforming, assuming
channel knowledge is available at both the transmitter and receiver.
Take bandwidth 𝐵 = 100 kHzand SNR 𝜌 = 10 dB.
Channel capacity with beamforming:
MIMO Diversity Gain: Numerical
• For MIMO beamforming, the channel capacity is
𝐶 = 𝐵 log 2 1 + 𝜎max 2 𝜌
where:
• 𝐵= bandwidth
• 𝜌= SNR in linear scale
• 𝜎max =largest singular value of 𝐻
𝐶 = 𝐵 log 2 1 + 𝜎max 2 𝜌
Since
𝜎max 2 = 𝜆max
where 𝜆max is the largest eigenvalue of 𝐻𝐻 𝐻 ,the formula becomes
𝐶 = 𝐵 log 2 1 + 𝜆max 𝜌
Channel capacity with beamforming:
MIMO Diversity Gain: Numerical
• Compute 𝐻𝐻 𝐻
0.7 0.9 0.8 0.7 0.3 0.1
𝐻𝐻 𝐻 = 0.3 0.8 0.2 0.9 0.8 0.3
0.1 0.3 0.9 0.8 0.2 0.9

1.94 1.09 1.06

𝐻𝐻 𝐻 = 1.09 0.77 0.45
1.06 0.45 0.91
• det(HH^H−λI)=0
det 𝐻𝐻 𝐻 − 𝜆𝐼 = 0
The eigenvalues of the matrix are:
• 𝜆1 = 3.17
• 𝜆2 = 0.40
• 𝜆3 = 0.05
The largest eigenvalue is
𝜆max = 3.17
Channel capacity with beamforming:
MIMO Diversity Gain: Numerical
Convert SNR from dB to Linear Scale
Given
𝜌 = 10 dB
Convert to linear scale:
𝜌 = 1010/10 = 10

Step 5: Compute Capacity

Substitute the values into the capacity formula:
𝐶 = 105 log 2 1 + 3.17 × 10
𝐶 = 105 log 2 1 + 31.7
𝐶 = 105 log 2 32.7
𝐶 = 105 × 5.03
𝐶 ≈ 5.03 × 105 bps
𝐶 ≈ 503 kbps
Case 4: Channel unknown at transmitter:
uniform power allocation
• We start from the assumption that the transmitter does not know
the channel matrix 𝐻. Since it has no knowledge of which antenna
path is strong or weak, it cannot favour one antenna over another.
Therefore, the transmitter treats all antennas equally and distributes
power uniformly.
• Suppose: Total transmit power (normalized SNR) is 𝜌
• Number of transmit antennas is 𝑀𝑡
• If power is divided equally among all transmit antennas, then each
antenna gets:
𝜌
𝑀𝑡
Channel unknown at transmitter: uniform
power allocation
• Now recall that the input covariance matrix is defined as
𝑅𝑥 = 𝐸 𝑥𝑥 𝐻 .
• For equal power allocation: each transmit signal has variance 𝜌/𝑀𝑡 ,
signals from different antennas are independent (uncorrelated).
• Therefore:
diagonal elements of 𝑅𝑥 =power on each antenna = 𝜌/𝑀𝑡 ,
off-diagonal elements = 0 (because antennas are uncorrelated).
So the covariance matrix becomes:
𝜌/𝑀𝑡 0 ⋯
𝑅𝑥 = 0 𝜌/𝑀𝑡 ⋯
⋮ ⋱
Channel unknown at transmitter: uniform
power allocation
This is exactly:
𝜌
𝑅𝑥 = 𝐼𝑀𝑡
𝑀𝑡
where 𝐼𝑀𝑡 is the identity matrix of size 𝑀𝑡 × 𝑀𝑡 .
So this equation comes directly from:
• The trace confirms the total power:
𝜌
Tr 𝑅𝑥 = 𝑀𝑡 × = 𝜌.
𝑀𝑡
• Hence the total transmit power remains 𝜌.
Channel unknown at transmitter: uniform
power allocation
• Substituting this covariance matrix into the general MIMO capacity
expression
𝐼 𝑋 𝑌 = 𝐵 log 2 det 𝐼 + 𝐻𝑅𝑥 𝐻𝐻 , (From case 1)
we get
𝜌
𝐼 𝑥 𝑦 = 𝐵 log 2
det 𝐼𝑀𝑟 + 𝐻𝐻𝐻 .
𝑀𝑡
Channel unknown at transmitter: uniform
power allocation
Now substitute this result into the mutual information formula:
𝜌
𝐼 𝑥 𝑦 = 𝐵 log 2 det 𝐼𝑀𝑟 + 𝐻𝐻 𝐻 .
𝑀𝑡
Using
1
𝐻𝐻 𝐻 → 𝐼𝑀𝑟 ,
𝑀𝑡
we get:
𝐼 𝑥 𝑦 = 𝐵 log 2 det 𝐼𝑀𝑟 + 𝜌𝐼𝑀𝑟 .
Factor the identity matrix:
𝐼𝑀𝑟 + 𝜌𝐼𝑀𝑟 = 1 + 𝜌 𝐼𝑀𝑟 .
So:
𝐼 𝑥 𝑦 = 𝐵 log 2 det 1 + 𝜌 𝐼𝑀𝑟 .
Channel unknown at transmitter: uniform
power allocation
Now use the determinant property:
det 𝑐𝐼𝑛 = 𝑐 𝑛 .
Hence:
𝑀𝑟
det 1 + 𝜌 𝐼𝑀𝑟 = 1 + 𝜌 .
Therefore:
𝑀𝑟
𝐼 𝑥 𝑦 = 𝐵 log 2 1 + 𝜌 .
Using the logarithm rule:
• log 𝑎𝑛 = 𝑛log 𝑎 ,
we finally get:
𝑰 𝒙 𝒚 = 𝑴𝒓 𝑩 𝒍𝒐𝒈 𝟐 𝟏 + 𝝆 .
𝑪 = 𝑴𝒓 𝑩 𝒍𝒐𝒈 𝟐 𝟏 + 𝝆 .
Diversity-Multiplexing Trade-offs

Multiple antennas in a MIMO system can improve

performance in two main ways:
 Multiplexing gain (capacity improvement)
 Diversity gain (reliability improvement)
In beamforming:
In spatial multiplexing: •The same signal is transmitted from multiple antenn
with suitable weights.
•The MIMO channel is decomposed into parallel •Signals combine coherently at the receiver.
subchannels. •This provides:
•Different data streams are transmitted simultaneously. •Array gain
•This increases channel capacity and data rate. •Diversity gain
•The performance of each subchannel depends on the •Higher received SNR
singular values of the channel matrix. •Better reliability
•Channels with small singular values have poor SNR •Beamforming creates a robust communication link w
and require strong channel coding. low error probability.
Diversity-Multiplexing Trade-offs

•Antennas do not have to be used entirely for multiplexing or diversity.

•Some spatial dimensions can provide diversity gain.
•Remaining dimensions can provide multiplexing gain.
•This leads to the diversity–multiplexing trade-off.

•The diversity–multiplexing trade-off studies the

relationship between:
 Data rate
 Probability of error
 System complexity
Diversity-Multiplexing Trade-offs
For systems where only the receiver has CSI:

•Full diversity and full multiplexing gain cannot generally be achieved simultaneously for finite block
lengths.

•Increasing multiplexing gain usually reduces diversity gain.

•Increasing diversity gain usually reduces multiplexing gain.

Frequency-Selective MIMO Channels
• When the bandwidth of a MIMO channel is large compared to the
multipath delay spread, the channel experiences intersymbol
interference (ISI).This situation is similar to ISI in SISO channels,
where delayed copies of transmitted symbols overlap with adjacent
symbols.
There are two main approaches to handle ISI in MIMO systems:
• Channel equalization
• OFDM (Orthogonal Frequency Division Multiplexing)
Frequency-Selective MIMO Channels
In MIMO equalization:
• An equalizer is used to reduce or remove ISI effects.
• Equalization is more complex than in SISO systems because:
• The channel must be equalized across both:
• Space (multiple antennas)
• Time (multipath delays)
• When combined with space-time coding:
• Equalizer design becomes even more difficult.
• Space-time codes are often:
• Nonlinear
• Noncausal
• In some special cases:
• The structure of the space-time code can simplify the problem.
• The MIMO equalization problem can be transformed into an equivalent SISO equalization problem.
• Then conventional SISO equalizers can be used.
Frequency-Selective MIMO Channels
An alternative approach is OFDM.
•OFDM converts a wideband frequency-selective channel into many narrowband subchannels.
•Each subchannel experiences approximately flat fading instead of frequency-selective fading.

Applying OFDM to MIMO systems results in:

•A parallel collection of narrowband MIMO channels.
•Space-time coding and modulation techniques can then be applied independently on each subchannel.

Advantages of MIMO-OFDM:
•Simplifies equalization
•Efficiently handles frequency-selective fading
•Converts a difficult wideband problem into multiple simpler narrowband problems

MIMO frequency-selective fading channels provide diversity in:

•Space
•Time
•Frequency
Smart Antennas
Smart antennas use:

•Sectorized antennas
•Phased array antennas

Directional antennas focus signal energy in specific directions instead of radiating equally in all directions.

Benefits of directional antennas:

•Increase signaling range

•Reduce intersymbol interference (ISI)
•Reduce flat fading
•Suppress interference from other users
•Improve overall system capacity
Smart Antennas

•Interference signals usually arrive from different directions.

•Directional antennas can:
•Null interference
•Attenuate interference
•This reduces unwanted received power.
•Reflected multipath signals also arrive from different angles.
•Directional antennas can suppress these reflections.
•This helps reduce:
•Delay spread
•ISI
•Flat fading
Smart Antennas
Sectorized Antennas
• Sectorized antennas provide high gain over a fixed angular region.
• Commonly used in cellular base stations.
• Purpose of sectorization:
• Reduce interference between users.
• Example:
• A base station covers 360∘ .
• It can be divided into:
• Three sectors of 120 ∘each.
• Each sector uses a separate antenna.
• Benefit:
• Interference is reduced approximately by the number of sectors.
• With 3 sectors:
• Interference reduces by a factor of 3.
• Disadvantage:
• Users moving between sectors require handoff.

• Therefore cellular systems often prefer sectorized antennas over fully directional antennas.
Smart Antennas
• Phased Array Antennas
Directional antennas commonly
use phased arrays. For an array with 𝑁antennas:
• In phased arrays: • Up to 𝑁nulls can be formed.
• Multiple antenna elements are used. • This can suppress 𝑁interferers.
• The phase of each antenna element is If the number of interferers is 𝑁𝐼 < 𝑁 :
adjusted. • 𝑁𝐼 antennas can null interference.
• By changing phases: • Remaining antennas can provide diversity gain.
• Beam directions can be controlled. Directional antennas require knowledge of:
• Null directions can also be controlled. • Desired signal direction
• Terms: • Interference directions
• Beam → direction of high antenna gain
• Null → direction of very low antenna
gain