' $

EE5131/EE6131 Wireless Communications

Lecture 7: Multicarrier Modulation

& %
1
' $

ISI Countermeasures
PL−1
• Frequency-selective fading channel with ISI: y(n) = l=0 hl x(n − l) + z(n)
• Equalization
– Signal processing at receiver to eliminate ISI (similar to MIMO detection)
– Can be very complex at high data rates, and performs poorly in
fast-changing channels
– Not that common in state-of-the-art wireless systems
• Spread spectrum
– Superimpose a fast (wideband) spreading sequence on top of data
sequence, allowing resolution for multipath combining (Rake Receiver)
• Multicarrier Modulation
– Break data stream into lower-rate substreams modulated onto narrowband
flat-fading subchannels, each with less severe ISI
& %
2
' $

Outline

• Multicarrier Modulation Basics

• Digital Implementation
– Orthogonal Frequency Division Multiplexing (OFDM)

– Vector Coding (EE6131 only)

• MIMO-OFDM (EE6131 only)

• Required reading: textbook, Chapter 4.3, 12.1-12.4, 12.6

& %
3
' $

Multicarrier Modulation

• Non-overlapping Subchannels
– Transmitter

& %
4
' $

– Receiver

• Overlapping Subchannels
– Transmitter (similar to that of non-overlapping case)

& %
5
' $

– Receiver

& %
6
' $

Mitigation of Subchannel Fading

• Frequency equalization (SNR per subchannel unchanged)

• Precoding (similar to channel inversion power control)
• Coding with interleaving over time and frequency
• Adaptive power and rate loading over time and frequency (similar to
adaptive modulation)

& %
7
' $

Digital Implementation of Multicarrier Modulation

• Orthogonal Frequency Division Multiplexing (OFDM)

– Adopt cyclic prefix to mitigate ISI
– Channel decomposition based on eigenvalue decomposition
– Channel not needed to be known at Tx
– Power loss due to cyclic prefix
• Vector Coding
– Adopt guard interval to eliminate ISI
– Channel decomposition based on singular value decomposition
– Channel needed to be known at Tx
– No power loss in guard interval
& %
8
' $

Block-Based Transmission

• Signals are transmitted in consecutive N -symbol blocks over time:

L−1
X
y(n) = hl x(n − l) + z(n), n = 0, . . . , N − 1 (1)
l=0

– hl denotes the complex channel gain for the lth delayed path,
l = 0, . . . , L − 1, where L is the number of resolvable paths
– E[|x(n)|2 ] ≤ P
– z(n) ∼ CN (0, σz2 )
– Assume time-invariant multipath channel (no fading)

& %
9
' $

OFDM Transmitter

& %
10
' $

OFDM Receiver

& %
11
' $

Matrix Representation

• Transmitted signal block of length N + µ with Cyclic Prefix (CP):

x[N − µ], . . . , x[N − 1], x[0], x[1], . . . , x[N − 1] (2)

• CP length satisfies µ ≥ L, to eliminate the inter-block interference

• Let x = [x[0], x[1], . . . , x[N − 1]]T represent the transmitted signal vector
without CP
• x is generated by OFDM modulation as

x = W HX (3)

• X = [X[0], X[1], . . . , X[N − 1]]T is the information signal vector

& %
12
' $

• W ∈ CN ×N is the Discrete Fourier Transform (DFT) matrix

 
1 1 ··· 1
 

1  1
−j2π
e N ···
−j2π(N −1)
e N 

W =√  .  (4)

N . . 

−j2π(N −1) −j2π(N −1)(N −1)
1 e N ··· e N

−j2πkm
where [W ]k,m = e N , m = 0, . . . , N − 1, k = 0, . . . , N − 1
• Check power constraint: E[kxk2 ] = E[kXk2 ] ≤ N P , since
W W H = W HW = IN
• In practice, IDFT/DFT is efficiently implemented as IFFT/FFT
• Received signal block of length N + µ is

y[−µ], . . . , y[−1], y[0], y[1], . . . , y[N − 1] (5)

& %
13
' $

• After CP removal, the resulted vector is y = [y[0], y[1], . . . , y[N − 1]]T

• The equivalent MIMO channel is
 
h0 0 ··· 0 hL−1 ··· h1
 
    h1 h0 0 0 h2

 
y[0]   x[0]
   .. .. .. .. ..  
 
  . . . . .  
 y[1]    x[1] 
  =  hL−1  +z
 ..   hL−2 0 0 0  .. 
 .    . 
   0  
hL−1 0 0 0 
y[N − 1]   x[N − 1]
 .. .. .. .. .. 
 . . . . . 
 
0 0 ··· hL−1 hL−2 ··· h0
(6)

where z = [z[0], . . . , z[N − 1]]T

• Alternatively, the MIMO channel is expressed as
y = Hx + z (7)
& %
14
' $

Eigenvalue Decomposition of Circulant Matrix

• For circulant matrix H, the following EVD exists:

H = W −1 ∆W (8)

where ∆ ∈ CN ×N is a diagonal matrix, with diagonal elements given by

L−1
X −j2πkl
H[k] = hl e N , k = 0, . . . , N − 1 (9)
l=0

– Since W W H = W H W = I N , W −1 = W H
– [H[0], . . . , H[N − 1]] is the N -point DFT of [h0 , . . . , hL−1 , 0, . . . , 0]
√
(adding N − L zeros), without the normalization by 1/ N
• Notice the eigenvalues H[k]’s are in general complex numbers, and thus
not equal to the singular values of H, which are always non-negative real
numbers
& %
15
' $

• Let the SVD of H be by U ΛV H , with singular values λ0 , . . . , λN −1

• We thus have
HH H = U Λ2 U H (10)

• Since H = W H ∆W , we also have

HH H = W H ∆∆H W (11)

• Since the eigenvalues of HH H must be unique (but eigenvectors may

not be unique), we obtain that
Λ2 = ∆∆H (12)

• Thus we have the following relationship between singular values and

eigenvalues of circulant matrix H:
λk = |H[k]|, k = 0, . . . , N − 1 (13)
& %
16
' $

MIMO Decomposition

• From (3) and (7), we have

y = Hx + z = HW H X + z = W H ∆W W H X + z = W H ∆X + z
(14)

• OFDM demodulation by applying DFT matrix to y, yielding

Y = W y = W W H ∆X + W z = ∆X + Z (15)

where Z = W z ∼ CN (0, σz2 I N )

• From (15), we see that the equivalent MIMO channel is decomposed into
N parallel SISO channels given by
Y [k] = H[k]X[k] + Z[k], k = 0, . . . , N − 1 (16)

where Y = [Y [0], . . . , Y [N − 1]]T , and Z = [Z[0], . . . , Z[N − 1]]T

& %
17
' $
2 1
PN −1
• Let E[|X[k]| ] = pk , k = 0, . . . , N − 1, with N k=0 pk ≤ P
• The receiver SNR for the kth subchannel/subcarrier is
|H[k]|2 pk
γk = , k = 0, . . . , N − 1 (17)
σz2
• The maximum achievable rate (in bps/Hz) for the OFDM system is
N −1
N 1 X
R= log2 (1 + γk ) (18)
µ + N N k=0
N
where the factor µ+N
accounts for the rate loss due to the CP insertion
• In the case of known CSIT, frequency-domain WF is optimal:
µ 2
¶+
σz
pk = ν − 2
, k = 0, . . . , N − 1 (19)
|H[k]|
PN −1
where ν is the water-level with which k=0 pk = N P
& %
18
' $

Capacity of Frequency-Selective Fading Channel

• First, consider the case of time-invariant multipath channel (no fading)

• For a fixed µ and letting N → ∞, we have the OFDM achievable rate
N
X −1 Z 2π µ 2
¶
1 1 |H(Ω)| S(Ω)
R→ log2 (1 + γk ) = log2 1 + dΩ (20)
N k=0
2π 0 σz2

1
R 2π
with 2π 0
S(Ω)dΩ ≤ P , where S(Ω) is called power spectral density
• H(Ω) is the Fourier Transform of the time-domain channel
[h0 , . . . , hL−1 ], given by
L−1
X
H(Ω) = hl e−jΩ , 0 ≤ Ω < 2π (21)
l=0

& %
19
' $

• In the CSIT-known case, optimal S(Ω) is given by WF power allocation:

µ 2
¶+
σz
S(Ω) = ν − 2
, 0 ≤ Ω < 2π (22)
|H(Ω)|
1
R 2π
where ν is determined from 2π 0 S(Ω)dΩ = P
• Thus, OFDM is asymptotically (when N → ∞) capacity optimal for
time-invariant multipath channel
• Next, consider the case of frequency-selective fading channel (with slow
fading)
• OFDM is asymptotically optimal for achieving the ergodic capacity of
frequency-selective fading channel with joint time-frequency WF power
1
R 2π
allocation subject to E[ 2π 0 S(Ω)dΩ] ≤ P (see Lecture 2)

& %
20
' $

Case Study: IEEE 802.11a Wireless LAN Standard

• Total bandwidth: 300MHz, divided into 15 20-MHz OFDM channels

• N = 64 subcarriers (SCs), in which only 48 SCs are used for data
transmission
• OFDM symbol period without CP: 64/(20 × 106 )=3.2us (why?)
• CP length: µ = 16
• OFDM symbol period with CP: 3.2us×(1+16/64)=4us
• QAM constellation size for each SC: M ∈ {2, 4, 16, 64}
• Code rate: r ∈ {1/2, 2/3, 3/4}
• Maximum throughput for each 20-MHz channel (M = 64, r = 3/4):
48×(3/4)×6/(4 · 10−6 )=54Mbps
& %
21
' $

Vector Coding
(EE6131 only, up to slide # 26)

• OFDM has power loss due to CP insertion

• Vector coding (VC) replaces CP by Guard Interval (GI), thus avoiding
power loss
• Consider VC based block transmission with N data symbols per block
and GI length equivalent to µ data symbols (similar to OFDM)

& %
22
' $

• The equivalent MIMO channel is

 
h0 0 ···
 
    h1 h0 ···

   
y[0]   x[0] z[0]
   .. ..    
 
  . . ···    
 y[1]    x[1]   z[1] 
  =  hL−1  +  (23)
 ..   hL−2 ···  ..   .. 
 .    .   . 
   0    
hL−1 ··· 
y[N + µ − 1]   x[N − 1] z[N + µ − 1]
 .. .. 
 . . ··· 
 
0 0 ···

• Alternatively, the MIMO channel is expressed as

y = Hx + z (24)

where H ∈ C(N +µ)×N

• Notice H is not circulant matrix for VC
& %
23
' $

• Let the truncated SVD of H be denoted by

H
H = Ũ Λ̃Ṽ (25)

where Λ̃ is an N × N diagonal matrix with diagonal elements λ1 , . . . , λN

1 H
• Apply eigenmode transmission, with x = Ṽ Σ X, Y = Ũ y
2

1 H
• Decomposed MIMO channel: Y = ΛΣ X + Z, with Z = Ũ z
2

1
PN
• VC achievable rate: N +µ i=1 log2 (1 + λ2i pi /σz2 )
PN
• Optimize power allocation Σ by WF under i=1 pi ≤ (N + µ)P (why?)
• Notice for VC, H needs to be known at Tx to obtain precoding matrix Ṽ
• In contrast, for OFDM, the precoding (IDFT) matrix W H is
independent of H, thus H needs not to be known at Tx: (DFT/IDFT
matrix is universal eigenvectors for circulant matrix)
& %
24
' $

MIMO-OFDM

• Consider the following MIMO frequency selective fading channel

L−1
X
y(n) = H l x(n − l) + z(n) (26)
l=0

– H l ∈ Cr×t : channel matrix for the lth path, l ∈ {0, . . . , L − 1}

– Assume sum-power constraint: E[kx(n)k2 ] ≤ P
– z(n) ∼ CN (0, σz2 I r )
• By applying OFDM modulation at each transmit antenna, and OFDM
demodulation at each receive antenna, the MIMO frequency selective
fading channel is converted to N parallel MIMO flat-fading channels:

Y [k] = H[k]X[k] + Z[k], k = 0, . . . , N − 1 (27)

& %
25
' $

where
L−1
X −j2πkl
H[k] = H le N , k = 0, . . . , N − 1 (28)
l=0

1
PN −1
• Power constraint: N k=0 E[kX[k]k2 ] ≤ P
• Assuming CSIT, eigenmode transmission can be applied at each SC to
convert the spatial MIMO channel into min(r, t) parallel SISO channels
• MIMO-OFDM is thus asymptotically capacity optimal for the MIMO
frequency-selective fading channel
• Ergodic capacity: eigenmode transmission plus power control by joint
space-time-frequency WF

& %
26
' $

Summary

• Multicarrier vs. Single-Carrier Modulation (an ongoing debate)

• MC digital implementation: OFDM vs. VC (OFDM wins for wireless)
• OFDM is asymptotically capacity optimal for frequency-selective fading
channel (but can we let N → ∞ in practice?)
• SVD/EVD-based channel decomposition is universal approach for
optimizing MIMO transmissions (over time and/or space)
• WF is universal approach to optimize power allocation over parallel
AWGN channels (in time, frequency, and/or space)

& %
27