Multi-Carrier Transmission

over Mobile Radio Channels

Jean-Paul M.G. Linnartz

Nat.Lab., Philips Research
 Outline
Introduction to OFDM
Introduction to multipath reception
Discussion of receivers for OFDM and MC-CDMA
Introduction to Doppler channels
Intercarrier Interference, FFT Leakage
New receiver designs
Simulation of Performance
Conclusions
 OFDM

OFDM: a form of MultiCarrier Modulation.

Different symbols are transmitted over different subcarriers
Spectra overlap, but signals are orthogonal.
Example: Rectangular waveform -> Sinc spectrum
 I-FFT: OFDM Transmission

Transmission of QAM symbols on parallel subcarriers

Overlapping, yet orthogonal subcarriers

User cos( ct)

symbols

cos( ct+ st)

Parallel-to-
Serial-to-

I-FFT
Serial-to-
parallel

Parallel

Serial
cos( ct+ i st)

cos( ct+ (N-1) st)

OFDM Subcarrier Spectra
OFDM signal strength versus
frequency.

Rectangle <- FFT -> Sinc

before channel

after channel

Frequency
 Applications

Fixed / Wireline:
ADSL Asymmetric Digital Subscriber Line

Mobile / Radio:
Digital Audio Broadcasting (DAB)
Digital Video Broadcasting - Terrestrial (DVB-T)
Hiperlan II
Wireless 1394
4G (?)
The Wireless Multipath Channel
 The Mobile Multipath Channel
Delay spread Doppler spread

Time

FT FT

Frequency Time

Frequency

Frequency
Effects of Multipath Delay and Doppler

Narrowband Wideband OFDM

QAM

Time
Time

Time

Frequency Frequency Frequency

 Effects of Multipath (II)

DS-CDMA Frequency MC-CDMA

Hopping

+
- + + - + -
Time

Time

Time
+
-
- - + - +
-
+
- - - + - +
+
Frequency Frequency Frequency
 Multi-Carrier CDMA

Various different proposals.

(1) DS-CDMA followed by OFDM
(2) OFDM followed by DS-CDMA
(3) DS-CDMA on multiple parallel carriers

First research papers on system (1) in 1993:

Fettweis, Linnartz, Yee (U.C. Berkeley)
Fazel (Germany)
Chouly (Philips LEP)
System (2): Vandendorpe (LLN)
System (3): Milstein (UCSD); Sourour and Nakagawa
 Multi-Carrier CDM Transmitter
Code
S/P N N N P/S
Matrix I-FFT
B C A

What is MC-CDMA (System 1)?

a form of Direct Sequence CDMA, but after spreading a Fourier
Transform (FFT) is performed.
a form of Orthogonal Frequency Division Multiplexing (OFDM),
but with an orthogonal matrix operation on the bits.
a form of Direct Sequence CDMA, but the code sequence is the
Fourier Transform of the code.
a form of frequency diversity. Each bit is transmitted
simultaneously (in parallel) on many different subcarriers.
 MC-CDM (Code Division Multiplexing) in Downlink

In the forward or downlink (base-to-mobile): all signals originate at

the base station and travel over the same path.

One can easily exploit orthogonality of user signals. It is fairly

simple to reduce mutual interference from users within the same
cell, by assigning orthogonal Walsh-Hadamard codes.

BS

MS 1
MS 2
 Synchronous MC-CDM receiver

Weigh I-Code
S/P N N N N P/S
FFT Matrix Matrix
Y W A C-1

The MC-CDM receiver

separates the various subcarrier signals (FFT)
weights these subcarriers in W, and
does a code despreading in C-1 :
(linear matrix over the complex numbers)

Compare to C-OFDM:
W := equalization or AGC per subcarrier
C-1 := Error correction decoder (non-linear operation)
 Synchronous MC-CDM receiver

Weigh I-Code
S/P N N N N P/S
FFT Matrix Matrix
Y W A C-1

Receiver strategies (How to pick W ?)

equalization (MUI reduction) w = 1/
maximum ratio combining (noise reduction) w =
Wiener Filtering (joint optimization) w = /( 2 + c)

Next step: W can be reduced to an automatic gain control, per

subcarrier, if no ICI occurs
 Synchronous MC-CDM receiver

Weigh I-Code
S/P N N N N P/S
FFT Matrix Matrix
Y W A C-1 B

Optimum estimate per symbol B is obtained from B = EB|Y

= C-1 EA|Y = C-1 A.
Thus: optimum linear receiver can implement FFT - W - C-1
Orthogonality Principle: E(A-A)YH = 0N, where A = WYH
Wiener Filtering: W = E AYH (EYYH)-1
EAYH diagonal matrix of signal power *
EYYH diagonal matrix of signal plus noise power w =
N0
* +
W can be reduced to an AGC, per subcarrier Ts
 MC-CDM BER analysis

Rayleigh fading channel

Exponential delay spread
Doppler spread with uniform angle of arrival
Perfect synchronisation
Perfect channel estimation, no estimation of ICI
Orthogonal codes

Pseudo MMSE (no cancellation of ICI)

 Composite received signal

Wanted signal
T N 1 N 1
x0 = b0 s n ,n wn ,n + m,n wn ,n c0 [n ]c0 [n m ]
N n =0 m 0 n =0

Multi-user Interference (MUI)

N 1 N 1
x MUI = Ts bk n ,n wn ,n c0 [n ]ck [n ]
k =1 n =0

Intercarrier interference (ICI)

N 1
x ICI = Ts a n n + ,n wn + ,n + c0 [n + ]
n =0 0
 Composite received signal

Wanted signal

T N 1
x0 = b0 s n,n wn,n
N n =0

Multi-User Interference (MUI)

2
2 Ts2 N 1
MUI = E ch E x MUI x *MUI = 2
E bk2 E ch n ,n wn ,n n ,n wn ,n
N k =1 nA+ nA

Intercarrier interference (ICI)

1 N 1 2 N 1 N 1 2
E [ c0 ( n )ck ( n m )] E ch m,n E ch wn ,n
2 2 2
ICI = E bk
N k =1 0 n =0 n =0
 BER for MC-CDMA

Avg. BER

10-1 (4) BER for BPSK versus Eb/N0

10-2 OFDM
(5) (1) 8 subcarriers
10-3 (1)
(2) 64 subcarriers
10-4 (2)
(3) infinitely many subcarriers
10-5 (4) 8 subc., short delay spread
(3)
AWGN (5) 8 subc., typical delay spread
5 10 15
Local-meanEn/N0
Eb/N0Eb/No (dB)
Local-mean Eb/N0
 Capacity
relative to non-fading channel

Coded-OFDM MC-CDM
Data Processing Theorem:
same as N fading channels COFDM = CMC-CDM
N
N0
COFDM = 2 exp 0 x 1 log
2 (1 + 2 x ) dx
0 P0Ts P0Ts
2
In practise, we loose a little.
In fact, for infinitely many
1 N0 N0
COFDM = exp E1 subcarriers,
ln 2 2 P0Ts 2 P0Ts

CMC-CDM = log2(1 + P0Ts/N0).

For large P0Ts/N0 on a Rayleigh
fading channel, OFDM has 0.4
where is MC-CDM figure of
bit less capacity per dimension
merit, typically -4 .. -6 dB.
than a non-fading channel.
 Capacity

6
Non-fading,
LTI
5
Rayleigh
Capacity: Bits per Subcarrier

MC-CDM
3

1 -* : Rayleigh

* : MC-CDMA

- : LTI
0
-5 0 5 10 15 20 25 30 35 40
Capacity per dimension
Local-meanversus
En/N0 (dB) local-mean EN/N0,
no Doppler.
 MC-CDMA in uplink
In the reverse or uplink (mobile-to-base), it is technically difficult to
ensure that all signals arrive with perfect time alignment at the
base station.

Frame mis-alignments cause severe interference

Different Doppler spectra for each signal
Different channels for different signals
Power control needed
BS

MS 1
MS 2
 OFDM and MC-CDMA in a
rapidly time-varying channel

Doppler spread is the Fourier-dual of a delay

spread
 Doppler Multipath Channel
Describe the received signal with all its
delayed and Doppler-shifted
components

Compact this model into a convenient

form, based on time-varying
amplitudes.

Make a (discrete-frequency) vector

channel representation

Exploit this to design better receivers

 Mobile Multipath Channel
Collection of reflected waves,
each with
random angle of arrival
random delay

Angle of arrival is uniform

Doppler shift is cos(angle)

Doppler Spectrum U-shaped power density

spectrum
 ICI caused by Doppler
0
10

Power or variance of ICI P0

-1
10

P1 P2 P3
Power, Variance of ICI

-2
10

-3
3rd tier subcarrier
10
2nd tier subcarrier
Neighboring subcarrier
-4
10
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Normalized Doppler [fm/fsub]

Doppler spread / Subcarrier Spacing

 BER in a mobile channel

0
Local-mean BER for
BPSK, versus antenna
10

10
-1
speed.
OFDM, 10 dB Local mean SNR of 10,
10
-2

MC-CDMA, 10 dB
20 and 30 dB.
OFDM, 20 dB
Comparison between
Local-Mean BER for BPSK

MC-CDMA and uncoded

-3
10

-4
OFDM, 30 dB OFDM for fc = 4 GHz
10

Frame durationTs=
MC-CDMA, 20 dB 30 dB
10
-5
896 s
-6
FFT size: N = 8192.
10

Sub. spacing fs = 1.17 kHz

10
-7

0 5 10 15 20 25 30 35 40 Data rate 9.14 Msymbol/s.

Antenna Speed (m/s)

Antenna Speed [m/s]

 Doppler Multipath Channel
Received signal r(t)

N 1 I w 1
r (t ) = an Di exp{ j ( c + n s )(t Ti ) + j i t} + n(t )
n =0 i =0

Channel model:
Iw reflected waves have OFDM parameters:
the following properties: N is the number of subcarriers
Di is the amplitude Ts is the frame duration
I is the Doppler shift an is the code-multiplexed data
Ti is the delay c is the carrier frequency
s is the subcarrier spacing
 Taylor Expansion of Amplitude
N 1 I w 1
r (t ) = a n Di exp{ j ( c + n s )(t Ti ) + ji t} + n(t )
n =0 i = 0

Rewrite the Channel Model as follows

N 1
r (t ) = anVn (t ) exp{ j ( c + n s )t} + n(t )
n =0

Tayler expansion of the amplitude

Vn(t) = vn(0)+ vn(1) (t- t) + vn(2) (t- t)2/2 + .. .
vn(q) : the q-th derivative of amplitude wrt time, at instant t = t.
vn(p) is a complex Gaussian random variable.

I w 1
v n( q ) = ( j i ) q Di exp{ j ( c + n s )Ti + ji t }
i =0
Random Complex-Gaussian Amplitude

It can be shown that for p + q is even

(1) q j p + q
E vn( p ) vm*(q ) (
= 2f D ) p+q ( p + q 1)!!
( p + q )!! 1 + j (n m)Trms s

and 0 for p + q is odd.

This defines the covariance matrix of subcarrier amplitudes and

derivatives,
allows system modeling and simulation between the input of the
transmit I-FFT and output of the receive FFT.
 DF Vector Channel Model
Received signal Y = [y0, y1, yN-1 ],

N 1 vn( q ) n( q)m T q
ym = {
a n exp j s t f n ( )} q!
f
+ nm
n =0 q =0
Lets ignore
f : frequency offset
t : timing offset System constants
(eg sinc) determined
We will denote = (0)
and = (1)

by waveform
For integer , :: 0 (orthogonal subcarriers)
models ICI following from derivatives of amplitudes Complex amplitudes
0 does not carry ICI but the wanted signal and derivatives
 DF-Domain Simulation
Simulation of complex-fading amplitudes of a Rayleigh
channel with Doppler and delay spread

Pre-compute an N-by-N matrix U, such that UUH is the channel

covariance matrix with elements n,m = Evn(0) vm*(0)
Simply use an I-FFT, multiply by exponential delay profile and FFT

Generate two i.i.d vectors of complex Gaussian random variables, G

and G, with unity variance and length N.

Calculate V = U G.

Calculate V(1) = 2 f T U G.
 DF Vector Channel Model
Received signal Y = [y0, y1, yN-1 ],

V . * A
Y = [ 0 I N 3 ] T +N
V '. * A
User data

0 1 .. N 1
Amplitudes & Derivatives
0 .. N 2
3 = 1
.. .. .. ..

N +1 N + 2 .. 0 FFT leakage
models ICI following from derivatives of amplitudes
0 does not carry ICI but the wanted signal
 Possible Receiver Approaches
Receiver

1) Try to invert adaptive matrix (Alexei Gorokhov)

0 DIAG(V ) + DIAG(V (1) )

2) See it as Multi-user detection: (J.P. Linnartz, Ton Kalker)
try to separate V .* A and V(1) .* A

[
Y = 0 I N ] V .* A
V ' . * AT + N

3) Decision Feedback (Jan Bergmans)
estimate iteratively V, V (1) and A
 Receiver 1: Matrix Inversion
Estimate amplitudes V and complex derivatives V (1)
create the matrix Q1 = DIAG(V)+ T DIAG(V(1) )
Invert Q1 to get Q1-1 (channel dependent)
Compute Q1-1 Y
V

X1 X2 Y -1 A
x + Q Slicer
A V V
X3
x 3 N
Channel
Estimator
V
Zero-forcing:
For perfect estimates V and V (1) , Q1-1 Y = A + Q1-1 N,
i.e., you get enhanced noise.

MMSE Wiener filtering inversion W

 Receiver 1: MMSE Matrix Inversion
Receiver sees Y = Q A + N, with Q=DIAG(V)+ T DIAG(V(1))

Calculate matrix Q = DIAG(V)+ T DIAG(V(1))

Compute MMSE filter W = QH [Q QH + n2 IN]-1.

Performance evaluation:

Signal power per subcarrier

Residual ICI and Noise enhancement from W
 Receiver 1: Matrix Inversion
10

0
Amplitudes
-10

-20 Determined by speed of antenna,

and carrier frequency
Magnitude in dB

-30

-40
First derivatives
-50

-60

-70
Amplitudes
Derivatives

-80
0 10 20 30 40 50 60 70
Subcarrier number

Simulation of channel for N = 64, v = 200 km/h fc = 17 GHz, TRMS = 1

s, sampling at T = 1 s. fDoppler = 3.14 kHz, Subcarrier spacing fsr
= 31.25 kHz, signal-to-ICI = 18 dB
 Receiver 1: Matrix Inversion
SNR of decision variable. Simulation for N = 64, MMSE Wiener
filtering to cancel ICI
30
Conventional OFDM
MMSE equalization

25

MMSE ICI canceller

20
Output SINR

15

Conventional OFDM
10

0
0 10 20 30 40 50 60 70
Subcarrier number
 Simplified Matrix Inversion
Rationale
ICI diminishes with increasing subcarrier difference
Approximate by band matrix with 2k+1 non-zero diagonals
Matrix Q is approximately Q = [I + ]
small, ~ diag(V(1) ./ V)
diagonal of amplitudes V
Approximate Q-1 = [I - ] 1

Complexity ~2kN
 Performance of (Simplified) Matrix Inversion

30

MMSE
Output SINR

25
k=4
20
Conv
15 OFDM

10

5 Conventional OFDM
MMSE equalization
simplified MMSE

0
5 10 15 20 25 30
Input SNR
N = 64, v = 200 km/h, fc = 17 GHz, TRMS = 1 s, sampling at T = 1 s.
fDoppler = 3.15 kHz, Subc. spacing fsr = 31.25 kHz:
Compare to DVB-T: v = 140 km/h, fc = 800MHz: fdoppler = 100 Hz while fsr = 1.17 kHz
 Receiver 1: Subconclusion

Performance improvement of 4 .. 7

Complexity can be reduced to ~2kN, k ~ 5 .. 10.

Estimation of V(1) to be developed, V is already being

estimated
 Receiver 3: Decision Feedback

Estimate

data,
amplitudes and
derivatives

iteratively
 Receiver 3: Decision Feedback

Iteratively do the following:

Compare the signal before and after the slicer
Difference = noise + ICI + decision errors
Invert to retrieve modulated derivatives from ICI
V(1).*A = -1 ICI
MMSE to minimize noise enhancements
Remove modulation 1/A
Smooth to exploit correlation in V(1)
Modulate with A
Feed through to estimate ICI
Subtract estimated ICI
 Receiver 3: DFE

Estimate V(1) in side chain

Estimated Amplitudes
V Pilot
Cancel V
Doppler
X1 X2 Y0 Y2 A
+
x + + Slicer
A - A.*V
X3 +-
x 3 N ICI
Z6
V M6
Channel Model 3 Z7
x 1/A
Z8
M7
Z9 =V
INT
Z10 A
x
 Implementational Aspects
Implementational considerations:
Estimated Amplitudes

Cancel Pilot V
Doppler
Y0 A
+
+
Y2
Slicer 1/A : table lookup
- A.*V
+-
ICI
20 taps FIR filter
FFT FIR (select from library depending on Doppler)

weigh 10X x 1/A

2 taps IIR filter bi-directional
IIR (select from library depending on Delay)
FFT V
INT

A FFT - multiply - I-FFT

x
 Implementational Aspects

Estimated Amplitudes

Cancel Pilot V

Amplitude of Filter Coefficients

40
Doppler
Y0 Y2 A
+ 30
+ Slicer
- 20

+-
A.*V 10
ICI 0

al
Optim
-10
FFT FIR -20

-30 IIR
weigh x 1/A -40

-50

IIR
~
-60
-1
FFT V -70

INT -20 -15 -10 -5 0 5 10 15 20

A
x Relative Subcarrier Number
Get derivatives & modulation
Smooth according to delay profile
Reconstruct ICI
 Performance of Receiver 3: DFE

Variance of decision variable after iterative ICI

cancellation versus variance in conventional receiver
in DFE receiver after ICI cancelling
2
10
Variance of decision variable

1
10
V a ri a n c e N e w S ys te m 3

0
10

-1
10

-2
10

-3
10
-2 -1 0 1 2
10 10 10 10 10
V a ri a n c e C o n ve n ti o n a l

Variance decision variable in conventional receiver

 Receiver 3: DFE
Decision Feedback
Sample run N=64 9 errors -> 4 errors
10
Amplitudes
5

Error Count 0

-5
Amplitude
Derivatives
-10

-15

-20

-25

-30
0 10 20 30 40 50 60 70
Subcarrier Number

N = 64 out of 8192 subcarriers, v = 30 m/s, fc = 600 MHz TRMS / NT= 0.03,

fDoppler = 60 Hz, Subcarrier spacing fsr = 1.17 kHz
 Conclusions

Modeling the Doppler channel as a set of time-varying subcarrier

amplitudes leads to useful receiver designs.

Estimation of V(1)is to be added, V is already being estimated

Basic principle demonstrated by simulation

Gain about
3 .. 6dB,
factor of 2 or more in uncoded BER,
factor 2 or more in velocity.

Promising methods to cancel FFT leakage (DVB-T, 4G)

More at http://wireless.per.nl
 Further Research Work
Optimise the receiver design and estimation of derivatives

Can we play with the waveform (or window) to make the tails of the
filter steeper?

Can we interpret the derivatives as a diversity channel?

Can estimation of derivatives be combined with synchronisation?

Isnt this even more promising with MC-CDMA?

Apply it to system design.