Introduction to
FMT Modulation and
Multiuser Multitone Wireless Uplink Systems
Andrea M. Tonello
Università di Udine
DIEGM - Department of Electrical, Mechanical, and Management Engineering
Italy
E-mail: [email protected] - Phone: +39 0432 558 288
Web: www.diegm.uniud.it/tlc/tonello
March 31, 2005
Outline
‰ Filtered Multitone (FMT) Modulation
ƒ Principles
ƒ Detection and Performance Limits
ƒ Synchronization
‰ Asynchronous Multiuser Multitone Systems
ƒ Multiuser DMT/OFDM
ƒ Multiuser FMT
ƒ Synchronization in Multiuser FMT
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
2
FMT Modulation
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
3
Multicarrier Architecture
− f0
f0
a ( 0) (lT0 )
Channel
x
g(nT)
RF
x
z (0 ) ( mT0 )
g(-nT)
z ( k ) (mT0 )
g(nT)
x
+
DAC
x
ADC
fk
a
Equalizer
RF
a ( k ) (lT0 )
( M −1)
aˆ ( 0) (lT0 )
g(-nT)
aˆ ( k ) (lT0 )
Equalizer
− fk
aˆ ( M −1) (lT0 )
z ( M −1) (mT0 )
(lT0 )
g(nT)
x
fM −1
g(-nT)
x
x (nT )
r (nT )
Equalizer
− f M −1
‰ Transmit a high data rate signal through a number of low rate sub-channels.
9 DMT (Discrete Multitone): well known OFDM scheme. It deploys a
prototype filter with rectangular impulse response.
9 FMT (Filtered Multitone): deploys a sub–channel prototype pulse with
time-frequency concentrated response.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
4
Filtered Multitone Architecture
‰ TX bandwidth:
W = 1 / T.
‰ Sub-carriers:
fk = k/(MT) , k=0,…,M-1.
‰ Sub-channel period:
T0=NT
‰
DMT – OFDM : Rectangular impulse response prototype pulse g(nT).
‰ FMT : Frequency concentrated prototype pulse, e.g., root-raised-cosine.
DMT - OFDM
Sub-Ch. Data Period:
T = MT
0
Tone Spacing:
∆f = 1 / MT
CS-FMT
Sub-Ch. Data Period:
T = MT
0
Tone Spacing:
∆f = 1 / MT
NCS-FMT
Sub-Ch. Data Period:
T = NT
0
Tone Spacing:
∆f=1 / MT
N/M =K>1
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
5
Non-Critically Sampled FMT Efficient Implementation
Channel
a (lT0 )
g1 (lT0 ;mT1 )
IFFT
a ( M −1) (lT0 )
aˆ (0) (lT0 )
z (0) (lT0 )
( 0)
P/S
RF
RF
DAC
ADC
g1- (mT1;lT0)
S/P
Eq.
FFT
aˆ ( M −1) (lT0 )
z ( M −1) (lT0 )
gM (lT0;mT1)
T0 = NT T1 = MT
gM -(mT1;lT0)
x ( nT )
Eq.
r ( nT )
‰ A possible efficient implementation is based on FFT and low-rate filtering
as proposed by Cherubini, Eleftheriou, Olcer, Cioffi, Comm. Mag. 2000.
‰ In NCS-FMT the sub-channel pulse is cyclically time-variant.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
6
Critically Sampled FMT Efficient Implementation
Channel
a (lT0 )
g1(lT0)
IFFT
a
( M −1)
(lT0 )
aˆ (0) (lT0 )
z (0) (lT0 )
( 0)
P/S
RF
RF
DAC
ADC
g1-(lT 0)
S/P
Eq.
FFT
aˆ ( M −1) (lT0 )
z ( M −1) (lT0 )
gM(lT 0)
gM-(lT0)
x (nT )
Eq.
r (nT )
CS-FMT
fk =
k −1
T0
T0 = MT
k = 1,..., M
g k (mT0 ) = g ((k − 1)T + mT0 )
m∈Z
Prototype pulse
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
7
Sub-channel Matched Filter Output
‰ RX front-end output sampled at symbol rate:
(k )
z ( k ) ( lT0 ) = a ( k ) ( lT0 ) g EQ
(0) +
∑a
m≠0
k
(k )
( lT0 − mT0 ) g EQ
( mT0 ) + ICI ( k ) + η ( k ) ( lT0 )
ƒ ∆t=0, ∆f=0: No ICI with band limited pulses but some ISI because of
the channel frequency selectivity or non-ideal Nyquist pulses.
ƒ ∆t≠0, ∆f=0: Increased ISI because of wrong sampling phase.
ƒ ∆t=0, ∆f≠0: ISI and some ICI when ∆f exceeds the frequency guards.
‰ When frequency concentrated pulses are deployed we get small ICI, however
we need to run sub-channel equalization.
‰ We need time/frequency synchronization to minimize the amount of ISI and ICI.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
8
Detection in FMT
‰ FMT requires some form of equalization.
‰ The optimal detector is a multi-channel MLSE/MAP equalizer.
‰ Independent sub-channel equalization via linear or DFE equalization
9 As we increase the number of tones we obtain narrower sub-channels.
9 The sub-channel equalizer has low complexity since the sub-channel
impulse response is short (sub-channel is narrow band).
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
9
Optimal FMT Detector
MF
k=...
y (t )
•
T0
MF
k=2
MF
k=1
ICI
T0
Joint
MAP
Detector
ISI
T0
The MAP detector computes the a posteriori probabilities of all data
symbols by observing the input signal y( t ) over a time window.
Ref: Tonello, BLTJ 2003, Proc. VTC 2002 Fall.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
10
Spectral Efficiency
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
11
Example of Sub-Channel Frequency Response
DMT
0
CS-FMT
-10
0
-10
-30
-20
-40
|H(f)| (dB)
|H(f)| (dB)
-20
-50
-60
-70
-80
0.5
-30
-40
-50
-60
0.505
0.51
0.515
fT
0.52
0.525
0.53
-70
-80
0.5
0.505
0.51
0.515
fT
0.52
0.525
0.53
NCS-FMT
0
M = 128 B = 25 MHz
-10
CS-FMT:
rectangular windowed pulses + 4 virtual
carriers
NCS-FMT: root raised cosine pulses N/M = 1.125 +
4 virtual carriers
CP-DMT:
|H(f)| (dB)
-20
-30
-40
-50
-60
-70
-80
0.5
0.505
0.51
0.515
fT
0.52
0.525
0.53
CP length = 30 chips + 16 virtual carriers
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
12
Complementary Distribution of the Achievable Bit Rate
Probability [ Achievable Bit Rate > K ]
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
40
45
------ DMT
------ CS-FMT
------ NCS-FMT
50
55
60
65
M bit/s
70
75
80
85
c Rayleigh exponential with τrms=100 ns
¦ Rayleigh exponential with τrms=40 ns
* Ricean exponential with R=5 dB, τrms=40ns
•
DFE sub-channel equalization in FMT.
•
FMT has higher spectral efficiency than DMT/OFDM.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
90
13
Performance Limits for FMT Modulation
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
14
Performance Limits
‰ Let us consider FMT modulation in time variant frequency selective fading.
‰ We can evaluate the matched filter bound performance in closed form:
- BER performance with an ideal equalizer
‰ The analysis yields very interesting insights:
- FMT modulation is a diversity transform and offers
¾ time-frequency diversity gains and coding gains that are a function
of the prototype pulse, and number of tones.
Ref: Tonello, IEEE Trans. on Wireless Comm. in press ; Tonello, Proc. WPMC 2003.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
15
Frequency Selective Time-Invariant Channel
ITU PB channel with quasi static fading
Bandwidth W=3.84 MHz. BPSK modulation.
BER
10
RECT
PB
M→
M=32
-4
10
-2
10
∞
M=128
-6
10
-6
10
128
16
-4
10
BER
RECT+CP (OFDM)
PB
-2
M=1
-8
-8
10
10
BER
10
M→
-4
10
SINC
PB
M→
∞
16
16
M=1
M=1
-8
10
0
6
12
18 24
Es/No (dB)
∞
128
128
-6
10
-2
10
30
36
0
6
12
18 24
Es/No (dB)
-4
10
BER
GAUSS B=0.33
PB
-2
-6
10
-8
30
36
10
MF Bound performance for FMT with rectangular, Gaussian, and sinc prototype pulse
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
16
Time-Variant Channel
Flat fading with Jakes' Doppler spectrum
Bandwidth W=24.3 kHz. BPSK modulation.
BER
RECT
FLAT
64
M=16
-4
10
Sta
tic
Sta
tic
RECT+CP (OFDM)
FLAT fd=100 Hz
-6
10
10
BER
-6
10
No
Fading M=128
-8
0
6
12
SINC
FLAT
-4
10
-6
16
No
Fading
64
30
-2
fd=100 Hz 10
Sta
tic
16
18 24
Es/No (dB)
-8
10
Sta
tic
-4
10
10
M=128
GAUSS B=0.1
FLAT fd=100 Hz
-2
36
0
6
12
-6
10
64
No
Fading
10
-4
10
16
No
Fading
-8
-2
fd=100 Hz 10
M=128
BER
-2
10
BER
128
10
64
-8
18 24
Es/No (dB)
30
36
10
MF Bound performance for FMT with rectangular, Gaussian, and sinc prototype pulse
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
17
Remarks
‰ In Frequency Selective Fading
9 FMT is a good choice complexity wise
9 Single Carrier modulation is a good choice performance wise.
‰ In Time Selective Fading
9 FMT is a good choice performance wise
9 Single carrier modulation is a good choice complexity wise.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
18
Synchronization
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
19
Time-Domain Synchronization
aˆ (0 ) (lT0 )
z (0) (lT0 )
g1-(mT 1;lT 0)
r ( nT )
Estimate/
compensate
time offset
Estimate/
compensate
frequency offset
S/P
MMSE Eq.
FFT
gM-(mT1;lT 0)
∆t
aˆ ( M −1) (lT0 )
z ( M −1) (lT0 )
MMSE Eq.
∆f
‰ Estimation and compensation of time/frequency offsets can be done in the time domain.
• Blind Synchronization. The drawback is that channel estimation cannot rely on
known training symbols.
• Cyclic Training Approach. We generate training sequences that exhibit a periodic
behavior at the tx-rx side, similarly in spirit to Schmidl and Cox method in OFDM.
• PN Training Approach. We generate PN training sequences.
Ref: Assalini, Tonello, Proc. WPMC 2003 ; Tonello, Rossi, Proc. WPMC 2004.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
20
Bit-Error-Rate Performance with Time-Domain Sync.
Timing Error
Rayleigh Fading Channel with τrms=50 ns
10
0
10
-1
10
-2
1 coeff
5 coeff
Rayleigh Fading Channel with τrms=100 ns
10
0
10
-1
10
-2
1 coeff
5 coeff
Uncompensated
BER
BER
Uncompensated
Compensated
SNR= 20 dB
10
SNR= 20 dB
-3
0
0.2
Compensated
0.4
0.6
∆t / T0
0.8
1
10
-3
0
0.2
0.4
0.6
∆t / T0
0.8
1
ƒ 4-PSK modulation, M=32 Sub-channels,Root-Cosine Pulses, Bandwidth 20
MHz.
ƒ Performance drastically decreases without time/frequency compensation.
ƒ Synchronization with random sequences and 1-5 Tap RLS equalizer.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
21
Bit-Error-Rate Performance Time-Domain Sync.
Frequency Error
Rayleigh Fading Channel with τrms=50 ns
10
0
10
-1
1 coeff
5 coeff
Rayleigh Fading Channel with τrms=100 ns
10
0
10
-1
1 coeff
5 coeff
Uncompensated
10
Compensated
BER
BER
Uncompensated
Compensated
-2
10
-2
SNR=20 dB
10
SNR=20 dB
-3
0
0.002 0.004 0.006 0.008
∆f x MT
0.01
10
-3
0
0.002 0.004 0.006 0.008
∆f x MT
0.01
ƒ The practical scheme with random training sequences performs well.
ƒ A single tap equalizer is a good choice for delay spreads of ~100 ns.
ƒ The performance of the synchronizer can be improved with a fine sync. method in
the frequency domain.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
22
Remarks
‰ OFDM is an elegant simple solution to cope with the channel frequency selectivity.
‰ FMT can yield higher spectral efficiency than DMT/OFDM.
‰ The sub-channel spectral containment of FMT makes it more robust to time and
carrier frequency offsets than OFDM.
‰ Time-Frequency acquisition is still of great importance.
‰ FMT can be more complex than DMT since it requires filtering and equalization:
- we save complexity by using a smaller number of tones than in OFDM.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
23
Multiuser Multitone Architectures
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
24
Asynchronous Multiple Access Channel
user 1
user 4
receiver
user 2
user 3
user NU
‰ Consider the Uplink
ƒ Users are asynchronous
ƒ Time offsets between users (propagation delays)
ƒ Carrier frequency offsets between users (oscillators, Doppler).
‰ Frequency Division Multiplexing is an Interesting Solution.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
25
Multitone Multiple Access
user
1
frequency
user
2
user
3
‰ M total tones
‰ Groups of tones are assigned to users
‰ Each user deploys multicarrier modulation over its set of tones
‰ Several tone allocation methods are possible:
9 Orthogonal or non-orthogonal, Static or dynamic…
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
26
Solutions
‰ Two efficient implementation solutions
9 Multiuser DMT : MU-DMT (OFDMA)
9 Multiuser FMT : MU-FMT
Ref: Tonello, Pupolin, Proc. WPMC 2001 ; Tonello, Bell Labs Tech. Journ. 2003.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
27
Received Composite Signal
NU M −1 ∞
y (t ) = ∑ ∑
(u ,k )
(u ,k )
a
(
lT
)
g
(t − ∆tu ,k − lT0 ; t )e
∑
R
0
j (2π∆fu ,k t +φu ,k )
+ η (t )
u =1 k = 0 l =−∞
Time-variant channel response for user u and sub-channel k
Time offset of user u and sub-channel k due to the propagation delay
Carrier Frequency offset due to Doppler and oscillators precision
Propagation delay:
∆t=6.6 us/km = 6.6 chips/km [B=1 MHz]
Doppler:
∆ f=0.93 Hz [v=1 km/h -- fc=1 GHz]
Oscillator:
∆ f=1000 Hz [fc=1 GHz -- 1 p.p.m]
The values depend upon the hardware and the mobility/coverage requirements
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
28
MAI-ICI-ISI Components
The interference depends upon
- sub-carrier spacing
- prototype filter
- tone allocation
Tone Allocation
Block
user 0
user 1
user 2
user 3
Interleaved
user 0
user 1
user 2
user 3
• Block allocation yields lower MAI than the interleaved.
• Tone interleaving is a better option to exploit the frequency diversity.
• Frequency guards can be inserted only with the block allocation in OFDM.
• The NCS-FMT architecture can comprise frequency guards.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
29
MU-OFDM Uplink Analysis
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
30
Interference in MU-OFDM
(M+µ)T
‰ The system is orthogonal in an ideal synchronous channel.
– Users’ time/frequency offsets generate MAI.
‰ A cyclic prefix can null the interference due to time misalignments.
‰ Interference from the carrier frequency offsets cannot be totally compensated.
Ref: Tonello, Laurenti, Pupolin, Proc. ICT 2000, Proc. VTC 2000 Fall.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
31
MU-OFDM: SIR on Adjacent Sub-Carrier
Block Multiplexing
K=64
M=1024
Interleaved Multiplexing
M=1024
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
32
SIR with Block Allocation in MU-OFDM
Block
The SIR rapidly decreases as the frequency offset and the time offset increase.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
33
SIR with Interleaved Allocation in MU-OFDM
Interleaved
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
34
SIR with Block Allocation and 1 Guard in MU-OFDM
Block with 1 Frequency Guard
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
35
MU-OFDM: Average Symbol Error Rate
Interleaved
Block
Block with One
Frequency Guard
4-PSK, 16 users, 12 tones/user, µ=64, W=4.096 MHz.
Excess time offsets and frequency offsets are independent and uniformly distributed between
[-∆ l , + ∆ l ] and [-∆ f , +∆ f ] respectively. The channel has an exponential power delay profile with d.s.=10 µs.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
36
Remarks
‰ MU-OFDMA is severely affected by users’ time/frequency offsets in the uplink.
‰ Time/Frequency guards reduce the spectral efficiency.
‰ Acquisition of timing and carrier frequency is difficult.
‰ Due to the spectral containment characteristics, FMT is a better option than
OFDM for the uplink.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
37
Optimal Multiuser Multitone Detection
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
38
Optimal MLSE/MAP Detection
MF
User 1, K=...
Time/Frequency Sync
User 1
MF
User 1, K=2
MF
User 1, K=1
y(t )
MF
User NU, K=...
Time/Frequency Sync
User NU
ICI
Joint
3D
MAP
Detector
ISI
MAI
MF
User NU, K=2
MF
User NU, K=1
• Optimal MAP Detection has a complexity that grows exponentially with the
number of users, tones, and sub-channel memory.
• With frequency concentrated pulses and orthogonal tone assignment it simplifies
to a bank of single channel MAP detectors.
Ref: Tonello, VTC 2002 Fall, Bell Labs Tech. Jour. 2003
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
39
Turbo Per-Symbol Detection
MF
User 1, K=...
Time/Frequency Sync
User 1
MF
User 1, K=2
MF
User 1, K=1
y ( nT )
MF
User NU, K=...
Time/Frequency Sync
User NU
MF
User NU, K=2
MF
User NU , K=1
Symbol
Detection
Symbol
Detection
Symbol
Detection
Symbol
Detection
Symbol
Detection
Symbol
Detection
Simple approach
• Symbol-by-symbol detection with iterative interference cancellation
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
40
Turbo Per-Symbol Decoding
MF
User 1, K=...
Time/Frequency Sync
User 1
MF
User 1, K=1
Symbol
Demapping
+
Soft/Hard IC
P/S
Deinterleaving
Symbol
Demapping
+
Soft/Hard IC
Decoding
User 1
Interleaving
y ( nT )
Interleaving
MF
User N U, K=...
Time/Frequency Sync
User NU
MF
User N U, K=1
Symbol
Demapping
+
Soft/Hard IC
Symbol
Demapping
+
Soft/Hard IC
P/S
Deinterleaving
Decoding
User NU
• Feedback from the decoders if we assume to use bit-interleaved codes
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
41
Example of Multiuser Multitone System
user 1
User 3
user 1
receiver
receiver
user 2
block
user 2
1
1
2
2
User 4
3
4
interleaved
1
1
2
2
3
4
•
16 Total carriers
•
QPSK – ½ Convolutional Code
•
∆t uniformly distributed in [- ∆tm/2, ∆tm/2]
•
∆f uniformly distributed in [- ∆fm/2, ∆fm/2]
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
42
Sub-Channel Pulses
DMT: Rectangular Pulse
FMT: Gaussian Pulse
MT
MT
1/(MT)
• Minimal sub-carrier spacing (no cyclic prefix)
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
43
BER Uncoded - Interleaved
2 Users - Interleaved - AWGN
∆tmax=T0/2 ∆f max=F0/8
4 Users - Interleaved - AWGN
∆tmax=T0/2 ∆f max=F0/8
-1
-1
10
10
#1
#1
#1
#1
-2
-2
10
Bit Error Rate
Bit Error Rate
10
-3
-3
10
10
-4
10
-4
2
6
10
SNR (dB)
14
18 2
6
10
SNR (dB)
14
18
10
Uncoded
------ RECT pulses
------ GAUSS pulses
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
44
BER Uncoded - Interleaved
2 Users - Interleaved - AWGN
∆tmax=T0/2 ∆f max=F0/8
4 Users - Interleaved - AWGN
∆tmax=T0/2 ∆f max=F0/8
-1
-1
10
10
#1
#1
#1
#6
-3
10
-2
10
#2
Bit Error Rate
10
Bit Error Rate
#1
#2
-2
#6
-3
10
#2
#2
#6
#6
-4
10
-4
2
6
10
SNR (dB)
14
18 2
6
10
SNR (dB)
14
18
10
Uncoded
------ RECT pulses
------ GAUSS pulses
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
45
BER Coded - Interleaved
2 Users - Interleaved - AWGN
∆tmax=T0/2 ∆f max=F0/8
4 Users - Interleaved - AWGN
∆tmax=T0/2 ∆f max=F0/8
-1
-1
10
10
#1
#1
#1
-2
10
#2
Bit Error Rate
Bit Error Rate
10
#1
#1
#2
-2
#1
#6
#6
#1
-3
10
-3
10
#2
#1
#2
#2
#6
#6
#2
-4
10
-4
2
6
10
SNR (dB)
14
18 2
6
10
SNR (dB)
14
18
10
Coded
------ RECT pulses
------ GAUSS pulses
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
46
Remarks
‰ FMT is a better solution than DMT for uplink asynchronous communications:
9 Single user detection with sub-channel equalization works fine if we use
appropriate frequency confined sub-channel pulses!
‰ Optimal performance can be achieved with multitone multiuser detection.
‰ Simple iterative per-symbol detection/decoding shows fast convergence to
the matched filter bound.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
47
Synchronization in MU-FMT
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
48
MU-FMT Synchronization Algorithm
‰
We assume a single user receiver with training sequences and a 1-3 taps RLS
equalizer per-sub-channel.
Ref: Tonello, Pecile, Proc. VTC 2005 Spring.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
49
Multiuser FMT - Synchronization Performance
0
32 Users - Delay Spread=100 ns
0
10
32 Users - Delay Spread=400 ns
10
-1
-1
10
10
Uncomp. 1 Tap
Uncomp. 3 Tap
Comp. 1 Tap
Comp. 3 Tap
BER
BER
Uncomp. 1 Tap
Comp. 1 Tap
SNR=20 dB
SNR=20 dB
-2
-2
10
10
0
0.02 0.04 0.06 0.08
∆ f,max x MT
0.1
0
0.02 0.04 0.06 0.08
∆ f,max x MT
0.1
ƒ 4-PSK Modulation, M=32 Sub-channels, Root-Cosine Pulses, Bandwidth 10 MHz ,
Frequency guards of 12.5 kHz. Training of length 15 symbols.
ƒ Delay spread 100 ns and 400 ns.
ƒ Users are time-asynchronous with a random phase, the frequency offsets are uniformly
distributed in [-∆ fmax, ∆ fmax]. They have a single sub-channel each.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
50
Remarks
‰ The practical scheme with training sequences performs well.
‰ The MU-FMT architecture is very robust to the multiple access interference.
‰ Single user detection is simply required.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
51
Final Conclusion
FMT is a promising technology that has great
potentiality for the broadband uplink scenario !
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
52
Some References
Multicarrier Systems Performance Analysis
•
•
•
•
•
A. Tonello, "Performance limits for filtered multitone modulation in fading channels", IEEE Trans. on Wireless Comm., in press, subm. April 2003.
A. Assalini, S. Pupolin, A. Tonello, “Analysis of the effects of phase noise in filtered multitone (FMT) modulated systems”, Proc. of IEEE Globecom 2004, Dallas,
November 2004.
A. Tonello, "Exact matched filter performance bound for filtered multitone modulation in fading channels", Proceedings of Wireless Personal Multimedia
Communications Symposium 2003, Yokosuka, October 2003.
A. Tonello, R. Bernardini, "On the exploitation of the time-frequency diversity in coded OFDM", Proceedings of Wireless Personal Multimedia Communications
Symposium 2003, Yokosuka, October 2003.
A. Tonello, "Performance limits of multicarrier based systems over fading channels with optimal detection", Proceedings of IEEE WPMC 02, Honolulu, October
2002.
Multicarrier Multiuser Systems
•
•
•
•
•
•
•
•
•
A. Tonello, "Asynchronous multicarrier multiple access: optimal and sub-optimal detection and decoding", Bell Labs Technical Journal Fall 2002, vol. 7 issue 3,
February 2003.
A. Tonello, A. Assalini, "An asynchronous multitone multiuser air interface for high speed uplink communications", Proceedings of IEEE Vehicular Technology
Conference 2003 Fall, Orlando, October 2003.
A. Tonello, "Multiuser detection/decoding in asynchronous multitone multiple access systems", Proceedings of IEEE WPMC 02, Honolulu, October 2002.
A. Tonello, "Multiuser detection and turbo multiuser decoding for asynchronous multitone multiple access systems", Proceedings of IEEE VTC 02-Fall,
Vancouver, September 2002.
A. Tonello, S. Pupolin, "Performance of single user detectors in multitone multiple access asynchronous communications", Proceedings of IEEE VTC 2002 Spring, Birmingham, Alabama, May 2002.
A. Tonello, S. Pupolin, "Discrete multi-tone and filtered multi-tone architectures for broadband asynchronous multi-user communications", Proceedings of
WPMC 2001,Aalborg, September 9-12, 2001.
A. Tonello, N. Laurenti, S. Pupolin, "Analysis of the uplink of an asynchronous multi-user DMT OFDMA system impaired by time offsets, frequency offsets,
and multi-path fading" , Proceedings of IEEE VTC 2000 - Fall, Boston, September 24-28, 2000.
A. Tonello, N. Laurenti, S. Pupolin, "Capacity considerations on the uplink of a DMT OFDMA system impaired by time misalignments, and frequency offsets",
in “Software Defined Radio”, Springer-Verlag, Del Re ed. 2000.
A. Tonello, N. Laurenti, S. Pupolin, "On the effect of time and frequency offsets in the uplink of an asynchronous multi-user DMT OFDMA system", Proceedings
of ICT 2000, Acapulco, May 22-25, 2000.
Synchronization in Multicarrier Systems
•
•
•
A. Tonello, F. Pecile, “Synchronization Algorithms for Multiuser Filtered Multitone (FMT) Systems”, to appear in Proceedings of IEEE VTC 2005 Spring,
Stockholm, Sweden.
A. Tonello, F. Rossi, “Synchronization and Channel Estimation for Filtered Multitone Modulation”, Proceedings of Wireless Personal Multimedia Communications
Symposium 2004, Abano Terme, September 2004.
A. Assalini, A. Tonello, "Time-Frequency synchronization in filtered multitone modulation based systems", Proceedings of Wireless Personal Multimedia
Communications Symposium 2003, Yokosuka, October 2003.
Multicarrier Multiple Antenna Systems
•
•
A. Tonello, "Orthogonal space-time discrete multitone and space-time filtered multitone coded architectures", Proceedings of IEEE Vehicular Technology
Conference 2004 Spring, Milano, May 2004.
A. Tonello, “A concatenated multitone multiple antenna scheme for multiuser uplink communications”, Proceedings of IEEE ITG Workshop on Smart Antennas,
Munich, March 2004.
© AMT - DIEGM UNIVERSITÀ DEGLI STUDI DI UDINE
53