Professional Documents
Culture Documents
Ho Chin Keong Communication Systems & Signal Processing Centre for Wireless Communications National University of Singapore
November 9, 2001
Topics
OFDM Transmitter Design Considerations Design Considerations in Transmission through Multipath Channel Additional OFDM Receiver Design Considerations
XI (0) XR (k)
up-convert to frequency 0
:
XR (k)
x' (t)
+
-sin(2nfot)
XI (k)
:
XR (N-1) XI (N-1) up-convert to frequency (N-1)f0
XI (k)
N 1
x (t) =
k=0 3
(1)
:
X (k)
x (t)
exp(j2nfot)
:
X (N-1) up-convert to frequency (N-1)f0
x(t) =
k=0 4
X(k) exp(j2kfot)
(2)
:
IFFT
:
P/S
x(n) = x(t = nT ) =
k=0
X(k) exp(j2nk/N )
(3)
becomes IDFT of {X(k)}k mod N implemented using IFFT, then use DAC
5 Ho Chin Keong: Overview on OFDM Design
WIRELESS OFDM symbol windowing - allows other OFDM symbols transmission COMMUNICATIONS
Time domain OFDM signal for 5 subcarriers modulated with {1,1,1,1,1}. 4
1 Frequency domain OFDM signal for 5 subcarriers modulated with {1,1,1,1,1}.
CENTRE FOR
3 2
Spectrum 0.8 0.6 0.4 0.2
0 6
0 subcarrier k
Time domain rectangular windowed OFDM signal 4 3 2 1 0 1 2 0 0.5 1 1.5 time / ( NT) 2 2.5 3
0.5 6 4 Spectrum 1 1.5
0.5
0 subcarrier k
before windowing: X(f ) = X(k)(f kfo) after windowing: X(f )W (f ) = X(k)(f kfo) sinc(f /fo) = X(k)sinc(f /fo k)
Note: X(f) from here on will be assumed to be windowed with rectangular window. 6 Ho Chin Keong: Overview on OFDM Design
Spectrum eciency for single carrier vs OFDM assume R is data rate of each (sub)carrier, bandwidth dened from null-to-null, N subcarriers
R for single carrier : = 2f o
but - windowing possible for single carrier to improve windowing not possible on a subcarrier to exploit implementation simplicity of IFFT
Digital to Analogue Converter (DAC) Need to perform upsampling for interpolation. Solutions: null subcarriers to ease lter design windowing - spectrum decays at 1/|f | for rectangular window digital ltering - usually FIR
Effects of raised cosine windowing (with 16 null subcarrriers) 5 3 overlapping window points 2 overlapping window points 1 overlapping window points
10
15
20
25
30
0.2
0.4
0.6
0.8
1 Frequency
1.2
1.4
1.6
1.8
x(n) is Gaussian distributed by law of large number - leads to high PAPR eect is more prominent as N increases requires expensive and high power amplier with high dynamic range, or else distorts time domain signal non-linearity or clipping introduces out-of-band spectral regrowth solutions
10
usually valid to treat channel as time invariant over each OFDM symbol channel changes after every OFDM symbol Frequency selective channel use discrete time baseband representation y(n) = x(n) h(n) + v(n) where v(n) is AWGN, is linear convolution ISI occurs which spills to other OFDM symbols
11
Mitigation of ISI
Time domain rectangular windowed OFDM signal with ISI 3.5 3 Transmitted OFDM symbol Received OFDM symbol
2.5
1.5
0.5
0.5
1.5
1 time / ( NT)
append guard interval (GI) at transmitter GI longer than maximum delay spread remove GI at receiver
12 Ho Chin Keong: Overview on OFDM Design
Introduce cyclic prex (CP) how to remove ISI within OFDM symbol? we dont ll appended GI with cyclic data to form CP dene c to be circular convolution modulo N , y(n) = x(n) h(n) ycp(n) = xcp(n) h(n) y(n) = x(n) c h(n)
Note: All discrete index from now on will be assumed to be modulo N .
(4)
13
ICI removed through cyclic prex (CP) removing cyclic prex, received time domain signal is
N 1
y(n) =
k=0
H(k)X(k) exp(j2nk/N ),
(5)
where H(k) =
Y (k) =
n=0
15
Time/frequency duality
single carrier in time selective channel vs OFDM in frequency selective channel y(n) = x(n)h(n) + v(n) Y (k) = X(k)H(k) + V (k)
single carrier in frequency selective channel vs OFDM with time selective channel y(n) = Ax(n)h(n) + ISI + v(n) Y (k) = BX(k)H(k) + ICI + V (k)
where A, B are complex attenuation factors ISI occurs for single carrier satisfying Nyquist pulse-shaping criterion when sampling oset occurs ICI occurs for OFDM when frequency oset occurs
16 Ho Chin Keong: Overview on OFDM Design
Synchronization in general timing synchronization errors- correctable in a pilot based system y(n + t) Y (k) exp(j2kt/N )
(7)
frequency synchronization errors when carrier frequency oset phase jitter Doppler spread frequency synchronization errors- introduces ICI y(n) exp(j2n/N ) Y (k + ) (8)
17
Windowing at receiver consider windowing after removing cyclic prex, before FFT yw (n) = y(n)w(n) = [x(n) c h(n)]w(n) the DFT is Yw (k) = [X(k)H(k)] c W (k)
N 1
(9)
=
l=0
X(l)H(l)W (k l)
(10)
18
y(n) =
k=0
(11)
dene w(n; ) = exp(j2n/N ) and thus W (k; ) is the dirichlet kernel using (9) and (10), we can re-write the DFT of (11) as Y (k) = H(k)X(k)W (0; ) + I(k) where the ICI is
N 1
(12)
I(k) =
l=0,l=k
H(l)X(l)W (k l; )
(13)
19
S pec trum of des ired s ubc arrier and ICI ICI des ired s ubc arrier
1.5
1.5
S pec trum
0.5
0.5
-6
-6
-4
-2
0 s ubc arrier k
ICI occurs when sampling at frequency k + instead of k . 20 Ho Chin Keong: Overview on OFDM Design
estimate using averaging of phase dierence of repeated samples, multiply complex term to de-rotate signal with increasing phase use windowing to reduce ICI frequency domain method: estimate , then express as matrix form y = W()diag{x}h + v = W()diag{h}x + v solve for h or x in the frequency domain using least squares by minimizing 2-norm of v
21 Ho Chin Keong: Overview on OFDM Design
(14) (15)
b(n)
d (n)
S/P
. . .
IDFT
. . P/S .
Add CP
S/P
. . DFT .
. . .
P/S
FEQ
Note: Some notations here are not inconsistent with this presentation.
22
p (n)
x (n)
x(n)
x(n)
h(n)
y(n)
y(n)
y (n)
(semi-)blind methods based on cyclic prex or cyclostationarity properties decision feedback - not popular, probably due to large processing delay in FFT training - substantial initialization, usually for xed line, not for packet based system pilots - most common method for wireless applications
23
10
Cyclostationarity based with o = 0.995 Subspace based with o = 0.98 Switch based with = 0.97
o
MSE
10
10
200
400 time n
600
800
1000
24
useful to think of OFDM symbols in a 2-dimensional (2D) time/frequency grid usually time correlation is higher than frequency correlation pilots used to sample 2D channel - need to satisfy sampling theorem (2 times of coherence bandwidth/time) 2D lters used for interpolation
25