Communication System Design

Practical multitone architectures
Lecture 4 Vladimir Stojanovi
6.973 Communication System Design Spring 2006 Massachusetts Institute of Technology
Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Loading with discrete information units

Chows algorithm (rounding of waterfilling results) Levin-Campello algorithm (greedy optimization)

Each additional bit placed on subchannel that requires the least incremental energy for its transport Incremental energy en (bn ) n (bn ) n (bn )
e (b ) g bn = log 2 1 + n n n
bit increment
Efficiency of bit distribution
Cant move a bit from one channel to another and reduce the total energy
6.973 Communication System Design
Levin-Campello (LC) algorithm- components
Efficientizing (EF) finding energy-efficient bit distribution
Always replace a bit distribution with a more efficient exhaustively search all single information unit changes at each step
Smallest energy to add a new bit Largest energy to subtract the bit Subtract bits from costly sub-channels and add to least costly sub-channels
Efficientizing example
1+0.9D-1 channel (Pe=10-6, gap=8.8dB, PAM/QAM)

PAM and single-sideband QAM
bn
bn + 1
Rate adaptive solution

Need also to satisfy the energy constraint E-tightness (ET)
Cant add another bit without violating the E constraint
If energy constraint violated subtract the most costly bit If energy less than max add the bit that costs the least to add
LC - Rate adaptive algorithm
ET example (continue from EF example)
Start from efficient distribution that blows up the energy constraint N = 8 1 = 8
Margin
N x
8 = 1.8 dB 5.32
Levin-Campello Rate Adaptive algorithm

Choose any bit distribution Make it efficient using (EF algorithm) Make it energy tight using (ET algorithm)
Dynamic rate adaptation

Change the loading when channel changes LC is a natural candidate Keep the ET bit distribution and perturb based on channel changes
Bit is moved from channel n to m
Tightly coupled with channel and noise estimation
Will cover in later lectures

Channel partitioning
Divide the channel into a set of parallel, independent channels

X1,k X2,k x XN-1,k XN,k 1(t) 2(t) n(t) h*(T-t) y(t) 1(T-t) 2(T-t) t=T t=T y1,k y2,k x(t) = x(t) N n=1 N k n=1 N k n=1 xn . n(t) xn,k . n(t - kT)
. . .
N-1(t) N(t) X1,k X2,k XN,k
x(t) h(t) +
. . .
N-1(T-t) N(-t)
t=T t=T
h(t) * x(t) = yN-1,k yN,k
xn,k . n(t - kT) * h(t) = xn,k . pn(t - kT)
n1,k
(MIMO)
Qk
n2,k nN,k
+ +
y1,k y2,k yN,k
(lower case y indicates normalizing ||p|| factors included)
k n=1 p m(t) * p* n(-t) qn,m (t) = . ||pm|| ||pn|| Qk = Q(kT)
Qk = Ik
Figure by MIT OpenCourseWare.
Generalized Nyquist criterion

No interference between symbols No interference between sub-channels

Modal modulation
Transmission with eigen-functions

r(t) = h(t) * h*(-t) n . n (t) = X1,k X2,k x XN-1,k XN,k N n=1 N n=1
(-TH / 2, TH / 2)
n
-T / 2 r(t - ) ()d n = 1,...,

1(t) 2(t) n(t) h*(T-t) y(t)
T/2
[-(T - TH) / 2, (T + TH) / 2)]

t=T t=T y1,k y2,k
1(T-t) 2(T-t)
. . .
N-1(t) N(t)
x(t) h(t) +
. . .
N-1(T-t) N(-t)
Per-channel MAP is optimal Problem is vector AWGN
t=T t=T
yN-1,k yN,k
k (t ) Channel eigenfunctions (channel dependent)

y(t) = N n=1 (nxn) . n (t)
x(t) = y(t) =
xnn(t)
~(t) xn . [r(t) * n (t)] + n
Convergence of multitone to modal modulation
Modal modulation is optimal for finite symbol time
Eigenfunctions hard to compute and channel dependent As both N->inf and [-T/2, T/2]->(-inf,+inf)
Multitone converges to Modal modulation
Set of eigenvalues of any autocorrelation function is unique This set determines the performance of MM through SNR Eigen-functions are not unique is also a valid eigen-function for inf symbol period Corresponding eigenvalues are R(2n/T) No ISI on any tone since symbol period is infinite Each tone is AWGN channel SBS detector is MAP optimal are then eigen-functions even on finite [-T/2,T/2] Can use extra bandwidth in the design to make the channel look periodic
If channel is periodic (does not exist in practice)

10
Finite symbol duration effects
ISI corrupts the neighboring symbol
Need to leave the guardband
Can try to create a more sinc-looking symbols in time by filtering the tones in frequency domain
Need excess bandwidth for filter roll-off
11
Discrete time channel partitioning
Digital realization
X0 m0 1/T x + x (t) (t) (LPF) y = Px + n h (t) n (t) + *(-t) y(t) (LPF)
A D C
f0* 1/T f1* y
Y0
X1
m1
Y1
D A C
XN-2
mN-2
* fN -2
YN-2
XN-1
mN-1
* fN -1
YN-1
yN-1 yN-2
p0 = 0 0 0
p1 p0 0
py py-1 p0
0 py p1
0 0 0 py
xN-1
x0 x-1 x-y
nN-1 nN-1 n0
y0
12
Vector coding
Creates a set of parallel channels by using SVD

singular-value decomposition of P
discrete transmitter waveforms discrete matched filters
since F is unitary N is also AWGN
13
VC example
1+0.9D-1, N=8
Need one sample guardband

Etot=(N+1)*E_dim=(8+1)*1=9
SVD on Gives singular values Sub-channel SNRs Waterfilling shows only 7 dimensions can be used Sub-channel energies SNRs are then Total SNR VC capacity
Would get 1.55bits/dim if N-> inf
14
DMT and OFDM
Discrete multitone (DMT) and Orthogonal frequency Division Multiplexing (OFDM) DMT used on slowly time-varying channels
Optimize bn and En per sub-channel But uses same bn and En on all channels Used on one-way broadcast channels In vector coding, M,F channel dependent Make the channel circular and make M,F channel independent - simplify hardware implementation
OFDM uses same channel partitioning as DMT

Forms of vector coding with added restrictions

15
DMT/OFDM channel partitioning
Make channel look circular
Repeat the tail of the symbol at its beginning
CP DMT
Add cyclic prefix to each transmitted symbol

xN ... xN 1
xN ... xN 1 x0 ...
VC
0...0
x0 ...
xN ... xN 1
SVD can be replaced by eigen-decomposition (spectral factorization) A discrete form of modal modulation While SNRs are unique, many choices for M and F
16
IDFT and DFT as orthogonal transformations
DMT and OFDM use IDFT and DFT as eigen-vectors
DFT
IDFT channel gain at tone n
Proof
qn is eigen-vector of P
17
DMT/OFDM implementation
X0
1 = N+v T
D A C
Y0
1 = N+v T (-t)
y(t)
(LPF)
A D C
X1
. . .
XN-2 XN-1
IDFT
parallel to serial & cyclic prefix insert
x(t) (t)
(LPF)
n(t) h(t) +
serial to parallel & cyclic prefix remover
Y1
DFT
. . .
YN-2 YN-1
Forces P to be cyclic matrix
XN-i = Xi if x(t)is real

Data rate penalty N /( N + )

18
One-tap frequency equalizer
Need to compensate for channel attenuation
To recover the original constellation distance

19
Rx/Tx Windowing
Rectangular window Raised cosine 5% Raised cosine 25%
Can overlap as long as sum is constant
20
1+0.9D-1 DMT example
N=8
Waterfilling with Etot=8
Waste one unit on CP
lower than VC (8.1dB)
For N=16 quickly reaches max of 8.8dB

21
Complexity of DFE, VC, DMT
Example 1+0.9D-1 channel MMSE-DFE, SNR=7.6 dB, 3ff, 1fb tap, 4 mac/sample VC, N=8, SNR=8.1 dB, 7*8/9=6.2MAC/sample DMT, N=8, SNR=7.6 dB, 8pt FFT/IFFT, 2.7MAC/sample

N=16, 3.8MAC/sample, SNR=8.8 dB DFE needs 10FF taps, 1FB tap, SNR=8.4 dB, 11MAC/sample
22
Asymmetric digital subscriber line (ADSL)

video switch
12 Mbps
ADSL ADSL Access ADSL Mux (DSLAM) ADSL

POTS (or ISDN) Switch
Split
1.5 Mbps
Split
ADSL
Telco CO
internet
138 kHz
frequency
1/T=2.208 MHz (CP = 40 samples) Each time domain symbol 2*256+40=552 samples

Hermitian symmetry creates real signal transmitted from 0-1.1MHz First 2-3 tones near DC not used avoid interference with voice Tone 256 also not used, 64 reserved for pilot
Nup=32, CP=5, each symbol 2*32+5=69 samples
Exactly 1/8 of downstream

1.1 MHz
Symbol rate T=250us Nd=256, 4.3125kHz wide
POTS
psd
UP Up
10 kHz 0 Hz
Down Down
23
802.11a Wireless LAN example

Up to 54Mbps symmetrically (<100m) 1-3 Tx power levels Complex baseband
40 mW
200 mW
800 mW
unlike ADSL which is real baseband
5.15 GHz
5.25 GHz 20 MHz
5.35 GHz
5.725 GHz
5.825 GHz
52 carriers total, each ~ 300 kHz
N=64 (-31 31) (so 128 dimensions)

not used
Symbol length = 80 samples, CP=16 Symbol rate 250kHz (T=4uS, T=50ns), CPguard=0.8us
6 9 12 18 24 36 48 54 BPSK BPSK 4QAM 4QAM 16QAM 16QAM 64QAM 64QAM
20 MHz
R (Mbps) constellation code rate

1/2 3/4 1/2 3/4 1/2 3/4 1/2 3/4
bn
1/2 3/4 1 3/2 2 3/2 3 9/2
bn
1/4 3/8 1/2 3/4 1 3/4 3/2 9/4
b
24 36 48 72 96 144 192 216
4 pilots
48 data tones
Broadcast channel cant optimize bit allocation
FCC demands flat spectrum so no energy-allocation The only knob is data rate selection
24
High-level system view
Data link control layer (DLC) Physical radio layer (PHY) Digital Guard Viterbi Demodulation interval decoder + + descrambler deinterleaver extraction + FFT Scrambler + forward error correction (FEC) coder Interleaver + modulation IFFT + guard interval insertion A/D Analog Analog front-end
D/A
Synchronization
25
Transceiver architecture
Pilot insertion Cyclic prefix Windowing Interleaver Upconvert Scrambler Descrambler Encoder Viterbi decoder Mapper Demapper DAC
FFT/IFFT
Channel estimator
LNA & Downconvert
Deinterleaver
Remove prefix
Synchronizer
AGC& ADC
[2]
26
Transmitter architecture detail
Data MAC Scrambler
Convolutional Encoder
Puncturer
Interleaver
Signal Mapper
IFFT
Cyclic Extend
Modulation & Upconvert
RF
Preamble
D/A
Transmitter (Digital Components)
Transmitter (Analog Components)
27
Scrambling

Need to randomize incoming data Enables a number of tracking algorithms in the receiver Provides flat spectrum in the given band
pseudo-random bit sequence (prbs) generator
What is the period of this pseudo-random sequence?
28
Interleaver

Protects the code from overload by burst errors Block interleaver

Block size is the #of coded bits in OFDM symbol (NCBPS) Two-step permutation Adjacent coded bits mapped
Onto nonadjacent sub-carriers Alternate between less and more significant bits in the constellation avoid long runs of low reliability LSBs
29
Convolutional Encoder
Rate 1/2 convolutional encoder
Punctured to obtain 2/3 and 3/4 rate
Omit some of the coded bits

Punctured Coding (r = 3/4) Source Data Output Data A Encoded Data A0 A1 A2 A3 A4 A5 A6 A7 A8 B0 B1 B2 B3 B4 B5 B6 B7 B8 X0 X1 X2 X3 X4 X5 X6 X7 X8
g0=1338
Input Data
Stolen Bit
Tb
Tb
Tb
Tb
Tb
Tb
Bit Stolen Data A0 B0 A1 B2 A3 B3A4 B5 A6 B6 A7 B8 (sent/received data) A0 A1 A2 A3 A4 A5 A6 A7 A8 B0 B1 B2 B3 B4 B5 B6 B7 B8
Bit Inserted Data
Inserted Dummy Bit
g1=1718

Output Data B Decoded Data y0 y1 y2 y3 y4 y5 y6 y7 y8
64-state (constraint length K=7) code Viterbi algorithm applied in the decoder
30
Signal mapper
BPSK, QPSK, 16-QAM, 64-QAM
Data divided into groups of (1,2,4,6) bits and mapped to a constellation point (i.e. a complex number) Gray-coded constellation mapings
Need the same average power for all mappings
Scale the output by KMOD

31
Pilot insertion and FFT/IFFT
Pilot insertion
Pilots BPSK, prbs modulated
FFT and IFFT shared
Just flip the Re and Im inputs
32
Spectral mask

Cannot use last 5 tones on each side Does not use extra windowing
Power Spectral Density (dB) Transmit Spectrum Mask (not to scale) -20 dBr
-28 dBr
Typical Signal Spectrum (an example)
-40 dBr -30 -20

-11 -9
fc
9 11
20
30 Frequency (MHz)
Transmit spectrum mask

33
Receiver architecture
I and Q from analog front-end Rx data to MAC
ADC
FIR
Remove DC of fset
Rotate
FFT
Channel
correct
Deinterleave
Viterbi
FIR
Frequency lock
Channel estimate and tracking
Rx gain to analog front-end
Autocorrelate Signal detect and AGC Symbol timing Pipeline control
34
Pilot tracking and channel correction
OFDM packet structure

8 + 8 = 16 s 10 x 0.8 = 8 s 2 x 0.8 + 2 x 3.2 = 8.0 s 0.8 + 3.2 = 4.0 s 0.8 + 3.2 = 4.0 s 0.8 + 3.2 = 4.0 s
t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 GI2
T1
T2
GI SIGNAL GI Data 1 GI Data 2

SERVICE + DATA DATA
Signal Detect, Coarse Freq. Channel and Fine Frequency RATE AGC, Diversity Offset Estimation Offset Estimation LENGTH Selection Timing Synchronize
to timing control
symbol timing adjust
training symbol pilots pilots long1 training from fft long2 training
pilot tracking core
angle adjust
magnitude adjust
fir
complex inverse
pilot magnitude multiply
rotate
data
composite channel correction channel to de-interleaver correction multiply

35
Synchronizer
Processing Data Path NCO

Input Data Z-16 Z-49 Moving Average JF(k) IP
crosscorrelator Output Symbols
FFT
IJ*
Plateau Detector
Tracking Data Path
Moving Average (16)
Ig.1(.) Jc(k)
Combine
36
Channel estimator
Equalizer Sync. + FFT Reference Sign Correction X(i,k) = H(i-D,k) Y(i,k) H(i-D,k) X(i,k) Residual Phase Correction Soft Demapper Deinterleaver Soft Viterbi
Channel Estimator Delay Buffer (D) HREF (k) Pilot Sign Correction Zero Forcing Y(i-D,k) H(i-D,k) = X(i-D,k) Mapper X(i-D,k) Interleaver Conv. Encoder Estimation Buffer (Circular Buffer)
Y(i-D,k)
37
First 802.11a chip
Image removed due to copyright restrictions.
[1]
38
References

[1] T.H. Meng, B. McFarland, D. Su and J. Thomson "Design and implementation of an all-CMOS 802.11a wireless LAN chipset," Communications Magazine, IEEE vol. 41, no. 8 SN - 0163-6804, pp. 160-168, 2003 [2] M. Krstic, K. Maharatna, A. Troya, E. Grass and U. Jagdhold "Implementation of an IEEE 802.11a compliant low-power baseband processor," Telecommunications in Modern Satellite, Cable and Broadcasting Service, 2003. TELSIKS 2003. 6th International Conference on vol. 1, no. SN -, pp. 97-100 vol.1, 2003. [3] J. Thomson, B. Baas, E.M. Cooper, J.M. Gilbert, G. Hsieh, P. Husted, A. Lokanathan, J.S. Kuskin, D. McCracken, B. McFarland, T.H. Meng, D. Nakahira, S. Ng, M. Rattehalli, J.L. Smith, R. Subramanian, L. Thon, Y.-H. Wang and R. Yu "An integrated 802.11a baseband and MAC processor," Solid-State Circuits Conference, 2002. Digest of Technical Papers. ISSCC. 2002 IEEE International vol. 1, no. SN -, pp. 126451 vol.1, 2002. [4] E. Grass, K. Tittelbach-Helmrich, U. Jagdhold, A. Troya, G. Lippert, O. Kruger, J. Lehmann, K. Maharatna, K.F. Dombrowski, N. Fiebig, R. Kraemer and P. Mahonen "On the single-chip implementation of a Hiperlan/2 and IEEE 802.11a capable modem," Personal Communications, IEEE [see also IEEE Wireless Communications] vol. 8, no. 6 SN - 1070-9916, pp. 48-57, 2001.
39

Communication System Design

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Communication System Design

Încărcat de

Drepturi de autor:

Formate disponibile

Practical multitone architectures

Lecture 4 Vladimir Stojanovi

6.973 Communication System Design Spring 2006 Massachusetts Institute of Technology

Loading with discrete information units

Chows algorithm (rounding of waterfilling results) Levin-Campello algorithm (greedy optimization)

Efficiency of bit distribution

6.973 Communication System Design

Levin-Campello (LC) algorithm- components

Efficientizing (EF) finding energy-efficient bit distribution

6.973 Communication System Design

1+0.9D-1 channel (Pe=10-6, gap=8.8dB, PAM/QAM)

PAM and single-sideband QAM

6.973 Communication System Design

Rate adaptive solution

Need also to satisfy the energy constraint E-tightness (ET)

Cant add another bit without violating the E constraint

6.973 Communication System Design

LC - Rate adaptive algorithm

ET example (continue from EF example)

Start from efficient distribution that blows up the energy constraint N = 8 1 = 8

Levin-Campello Rate Adaptive algorithm

6.973 Communication System Design

Dynamic rate adaptation

Bit is moved from channel n to m

Tightly coupled with channel and noise estimation

Will cover in later lectures

6.973 Communication System Design

Divide the channel into a set of parallel, independent channels

h(t) * x(t) = yN-1,k yN,k

xn,k . n(t - kT) * h(t) = xn,k . pn(t - kT)

y1,k y2,k yN,k

(lower case y indicates normalizing ||p|| factors included)

k n=1 p m(t) * p* n(-t) qn,m (t) = . ||pm|| ||pn|| Qk = Q(kT)

Figure by MIT OpenCourseWare.

Generalized Nyquist criterion

No interference between symbols No interference between sub-channels

6.973 Communication System Design

Transmission with eigen-functions

-T / 2 r(t - ) ()d n = 1,...,

[-(T - TH) / 2, (T + TH) / 2)]

Per-channel MAP is optimal Problem is vector AWGN

k (t ) Channel eigenfunctions (channel dependent)

6.973 Communication System Design

Convergence of multitone to modal modulation

Modal modulation is optimal for finite symbol time

Multitone converges to Modal modulation

If channel is periodic (does not exist in practice)

6.973 Communication System Design

Finite symbol duration effects

ISI corrupts the neighboring symbol

Need to leave the guardband

Need excess bandwidth for filter roll-off

6.973 Communication System Design

Discrete time channel partitioning

f0* 1/T f1* y

Figure by MIT OpenCourseWare.

6.973 Communication System Design

Creates a set of parallel channels by using SVD

discrete transmitter waveforms discrete matched filters

since F is unitary N is also AWGN

6.973 Communication System Design

Need one sample guardband

Would get 1.55bits/dim if N-> inf

6.973 Communication System Design