1. Wideband Communications
Lecture 8-9:
DMT
Aliazam Abbasfar

2. Outline
DMT examples
Channel estimation
Noise estimation

3. ADSL/VDSL ADSL : The most popular broadband Internet service
Over telephone lines
ITU .G992.1
DMT : T = 250 usec
Down stream
256 tones, KHz spacing, real baseband
(ADSL2+ /VDSL -> 512/4096 tones)
1/T’ = MHz ( BW = MHz)
N + v = (Hermitian symmetry )
2-3 tones are not used (phone line)
Tone 64 is pilot ( known QPSK data), Tone 256 not used
Pmax = 20.5 dBm
Up stream
Upstream transmission uses 32 tones to frequency 138 KHz
1/T’ = 276 KHz ( BW = 138 KHz)
N + v = (Hermitian symmetry )
1st tone not used (phone line)
Pmax = 14.5 dBm
Upto 12/1.5 Mbps down/upstream
Bit loading to optimize data rate
bmax = 15 (per tone)

4. Isaksson’s Zipper Bidirectional communication (full duplex)
N = 10, v=2, D = 3
Suffix = 2 D
Suffix = D
Time advance at TX

5. WiFi Wireless LAN 802.11a/g @ 5/2.4 GHz COFDM : T = 4 usec
64 tones, KHz spacing, complex baseband
1/T’ = 20 MHz BW = 20 MHz (15.56)
N + v =
Tones : -31 to 31 (48 tones for data )
-31 to -27, 0, 27 to 32 are not used
-27, -7, 7, and 21 are pilot
Data rate = k * 48 * 250 KHz = 12k Mbps k = bn : bits/tone
Upto 54 Mbps
Variable coding
No bit loading
bn is constant for all tones
Pmax = 16/23/29 dBm

6. Digital Video Broadcast (DVB)
Digital TV broadcast
Single frequency network (SFN)
Improves coverage
Creates ISI
COFDM
2048 or 8192 tones, 4.464/1.116 KHz spacing
complex baseband
1/T’ = MHz BW = 20 MHz (15.56)
N T’ = 8192 T’ = 896 usec (1/1.116 KHz)
(N+v) T’ = 924/952/1008/1120 usec
N T’ = 2046 T’ = 224 usec (1/4.464 KHz)
(N+v) T’ = 231/238/252/280 usec
4/16/64 QAM
Coding : 172/204 * 1/2, 2/3, 3/4, 5/6, or 7/8
Data rates : 4.98  Mbps
Carries 2-8 TV channels

7. Channel estimation Channel model (n=0, 1,.., N-1)
yn = xn * pn + un
Yn = Pn Xn + Un
P = Q p
Time/Freq. domain estimates and errors
pn , Pn
en = yn – xn * pn
En = Yn - Pn Xn
MSE = E[|e|2] /N = E[|E|2] /N
MSE = E[|u|2] /N + E[|p-p|2] Ex
MSE = E[|U|2] /N + E[|P-P|2] Ex
MSEch = E[|p-p|2] Ex = E[|P-P|2] Ex
SNR = E[|x|2 ] /N |p|2 / E[|u|2] /N = Ex |p|2 / s2u
SNRn = |Pn|2 E[|Xn|2] / E[|Un|2] = |Pn|2 E[|Xn|2]/ s2n
SNRch,n >> SNRn  E[|Un|2] >> E[|Pn - Pn|2]

8. Channel estimation (2) Estimation method
Average over L trials
Estimation error
E[ |En|2] = E[|Un|2] + E[|P-P|2] /N
= E[|Un|2] (1+ 1/L)
SNRch,n = L SNRn
L = 40  1/L = 0.1 dB  0.1 dB accuracy
Time–domain constraint improves the estimation

9. FEQ Frequency domain equalizer Adaptive methods Yn x Wn = Xn Wn = 1/Pn
Zero Forcing
MMSE

10. Noise estimation Noise variance estimator 0.1 dB accuracy  L=3200
Mean = true estimate
Variance
0.1 dB accuracy  L=3200
%10 chance of more deviation
%1  12,800
%0.1  28,800
Noise tracking
First order update
L = 6400  m’ = 6x10-4
Note : 1/gn = k s2

11. Windowing DMT/OFDM has a windowing for each symbol Windowing in time =
Duration = T
Rectangular/Raised cosine
Needs extra guard time
excess bandwidth
Windowing in time =
Convolution in frequency
At TX
Reduces out of band emission
At RX
Reduces narrowband noise
Lower sidelobes improves gn

12. Reading
Cioffi Ch. 4.7, 4.9