High-Speed Wireline

Communication Systems:
Semester Wrap-up
Ian C. Wong, Daifeng Wang, and
Prof. Brian L. Evans
Dept. of Electrical and Comp. Eng.
The University of Texas at Austin
http://signal.ece.utexas.edu

http://www.ece.utexas.edu/~bevans/projects/adsl
 Outline
• Asymmetric Digital Subscriber Line (ADSL) Standards
– Overview of ADSL2 and ADSL2+
– Data rate vs. reach improvements
– ADSL2+
• Multichannel Discrete Multitone (DMT) Modulation
– Dynamic spectrum management
– Channel identification
– Spectrum balancing
– Vectored DMT
• System Design Alternatives and Recommendations

2
 1
ADSL2 and ADSL2+ - the new standards
• ADSL2 (G.992.3 or G.dmt.bis, and G.992.4 or G.lite.bis)
– Completed in July 2002
– Minimum of 8 Mbps downstream and 800 kbps upstream
– Improvements on:
• Data rate vs. reach performance
• Loop diagnostics
• Deployment from remote cabinets
• Spectrum and power control
• Robustness against loop impairments
• Operations and Maintenance
• ADSL2+ (G.992.5)
– Completed in January 2003
– Doubles bandwidth used for downstream data (~20 Mbps at 5000 ft)
1
Figures and text are extensively referenced from [ADSL2] [ADSL2white] 3
 Data rate vs. reach performance improvements
• Focus: long lines with narrowband interference
• Achieves 12 Mbps downstream and 1 Mbps upstream
• Accomplished through
1. Improving modulation efficiency
2. Reducing framing overhead
3. Achieving higher coding gain
4. Employing loop bonding
5. Improving initialization state machine
6. Online reconfiguration

4
 1. Improved Modulation Efficiency
• Mandatory support of Trellis coding (G.992.3, §8.6.2)
– Block processing of Wei's [Wei87] 16-state 4-dimensional trellis code
shall be supported to improve system performance
– Note: There was a proposal in 1998 by Vocal to use a Parallel
concatenated convolutional code (PCCC), but it wasn’t included in the
standard (http://www.vocal.com/white_paper/ab-120.pdf)
• Data modulated on pilot tone (optional, §8.8.1.2)
– During initialization, the ATU-R receiver can set a bit to tell the ATU-
C transmitter that it wants to use the pilot-tone for data
– The pilot-tone will then be treated as any other data-carrying tone
• Mandatory support for one-bit constellations (§8.6.3.2)
– Allows poor subchannels to still carry some data

5
 2. Reduced framing overhead
• Programmable number of overhead bits (§7.6)
– Unlike ADSL where overhead bits are fixed and consume 32 kbps of
actual payload data
– In ADSL2, it is programmable between 4-32 kbps
– In long lines where data rate is low, e.g. 128 kbps,
• ADSL: 32/128 = 25% is overhead
• ADSL2: as low as 4/128 = 3.125% is overhead

6
 3. Achieved higher coding gain
• On long lines where data rates are low, higher coding gain
from the Reed-Solomon (RS) code can be achieved
• Flexible framing allows RS code to have (§7.7.1.4)
• 0, 2, 4, 6, 8, 10, 12, 14, or 16 redundancy octets
• 0 redundancy implies no coding at all (for very good channels)
• 16 would achieve the highest coding gain at the expense of higher
overhead (for very poor channels)

7
 4. Loop Bonding
• Supported through Inverse Multiplexing over ATM (IMA)
standard (ftp://ftp.atmforum.com/pub/approved-specs/af-phy-0086.001.pdf)
– Specifies a new sublayer (framing, protocols, management) between
Physical and ATM layer [IMA99]

8
 5. Improved initialization state machine
• Power cutback
– Reduction of transmit power spectral density level in any one direction
– Reduce near-end echo and the overall crosstalk levels in the binder
• Receiver determined pilots
– Avoid channel nulls from bridged taps or narrow band interference
from AM radio
• Initialization state length control
– Allow optimum training of receiver and transmitter signal processing
functions
• Spectral shaping
– Improve channel identification for training receiver time domain
equalizer during Channel Discovery and Transceiver Training phases
• Tone blackout (disabling tones)
– Enable radio frequency interference (RFI) cancellation schemes
9
 6. Online reconfiguration (§10.2)
• Autonomously maintain operation within limits set by
control parameters
– Useful when line or environment conditions are changing
• Optimise ATU settings following initialization
– Useful when employing fast initialization sequence that requires
making faster estimates during training
• Types of online reconfiguration
– Bit swapping
• Reallocates data and power among the subcarriers
– Dynamic rate repartitioning (optional)
• Reconfigure the data rate allocation between multiple latency paths
– Seamless rate adaptation (optional)
• Reconfigure the total data rate

10
 ADSL2+ (G.992.5)
• Doubles the downstream bandwidth
• Significant increase in downstream data rates on shorter
lines

11
 Outline
• Asymmetric Digital Subscriber Line (ADSL) Standards
– Overview of ADSL2 and ADSL2+
– Data rate vs. reach improvements
– ADSL2+
• Multichannel Discrete Multitone (DMT) Modulation
– Dynamic spectrum management
– Channel identification
– Spectrum balancing
– Vectored DMT
• System Design Alternatives and Recommendations

12
 Dynamic Spectrum Management
• Allows adaptive allocation of spectrum to various users in
a multiuser environment
– Function of the physical-channel
– Used to meet certain performance metrics
– One can treat each DMT receiver as a separate user
• Better than static spectrum management
– Adapts to environment rather than just designing for worst-case
– E.g. ADSL used static spectrum management (Power Spectral Density
Masks) to control crosstalk
– Too conservative: limited rates vs. reach

13
 Dynamic Spectrum Management
• Channel Identification Methods
– Initialization and training
– Estimation of the channel transfer function
• Spectrum Balancing
– Distributed power control (iterative waterfilling)
– Centralized power control (optimal spectrum management)
• Vectored Transmission Methods

14
 Training Sequences
• Training Sequence
– Goal: estimate the channel impulse response before data transmission
– Type: periodic or aperiodic, time or frequency domain
– Power spectrum: approximately flat over the transmission bandwidth
– Design: optimize sequence autocorrelation functions
• Perfect Training Sequence
– All of its out-of-phase periodic autocorrelation terms are 0 [1]
• Suggested training sequences for DMT
– Pseudo-random binary sequence with N samples
– Periodic by repeating N samples or adding a cyclic prefix

[1] W. H. Mow, “A new unified construction of perfect root-of-unity sequences,” in Proc.

15
Spread-Spectrum Techniques and Applications, vol. 3, 1996, pp. 955–959.
 Training Sequences
• y=Sh+n
MIMO is multiple-input multiple-output
– h: L-tap channel
– S: transmitted N x L Toeplitz matrix made up of N training symbols
– n: additive white Gaussian noise (AWGN)
Domain Method Minimum Complexity Optimal
MSE Sequence*
Time Periodic (LS)[1] Yes High (2N) Yes
Aperiodic [2] No Medium (N2) Yes
L-Perfect (MIMO) Almost Low (N log2N) Sometimes
[3]
Frequency Periodic [4] No Low (N log2N) Sometimes

* impulse-like autocorrelation and zero crosscorrelation

[1] W. Chen and U. Mitra, "Frequency domain versus time domain based training sequence optimization," in Proc. IEEE Int. Conf.
Comm., pp. 646-650, June 2000.
[2] C. Tellambura, Y. J. Guo, and S. K. Barton, "Channel estimation using aperiodic binary sequence," IEEE Comm. Letters, vol.
2, pp. 140-142, May 1998.
[3] C. Fragouli, N. Al-Dhahir, W. Turin, “Training-Based Channel Estimation for Multiple-Antenna Broadband Transmissions,"
IEEE Trans. on Wireless Comm., vol.2, No.2, pp 384-391, March 2003
16
[4] C. Tellambura, M. G. Parker, Y. Guo, S . Shepherd, and S . K. Barton, “Optimal sequences for channel estimation using
Discrete Fourier Transform techniques,” IEEE Trunsuctions on Communicutions, vol.47, no.2, pp. 230-238, Feb. 1999
 Training-Based Channel Estimation for MIMO
• 2 x 2 MIMO Model
Duplex Channel
TX 1 RX 1
h11

h21 h12
TX 2 RX 2
h22

 y1   h11 ( L ) h12 ( L ) 
y   Sh  z [ S1 ( L, N t ) S 2 ( L, N t )]  h ( L ) z
y
 2  21 h22 ( L ) 
where y and z are of dimension 2( N t  L  1) 1
T
hij ( L )  hij (0)  hij ( L  1)  , i or j 1, 2
 si ( L  1)  si (0) 
 s ( L)  si (1) 
S i ( L , N t )  i  , i 1, 2
    
 
 si ( N t  1)  si ( N t  L )  17
 Crosstalk Estimation
• Noises are “unknown” crosstalkers and thermal/radio
– Power spectral density N(f)
– Frequency bandwidth of measurement
– Time interval for measurement
– Requisite accuracy
• Channel ID 1
– Estimate gains at several frequencies
– Estimate noise variances at same frequencies
– SNR is then gain-squared/noise estimate
• Basic MIMO crosstalk ID Transmitter
User i
– Near-end crosstalk (NEXT)
NEXT FEXT
– Far-end crosstalk (FEXT)
Receiver
User j

18
 Spectrum Balancing
• Decides the spectral assignment for each user
– Allocation is based on channel line and signal spectra
– For single-user, ‘water-filling’ is optimal
– For the multiuser case, performance evaluation and/or optimization
becomes much more complex
• Methods
– Distributed power control
• No coordination at run-time required
• Set of data rates must be predetermined
– Centralized power control
• Coordination at central office (CO) transmitter is required

19
 Distributed Multiuser Power Control
[Yu, Ginis, & Cioffi, 2002]
• Iterative waterfilling approach

20
 Centralized Optimal Spectrum Management
[Cendrillon, Yu, Moonen, Verlinden, & Bostoen, to appear]

• Rate-adaptive problem with rate constraints

21
 Comparison among methods
10K ft
CO
10K ft
7K ft RT

22
 Vectored Transmission Methods
• Signal level coordination
– Full knowledge of downstream transmitted signal and upstream
received signal at central office
– Block transmission at both ends fully synchronized
• Channel characterization
– MIMO on a per-tone basis
DS-Precoding

Tx Rx

Rx Tx

CO US-Successive RT
Crosstalk-Cancellation 23
 Upstream: Successive Crosstalk Cancellation

= +

K vector of K£K MIMO

received samples channel matrix for tone i

= +

uncorrelated components

24
 Downstream: MIMO Precoding

Transmitted signal
£
Original symbols

Channel

=
Received signal
crosstalk-free
• We can also use Tomlinson-Harashima precoding
(as used in High-speed DSL) to prevent energy increase
25
 Comments
• Because of limited computational power at downstream Tx
(reverse of that in typical DSL/Wireless systems)
– Successive crosstalk cancellation at Rx makes more sense
• Do the QR decomposition also at Rx
• Don’t need to feedback channel information, since it is used at the
receiver only
• Transmit optimization procedures can also be done at Rx
– It is actually simpler since we can assume that the cross-talk is
cancelled out
• Just do single-user waterfilling for each separate user (loop)
– Optimal power allocation settings fed back to transmitter

26
 Outline
• Asymmetric Digital Subscriber Line (ADSL) Standards
– Overview of ADSL2 and ADSL2+
– Data rate vs. reach improvements
– ADSL2+
• Multichannel Discrete Multitone (DMT) Modulation
– Dynamic spectrum management
– Channel identification
– Spectrum balancing
– Vectored DMT
• System Design Alternatives and Recommendations

27
 Training-Based Channel Estimation for MIMO
• Linear Least Squares
– Low complexity but enhances noise. Assumes S has full column rank

 hˆ11 hˆ12  H -1 H
ĥ=   =(S S) S y
 hˆ21 hˆ22 

• MMSE
– zero-mean and white Gaussian noise: R z E  zz  2 I Nt - L 1
H 2


  
h  hˆ  2 2Tr ((SH S) -1 )
H
ˆ
MSE E h  h
 
2 L
2
H
 S1
H
S1 S
H
2 S1

MMSE = , iff S S=  H  ( N t  L  1)I 2 L
Nt  L 1  S1 S2
H
S2 S2 
– Sequences satisfy above are optimal sequences
– Optimal sequences: impulse-like autocorrelation and zero
crosscorrelation
28
 Simple Channel Estimation for MIMO
Method Computational Time
• How to design s1(L,Nt) and s2(L,Nt) ? Complexity
Simple Low High
• Simple and intuitive method ( 2 X 2 ) Design TS High Low

– Sending the training data at only one TX( turn off another TX) during
one training time slot, i.e.
yt0 ,1 yt0 ,2
time0 :  s1 s2   s 0  h11  , h12 
s s
yt1 ,1 yt1 ,2
time1:  s1 s2   0 s   h21  , h22 
s s
– Very Low Complexity and even No Need to Design Training
Sequences
– But Time Consuming
• Design training sequences to estimate the channel during
one training time slot
29
 Design Training Sequences for MIMO
• Recommendation Design Method I
– Design instead a single training sequence s (2L, Nt+L+1)
– s1=[s(0)…s(Nt)], s2=[s(L)…s(Nt+L)] Method Computationa MMSE
l Complexity
– MMSE but High searching complexity I High Yes
• Recommendation Design Method II II Low Almost

– A sequence s produces s1 and s2 with 0 cross correlation by encoding

 y1    S*2 S1*   h1   z1 
 y       
 2   S1  S2   h 2   z 2 
S
– Lower MSE and Only s with good auto-correlation properties
p
– Trellis Code: s1 (k ) s (k ), s2 (k ) (  1) k  1 s ( k  1)
– Block Code: [ s1  s2* ]  s1 [ s2 s1* ]  s2 ~ time-reversing

1)  =S , S =S
S * complex
1 1 2 1
SH S ( N t conjugation
 L  1)I 2 L
2) S1 =S1 , S2 =S 1 30
 Choice of Multichannel Method
• Choice of methods is a performance-complexity tradeoff
• Loop bonding simplest to implement, but poor performance
• Spectrum balancing methods
– Iterative waterfilling at the receiver can be implemented pretty easily
• Pre-determine target rates through offline analysis
• No coordination needed among the loops
• Just feedback the power allocation settings to corresponding Tx
– Optimal spectrum management
• We can simply maximize rate-sum (all weights=1)
• Coordination at Rx is needed (jointly optimize across loops)
• Vectored transmission
– Coordination on both sides are required
– Run-time complexity is not too bad: O(K3) QR-Decomposition only
need to be done at training
– Transmit optimization is also simpler than spectrum balancing methods 31
 Comparison

Loop Iterative Optimal Vectored-

Bonding Waterfilling Spectrum DMT
Balancing
Design Low Medium Medium High
Complexity
Computational Low Medium Very high High
Complexity
Coordination Low Medium High Very high
Required
Data-rate Low Medium High Very High
performance

32
Backup Slides

33
 ADSL2 improvements over ADSL
• Application-related features
– Improved application support for an all digital mode of operation and
voice over ADSL operation;
– Packet TPS-TC1 function, in addition to the existing Synchronous
Transfer Mode (STM) and Asynchronous TM (ATM)
– Mandatory support of 8 Mbit/s downstream and 800 kbit/s upstream for
TPS-TC function #0 and frame bearer #0;
– Support for Inverse Multiplexing for ATM (IMA) in the ATM TPS-
TC;
– Improved configuration capability for each TPS-TC with configuration
of latency, BER and minimum, maximum and reserved data rate.

1
Transport Protocol Specific-Transmission Convergence 34
 ADSL2 improvements over ADSL (cont.)
• PMS-TC1 related features
– A more flexible framing, including support for up to 4 frame bearers, 4
latency paths;
– Parameters allowing enhanced configuration of the overhead channel;
– Frame structure with
• Receiver selected coding parameters;
• Optimized use of RS coding gain;
• Configurable latency and bit error ratio;
– OAM2 protocol to retrieve more detailed performance monitoring
information;
– Enhanced on-line reconfiguration capabilities including dynamic rate
repartitioning.

1
Physical Media Specific-Transmission Convergence
2
Operations, Administration, and Maintenance 35
 ADSL2 improvements over ADSL (cont.)
• Physical Media Dependent (PMD) related features
– New line diagnostics procedures for both successful and unsuccessful
initialization scenarios, loop characterization and troubleshooting;
– Enhanced on-line reconfiguration capabilities including bitswaps and
seamless rate adaptation;
– Optional short initialization sequence for recovery from errors or fast
resumption of operation;
– Optional seamless rate adaptation with line rate changes during
showtime;
– Improved robustness against bridged taps with RX determined pilot;
– Improved transceiver training with exchange of detailed transmit signal
characteristics;
– Improved SNR measurement during channel analysis;
– Subcarrier blackout to allow RFI measurement during initialization and
SHOWTIME;
– Improved performance with mandatory support of trellis coding, one-bit
constellations, and optional data modulated on the pilot-tone
36
 ADSL2 improvements over ADSL (cont.)
• PMD related features (cont.)
– Improved RFI robustness with receiver determined tone ordering;
– Improved transmit power cutback possibilities
– Improved Initialization with RX/TX controlled duration of init. states;
– Improved Initialization with RX-determined carriers for modulation of
messages;
– Improved channel identification capability with spectral shaping during
Channel Discovery and Transceiver Training;
– Mandatory transmit power reduction to minimize excess margin under
management layer control;
– Power saving feature with new L2 low power state and L3 idle state;
– Spectrum control with individual tone masking under operator control
through CO-Management Information Base;
– Improved conformance testing including increase in data rates for
many existing tests.
37
 Bibliography
[ADSL2] ITU-T Standard G.992.3, Asymmetric digital subscriber line transceivers 2
(ADSL2), Feb. 2004
[ADSL2white] ADSL2 and ADSL2plus-The new ADSL standards. Online:
http://www.dslforum.org/aboutdsl/ADSL2_wp.pdf, Mar. 2003
[Wei87] L.-F.Wei, “Trellis-coded modulation with multidimensional constellations,” IEEE
Trans. Inform. Theory, vol. IT-33, pp. 483-501, July 1987.
[IMA99] ATM Forum Specification af.phy-0086.001, Inverse Multiplexing for ATM (IMA),
Version 1.1., Mar. 1999

38