1. BLIND CROSSTALK CANCELLATION FOR DMT SYSTEMS
Nadeem Ahmed Nirmal Warke
ECE Dept DSPS R&D Center
Rice University Texas Instruments

2. Motivation
New multimedia and networking applications => increasing demand for bandwidth
DSL is cost effective broadband solution
100 MHz
10 MHz
1 MHz
100 kHz
10 kHz
POTS
ADSL
VDSL
ISDN
HDSL

3. Motivation
Increasing density of DSL deployment => Increased crosstalk
Crosstalk typically increases with frequency => significant impairment for high speed DSL
Binder
ADSL lines
HDSL lines
POTS

4. Types of Crosstalk Near-End Crosstalk (NEXT):
Interference that arises when signals are transmitted in opposite directions
Far-End Crosstalk (FEXT):
Interference that arises when signals are transmitted in the same direction

5. DSL System Model FEXT signals travel the entire length of the channel
FDD modems virtually eliminate self-NEXT.
Main source of crosstalk comes from other services (i.e. HDSL, T1, etc), which are much stronger than self-FEXT.

6. Crosstalk Power on Line

7. Combating Crosstalk Crosstalk Avoidance Crosstalk Cancellation
Varying transmit spectra
Modified bit-loading algorithm
Block coding across modems at CO
Crosstalk Cancellation
Treat as multiuser detection problem
Using DFE’s
Exploit symbol rate differences

8. Varying Transmit Spectra
Design optimal transmit spectra which vary with channel, noise and interference
Designed to reject self-NEXT in a manner which maximizes overall data rate
Maintains spectral compatibility with other services

9. Modified Bit-Loading Algorithm
Modify the bit-loading algorithm
Change order of placing power in bins
Factor NEXT into algorithm
Minimizes NEXT within cable binder and extend reach of service

10. Block Coding Across COs
Block coding to eliminate NEXT
If code blocks are greater than a minimum length, NEXT can be completely eliminated
Need control of a service i.e., all DSL modems
only useful for self-NEXT rejection

11. Multi-User Detection
Use multiuser detection techniques to cancel crosstalk
Jointly detect desired and crosstalk signals
Published results for Home LAN interference cancellation from VDSL

12. DFE For Self-NEXT/FEXT
Use DFE to remove cyclo-stationary crosstalk
Assumes crosstalk has same sampling rate as source
Useful for self-NEXT and self-FEXT cancellation

13. Excess Band Crosstalk Cancellation
Crosstalkers like ISDN, HDSL, T1 have large excess band
Algorithm
Exploits lower symbol rate of crosstalker relative to the sampling rate of DSL
Crosstalker estimated in excess band and cancelled in main band

14. Practical Issues
Most methods require knowledge of crosstalk coupling function
How do you reliably estimate the coupling function-
Use models?
Based on training data?
Very difficult problem

15. Excess Band Crosstalk Cancellation
Paper by Zeng et al on Crosstalk Cancellation for DMT Systems

16. Excess Band Crosstalk Cancellation
Brick wall filters cannot be realized
After D/A conversion, filter cannot remove all of image energy
If crosstalk signal is oversampled with respect to xDSL, excess band can be observed
Estimate crosstalk signal in excess band and predict crosstalk in main band

17. Mathematical Formulation
DMT Modulation
System Impaiments- crosstalk and noise
DMT Demodulation

18. Mathematical Formulation
Partition into 2 freq. Bands:
2 => main band 1 => excess band
Demodulate DMT signal in excess band and subtract to estimate crosstalk signal

19. Cancellation Algorithm
Let x = M.r be a linear estimate of crosstalk signal component x
MMSE Estimate:
Hence crosstalk signal in main band is,
Project onto
main band M

20. Blind Cancellation
If = .C and x = b, channel is assumed to be known => Zeng’s solution
Instead, let = and x = C.b => Blind Approach
Solution uses crosstalk statistics i.e. autocorrelation information
Estimate coupling function and crosstalk data simultaneously

21. Dependence on crosstalk symbol delay
Relative crosstalk symbol delay varies with DMT frame => varies with DMT frame
where,

22. Blind Cancellation- Practical Solution
Autocorrelation can be easily estimated during training and/or quiet periods
Crosstalk cancellation matrix can be pre-computed and stored
Steady state operation involves product of cancellation matrix with vector r
Practical to implement

23. Crosstalk Simulations
Consider an ADSL system:
Transmission bandwidth: (25.875, 1104) kHz
256 tones over 1104 kHz bandwidth
AWGN at –140 dBm/Hz
Crosstalk: 1 HDSL (f_N=192kHz) and 1 T1 (f_N=772kHz)
Assumption
Assume crosstalk symbol delay is known to within some finite precision

24. Crosstalk Measurements
Used vector signal analyzer
12 wire twisted pair cable binder (4000 ft)
Used periodic chirp as input signal
Captured magnitude and phase of transfer function

25. Channel Measurements
4000 ft, 24AWG, 21 pair wire binder

26. NEXT Coupling Functions
From 1 into 2
From 11 into 5
4000 ft, 24AWG, 21 pair wire binder

27. HDSL Crosstalk Cancellation
15/12dB average crosstalk energy reduction for Q(T/4)/Q(T/2)

28. HDSL Crosstalk Cancellation
1500/1000ft average reach improvement at 1Mbps for Q(T/4)/Q(T/2)

29. HDSL+T1 Crosstalk Cancellation
require 2x oversampled receiver
12/7dB average crosstalk energy reduction for Q(T/4)/Q(T/2)

30. HDSL+T1 Crosstalk Cancellation
2000/1500ft average reach improvement at 1Mbps for Q(T/4)/Q(T/2)

31. Conclusions
Blind crosstalk cancellation method uses statistical properties of received signal
Signal cancellation matrix can be pre-computed (steady state operation involves inner products)
Simulations show significant gain for realistic ADSL system
Performance is robust to jitter in crosstalk symbol timing estimate

32. Future Work Investigate methods for estimating crosstalk symbol timing
Study effect of incorrect DMT decisions in excess band on cancellation performance (multiple crosstalkers)
Investigate alternative crosstalk cancellation methods