Download presentation

Presentation is loading. Please wait.

Published byDarlene Climo Modified about 1 year ago

1
Crosstalk and Loop Make-Up Identification for DSL Systems Dr. Stefano Galli Senior Scientist, Telcordia Technologies University of British Columbia, Vancouver, Distinguished Lecture, Feb. 27, 2006

2
UBC 2006– 2 Talk Outline DSL issues DSL Spectrum Management Enablers – Crosstalk Identification – Loop Make-Up Identification Results on Iterative Algorithms for Optimization Simulation results

3
UBC 2006– 3 Copper Impairments “POTS” faults – grounds, shorts, – load coils, balance, dBrn, loop length Loop loss – Increases with frequency – Bridged tap: up to 10 dB spectral dips Crosstalk – Coupling between different systems on different pairs – Often most significant noise – Multi-pair feeder and distribution cabling Radio ingress noise = Electromagnetic interference (EMI) – Narrow band frequency spikes – Unshielded drop and inside wire Impulsive noise hits – Short bursts (10s of microseconds) of high power noise – Long-term (hour) error monitoring Non-linear distortion – From phones with no microfilter, some protectors

4
UBC 2006– 4 Crosstalk is the major impairment of xDSL Near End Crosstalk (NEXT) Far End Crosstalk (FEXT)

5
UBC 2006– 5 Managing crosstalk As DSL rollout continues, the problem of crosstalk will become more and more important for at least two reasons: – Spectral compatibility issues necessity of impartial third party for monitoring and resolution – Performance issues necessity of multiuser detection, cancellation, or dynamic management Systems engineered to face worst case xtalk

6
UBC 2006– 6 Dynamic Spectrum Management (DSM) CO CO-based ADSL ADSL, ADSL2+, or VDSL Remote Terminal (RT) CORT Power Frequency CO RT Power Frequency Power Frequency CO RT Frequency Power CO RT Today – equal transmit spectra really bad Crosstalk Separate Frequency bands Power Back-off Joint Optimization Tomorrow

7
UBC 2006– 7 – Crosstalk is NOT random noise – we can control it! Identify crosstalk sources and couplings – Spectrum balancing Balance transmit power levels Adapt transmit spectra to optimize performance AND minimize crosstalk BIT RATES APPROXIMATELY DOUBLE – Vectoring Real-time signal coordination, crosstalk cancellation BIT RATES APPROXIMATELY TRIPLE DSL #1 DSL #2DSL #3 XT 1,2 XT 1,3 XT 2,3 Managing crosstalk

8
UBC 2006– 8 Input data from DSL Modems & DSLAMs Double-Ended Loop Test (DELT) – Requires working DSL service – Now: proprietary TL1 & SNMP interfaces Bit rates, SNR Margin, Gross Attenuation, Bit loading spectrum Different formats, often lack accuracy ADSL2 & 2+ Diagnostics (ITU G &.5) - products just out Standardized formats, accuracy For each of the 255 subcarriers to 1.1 MHz: Channel Transfer Function H(f), Quiet Line Noise PSD QLN(f), and Signal ‑ to ‑ Noise Ratio SNR(f) Aggregate: Line Attenuation, Signal Attenuation, Signal ‑ to ‑ Noise Margin, Attainable Net Data Rate, Aggregate Transmit Power Single-ended loop test (SELT), G.selt ITU-T project, vendor implementations – Single-ended modem-based loop & noise tests – no service or end- user modem required

9
UBC 2006– 9 Extract data from ADSL modems ADSL Discrete multi-tone (DMT) modem data: tones - Nearly as good as continuous spectra!

10
UBC 2006– 10 SELT Approach It is targeted both to the estimation of the pair-to-pair couplings and to the identification of the source. It is useful for crosstalk cancellation/multiuser detection. Should not only be modem-based: it may be used before providing service, measures crosstalk across all bandwidth. Not much literature on xtalk identification. First papers are recent: Zeng, C. Aldana, A. Salvekar, J. Cioffi, “Crosstalk Identification in DSL Systems”, IEEE JSAC, Aug S. Galli, C. Valenti, K. Kerpez, “A Frequency-Domain Approach to Crosstalk Identification in DSL Systems”, IEEE JSAC, Aug Crosstalk Identification

11
UBC 2006– 11 Traditional approach is in terms of power sums (sum of the pair- to-pair NEXT coupling powers of the other pairs in the binder group). Engineering is made in terms of the 1% worst case: linear in the log-log scale, 15 dB per decade of frequency. Power-Sum Models !

12
UBC 2006– 12 Individual pair-to-pair couplings: situation is more complex No smooth curves, high variability, no known models.

13
UBC 2006– 13 Worst 25 measured pair-to-pair crosstalk couplings out of 300 Dark black line = 99% worst-case model

14
UBC 2006– 14 Perform a vast measurement campaign of the pair-to-pair couplings on several cables. Create a set of pair-to-pair couplings, choosing them on the basis of a specific criterion create a dictionary of ptp couplings. Multiply the ptp couplings of the dictionary by the PSDs of all the possible xDSLs (ISDN, ADSL, HDSL, etc.) create a dictionary of crosstalk PSDs profiles. Measure xtalk PSD and search for the “closest” xtalk PSD profile in the dictionary. Finding the “closest” PSD profile gives us the most likely xtalk source and ptp coupling. Crosstalk Identification Algorithm

15
UBC 2006– 15 Basis set members for T1 NEXT

16
UBC 2006– 16 Vector Choice Methods Power: ranking the ptp couplings on the basis of their dB sum across all frequencies. Singular Value Decomposition: choosing only the linear independent vectors in the dictionary. Search Methods Correlation: Perform a statistical correlation between measured xtalk PSD and xtalk PSD profiles in the dictionary. Multiple Linear Regression: best suited for the identification of multiple disturbers. Matching Pursuit: mathematical formulation of the problem equivalent to the problem of finding the best sparse representation of a vector on the basis of an overcomplete dictionary.

17
UBC 2006– 17 : set of N frequency samples of the measured crosstalk PSD caused by un unknown DSL disturber : set of N frequency samples of the k-th basis crosstalk PSD profile, with 1 k P Crosstalk ID: Problem Statement Find the single disturber that generates crosstalk Y given the set of all the crosstalk PSD profiles X, i.e. find relationship between Y and X Classical regression problem:

18
UBC 2006– 18 The regression coefficients a (k) and b (k) are determined by the condition that the sum of the squared residuals S (k) is minimum: It is possible to show that the sum of squared residuals can be expressed in terms of the correlation coefficient: Minimizing sum of squared residuals is equivalent to finding the maximum correlation coefficient

19
UBC 2006– 19 vector containing the measured crosstalk PSD from unknown disturbers across all N frequencies full rank NxP matrix containing all the PSD profiles vector of weighting coefficients vector containing the residuals over all the frequency points

20 UBC 2006– 20 N<

21
UBC 2006– 21

22
UBC 2006– 22 No SVD (P = 800) q=[1,1,1,1,1,1,1,1] ( = 8) 1) ISDN14/14 2) HDSL24/2423/24 3) T138/3836/38 4) ADSL Dn37/37 5) SDSL 40026/26 6) SDSL /2823/27 7) SDSL /3126/31 8) HDSL2 Up27/2729/29 Totals223/225214/226 ID rate (%)99.1%94.7% SVD allows to drastically reduce computational complexity.

23
UBC 2006– 23 Loop Response Estimation TDR wideband reflectometry signals estimate bridged taps, segment lengths and gauges Entire DSL band spectral response – Better than just one frequency or one length number – Determine Bridged Tap effects Example: Actual loop Vs ft equivalent 26 gauge (EWL) 1400 ft 813 ft 22 Gauge 650 ft

24
UBC 2006– 24 Necessity of Loop Qualification Not all local loops can support DSL technology Before DSL can be deployed, local loops must be tested to see whether they can support it or not This may be obtained on the basis of a detailed loop characterization Detailed loop characterization is difficult to obtain because: 1) information kept for POTS service was not detailed 2) loop records are often on paper 3) records are often wrong It is necessary to perform measurements

25
UBC 2006– 25 Single-ended testing: Requires test equipment at the central office only Measures can be performed at the Central Office without involving on-site technicians Double-ended testing: Requires equipment at both ends of the loop Involves dispatching a technician to the customer’s location Transfer function easily estimated Types of Loop Qualification Single-ended testing is more complicated but: It is less expensive loop identificationIt may allow us to unveil the loop make-up, not only the transfer function “loop identification” implies loop qualification

26
UBC 2006– 26 Telco DSL CO 5900’ 26 gauge 2000’ 24 gauge 500’ 24 gauge 500’ 22 gauge 1500’ 26 gauge 1500’ 26 gauge Estimate loop make-up stick diagram from single-ended CO-based TDR measurement TDR Loop Make-Up Identification The test equipment transmits a set of “ad-hoc” designed signals echoWhen a signal travels on a loop and encounters medium discontinuities (gauge changes, bridged taps, end of line) part of the signal is reflected back, i.e. an echo is generated. Unwanted spurious echoes are also always present The test equipment will process the received echoes

27
UBC 2006– 27 Echo Modeling

28
UBC 2006– 28 Observation of the superposition of all the echoes (Test signal: 200 ns Square Pulse, 1 V amplitude) Echo Modeling

29
UBC 2006– 29 The following unknown quantities must be estimated: Time of arrival of the echo: i Amplitude of the echo: a i Waveform of the echo (shape): e (i) (t) Number of non-negligible echoes N Echo Modeling Problems: Some echoes are spurious and some are not Some echoes overlap

30
UBC 2006– 30 The shape of echo (real or spurious) generated at a discontinuity depends on the following quantities: The insertion loss of the echo path: The reflection coefficient: The transmission coefficient: f (f) Echo Modeling

31
UBC 2006– 31 The reflection coefficient for the spurious echoes: Gauge Changes: Bridged Taps: Echo Modeling

32
UBC 2006– 32 Echo Modeling – Experimental Validation

33
UBC 2006– 33 Echo Modeling – Experimental Validation

34
UBC 2006– 34 Traditional Signal Processing Approach The problem of detecting and resolving signals generated by D sources has been usually addressed assuming the availability of an array of M>D sensors. It is a combined detection/estimation problem: determine the number of sources and then estimate their location in time. In our case the sources are the discontinuities, the signals are the echoes and the location in time is the position of the discontinuity.

35
UBC 2006– 35 Traditional Signal Processing Approach Problem: Here we have only one sensor!!! Let’s try to turn our single sensor case into a multi-sensor one and use MUSIC, ESPRIT, or WSF.

36
UBC 2006– 36 Similar to the multi-sensor problem but now the array manifold depends on the shape of the echo Traditional Signal Processing Approach Problems: The manifold S requires the knowledge of the shape of the echo and that all the echoes have the same shapes. The real and spurious echoes would still be undistinguishable from each other.

37
UBC 2006– 37 Echo Signatures of Discontinuities

38
UBC 2006– 38 Not much literature on Loop Make-Up identification. First papers are recent: S. Galli, D. L. Waring, “Loop Make-up Identification via Single Ended Testing: Beyond Mere Loop Qualification”, IEEE J. Select. Areas Commun., vol. 20, no. 5, June T. Bostoen, P. Boets, M. Zekri, L. Van Biesen, T. Pollet, and D. Rabijns, "Estimation of the Transfer Function of a Subscriber Loop by Means of a One-Port Scattering Parameter Measurement at the Central Office," IEEE J. Select. Areas Commun., vol. 20, no. 5, June 2002.

39
UBC 2006– 39 The identification process is based on analyzing TDR measurements in such a way that the measurements are successively mapped to gradually augmented loop make- up topologies until the error between the measured TDR trace and the simulated TDR waveform of a set of hypothesized loop topologies becomes sufficiently small. S. Galli, K. Kerpez, "Single-Ended Loop Make-Up Identification - Part 1 and 2," IEEE Transactions on Instrumentation and Measurement, vol. 55, no. 2, April A Novel Step-by-Step Maximum Likelihood Technique

40
UBC 2006– 40 A Novel Step-by-Step Maximum Likelihood Technique Rationale: Exploit the “deterministic” nature of the twisted-pair. The echoes from near discontinuities “hide” the echoes from far discontinuities de-embedding 1) Hypothesize all “sensible” topologies and generate corresponding waveform according to echo model 2) Choose topology whose waveform best matches measured data, and identify discontinuity 3) Augment chosen topology using auxiliary topologies (infinite length), generate corresponding waveform, and subtract it from measured data to obtain a de-embedded TDR trace 4) Identify the next discontinuity 5) Go to 2 using de-embedded trace as measured data until last echo is found

41
UBC 2006– 41

42
UBC 2006– 42

43
UBC 2006– 43 Example of Loop ID

44
UBC 2006– 44 Example of Loop ID

45
UBC 2006– 45 Example of Loop ID

46
UBC 2006– 46 Example of Loop ID

47
UBC 2006– 47 Example of Loop ID

48
UBC 2006– 48 CO 5900’ 26 AWG 2000’ 24 AWG 500’ 24 AWG 500’ 24 AWG 500’ 26 AWG Current loop estimate CO 5900’ 26 AWG 2000’ 24 AWG 500’ 24 AWG 500’ 24 AWG 500’ 26 AWG Previous loop estimate State space Reduced- state Viterbi estimation Enhancement: Multiple Estimate Path Search

49
UBC 2006– loops representing the variety at a CO. Loops picked so that 5%, 10%,..., 95% of all loops at the wire center were shorter. Loops include bridged tap, gauge change, etc. Experimental results: EWL error

50
UBC 2006– 50 Experimental results: DSL bit rate error

51
UBC 2006– 51 Simulation results DSM Static spectrum management Different actual pair-to-pair crosstalk coupling functions Presented at the IEEE ICC 2005 conference

52
UBC 2006– 52 Conclusions Crosstalk is not random noise, we can control it Efficient optimization algorithms need: – Crosstalk identification – Loop make-up identification With optimization, bit rates increase several times

53
UBC 2006– 53 Back-Up Slides

54
UBC 2006– 54 Difference in NEXT loss between coupling lengths of 1 kft and 18 kft.

55
UBC 2006– 55 The probability histogram of working lengths measured in the 1983 loop survey.

56
UBC 2006– 56 The probability histogram of total bridged tap lengths, measured in the 1983 loop survey (excluding zero lengths)

57
UBC 2006– 57 Distance from CO in kft (km)% 26-gauge% 24-gauge% 22-gauge % 19- gauge (0) (1.5) (3.0) (4.6) (6.1) (7.6) (9.1) Overall Cable gauge statistics from the 1983 and the Bellcore loop surveys. In the Bellcore loop surveys, only 0.1% of cabling overall was 19 gauge, so 19 gauge is omitted from the table

58
UBC 2006– 58 Broadband Test Head prototype architecture

59
UBC 2006– 59 Schematic diagram of our differential TDR

60
UBC 2006– 60 Measured TDR traces obtained probing a 975 m (3,200 ft) of AWG26 followed by 975 m (3,200 ft) of AWG24. (Left) Echo response to our differential probing TDR (Right) Echo response to conventional unbalanced probing

61
UBC 2006– 61

Similar presentations

© 2016 SlidePlayer.com Inc.

All rights reserved.

Ads by Google