Loop Qualication for VDSL2

MATTIAS ERNELLI
Masters Degree Project
Stockholm, Sweden April 2008
XR-EE-KT 2008:3
Acknowledgements
This master thesis project was carried out at the Fixed Networks department
at Telenor AB in Stockholm 2007. I would like to thank Per

Odling and Per
Ola Borjesson at The Faculty of Engineering LTH, at the University of Lund
for their academic support regarding the theoretical part of the project. Their
comments and input regarding the outline of the report has been very helpful
and made me focus on the more important aspects of my work.
I also want to thank Miguel Berg and Per-Erik Eriksson at the Ericsson signal
processing department for their support and input regarding various standards
related issues.
My colleagues at the Telenor Network department Fredrik Tingsborg and
Thomas Rahm for letting me use the DSL lab and helping me with the equip-
ment.
And nally my wife Karin for her patience during late evenings and never
ending promises that the report will soon be nished.
1
Abstract
Broadband access technology is rapidly developed due to higher bandwidth
demand. One of the driving factors is triple play, where Internet access, IPTV
and IP telephony is delivered over the same shared medium. Digital Subscriber
Line (DSL) technology has been the fastest growing access technology. DSL uses
the local telephone loop between the Central Oce (CO) and the Customer to
carry data trac. The twisted pair copper cable can carry information at a
much higher data rate than the Public Switched Telephone Network (PSTN).
The next generation of mainstream copper access technology will use VDSL2
transmission. VDSL2 uses the same basic modulation as ADSL, ADSL2 and
ADSL2+, which is Discrete Multitone modulation, based on the same coding
and modulation parameters as ADSL2+ but uses frequencies up to 30 MHz.
The capacity of a VDSL2 connection is very dependant on the signal condition
of the local loop and diers between individual pairs in a single binder. Loop
pre-qualication is very important so that an upgrade to VDSL2 is only oered
to those subscribers that will actually achieve a higher data rate, and also that
upgrades is not prohibited for customers that has a potential capacity gain.
The ADSL2/ADSL2+ standard denes a loop diagnostics mode where the
central oce equipment, DSLAM, and customer-premises equipment, DSL Mo-
dem, performs a similar test sequence as the training sequence performed when
the link is activated. The loop characteristic parameters such as attenuation
and noise level is then available for analysis. This masters degree thesis project
evaluates the ADSL2+ built in loop diagnostics function to se if it can be used
as a loop qualication method for high-speed data service such as VDSL2.
2
Contents
Abbreviations and Acronyms . . . . . . . . . . . . . . . . . . . . . . . 6
1 Introduction 7
1.1 Outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.1 Overview to DSL technology and its applications . . . . . 7
1.1.2 Communications theory . . . . . . . . . . . . . . . . . . . 7
1.1.3 Capacity estimation of the local loop . . . . . . . . . . . . 7
1.1.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Overview of DSL technology . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 Digital communication over PSTN . . . . . . . . . . . . . 8
1.2.2 The Digital Subscriber Loop . . . . . . . . . . . . . . . . 8
1.3 Communication systems theory . . . . . . . . . . . . . . . . . . . 9
1.3.1 Digital Communication systems . . . . . . . . . . . . . . . 9
1.3.2 The Additive White Gaussian Noise channel . . . . . . . 10
1.3.3 Probability of error in the AWGN channel . . . . . . . . . 11
1.3.4 Probability of error for M-ary PAM . . . . . . . . . . . . 12
1.3.5 Quadrature Amplitude Modulation . . . . . . . . . . . . . 15
1.3.6 Probability of symbol error in QAM . . . . . . . . . . . . 16
1.3.7 Error performance of QAM . . . . . . . . . . . . . . . . . 16
1.4 Coding and error correction . . . . . . . . . . . . . . . . . . . . . 18
1.4.1 Trellis code . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.4.2 Forward Error Correction . . . . . . . . . . . . . . . . . . 18
1.5 Channel equalisation and Inter Symbol Interference . . . . . . . . 18
1.5.1 Inter Symbol Interference . . . . . . . . . . . . . . . . . . 19
1.5.2 Channel equalisation . . . . . . . . . . . . . . . . . . . . . 19
1.5.3 Decision Feedback Equalization . . . . . . . . . . . . . . . 20
1.6 Discrete Multitone Modulation . . . . . . . . . . . . . . . . . . . 20
1.6.1 DMT Implementation using Fast Fourier Transform . . . 20
1.6.2 The benets of using DMT in DSL transmission . . . . . 21
1.6.3 Cyclic extension and ISI . . . . . . . . . . . . . . . . . . . 24
1.6.4 Time domain equalization . . . . . . . . . . . . . . . . . . 24
2 Modelling the local loop 25
2.1 The local loop distribution network . . . . . . . . . . . . . . . . . 25
2.1.1 The topology of the loop plant . . . . . . . . . . . . . . . 25
2.2 Transmission line characterisation . . . . . . . . . . . . . . . . . . 27
2.2.1 ABCD model of the transmission line . . . . . . . . . . . 27
2.2.2 The Telegraphers equations . . . . . . . . . . . . . . . . . 28
2.3 Distributed RLCG parameters . . . . . . . . . . . . . . . . . . . 31
2.3.1 Measurement Procedure . . . . . . . . . . . . . . . . . . . 32
2.3.2 Cable models . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3.3 The MAR model . . . . . . . . . . . . . . . . . . . . . . . 33
2.3.4 MAR model . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.3.5 BT vs. the MAR model . . . . . . . . . . . . . . . . . . . 34
3
2.4 Noise . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.1 Background noise . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.2 Crosstalk noise . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4.3 FEXT Modelling . . . . . . . . . . . . . . . . . . . . . . . 37
2.4.4 Radio Frequency Ingress . . . . . . . . . . . . . . . . . . . 37
2.4.5 Impulse noise . . . . . . . . . . . . . . . . . . . . . . . . . 38
3 Loop measurement 39
3.1 Channel characterisation . . . . . . . . . . . . . . . . . . . . . . . 39
3.1.1 Single Ended Line Testing . . . . . . . . . . . . . . . . . . 39
3.1.2 Dual Ended Line Testing . . . . . . . . . . . . . . . . . . 39
3.1.3 ADSL2 Training sequence . . . . . . . . . . . . . . . . . . 39
3.1.4 ADSL2+ training sequence . . . . . . . . . . . . . . . . . 41
3.1.5 Loop Diagnostics mode . . . . . . . . . . . . . . . . . . . 41
3.1.6 Loop diagnostic parameters . . . . . . . . . . . . . . . . . 42
3.1.7 Reference measurements . . . . . . . . . . . . . . . . . . . 43
3.1.8 Noise reference measurement . . . . . . . . . . . . . . . . 46
3.1.9 Noise level accuracy . . . . . . . . . . . . . . . . . . . . . 47
3.1.10 Noise level correction . . . . . . . . . . . . . . . . . . . . . 47
3.1.11 Measuring the Modem Background Noise . . . . . . . . . 47
3.1.12 Measuring the line background noise . . . . . . . . . . . . 48
4 Channel capacity estimation 51
4.1 Fitting the Loop channel model . . . . . . . . . . . . . . . . . . . 51
4.1.1 Fitting the cable model . . . . . . . . . . . . . . . . . . . 51
4.1.2 Error analysis of the cable model . . . . . . . . . . . . . . 52
4.1.3 Further analysis of the model error . . . . . . . . . . . . . 54
4.1.4 Results from tting the optimized cable model to simula-
tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.1.5 Correctness of the optimized model . . . . . . . . . . . . . 56
4.1.6 Importance of cable model optimization . . . . . . . . . . 58
4.2 Fitting the noise model . . . . . . . . . . . . . . . . . . . . . . . 58
4.2.1 Validity of the Noise Model . . . . . . . . . . . . . . . . . 60
4.3 Channel capacity estimation . . . . . . . . . . . . . . . . . . . . . 60
4.3.1 Calculating the attainable rate . . . . . . . . . . . . . . . 61
4.3.2 VDSL2 proles . . . . . . . . . . . . . . . . . . . . . . . . 62
4.3.3 VDSL2 band plans . . . . . . . . . . . . . . . . . . . . . . 62
5 Implementation 64
5.1 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.1.1 DELT measurements . . . . . . . . . . . . . . . . . . . . . 64
5.1.2 HLOG/HLIN measurement correction . . . . . . . . . . . 64
5.1.3 QLN measurement correction . . . . . . . . . . . . . . . . 64
5.2 Fitting the loop model . . . . . . . . . . . . . . . . . . . . . . . . 64
5.2.1 The FTW xDSL simulation tool . . . . . . . . . . . . . . 64
5.2.2 Fitting the cable model . . . . . . . . . . . . . . . . . . . 65
5.2.3 Fitting the Noise model . . . . . . . . . . . . . . . . . . . 65
5.3 Calculating the attainable rate . . . . . . . . . . . . . . . . . . . 65
6 Results 67
6.1 Lab measurements . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.1.1 Test set-up . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.1.2 QLN measurement . . . . . . . . . . . . . . . . . . . . . . 67
6.1.3 VDSL2 estimation . . . . . . . . . . . . . . . . . . . . . . 69
6.1.4 VDSL2 measurements . . . . . . . . . . . . . . . . . . . . 69
4
6.1.5 Deviations between real and calculated rate . . . . . . . . 69
6.2 Field measurements . . . . . . . . . . . . . . . . . . . . . . . . . 70
6.2.1 Capacity results . . . . . . . . . . . . . . . . . . . . . . . 70
6.2.2 Model dependency . . . . . . . . . . . . . . . . . . . . . . 71
7 Conclusions 73
7.1 Further work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.1.1 Improving the DELT results . . . . . . . . . . . . . . . . . 73
7.1.2 VDSL2 DELT . . . . . . . . . . . . . . . . . . . . . . . . 73
7.1.3 Measuring background noise . . . . . . . . . . . . . . . . . 74
7.1.4 Estimating the loop topology . . . . . . . . . . . . . . . . 74
5
Abbreviations and Acronyms
2B1Q Coding method used in ISDN, equvivalent to PAM-4 modulation
ADSL Asymmetric Digital Subscriber Line transmission
ADSL2 Next generation ADSL where several optional features is mandatory
ADSL2+ Improvement over ADSL2 by extending the tranmission band to 2.2MHz
AFE Analog Front End, performs gain adjustment and ADC
ATM Asynchronous Transfer Mode
AWGN Additive White Gaussaian Noise, common noise model in digital communica-
tion
BER Bit Error Rate
CO Central Oce, where the telecom operator aggerages all subscriber loops
DELT Dual Ended Line Testing, loop test method which uses active equipment in
booth ends
DFE Decision Feedback Equalizer
DMT Discrete Multitone, a modulation method
DSL Digital Subscriber Line, technology that uses the telephone local loop for data
transmission.
DSLAM Digital Subscriber Line Access Multiplexer, equipment used to connect multiple
lines in the CO.
FDM Frequency Division Multiplexing, transmission scheme to provide channel sep-
aration by using dierent frequencies
FEC Forward Error Correction
FEXT Far-End Crosstalk.
FFT Fast Fourier Transform, an eecient algorithm to perform a Discrete Fourier
Transform.
HDSL High bitrate Digital Subscriber Line.
IFFT Inverse Fast Fourier Transform.
IL Insertion Loss, the increased loss in a transmission system by inserting a cable
or device in the signal path.
INP Impulse Noise Protection.
IPTV Digital TV transmission using IP networks.
ISDN Integrated Services Digital Network.
ISI Inter Symbol Interference.
NEXT Near-End Crosstalk.
OFDM Orthogonal frequency division multiplexing
PAM Pulse Amplitude Modulation
POTS Plain Old Telephone Service, basic analog voice grade telephone service.
PSD Power Spectral Density
PSTN Public Switched Telephone Network
QAM Quadrature Amplitude Modulation
QLN Quiet Line Noise
RFI Radio Frequency Ingress
RS Reed Solomon error correction code
SNR Signal to Noise Ratio
UTP Unshielded Twisted Pair, the type of cable used in subscriber loops
VDSL Very high speed Digital Subscriber Line, uses booth single carrier QAM and
DMT
VDSL2 Next generation VDSL technology, uses DMT and FDM.
6
Chapter 1
Introduction
1.1 Outline of the thesis
1.1.1 Overview to DSL technology and its applications
The report is arranged as follows. First this introduction section that explains
the concept of DSL and gives a short background to the history and evolution
of DSL technology.
1.1.2 Communications theory
Then follows an introduction to communication systems theory covering those
topics that is relevant to Discrete Multitone (DMT) modulation and the prin-
ciple of DMT modulation is explained.
1.1.3 Capacity estimation of the local loop
The main topic of this project, capacity estimation of the twisted pair local loop
for VDSL2 is studied, which covers
The electrical characteristics of the UTP transmission line.
Models for simulating signal transmission over UTP.
The various sources of disturbance and crosstalk.
The established model for FEXT, far end crosstalk.
Channel characterisation through loop diagnostics.
Qualication of the loop diagnostics result
Fitting the UTP channel model to measurements
Stability and correctness of tted model
Capacity and rate calculation of the modelled channel
1.1.4 Results
Then the established method to estimate the capacity of the local loop is tested
in a lab setup. Finally the results from example loops from the Telenor network
is presented.
7
1.2 Overview of DSL technology
1.2.1 Digital communication over PSTN
Digital Subscriber Line, DSL, technology is an access technology that uses the
higher capacity of the twisted pair (TP) local loop that connects the Subscriber
to the PSTN (Public Switched Telephone Network). The PSTN, originally
designed for voice communication, can only transmit signals in a narrow fre-
quency band, wide enough to reproduce spoken language with minor impair-
ments. When frequency division multiplexing was introduced on long distance
interconnects, each channel was band limited to a frequency range of 300 to
3400 Hz. Later when the PSTN went digital, the sampling frequency of 8 kHz,
and 8-bits of resolution established the upper limit for data communication
over PSTN to 64 kbit/s. During the massive growth of Internet users during
the 90s, analogue modem technology achieved the capacity of 56 kbit/s in the
downstream direction, utilising the all-digital connection between the Internet
Service Provider (ISP) modem pool and Central Oce (CO) equipment.
1.2.2 The Digital Subscriber Loop
By installing communication equipment in the central oce, the higher band-
width and signal to noise ratio of the local loop compared to the limited capacity
of the PSTN, can be utilised. The local loop becomes a Digital Subscriber Loop.
Integrated Services Digital Network
Basic Rate Integrated Services Digital Network (ISDN) uses two binary, one
quaternary (2B1Q) modulation with a symbol rate of 80 kbaud providing a
symmetric line rate of 160 kbit/s that carries two 64 kbit B-channels and one
16 kbit D-channel. Duplex communication is achieved by echo cancellation.
High bitrate Digital Subscriber Line
High bitrate Digital Subscriber Line (HDSL) uses 2B1Q or Carrierless Ampli-
tude Phase (CAP). The data rate when using 2B1Q modulation is either 784,
1168 or 2320 kbit/s depending on if 1, 2 or 3 lines are bonded together. Sin-
gle pair HDSL (SHDSL) provides symmetric data rate from 192 to 2304 kbit/s
using one pair or up two 4608 kbit/s using two pairs.
Asymmetric Digital Subscriber Line
Asymmetric Digital Subscriber Line (ADSL) was the rst mainstream broad-
band access technology deployed. ADSL uses Discrete Multitone (DMT) mod-
ulation with 256 carriers, each using a bandwidth of 4.3125 kHz thus operating
up to 1104 kHz. Each carrier uses Quadrature Amplitude Modulation (QAM)
and can transmit from to 1 to 15 bits of data per symbol (8 bits/carrier in early
implementations). The symbol rate is 4000 baud/s, which provides a maximum
uncoded line rate from 7.168 Mbit/s to 13.4 Mbit/s, if frequency division multi-
plexing is used for duplexing. The rst 32 tones are used for upstream data and
POTS. Thus providing 25 upstream carriers capable of transmitting 800 kbit to
1.5 Mbit uncoded line rate. Trellis coded modulation is optional.
ADSL uses Asynchronous Transfer Mode (ATM) as frame bearer. Various
data services is mapped in to ATM Private Virtual Channels (PVCs). The
overhead by using the ATM frame structure reduces the eective payload by
about 10%.
8
1011 1011
Transmitter Channel Receiver
Detects message Distorts signal
Encodes message
Message
Signal Signal
Message
Noise
Figure 1.1: Basic communication system
ADSL2
ADSL2 is an enhancement over ADSL. In ADSL2 a QAM-constellation size
of 15 bits and trellis coding is mandatory. ADSL2 also supports an increased
upstream band using up to tone 64, thus extending the upstream data rate to
3 Mbit/s on the expense of a reduced downstream rate.
ADSL2+
ADSL2+ improves the data rate even more by using 512 subcarriers. The
transmission band is extended up to 2.208 MHz and the maximum downstream
data rate is 24 Mbit/s.
VDSL2
VDSL2 is the latest broadband access technology. VDSL2 extends the transmis-
sion band up to 30 MHz. For DSL transmission on subscriber loops 8.5 MHz,
12 MHz or 17 MHz proles are provided. Where the 8.5 MHz and 12 MHz is
targeted for deployment in the Central Oce. VDSL2 uses the same modulation
as ADSL2/ADSL2+. For the proles up to 12 MHz, transmission is divided in
two downstream bands, DS1 and DS2 and up to three upstream bands US0,
US1 and US2. The bidirectional data rate on VDSL2 ranges from 50 MBit/s to
150 Mbit/s depending on selected transmission prole and the conditions on the
local loop. The later, which is the most important factor regarding successful
VDSL2 deployment is what will be the main focus on in this project.
1.3 Communication systems theory
1.3.1 Digital Communication systems
The basic principle of a digital communication system is to transmit a message
from a source to a user over a physical medium, a channel. By representing the
information being transmitted with a discrete set of symbols, e.g. digitally coded
information, a communications system can be designed for data transmission
with a very low probability of error. Digital Communication theory covers
the problem of achieving error free communication as close as possible to the
theoretical capacity of a given channel. Figure 1.1 shows the basic scenario for
a communication system.
9
A digitally encoded message is transmitted over some sort of channel. The
channel can be an electric wire or free space if the signal is carried by radio
waves. The message to be sent is mapped to a set of symbols. Each symbol
can transmit a certain number of data bits depending on the size of the symbol
set. Given a set of M symbols the number of bits represented by each symbol
is then
N = log
2
M (1.1)
The data bits in the message is rearranged in chunks to t the number
of bits a given symbol-set can represent and is transmitted sequentially. For
example the 2B1Q coding maps two data bits to a symbol set of 4 symbols.
The number of symbols transmitted in a given time frame is called baud rate
given as B = baud/s. Since each symbol is associated to N number of data bits,
the transmission rate becomes B N bits/second.
The channel always distorts the signal between the transmitter and the re-
ceiver in some way. For wire line channels the distortion mainly consists of
frequency dependant attenuation, caused by resistive loss in the wire and phase
dispersion. In radio transmission the channel distortion consists of at or fre-
quency selective fading and dispersion caused by multipath propagation. The
Channel distortion is expressed by its transfer function H(f). Apart from chan-
nel distortion, a communication system is also aected by noise. In most cir-
cumstances the channel noise can be modelled as white and Gaussian. In the
case of thermal noise in the receiver circuit its Power Spectral Density is given
by:
S
n
(f) =
hf
2(e
hf
kT
1)
[W/Hz] (1.2)
Where Plancks constant h = 1.06 10
34
[J/s], Boltzmanns constant k =
1.38 10
23
[J/k] and T denotes the temperature in Kelvin. S
n
(f) can be con-
sidered constant with respect to frequency for most practical communication
systems [1], and is then expressed as
N
0
2
=
kT
2
[W/Hz] (1.3)
Other sources of noise will also aect the performance of a communication
system. For DSL communication the dominant noise source is crosstalk from
adjacent copper pairs transmitting signals in the same frequency range as the
signal being transmitted. Since multiple disturbers contribute to the crosstalk
in a single wire, it can be modelled as Gaussian noise. We will study crosstalk
noise further when the capacity of the UTP local loop is analysed.
1.3.2 The Additive White Gaussian Noise channel
The Additive White Gaussian Noise (AWGN) channel is a simplied mathe-
matical model of a communication channel in which the transmitted signal is
only aected by white noise. Assuming an optimal receiver, the received signal
r(t) is given by
r(t) = s
m
(t) +n(t), 0 t T (1.4)
Where s
m
(t) is the transmitted signal waveform, being one of M possible
waveforms from the set of symbols. n(t) is white noise added to the transmit-
ted signal in the channel and the receiver. In the case of an N dimensional
symbol set, using a linear combination of N orthogonal signal waveforms, the
10
matched lter or correlation-type demodulator in the receiver converts the re-
ceived waveform r(t) into an N dimensional vector r(t). for instance QAM uses
two orthogonal carriers, each modulated by M-ary Pulse Amplitude Modulation
(PAM) signals in which case the detected signal is a two dimensional vector rep-
resenting M
QAM
= m
1
m
2
possible symbols.
The AWGN channel is a sucient model for DMT transmission as we will see
later, since each narrowband DMT channel is closely approximated with the
AWGN model.
1.3.3 Probability of error in the AWGN channel
With the presence of noise on a communication channel, the probability of
error becomes non-zero. The signal received over an AWGN channel is given
by Equation (1.4) that consists of the transmitted symbol waveform and an
additive noise component. For a system with binary modulation, e.g. antipodal
signalling with M = 2, the signal consists of s
1
= g
T
(t) and s
2
= g
T
(t),
where g
T
(t) is an arbitrary pulse with energy
s
. Which in the case of antipodal
signalling is the same as energy per bit
b
. The received signal from the matched
lter or correlation-type demodulator is then
r =
_

s
+n if s
1
is transmitted

s
+n if s
2
is transmitted
(1.5)
The optimum detector, given that both signals are equiprobable is then
s =
_
s
1
if r 0
s
2
if r < 0
(1.6)
The probability of error is the probability that the noise component is greater
than the signal component and is of opposite sign. If for example s
1
was trans-
mitted. The probability of error is the probability that r < 0 in which case
the detector assumes that s
2
was transmitted, which happens if n <

s
. In
the case of white Gaussian noise, n is the additive noise component which has
0 mean and variance
2
n
= N
0
/2 where N
0
/2 is given in (1.3). Although N
0
/2
is given [W/Hz] it is assumed that the noise and signal waveform occupies the
same bandwidth. Then the signal to noise ratio (SNR) is expressed as
SNR =

s
N
0
/2
=
2
s
N
0
(1.7)
where
s
denotes the average energy of the transmitted signal in [W/Hz].
Thus the probability of symbol error and bit error becomes independent of the
bandwidth of the channel and signal, this is the case in the AWGN channel
model. The PDF of White Gaussian Noise with variance
2
n
is
f(n) =
1
_
2
2
n

n
2
2
2
n
(1.8)
For the received signal r the conditional PDF then becomes
f(r|s
1
) =
1

N
0
e

(r

s
)
2
N
0
(1.9)
f(r|s
2
) =
1

N
0
e

(r+

s
)
2
N
0
(1.10)
Where (1.9) is the PDF of the received signal centred around

s
and the
PDF given by (1.9) is centred around

s
, which is also illustrated in Figure
1.2.
11
r
0

s
Figure 1.2: Conditional PDFs of antipodal signals
The area of (1.9) in the interval where r < 0 is equal to the symbol error
probability for symbols transmitted using s
1
. As well as the area of (1.9) where
r 0 is equal to the error probability for symbols transmitted using s
2
. The
probability of error when transmitting symbol s
1
is then
P(e|s
1
) =
_
0

N
0
e

(r

s
)
2
N
0
dr (1.11)
=
1

2
_

2
s
N
0
e

x
2
2
dx (1.12)
= Q
__
2
s
N
0
_
(1.13)
Where Q(x) is related to the Gaussian error function by
Q(x) = 1 (x) (1.14)
= 1
1

2
_
x

u
2
2
du (1.15)
=
1

2
_

x
e

u
2
2
du (1.16)
Where (1.16) is identied in (1.12). There exists no closed form for Q(x)
so the integral has to be solved numerically. Q(x) is equal to the probability
that the outcome of a random variable with normal distribution, zero mean and
variance 1, is larger than x.
Q(x) = P(X > x), X N(0, 1) (1.17)
1.3.4 Probability of error for M-ary PAM
In M-ary PAM, M dierent levels of the signal waveform is used to represent a
symbol. When the channel transfer function is unity, aected by white Gaussian
noise and all M symbols are equally probable, then the optimal constellation
for M-ary PAM is shown in Figure 1.3.
The signal points is evenly spread with distance d in between, symmetrically
around 0. The average energy per symbol, given all symbols is equiprobable is
12
0

av

av
d d d d d d d
Figure 1.3: Signal constellation for M-ary PAM, (M = 8)

av
=
1
M
M

m=1

m
=

g
M
M

m=1
A
2
m
(1.18)
Where
g
is the energy of the signal waveform g
T
(t) and A
m
is the amplitude
scale of the signal waveform used for each symbol. Each symbol is transmitted
using the signal
s
m
(t) = A
m
g
T
(t) (1.19)
With the energy

m
= A
2
m

g
(1.20)
When the signal amplitude is symmetric around 0, A
m
is then
A
m
= (2mM 1), m = 1, 2, ..., M (1.21)
The energy per symbol given by (1.18) becomes

av
=

g
M
M

m=1
(2mM 1)
2
=
g
(M
2
1)/3 (1.22)
The detection of M-ary PAM symbols consists of dividing the received sig-
nal level using M 1 thresholds, placed in the midpoint between two symbol
amplitude levels, as shown in Figure 1.4.
In the detector, each threshold interval T
i1
r T
i
maps to a correspond-
ing symbol s
m
for symbols m = 2, ..., M 1. Except for the symbols at s
1
and
s
M
, e.g. the left and rightmost symbols in the Figure 1.4. For those symbols
the detector maps
s =
_
s
M
if r T
M1
s
1
if r < T
1
(1.23)
Where the thresholds T
m
, m = 1, 2, ..., M is equal to
T
m
= (A
m
+ 1)
2

g
= (2mM)
2

g
(1.24)
The probability of error for symbols s
m
, m = 2, 3, ..., M 1 is then twice as
high as for binary PAM with the same distance between the symbol points, as
can be seen in Figure 1.5
13
0
s
3
s
2
s
1
s
4
s
5
s
6
s
7
s
8
T
3
T
2
T
1
T
4
T
5
T
6
T
7
Figure 1.4: Detector thresholds for M-ary PAM, (M = 8)
0
s
3
s
2
s
1
s
4
s
5
s
6
s
7
s
8
T
3
T
2
T
1
T
4
T
5
T
6
T
7
Figure 1.5: Conditional PDF of received signal for s
3
, error area marked in grey
14
k bits
k/2 bits
PAM
PAM
k/2 bits
DET
DET
s
n
cos(
c
t)
sin(
c
t)
cos(
c
t)
sin(
c
t)
k bits
k/2 bits
k/2 bits
Figure 1.6: Principle of a QAM transmission system
The distance between two signal points d = 2

g
gives the symbol error
probability as
P
e,s
m
= 1
_

g
/N
0

g
/N
0
1

r
2
2
dr, m = 2, ..., M 1 (1.25)
=
_

g
/N
0

r
2
2
dr +
_

g
/N
0
1

r
2
2
dr (1.26)
= 2
_

g
/N
0
1

r
2
2
dr (1.27)
= 2Q
__
2
g
N
0
_
(1.28)
Since the detection of the two symbols s
1
, s
M
has the error probability given
by (1.13), the symbol error probability for M-ary PAM becomes
P
e
=
(M 2)
M
2Q
__
2
g
N
0
_
+
2
M
Q
__
2
g
N
0
_
=
2(M 1)
M
Q
__
2
g
N
0
_
(1.29)
With SNR expressed using average energy per symbol, as given in (1.22),
the probability of symbol error is
P
e,PAM
=
2(M 1)
M
Q
_
3
av
(M
2
1)N
0
_
(1.30)
1.3.5 Quadrature Amplitude Modulation
QAM is a modulation method where usually two baseband PAM signals mod-
ulates two sinusoidal carrier signals. The two carrier signals have the same
frequency f
c
but are separated in phase by 90 degrees. In the detector, the
received signal is correlated with two signals with the same frequency as the
carrier signal in the modulator, the two signals in the detector is also separated
in phase by 90 degrees. Figure 1.6 shows the principle of a QAM transmission
system.
The two QAM modulated signals represents a 2 dimensional symbol space
divided in M regions. The smallest constellation is QAM-4 where each symbol
is located in one of the 4 quadrants of the symbol space. A QAM constellation
15
QAM-4
even
QAM-8
odd
QAM-16
even
QAM-32
odd
Figure 1.7: QAM constellations used in ADSL/VDSL2
can be either odd or even. Even constellations consists of two orthogonal M-ary
PAM signals and odd constellations consists of one M-ary and one orthogonal
N-ary PAM signal. Even constellations are usually always square. Odd con-
stellations can be rectangular, or symmetrical. Figure 1.7 shows some examples
of the QAM constellations used in DMT ADSL/VDSL, that is symmetrical for
both odd and even sizes. For rectangular constellations, an M-ary QAM symbol
is represented by the signal
s(t) = X
n
g
T
(t)cos(2f
c
t) Y
n
g
T
(t)sin(2f
c
t), 0 t T (1.31)
Where X
n
and Y
n
maps to one of the points in the M-ary QAM constellation
and g
T
is the transmitter signal waveform. f
c
is the center frequency of the
carrier.
1.3.6 Probability of symbol error in QAM
Since the two carriers in a QAM system is orthogonal, the transmission of two
M-ary PAM signals over an AWGN channel can be seen as two independent
uses of the channel. Both signals is aected by the channel transfer function
H(f) and additive noise N
0
/2. The probability of a QAM symbol error is the
probability that either one of the two detected M-ary PAM symbols is erroneous.
For square constellations, the probability of a symbol error is the same for
both M-ary PAM signals. With P
e
for M-ary PAM given by Equation (1.30),
the probability for a QAM-M symbol error using a square constellation becomes
P
e,QAM
even
= 2 P
e,PAM
=
4(

M 1)

M
Q
_
3
a
v
(M 1)N
0
_
(1.32)
and for odd constellations it is bounded by
P
e,QAM
odd
< 4(1
3
2M
) Q(

6
av
(2M 1)N
0
) (1.33)
1.3.7 Error performance of QAM
The theoretical maximum capacity C in bits per second, of a communication
system transmitting information over an AWGN channel is given by
16
0 5 10 15 20 25 30 35 40 45 50 55 60
8
7
6
5
4
3
2
1
0
SNR [dB]
l
o
g
1
0

P
e
QAM4
QAM16
QAM64
QAM256
QAM1K
QAM4K
QAM16K
Student Version of MATLAB
Figure 1.8: Symbol error rate for even QAM as a function of SNR
Table 1.1: SNR requirement for symbol error rate 10
7
QAM SNR Gap
size [dB] [dB]
4 14.5 9.76
16 21.6 9.88
64 27.9 9.92
256 34.0 9.94
1024 40.0 9.95
4-K 46.1 9.95
16-K 52.1 9.96
C = W log
2
(1 +SNR) = W log
2
(1 +
P
N
0
W
)[bits/s] (1.34)
Where W is the bandwidth of the signal being transmitted in [Hz], and
detected using an optimal detector. This is often referred to as the Shannon
capacity. Equation (1.34) can also be expressed in [bits/Hz] if the received
signal and noise level varies with frequency
b(f) = log
2
(1 +
S(f)
N(f)
) = log
2
(1 +SNR(f))[bits/Hz] (1.35)
The channel capacity when using QAM transmission without coding depends
on the error rate accepted. Figure 1.8 shows the symbol error rate as a function
of SNR.
The target error rate for ADSL/VDSL is 10
7
according to the standard.
Table 1.1 shows the required SNR for dierent QAM constellations to achieve
that error rate. The SNR Gap is also given which is the dierence in dB between
the channel capacity for a given SNR according to (1.35) and the required SNR
17
for QAM, which asymptotically goes to 9.96 dB. SNR Gap is usually denoted
using .
When evaluating the performance of a transmission system, the term errored
second is usually used. An errored second is one or more errors within one
second of channel use. If a symbol error occurs, one or more bits of information
will be corrupt, but it will be counted as one errored second. The bit error
rate is of interest when block coding is used to improve the performance of the
communication system.
1.4 Coding and error correction
1.4.1 Trellis code
Trellis code is used in ADSL2+/VDSL2 to improve the coding eciency without
increasing the bandwidth. Trellis codes uses detection of several symbols trans-
mitted sequentially or in parallel (by using larger constellations) to improve the
detection. The data rate is increased by adding a convolutional code. By using
a maximum likelihood decoder, e.g. a viterbi decoder, the received sequence is
compared to all possible sequences and the most probable is selected.
A 4-D 16-state Trellis Code is used in ADSL2, ADSL2+ and VDSL2 to
obtain an asymptotic coding gain of about 4.5 dB [4]. That is, for a given SNR
the coding Gap can be reduced from 10 dB to about 5.5 dB, and still maintain
a symbol error probability of 10
7
.
1.4.2 Forward Error Correction
Apart from improving the coding gain by using Trellis coded modulation, an
outer Reed Solomon (RS) Forward Error Corretion (FEC) block code is also used
in ADSL2+/VDSL2. The outer block code is usually combined with a block
interleaver to dilute the data in one coded block over several DMT symbols.
Thus make the transmission more immune to wide band noise impulses, which
occur intermittently and contains much more energy than the stationary noise
present on the channel. Dierent combinations of RS parity check words and
interleave depth provide an INP (Impulse Noise Protection) value that measures
the number of lost DMT symbols that can be fully recovered.
FEC coding also provides an additional coding gain to the trellis code. The
RS code in ADSL2+/VDSL2 uses GF(256) with the number of redundancy
octets between 0 and 16. The error correction capability, without using erasures,
is R/2, where R is the number of redundancy octets per FEC codeword. The
size of an RS codeword in ADSL2/ADSL2+ and VDSL2 ranges from 32 to 255.
The largest protection is given by the FEC conguration (n, k) = (32, 16) which
has an overhead of 50%. It can correct 8 erroneous octets out of 32 transmitted
octets.
1.5 Channel equalisation and Inter Symbol In-
terference
The AWGN channel is a simplied model of a transmission system. Very few real
communication channels share the properties of an AWGN channel, unless only
a fraction of the channel capacity is utilised. All practical channels introduces
some sort of distortion that aects the shape of the signal g
T
(t) being trans-
mitted. If the channel transfer function C(f) is not considered in the detector,
not all of the available energy in the received signal r(t) will be recovered, thus
18
0 1 2 3 4 5 6
1
0.5
0
0.5
1
0 1 2 3 4 5 6
1
0.5
0
0.5
1
0 1 2 3 4 5 6
1
0.5
0
0.5
1
0 1 2 3 4 5 6
1
0.5
0
0.5
1
Student Version of MATLAB
Figure 1.9: Eects of ISI when transmission rate increases
optimal performance will not be the case. Apart from suboptimal detection,
another eect from the channel transfer function is Inter Symbol Interference.
1.5.1 Inter Symbol Interference
Inter Symbol Interference (ISI) is caused by dierent propagation time for dif-
ferent frequencies e.g. the phase delay is not a linear function of the frequency.
Both causes dispersion in which the energy of the transmitted symbol is spread
out in the time domain. If the duration of the transmitted signal g
T
(t) is T
S
,
then the duration of the received signal r(t) is T
S
+T
D
. The time dispersion T
D
is the duration of the impulse from the channel. If T
D
0 the signal energy in
the symbol received at t = T
0
will contain energy from the symbol transmitted
at t = T
0
T
S
, and thus reduce the average signal energy in the demodulated
signal and the result is lower average SNR than a channel without ISI would
have. Figure 1.9 shows the eects of increasing ISI for 6 consecutive symbols
when the transmission rate is increased.
ISI is mitigated by either using a channel equaliser that shortens the channel
impulse response or by the use of Decision Feedback Equalisation. Both methods
use an estimation of the channel transfer function that is usually determined by
analysing a known training sequence sent by the transmitter prior start of data
transmission.
1.5.2 Channel equalisation
With the estimated channel transfer function given as

G(f) a simple channel
equalisation is performed by using a Zero Forcing Equalizer (ZFE). The ideal
ZFE is simply the inverse of the channel transfer function
F(f) =
1

G(f)
(1.36)
19
The drawback of the ZFE is that noise is not accounted for. The ZFE lter
may amplify noise to such an extent that the error performance is degraded. The
performance is much better when using a minimum mean square error (MMSE)
Equalizer that optimizes the receiver lter based on error performance.
1.5.3 Decision Feedback Equalization
Decision Feedback Equalization (DFE) is a non-linear ISI mitigation method
where the residual of the previously transmitted symbol is subtracted from the
currently received signal. DFE requires that the receiver knows the channel
impulse response from some training sequence. The drawback of DFE is that if
an incorrectly detected symbol is fed back into the detector, the performance will
be severely degraded and the detector can lock up and generate long sequences
of incorrect decisions. The Tomlinson-Harashima precoding applies the decision
feedback in the transmitter by feeding back the transmitted symbol through an
inverse channel response function, thus transmitting an ISI cancellation signal.
The performance is then improved over the ZFE since noise is not enhanced,
but if the channel transfer function contains zeroes, the inverse channel function
may be unstable.
1.6 Discrete Multitone Modulation
Much of the problems regarding channel equalisation, ISI, and ecient modu-
lation when the SNR varies with frequency, is solved by using DMT. The basic
concept of DMT is to divide the transmission spectrum into N equally sized
channels occupying the same bandwidth f. DMT is basically the same as
orthogonal frequency division multiplexing (OFDM) with the dierence that in
DMT each subchannel uses a QAM constellation size that is adapted for the
SNR condition for that particular channel. In conventional OFDM all subchan-
nels usually uses the same modulation parameters.
The generic multi carrier modulator consists of N orthogonal sub-carriers,
each separated by f. Mathematically, a set of functions represents an or-
thonormal basis if
_

m
(t)
n
(t)dt =
mn
, where
mn
=
_
1 m = n
0 m = n
(1.37)
This is satised by the basis function

n
(t) = e
j2nft
(1.38)
This is also the same principle as Frequency Division Multiplexing, although
applied within a single channel.
1.6.1 DMT Implementation using Fast Fourier Transform
DMT and OFDM is usually implemented using Inverse Fast Fourier Transform
(IFFT) in the transmitter and Fast Fourier Transform (FFT) in the receiver.
FFT/IFFT is a widely studied algorithm for which various ecient implemen-
tations exists both in software and hardware designs. With the basis function
for each carrier given by Equation (1.38), modulation is performed by mapping
binary data onto QAM constellations represented by a complex number. Each
one of the N subcarriers uses a QAM constellation determined by the SNR for
that subcarrier, based on a target error rate. The modulated subcarrier is then
given by
20

n
(t) = (X
n
+jY
n
) e
j2nft
(1.39)
Where X is the real part of the constellation and Y the imaginary part.
X and Y maps into a QAM constellation as shown in Figure 1.7. With the
modulated carrier given by (1.39) a DMT modulator can be implemented using
IFFT. In ADSL and VDSL2, the modulation of one DMT symbol carrying N
carriers is dened as
x
n
=
2N1

i=0
e
j2
ni
2N
Z
i
, for n = 0 to 2N 1 (1.40)
Where Z
i
, i = 0, ..., N carries the QAM constellation mapped onto each
carrier, and
Z
i
= conj(Z
2Ni
), for i = N + 1 to 2N 1
Z
0
= 0 and Z
N
being real. The result x
n
is a sequence of real values that can
be transmitted over an analogue channel using a digital to analogue converter.
In the receiver, after the analogue to digital converter, the sequence is reversed
and the complex values Z
i
, i = 0, ..., N is reconstructed by the FFT transform
and the encoded data is detected using N parallel QAM detectors.
1.6.2 The benets of using DMT in DSL transmission
DMT enables the possibility to adjust the modulation and data rate to a channel
where the SNR varies with frequency. On long telephone loops the dierence in
attenuation between the low frequency part of the signal and the high frequency
part can dier as much as 90 dB. Also the noise, which mainly consists of
crosstalk from adjacent lines, changes with frequency. The result is a channel
where the capacity, expressed in bits/Hz bandwidth changes with frequency.
Figure 1.10 shows the attenuation, received signal, receiver noise, and resulting
SNR for a typical DSL channel. In the lower frequency band, the SNR is limited
by the crosstalk from other disturbers. In the upper part the attenuation of the
loop brings the received signal below the noise oor.
Waterll bit allocation
In most communication systems, transmitter power is a limiting factor. Also,
the Power Spectral Density (PSD) of the transmitted signal may not exceed the
limits determined by the transmitter PSD mask. To eciently utilise a channel
given those constraints, the energy has to be distributed over all subcarriers
in an optimum way. The number of bits subcarrier n can transmit, per Hz of
bandwidth, with a given signal to noise ratio (SNR) and gap is
b
n
=
1
2
log
2
(1 +
SNR
n

) (1.41)
With QAM modulation, each subchannel transmits two orthogonal, inde-
pendently modulated signals, thus the capacity per subchannel is
b
n
= log
2
(1 +
SNR
n

) (1.42)
Given the transmitted energy
n
, the receiver noise PSD
2
n
, the channel
gain g
n
, usually between 0 and 1 since all passive channels attenuate the signal,
the number of bits each subchannel can carry is
21
0 276 552 828 1104 1380 1656 1932 2208
160
140
120
100
80
60
40
20
Frequency [kHz]
[
d
B
]
|H(f)|
2
S(f)
N(f)
0 276 552 828 1104 1380 1656 1932 2208
0
10
20
30
40
50
60
S
N
R
(
f
)

[
d
B
]
Frequency [kHz]
Student Version of MATLAB
Figure 1.10: Characteristics of 3 km long DSL loop
b
n
= log
2
(1 +

n
g
n

2
n
)[bits] (1.43)
and the total number of bits per transmitted symbol, the sum of all N
subchannels is
b =
N
n=1
b
n
[bits] (1.44)
To maximize the expression in Equation (1.44), given an output power con-
straint
N
n=1

n
, the waterll algorithm solves that problem by maintaining
the relation

n
+

2
n
g
n
= K (1.45)
Where the constant level K is the sum of /g
n
and
n
. Figure 1.11 illus-
trates the waterll analogy, where the level K represents the waterline if the
available energy was poured as water into the subchannels. The depth of
each subchannel is the inverse SNR scaled by the coding gap. Where high SNR
subchannels will receive more energy than low SNR subchannels. Subchannels
where SNR minus coding Gap is negative, will not receive any energy at all.
The Waterll algorithm iteratively removes unusable subchannels. The result
is that energy is distributed between the subchannels capable of transmitting
information, in a way that maximizes the capacity.
Usually the number of bits per subchannel given by Equation (1.42) requires
rounding since the encoder/decoder can only work with xed sized constella-
tions. The integer constraint on the number of bits per subchannel results in a
saw tooth shaped energy level for a channel with monotonically decreasing SNR,
which is illustrated in Figure 1.12. Therefore the transmitter PSD mask allows
for the extra energy required when rounding towards larger bit constellations
22
1 2 3 4 5 6 7 8 9
140
120
100
80
60
40
20
10 log
10
|g
n
|
2
10 log
10

2
n
1 2 3 4 5 6 7 8
0
1
2
3
4
5
x 10
7

n
2
/ g
n

n
Student Version of MATLAB
Figure 1.11: Example of waterll bit allocation
0 5 10 15 20 25 30
0
10
20
30
40
50
60
b
i
t
s
,
[
d
B
]
10 log
10
SNR
n
b
n
0 5 10 15 20 25 30
10
8
6
4
2
0
2
[
d
B
]
10 log
10

n
Student Version of MATLAB
Figure 1.12: Monotonically decreasing SNR and resulting Energy per subcarrier
23
0 10 20 30 40 50
5
0
5
10
15
20
x 10
3
I
m
p
u
l
s
e

R
e
s
p
o
n
s
e
t [s]
Student Version of MATLAB
Figure 1.13: Impulse response of a 2 km 0.5 mm cable, simulated using the
PE05 mar cable model
1.6.3 Cyclic extension and ISI
DMT uses a low symbol rate, since each DMT symbol carries N parallel sub-
channels simultaneously. ISI can then be mitigated by introducing a guard pe-
riod between each transmitted symbol. In OFDM and DMT the guard period
consists of a cyclic prex. The transmitted symbol consists of the 2N 1 sam-
ples from the IFFT transform prepended with the last K samples. Where K in
ADSL is 4/642N. The symbol rate in ADSL2 are 4000 data symbols per second.
The transmitter sends one sync symbol after 68 data symbols, thus requiring
the symbol rate to be increased to 69/68 4000 = 4059 symbols per second. The
cyclic prex causes the transmitter to use a sampling rate that is (1+4/64) mul-
tiplied with the symbol rate times twice the number of subcarriers. This causes
the subcarriers to be separated by f = (1 +4/64) 4000 69/68 = 4.3125kHz.
The cyclic prex also causes the transmitted symbol to be periodic so the re-
ceiver FFT can reconstruct the original constellation points Z
i
, i = 0, ..., N.
1.6.4 Time domain equalization
The cyclic prex reduces the impact of the channel impulse response from the
previous transmitted symbol. For ADSL2+ the cyclic prex introduced can
handle impulse responses with a duration of (4/64) 2N = 64 samples, which
corresponds to 64/f
s
= 64/2.208 10
6
= 30s.
The simulated impulse response of a 2 km 0.5 mm twisted pair copper cable
is shown in Figure 1.13. For longer loops the duration of the impulse response
from the loop may exceed the guard interval introduced by the cyclic prex.
The solution is to use a Time Domain Equalizer, TEQ that shortens the
impulse response of the channel to t within the guard interval.
24
Chapter 2
Modelling the local loop
The upper capacity limit of a channel is determined by its signal to noise ra-
tio over all frequencies used for communication and its given by the Shannon
capacity
C =
_
f
2
f
1
log
2
(1 +
S(f)
N(f)
)df (2.1)
Estimating the capacity of the twisted pair local loop is then equal to deter-
mine the signal attenuation over all frequencies used in a modulation scheme as
well as estimating the noise power over the same frequency range.
2.1 The local loop distribution network
Each telephone subscriber is connected to the Central Oce by a twisted pair
copper cable. The use of a twisted pair instead of a single wire with a common
ground was early discovered to give a better performance when it comes to
noise immunity. By injecting the signal into the cable using an unbalanced
to balanced circuit, and receiving the dierential signal and convert it from
balanced to unbalanced, common mode induced disturbances and crosstalk is
cancelled out. Figure 2.1 shows the principle of a twisted pair transmission line.
The upper pair shows how common mode noise ingress is cancelled by the use of
dierential signalling. The lower pair shows how continous crosstalk coupling is
cancelled out by the alternating polarity due to the twists. The level of crosstalk
immunity between individual pairs in a multi pair binder is very dependent on
the twist ratio. In the optimum case, all pairs should have unique twist ratios,
but that is not possible for practical reasons. The twisted pair copper cable
usually consists of polyethylene insulated copper wire of approximately 0.5 mm
diameter.
2.1.1 The topology of the loop plant
The local loop plant topology is shown in Figure 2.2. The central oce in a
dense populated area can serve 10,000 to 20,000 customers. The distribution
of the subscriber loop is carried in segments of feeder cable. The feeder cable
closest to the CO carries hundreds or thousands of pairs, while the more remote
segments consist of 100 pair binders or less. For long loops, thicker cable is
used on the more distant segments to keep loop resistance under a certain limit.
Therefore a mixture of cable gauges can be expected for longer loops.
25
+
-
+
-
+
+
+
-
+
-
Figure 2.1: The principle of twisted pair transmission
Customer
Equipment
X-connect
Central Office Cabinet Cabinet
Feeder Drop Wire Feeder
CPE
Figure 2.2: The topology of a telephone local loop plant
26
+
-
V
S
Z
S
V
1
+
-
V
2
+
-
I
1
I
2
Z
L
[
A B
C D
]
Source Load Two port network
Figure 2.3: Two-port network model
2.2 Transmission line characterisation
Transmission lines is characterised by its frequency dependant attenuation and
phase delay, the later causing propagation time dispersion. The electrical prop-
erties of the transmission line is often described by a two-port linear circuit
using ABCD parameters in matrix form.
2.2.1 ABCD model of the transmission line
The following relation denes the ABCD model between the input voltage V
1
,
input current I
1
, output voltage V
2
and output current I
2
_
V
1
I
1
_
=
_
A B
C D
_

_
V
2
I
2
_
=
_
V
2
I
2
_
(2.2)
The parameters A, B, C and D is dened by the following relations
Open load voltage ratio
A =
V
1
V
2

I
2
=0
(2.3)
Shorted load impedance
B =
V
1
I
2

V
2
=0
(2.4)
Open load admittance
C =
I
1
V
2

I
2
=0
(2.5)
Shorted load current ratio
D =
I
1
I
2

V
2
=0
(2.6)
The ABCD parameters is also frequency dependant and complex, since the
voltages and currents is complex and frequency dependant. The ratio
T(f) =
V
2
(f)
V
1
(f)
=
V
2
(f)
A(f)V
2
(f) +B(f)I
2
(f)
=
1
A(f) +B(f)
I
2
(f)
V
2
(f)
(2.7)
27
V(x)
+
-
V(x+dx)
+
-
I(x)
I(x+dx)
G dx
R dx L dx
C dx
Figure 2.4: Incremental section of a transmission line
is related to the transfer function H(f) between input voltage V
S
with in-
ternal impedance Z
S
and output voltage V
2
= V
L
across load Z
L
.
H(f) =
V
L
(f)
V
S
(f)
=
V
L
(f)
V
2
(f)
V
2
(f)
V
S
(f)
=
Z
1
Z
1
+Z
S
T(f) (2.8)
Where Z
1
= V
1
/I
1
is the input impedance to the network. The transfer
function depends on the source and load impedances. According to circuit the-
ory, the maximum power transfer occurs when the source and load impedance
is the conjugate of each other. As we will see, the impedance of a twisted pair
transmission line is dominated by the real part {Z
1
} and is more or less con-
stant above 300 kHz. Therefore transmission lines is usually terminated by real
impedances, e.g. simple resistors where R = Z
L
= Z
S
= {Z
0
} where Z
0
is the
average real part of the characteristic input impedance over the frequency range
used for signal transmission. Z
0
is given by the solution of the transmission line
dierential equation that we will look at next.
One property of two-port representation of transmission lines using ABCD
matrices is that multiple cascaded line segments can be modelled by multiplying
the two-port ABCD matrix for each line segment. For example, the characteris-
tics of a loop consisting of N line segments, each with a two-port representation

n
can be calculated as
=
1

2
...
N
(2.9)
Where
1
is the two port model of the rst line segment,
2
the second
and so on starting from the source. The order of multiplication is important.
A single homogeneous line segment is symmetrical, but the cascade of multi-
ple line segments with dierent characteristics has dierent two-port behaviour
depending on which end the source and load is connected.
2.2.2 The Telegraphers equations
The two-port characterization of a transmission line and its corresponding ABCD
parameters is derived by looking at an incremental line segment, as shown in
Figure 2.4
The current and voltage over the incremental line segment is related by the
following pair of dierential equations
28

dV
dx
= (R +jL)I(x, ) (2.10)

dI
dx
= (G+jC)V (x, ) (2.11)
I and V are dependant on time and distance x. j represents rst derivative
with respect to time. R, L, C and G are frequency dependant, and should be
written as R(f), L(f), C(f) and G(f), but are considered constant for a given
frequency in a homogeneous transmission line segment. The solution to (2.10)
and (2.11) is the following equivalent set of second-order dierential equations
d
2
V
dx
2
=
2
V (x, ) (2.12)
d
2
I
dx
2
=
2
I(x, ) (2.13)
Which is identied as a wave solution where the propagation constant is
expressed as
= +j =
_
(R +jL)(G+jC) =

ZY (2.14)
This is a frequency dependant constant, since R, L, C and G is frequency
dependant. The solutions to the dierential equations is a sum of positive and
negative going waves, which vary with position according to e
x
. The real part
of is the attenuation constant, which represents the loss in the transmission
line. The imaginary part, represents phase constant. The phase constant
relates the wavelength with the phase velocity for each frequency component
of the signal. When the phase constant varies with frequency, dierent fre-
quency components travel along the transmission line with dierent velocity,
thus spreading the signal in time. This causes time dispersion and is one of the
sources for ISI, Inter Symbol Interference as discussed in section 1.5.1.
When DMT transmission is used, the time dispersion caused by the trans-
mission line is not a major issue, since the cyclic prex separates two consecutive
symbols in time. The attenuation on the other hand is the largest impairment for
DSL communication. The signal amplitude is attenuated as e
x
. Even though
subscriber loops uses heavier gauge on the remote sections on long loops, the
skin eect causes attenuation to increase with frequency, which we will se later
in section 2.3.
The solution in (2.12) and (2.13) can be modelled as the sum of two opposite
going voltage/current waves
V (x) = V
+
0
e
x
+V

0
e
x
(2.15)
I(x) = I
+
0
e
x
+I

0
e
x
(2.16)
If any of these two solutions is inserted into the appropriate rst order volt-
age/current equations in (2.10) or (2.11), the ratio of the positive-going voltage
to the positive-going current, or the (negative) negative-going voltage to the
negative-going current, is equal to the characteristic impedance of the transmis-
sion line according to
Z
0
=
V
+
0
I
+
0
=
V

0
I

0
=
R +jL

R +jL
G+jC
=
_
Z
Y
(2.17)
29
A uniform segment of a transmission line of length d, e.g. one where the
RLCG parameters is constant along the line for a certain frequency, has the
solution
V
L
= V (d) = V
+
0
e
d
+V

0
e
d
(2.18)
I
L
= I(d) = I
+
0
e
d
+I

0
e
d
(2.19)
Since V
+
0
is related to I
+
0
and V

0
is related to I

0
by Z
0
, given in (2.17),
the Equation (2.19) can be written as
I
L
=
V
+
0
Z
0
e
d

0
Z
0
e
d
(2.20)
And thus V
+
0
and V

0
is expressed as
V
+
0
=
1
2
(V
L
+I
L
Z
0
)e
d
(2.21)
V

0
=
1
2
(V
L
I
L
Z
0
)e
d
(2.22)
The solution above solves the boundary condition for Equation (2.18) when
d = 0 as
V (0) = V
+
0
+V

0
(2.23)
=
1
2
(V
L
+Z
0
I
L
)e
d
+
1
2
(V
L
Z
0
I
L
)e
d
(2.24)
= V
L
cosh(d) +Z
0
I
L
sinh(d) (2.25)
Similarly the solution to I(0) becomes
I(0) = I
+
0
+I

0
(2.26)
=
1
2
(I
L
+
1
Z
0
V
L
)e
d
+
1
2
(I
L

1
Z
0
V
L
)e
d
(2.27)
= I
L
cosh(d) +
1
Z
0
V
L
sinh(d) (2.28)
This is identied as the ABCD parameters in the relation given in (2.2),
thus the A, B, C and D two-port representation of a transmission line with
distributed parameters R, L, C, G becomes
_
A B
C D
_
=
_
cosh(d) Z
0
sinh(d)
1
Z
0
sinh(d) cosh(d)
_
(2.29)
Where is a function of the frequency dependant R, L, C, G parameters and
is given in (2.14)
The transfer function can then be calculated directly from the ABCD pa-
rameters using
H(f) =
V
L
(f)
V
S
(f)
=
Z
L
A Z
L
+B +C Z
S
Z
L
+D Z
S
(2.30)
But more common is the use of T
IL
(f), insertion loss (IL), since it is the in-
sertion of the cable between sender and receiver that results in signal loss, not
30
the impedance matching source and load impedance. Given a transfer function
H(f) the insertion loss is given by
T
IL
= H(f)
Z
L
+Z
S
Z
L
(2.31)
If for example the two-port element is described by a unity matrix (
_
1 0
0 1
_
)
the insertion loss becomes
T
IL
= H(f)
Z
L
+Z
S
Z
L
=
Z
L
Z
L
+Z
S
fracZ
L
+Z
S
Z
L
= 1 (2.32)
While the transfer function is
H(f) =
Z
L
Z
L
+Z
S
T
IL
=
Z
L
Z
L
+Z
S
(2.33)
Which is the generated power divided between source and load impedance,
and when the line is terminated with its characteristic impedance, it results in
6 dB loss of signal power in the source impedance.
2.3 Distributed RLCG parameters
The twisted pair transmission line consist of two polyethylene (PE) insulated
copper cables twisted together. The diameter of the copper cable varies between
0.4 mm and 0.91 mm. With thicker cables resulting in lower DC resistance per
meter. The theoretical electrical properties [5] of a two-wire transmission line
in terms of its distributed RLCG parameters, given its copper core diameter a
and the distance between the conductors D is
L =

cosh
1
(
D
2a
) [H/m] (2.34)
C =

cosh
1
(D/2a)
[F/m] (2.35)
G =

cosh
1
(D/2a)
[S/m] (2.36)
R =
1
a
_
f
c

c
[/m] (2.37)
These equations is normally not useful when considering real twisted pair
cables. The proximity eect causes the surface current to be non-uniform, which
aects both inductance L and resistance R. Also the twists along the cable
causes the distance D between the pairs to vary. In the equations above it is only
R that is frequency dependant. R increases with the square root of frequency.
The increase in resistance with frequency is caused by the Skin eect, which
occur due to Eddy currents induced in the copper core which forces the current
in the wire to travel along its circumference, thus the eective conducting area
of the cross section of the wire is decreased. The skin eect aects both the
Resistance as well the Inductance. C and G is normally considered constant,
and G is usually neglected. Allthough conductance does increase with high
frequencies so for wide band DSL transmission, such as VDSL2 it must be
taken in account. In practice all four RLCG parameters is frequency dependant
which makes the equations above unsuitable to model a twisted pair cable using
DSL transmission.
31
To properly model the characteristics of a twisted pair transmission line, the
frequency dependant RLCG parameters is best established through measure-
ments.
2.3.1 Measurement Procedure
When measuring twisted pair cable characteristics, the measurement proce-
dure consists of measuring the open circuit impedance Z
OC
and short circuit
impedance Z
SC
for a cable of length l. The characteristic impedance is then
calculated as [2]
Z
0
=
_
Z
OC
Z
SC
(2.38)
And the propagation constant is derived from
=
1
l
tanh
1
_
_
Z
SC
Z
OC
_
(2.39)
The relation
=
_
(R +jL)(G+jC) (2.40)
Z
0
=

R +jL
G+jC
(2.41)
Leads to the following relation
R = {Z
0
} = {R +jL} (2.42)
L =
1

{Z
0
} =
1

{R +jL} (2.43)
G = {

Z
0
} = {G+jC} (2.44)
C =
1

{

Z
0
} =
1

{G+jC} (2.45)
Cable measurements is performed on a cable of length l = 10m for fre-
quencies below f = 2MHz. For measurements between f = 2MHz and
f = 30MHz, a cable of length l = 1m is used.
2.3.2 Cable models
The resulting R, L, C and G values derived from measurements is then tted into
a parameterised smooth curve. Several empirical cable models exist and one of
the more popular is the BT model. In the BT model [6] the primary parameters
are given by tting the following curves to the measured cable impedance
R(f) =
1
1
4

r
4
oc
+a
c
f
2
+
1
4

r
4
os
+a
s
f
2
(2.46)
Equation (2.46) models the frequency dependence of a steel enforced copper
wire. This model is also referred to as BT#1. For plain copper wires (BT#0),
the expression is simplied
R(f) =
4
_
r
4
oc
+a
c
f
2
(2.47)
32
Which shows that R is proportional to

f when f > , which is also
shown in Equation (2.37).
Inductance is modelled according to the BT model as
L(f) =
l
0
+l

x
b
1 +x
b
(2.48)
x
b
= (
f
f
m
)
b
(2.49)
Where l
0
and l

are the low frequency and high frequency inductance, b is

a parameter chosen to t the transition between low and high frequencies in the
measured inductance values.
Even though C(f) is more or less constant and G(f) usually negligible for
f < 1MHz , conductance becomes notable for frequencies where f > 10MHz.
C(f) and G(f) is given by the BT model as
C(f) = c

+c
0
f
c
e
(2.50)
G(f) = g
0
f
g
e
(2.51)
Where c

is the contact capacitance and c

0
and c
e
is used to t the mea-
surements. g
0
and g
e
are constants chosen to t the measurements.
2.3.3 The MAR model
Musson [7] pointed out that the established BT model for cable simulations
lacked physical signicance. All practical models are empirical in the sense that
they t smooth curves onto measurements. What Musson emphasized was that
R and jL is the real and imaginary part of a physical impedance and thus
must be related by the Hilbert transform, which origin is that an eect cannot
precede its cause. Any model where R and jL as well as G and jC is not
related by the Hilbert transform, is not guaranteed to have a causal time domain
behaviour, which is also shown in [8].
2.3.4 MAR model
Musson suggested that the measured cable parameters should be tted according
to a model in which R and jL is related by the Hilbert transform, the suggested
model is the MAR1 [7] model with the following parameters
Z
S
= jL

+R
0
(
1
4
+
3
4
_
1 +
as(s +b)
s +c
) (2.52)
Y
P
= C
f
(j +tan) = jC
1MHz
(jf 10
6
)

(2.53)
Where
L

= High frequency inductance per km.

R
0
= DC resistance per km.
s =
410
7
jf
3
4
2
R
0
10
3
= j
f
447.6R
0
a = Proximity factor for skin eect
b and c are skin eect shape coecients
= Shunt capacity loss angle
C
f
=
C
1MHz
(f10
6
)
2

33
Table 2.1: Residential Background Noise Power [dBm/Hz]
Location 8-100 kHz 100-500 kHz 500 kHz - 2 MHz
1 -150 -148 -133
2 -136 -153 -146
The MAR1 model has been simplied in the MAR2 model where b, and c is
xed to b = 2, c = 2.765.
2.3.5 BT vs. the MAR model
Both cable models are empirical in the sense that they t smooth curves on
to measured cable parameters. Mussons argument for using the MAR model
is that by maintaining the relation between the real and imaginary part of Z
S
and Y
P
by the Hilbert transform, the resulting model parameters will be less
sensitive to measurement noise. According to Musson, it is essential in Max-
imum Likelihood estimation that one establish the set of possible parameters
to estimate. Modelling R and L independent of each other is not possible in
a physical sense, and should not be allowed by the model. Although the BT
model do resemble real cable measurements the main argument for using the
MAR model is that it is Causal which makes it possible to manufacture test
equipment that realises the standardised models.
2.4 Noise
The capacity of a communication channel, according to (1.34), is a function of
the ratio between signal and noise at the receiver. In the preceding sections we
have focused on the attenuation of the signal when transmitted over the twisted
pair copper cable. The noise at the receiver end consists of thermal noise from
the receiver electrical circuit and various induced signals in the copper pair. The
thermal noise in the receiver circuit can be calculated from Equation (1.2) and
is -174 dBm/Hz in room temperature, but it also depends on the design of the
analogue front end.
2.4.1 Background noise
The thermal noise is usually much weaker than the background noise induced
into the copper pair. Some measurements have been performed. In [3] a study
made by Bellcore at two locations in New Jersey is mentioned. The result
of those measurements is listed in table 2.1. Based on the results from the
Bellcore study, the background noise level on twisted pair telephone loops has
been assumed to be -140 dBm/Hz. For most DSL installations, crosstalk from
adjacent lines is the dominating noise source, but the background noise level
sets the capacity limit for long loops.
2.4.2 Crosstalk noise
Crosstalk noise arises in DSL because the individual copper pairs in a binder
radiates energy electromagnetically. The eld surrounding wire i, couples to
surrounding wires j, through mutual inductance denoted by M
ij
and coupling
capacitance denoted by E
ij
, which is illustrated in Figure 2.5.
34
Pair 1 wire 1
Pair 1 wire 2
Pair 2 wire 3
Pair 2 wire 4
E
13
E
24
E
14
E
23
M
13
M
24
M
23
M
14
Figure 2.5: Distributed mutual coupling between twisted pairs
The balanced transmission mode of the signal as well as the twisting of the
cable pair mitigates electromagnetic coupling. The balanced signal transmis-
sion suppresses common mode noise signals induced in the cable from external
disturbers, such as radio waves and electrical equipment. By twisting the cable,
induced balanced signals is supposed to cancel itself out along the cable. Unfor-
tunately twisted pair cables are not perfectly homogeneous. RLCG parameters
dier along the cable, so the coupling varies between the twists. Also the indi-
vidual pairs move around along the binder, which alters the inter pair coupling
even more.
The frequency dependant coupling function between pair 2 to pair 1, X
21
(f)
can be found by a generalisation of a two port model. The induced noise, at
position x along the cable, is then given by
N
1
(f, x) = X
21
2jf V
2
(f, x) (2.54)
Where X
21
is a generalised coupling function, determined by all M and E
parameters for the two pairs. N
1
is the induced voltage on pair 1 and V
2
is
the voltage in the disturbing pair 2 at position x. The factor 2jf represents
the time derivative of the disturbing voltage, since the inductive and capacitive
coupling only transmits alternating current/voltage.
Equation (2.54) gives the induced noise voltage along the cable binder be-
tween two individual pairs. The resulting noise signal at the end of the cable
is characterised as either Near End Cross Talk (NEXT) or Far end Cross Talk
(FEXT). NEXT arises in the receiver at the disturbed pair, originating from
the transmitter on the disturbing pair in the same end. FEXT is the received
noise from a disturbing pair originating from a transmitter in the far end of the
cable. Figure 2.6 and Figure 2.7 illustrates the dierence between FEXT and
NEXT.
NEXT is mainly a problem in DSL transmission techniques where the same
spectrum is shared between upstream and downstream trac, through echo
cancellation or time division multiplexing. NEXT also occur between spectrally
incompatible modulation techniques where the upstream and downstream spec-
trum overlap. VDSL2 is spectrally compatible with ADSL2+ and uses frequency
division multiplexing so when estimating crosstalk noise we will only consider
FEXT.
35
Pair 2
Transmitter
Pair 1
Receiver
Figure 2.6: Inter pair coupling causing NEXT
Pair 2
Transmitter
Pair 1
Receiver
Figure 2.7: Inter pair coupling causing FEXT
Pair 2
Transmitter
S
2
(f)
Pair 1
Receiver
d
x
H
2
(f,x)
H
1
(f,d-x)
X
21
(f)
Figure 2.8: Model of the signal coupling along the cable causing FEXT
36
2.4.3 FEXT Modelling
Figure 2.8 show how the signals contributing to FEXT is coupled between two
adjacent pairs along the cable binder. Since the signal in the disturbing pair
2 is attenuated along the cable, the energy transferred through the coupling
function X
21
(f) is a function of transmitter energy S
2
(f), cable attenuation up
to x given by H
2
(f, x). The induced signal in pair 1 then travels further down
the cable and is attenuated according to H
1
(f, d x) for the remaining section
of twisted pair cable. The total FEXT contribution from pair 2 to pair 1 is then
given by
F(f, d) =
_
d
0
X
21
2jf V
2
(f) T
2
(f, x) T
1
(f, d x) dx (2.55)
Where T
1
and T
2
is the insertion transfer function of the line up to the point
of coupling and from that point onward to the far-end receiver respectively.
Taking the squared magnitude of the crosstalk, since it is viewed as noise and
assuming that both pairs are terminated by its own characteristic impedance,
and also assuming that pair 1 and pair 2 has the same RLCG characteristics,
Equation (2.55) can be rewritten as
|F(f, d)|
2
= (4
2
f
2
) |X
21
(f)|
2
|V
2
(f)|
2

_
d
0
e
2d
dx (2.56)
= (4
2
f
2
) |X
21
(f)|
2
|V
2
(f)|
2
d e
2d
(2.57)
FEXT increases with the square of frequency of the transmitted signal.
When considering a multi-pair cable binder, each individual pair to pair cou-
pling function X
ij
(f), varies signicantly with frequency, but the sum of several
cross talkers can be approximated as
S
FEXT
(f) = k
FEXT
f
2
d p |H(f, d)|
2
S
2
(f) (2.58)
Which is also the empirical FEXT model accepted by ETSI [9]. d is the
length of the cable in meter, p is a meter to feet conversion factor (1/0.3048).
|H(f, d)|
2
is the magnitude of the cable transfer function. S
2
(f) is the PSD of
the disturber. The factor k
fext
is given by
k
FEXT
= (
N
49
)
0.6
8 10
20
(2.59)
Where N is the number of disturbers in the same binder. The calculated
S
FEXT
(f) according to the model corresponds to a 1% worst case value. The
model is normally used for design purposes and simulation, thereby focus on
modelling the 1% worst case.
When estimating the performance for VDSL2 based on noise measurements
from ADSL2+ and ADSL&2 disturbers k
FEXT
will be estimated by tting the
model onto measured noise PSD.
2.4.4 Radio Frequency Ingress
External interference such as RF signals and impulse noise is not considered in
this study. The sensitivity to Radio Frequency Ingress (RFI) for a given copper
pair depends on how well the pair is balanced. However, balance changes with
frequency so a copper pair that shows no signs of RFI in the ADSL2+ band may
contain severe RFI in the VDSL2 frequency range. Table 2.2 lists the frequency
ranges used by HAM radio that could interfere with VDSL2 [13].
37
Table 2.2: International HF amateur radio bands
Band start Band stop
(kHz) (kHz)
1 810 2 000
3 500 3 800
7 000 7 100
10 100 10 150
14 000 14 350
18 068 18 168
21 000 21 450
24 890 24 990
28 000 29 100
2.4.5 Impulse noise
Impulse noise is dened as non-stationary noise injected into the local loop ei-
ther from an external disturber or as crosstalk between electrical wires in close
proximity to the telephone cable binder. It can also originate from phones ring-
ing in adjacent pairs in the binder. The characteristics of impulse noise is a
short-duration, high amplitude wideband signal that occurs intermittently. Im-
pulse noise usually have a duration of 10-100 s but can be up to 2-3 ms, thus
aecting multiple DMT symbols. The most ecient method to mitigate impulse
noise is Forward Error Correction (FEC) and interleaving. In the ADSL2+ stan-
dard [10] impulse noise protection is specied by the number of DMT symbols
that can be recovered and is given by
INP = (S D)
1
2

R
N
FEC
(2.60)
Where Impulse Noise Protection INP is the number of DMT symbols that
can be recovered by the error correction. D is the interleaver depth in number
of Reed Solomon (RS) FEC codewords. R is the number of redundancy octets
per RS FEC codeword. N
FEC
is the size of the RS FEC codeword. S is given
by
S =
8 N
FEC
L
(2.61)
Where L is the number of bits transmitted by each DMT frame. As Equation
(2.61) shows, when the transmission rate increases, e.g. the number if bits
transmitted in each DMT frame increases, the data transmission becomes more
vulnerable to impulse noise. VDSL2 increases the data rate by a factor of 3
or 4 depending on the selected prole but uses the same FEC conguration
as ADSL2. To maintain the same impulse noise protection larger interleave
memory is used.
38
Chapter 3
Loop measurement
3.1 Channel characterisation
Characterisation of the local loop is vital to successful DSL deployment. Several
methods of loop characterisation exist. First there is capacity estimation based
on calculations using key parameters such as loop length and topology as input.
Those parameters are sometimes based upon installation data where the type
of copper cable and distance is known, or by open loop measurements where
conductance and capacitance between the cables in a pair is measured. Using
information about the local loop, coarse capacity estimations can be made using
the models presented so far.
3.1.1 Single Ended Line Testing
Single Ended Line Testing (SELT), is based on Time Domain Reectometry
(TDR). During a TDR measurement, a pulse is transmitted from one end of
the pair, usually the CO side. If the pair is not properly terminated, either
by leaving the remote end unconnected, e.g. open ended, or with a telephone
handset on hook, the injected pulse will be reected back. By measuring the
propagation time and the attenuation of individual frequencies in the trans-
mitted pulse. The gathered data can be used to calculate the loop length,
attenuation and cable parameters. If the quiet line noise is measured in the CO
end, a rough estimate of the attainable data rate can be calculated. SELT mea-
surements is very useful when it comes to fault diagnostics where the DSLAM
and Modem fails to establish a connection, for example if the cable is broken or
incorrectly patched.
3.1.2 Dual Ended Line Testing
Loop characterisation using Dual Ended Line Testing, DELT, requires active
equipment in both ends of the loop. Traditionally such measurements requires
a technician to go to the remote location and connect measurement equipment
and perform the measurement between the remote end and the central oce,
where remote controlled equipment is already present. Such measurements has
only been aordable when qualifying leased lines for DSL access in the past.
3.1.3 ADSL2 Training sequence
In the ADSL2 standard Dual Ended Line Testing functionality is built into the
modem and DSLAM. DELT is performed as part of the initialisation/training
39
Table 3.1: ADSL2/2+ training sequence
Training phase
1 Handshake procedures
2 Channel Discovery
3 Transceiver Training
4 Channel Analysis
5 Exchange
sequence if either the ATU-C or ATU-R requests DELT procedure to be entered.
During a normal training sequence the following steps is performed:
Handshake procedure
In phase 1, the ATU-R (modem transceiver) and ATU-C (DSLAM transceiver)
detects each other and agree on a transmission mode. From here on we assume
that both ATU-R and ATU-C agree on G.993.2 (ADSL2+) there are schemes
to fall back, when line conditions are severe, and try lower levels of transmission
modes. The initial handshake is carried out using the G.994.1 standard [12]. It
is also in phase 1 that a decision to initiate loop diagnostics mode is taken.
Channel Discovery
In phase 2, the channel is discovered. Here both ATUs may perform coarse
timing recovery and perform channel probing to determine a power cutback
based on hook status. The ATU-R can also choose a proper subcarrier for
timing reference. Channel discovery is initiated with QUIET state on both
transceivers. During the QUIET state both receivers measure the Quiet Line
Noise (QLN) over a minimum of 512 DMT symbols ( 128 ms). Then chan-
nel discovery is carried out by sending the COMB test signal. The COMB
signal consists of repeatedly transmitted multi tone-symbols with 16 active car-
riers in the downstream direction and 5-10 carriers in upstream (Depending on
transmission mode). The carriers is chosen not to interfere with underlying
POTS service. The carriers is modulated with a 4-QAM constellation and car-
ries random data from a pseudo random bit sequence (PRBS) of length 511 for
downstream and 63 for upstream. The mapping of PRBS bits to each real and
imaginary part of each modulated carrier follows the same scheme as for the
REVERB symbol.
At the end of channel discovery, Both transceivers agrees on a power cutback
level (PCB) on both upstream/downstream to avoid saturation of the receiver
analouge front end (AFE) in each end. This is necessary for very short loops.
Transceiver Training
Phase 3, Transceiver Training. First the REVERB signal is transmitted. The
REVERB signal is generated using the same PRBS as used for the COMB signal
but now all DMT downstream carriers and upstream carriers are modulated.
The signal level used is also reduced from the NOMPSD level to REFPSD
which is NOMPSD - PCB as determined during the channel discovery phase.
The REVERB signal is stationary, e.g. it does not change between symbols.
That way both receivers can train their Automatic Gain Controller and Time
Domain Equalisers. After the REVERB signal, the ATU-C sends the pilot
tone C-TREF. it consists of a single tone symbol where the subcarrier chosen
40
by the ATU-R to be used for pilot modulates the 4-QAM {0,0} constellation
point. Also during phase 3, both transceivers are given opportunity to transmit
a vendor proprietary echo cancellation training signal.
After phase 3 both transmitters now enable the cyclic prex on all DMT
symbols transmitted.
Channel Analysis
Phase 4, channel analysis phase. During the channel analysis phase both transceiverss
transmit the MEDLEY symbol, which is generated using the same PRBS as the
REVERB symbol except that the PRBS is not restarted between each symbol.
The duration of the PRBS is 511 bits downstream and 63 bits upstream so
each symbol generated will shift the PRBS sequence by 1 bit. The MEDLEY
state always lasts for a multiple of 512 symbols so all subcarriers will have gone
through the same 4-QAM constellation sequence, although at dierent points in
time. The receiving ATU aggregates the signal level on each subcarrier, during
the entire MEDLEY sequence so the received aggregated signal will accurately
indicate the loop attenuation on a per carrier sense.
After the channel analysis phase, both ATUs calculate the optimum trans-
mitter conguration for the other end using proprietary bin loading algorithms.
Exchange
After the Channel Analysis phase, both ATUs exchange transmission related
parameters to congure the ATU on the other end in terms of tone order, bit
constellation and gain scaling for each carrier, as well as performance related
parameters such as total aggregate transmit power and SNR margin.
3.1.4 ADSL2+ training sequence
ADSL2+ uses the same training sequence as ADSL2 except that in ADSL2+ the
DMT symbol used for downstream communication uses 512 subcarriers instead
of 256, which means that the PRBS used for the REVERB and MEDLEY
will be repeated once during each generated DMT symbol. Also the REFPSD
level used for downstream transmission is limited by the ADSL2+ transmit
spectral mask, which depends on the operation mode and is specied in the
corresponding annex in [11]. In the channel analysis phase, the measured PSD
of the received signal is not only a function of the loop attenuation but the loop
attenuation and transmit PSD.
3.1.5 Loop Diagnostics mode
ADSL2 and ADSL2+ oer a built in diagnostics mode with the purpose of
analysing the local loop to identify sources of various impairments such as
crosstalk, RFI, bridged taps, impedance mismatch or bad connections. The loop
diagnostics mode is entered from the G.994.1 initialisation 3.1.3. During loop di-
agnostics mode the same sequence is performed as during initialisation/training
up to the MEDLEY state. Each variable length state have a xed duration,
equal to the maximum duration of the same state during initialisation, except
for the QUIET state which has a longer duration during Loop Diagnostics than
the maximum duration during training. After the C-EXCHMARKER and R-
EXCHMARKER, which starts the Exchange phase during normal initialisation,
loop diagnostics specic states is entered. During the loop diagnostics specic
states, the transceives in both ends exchange the loop characteristics parameters
that has been measured.
41
Table 3.2: Parameters determined during Initialisation and Loop Diagnostics
mode
H
lin
(i f) Channel Characteristics per subcarrier, linear X +jY
H
log
(i f) Channel Characteristics per subcarrier, 20 log
10
(|X +jY |)
QLN(i f) Quiet Line Noise per subcarrier, [dBm/Hz]
SNR(i f) Signal-to-Noise Ratio per subcarrier
LATN Loop Attenuation
SATN Signal Attenuation
SNRM Signal-to-Noise Ratio Margin
ATTNDR Attainable Net Data Rate
ACTATP Actual Aggregate transmit power (far-end)
3.1.6 Loop diagnostic parameters
Table 3.2 lists the parameters determined during training/initialisation and
Loop diagnostics mode.
H
lin
is given by
H
lin
(i f) = X +jY = scale/2
15
(a(i) +jb(i))/2
15
(3.1)
Where scale is a 16 bit unsigned integer, a and b are 16 bit signed integers.
The range of Hlin is -2 to 2 for the real and imaginary part and accommodates
for a dynamic range of +6 dB to -90 dB. This is necessary since short loops
can result in signal levels above 0 dB. The resolution of Hlin decreases with the
received signal level.
H
log
is calculated as
H
log
(i f) = 20 log
10
|H(i f)| = 6 m(i)/10 (3.2)
Where |H(f)| is the magnitude of the loop attenuation for a given frequency.
m(i) is represented by a 10 bit unsigned integer which gives H
log
a dynamic
range of +6 dB to -96.2 dB in 0.1 dB steps. The resolution of H
log
is con-
stant over the whole parameter range. The receiving end calculates H
log
(f)
and H
lin
(f) during the transmission of at least 256 REVERB symbols. Apart
from measurement noise, the aggregated measurement over multiple symbols
increases the resolution of the measured attenuation. In [2] a linear relationship
between the accuracy of the estimated channel gain and number of training
symbols is given. With 256 measurements the accuracy of the estimated gain is
improved by 24 dB over the noise. The limiting factor when measuring atten-
uation on long loops will be the resolution of the Analog to Digital Converter
(ADC).
QLN in dBm/Hz is given by
QLN(i f) = 23 n(i)/2 (3.3)
Where n(i) is an 8 bit unsigned integer. This data format supports a QLN(f)
granularity of 0.5 dB and a dynamic range of -23 to -150 dBm/Hz. QLN is mea-
sured during the QUIET state over at least 256 DMT symbols during initialisa-
tion and at least 1 second (4000 DMT symbols) during Loop Diagnostics mode.
QLN(f) represents the estimation of the variance of the narrowband noise in the
centre of each subcarrier, which is given by
2
n
. The variance of that estimation,
assuming that the noise is Gaussian is then [2]
var(
2
n
) =
2
L

4
n
(3.4)
42
The estimated QLN has a Std. deviation of
_
2/L
2
n
which expressed in
dB becomes 10 log
10
(1+
_
2/L)dB per sigma point. For a measurement length
of 256 symbols this equals a std deviation of 0.36 dB. For a 99.7% condence up
to 1.02 dB of error in QLN measurement must be expected. In Loop diagnostics
mode, the error is reduced to 0.28 dB when QLN is measured over 4000 symbols,
which is below the 0.5 dB granularity of the QLN parameter.
SNR is measured during the MEDLEY state and is dened as
SNR(i f) = 32 +snr(i)/2[dB] (3.5)
Where SNR(i) is an unsigned 8 bit integer which results in a granularity of
0.5 dB and SNR dynamic range of -32 to 95 dB.
LATN is the average Loop Attenuation over all subcarriers and is given by:
LATN[dB] = 10 log
10
1
NSC

i = 0
NSC1
|H(i f)|
2
| (3.6)
Where NSC is the number of used subcarriers used. In diagnostics mode,
H is the HLIN parameter measured during the REVERB state. During normal
initialisation, H is taken from HLOG parameters converted to linear scale.
SATN is the dierence in dB between the aggregated transmitted power
(NOMACTP lowered by the power cutback PCB) and aggregated received
power. LATN and SATN can be used to verify the Hlog and HLin parame-
ters as we will see later.
ATTNDR is the estimated attainable data rate determined from the sum
ATTNDR = (
NSC1

i=0
[log
2
(1 + 10
(SNR(i)TARSNRM)/10
)]) 4000 (3.7)
Where the Gap is 9.75 dB which corresponds to a 4-QAM modulated signal
with a Bit Error Rate (BER) equal to 10
7
and TARNSNRM is the target SNR
margin per subcarrier, usually 6 dB. This capacity estimation does not take
in account any bin loading algorithm and is a coarse estimate. The log
2
value
shall be within [0..BIMAX] where BIMAX is the maximum number of bits per
carrier, which for ADSL2/2+ is 15 and shall be rounded to the nearest integer.
ACTATP is an estimate of the total power delivered into the loop from the
transmitter at the point of insertion (tip and ring), it should take into account
the transmit lter function when calculating the aggregate transmit power as
a function of NOMATP (Nominal aggregate transmit power) lowered by the
power cutback.
3.1.7 Reference measurements
When measuring the characteristics of the local loop using DELT, initial ref-
erence measurements are necessary. By connecting a DSL modem directly to
the DSLAM using at short (few meters) cable a reference measurement on the
0-loop can be performed. Figure 3.1 shows the result of DELT HLIN measure-
ments over a 0-loop on two modems using dierent chipsets, connected to two
DSLAMs using dierent chipsets.
The reported LATN (Loop attenuation) is shown together with each mea-
surement. As the graph shows, the reported HLINpsds values do not conform
very well to the 0-loop where the Texas chipset is used. With the Connexant
chipset in the CO, minor eects from impedance mismatch can be visible. The
oset between the reported LATN (loop attenuation) and HLIN can be cor-
rected for by using the relation in (3.6). For the Texas chipset, a correction for
43
0 276 552 828 1104 1380 1656 1932 2208
60
50
40
30
20
10
0
10
Frequency [kHz]
2
0

l
o
g
1
0

|
H
l
i
n
(
f
)
|

[
d
B
]
H
lin
, CO:Connexant, RE:Texas , LATN:0.5 dB
H
lin
, CO:Connexant, RE:Broadcom, LATN:3.0 dB
H
lin
, CO:Broadcom , RE:Texas , LATN:0.5 dB
H
lin
, CO:Broadcom , RE:Broadcom, LATN:1.5 dB
Student Version of MATLAB
Figure 3.1: Reference measurements on 0-loop between dierent vendors
the transmit PSD mask must be applied, since the measured HLIN is not re-
ported according to [10]. In section 8.12.3.1 it is stated that the reported H(f)
should be equal to U2/U1 where U1 is the received signal without the loop and
U2 the received signal with the loop in-between CO and RE. With a 0-loop
inserted, H(f) should be unity for all frequencies. The level of conformance in
the implementation of DELT varies between vendors and rmware releases. To
be able to use the DELT measurements in loop qualication, it is necessary to
adjust the reported H
lin
and H
log
values according to the reported overall loop
attenuation LATN. The adjusted H(f) for the 0-loop measurement is shown
in Figure 3.2.
The reported loop attenuation and the actual graph for HLIN do not truly
reect the 0-loop even with the correction applied. Eects of the transmit
lters, receive lters and eventual gain adjustments and adaptive hybrids is not
properly compensated for by the ATU-C and ATU-R.
When tting the cable model to the measured loop characteristics function
H(f), the model will only be optimized in the frequencies between 276 kHz
and 1104 kHz. Within that range, the receive and transmit lters does not
aect the measured loop characteristics. Also the eect of the PSD mask for
ADSL2+, which is not properly compensated for by the Texas chipset, is not
present within that frequency range.
Another argument to limit the cable t optimization to frequencies below
1104 kHz is the eect of noise and parameter resolution. When measuring long
loops ( 3000 m) the attenuation for the higher frequencies exceed the SNR
of the transmitted signal. Figure 3.3 shows the result of a DELT measurement
on a 3100 meter long loop. The H
lin
parameter becomes less usable below -50
dB because of its linear resolution. Below -50 dB each linear step (1/32768)
represents a step larger than 0.1 dB on the logarithmic scale so below -50 dB
measurement results should be taken from H
log
(f) instead.
44
0 276 552 828 1104 1380 1656 1932 2208
60
50
40
30
20
10
0
10
Frequency [kHz]
2
0

l
o
g
1
0

|
H
l
i
n
(
f
)
|

[
d
B
]
H
lin
, CO:Connexant, RE:Texas , LATN:0.5 dB
H
lin
, CO:Connexant, RE:Broadcom, LATN:3.0 dB
H
lin
, CO:Broadcom , RE:Texas , LATN:0.5 dB
H
lin
, CO:Broadcom , RE:Broadcom, LATN:1.5 dB
Student Version of MATLAB
Figure 3.2: Reference measurements on 0-loop, normalised to reported LATN
0 276 552 828 1104 1380 1656 1932 2208
100
90
80
70
60
50
40
30
20
10
0
Frequency [kHz]
H
(
f
)

[
d
B
]
20 log
10
|H
lin
(f)|
H
log
(f)
Student Version of MATLAB
Figure 3.3: DELT on 3100 m (Electrical PE05 length), H
lin
and H
log
not nor-
malized
45
DSLAM MODEM
100 ohm
1:1
100 ohm
1:1
Noise Gen
Zout 50 ohm
Figure 3.4: Noise Reference measurement set-up
Table 3.3: Noise injection circuit loop attenuation
Modem Without source load With 50 ohm source load
Chipset LATN [dB] LATN [dB]
Texas 0.5 7.0
Broadcom 1.5 8.0
3.1.8 Noise reference measurement
To verify the reported QLN (Quite Line Noise) from the modem during DELT,
an initial noise reference measurement was performed. The reference noise was
generated using an Agilent 33250 Arbitrary Waveform Generator and inserted
into the 0-loop using the circuit diagram shown in Figure 3.4. The Waveform
generator was set to generate white noise with output bandwidth 0 to 50 MHz.
Maximum output power is 20 dBm, which corresponds to -57 dBm/Hz. Usu-
ally the noise source should be connected to the insertion point using High
Impedance to avoid loading the loop, but that also attenuates the inserted sig-
nal. The two baluns generate a parallel load of 50 ohm (100//100). With
a 5k resistor between the generator and the insertion point, an additional 40
dB attenuation would be added and the maximum power of the inserted noise
would be -97 dBm/Hz (100 dBm/Hz into each balun). The CO transceiver, and
crosstalk from the multi pair connector cable, generates approximately -115
dBm of background noise, so it would not be possible to measure any reference
levels other than the maximum output power from the signal generator.
One solution is to generate a custom noise pattern with narrower bandwidth
more suitable to the ADSL2+ frequency range, but when performing noise refer-
ence measurement, loading the loop is not an issue. The 5k resistor was removed
and the maximum inserted noise power with the output set to 20 dBm is then
-57 dBm/Hz (-60 dBm/Hz into each balun). Without the 5k resistor between
the noise generator and the noise insertion circuit, the output impedance of the
noise generator loads the insertion point with a shunt impedance of 50ohm. The
resulting signal attenuation is then 6 dB. This is also veried by the LATN pa-
rameter from the DELT measurement shown in Table 3.3 which also indicates
an increased loop attenuation of about 6 dB.
46
0 276 552 828 1104 1380 1656 1932 2208
150
140
130
120
110
100
90
80
70
60
50
Frequency [kHz]
Q
L
N

[
d
B
m
/
H
z
]
Broadcom: background noise
Broadcom: [100 90 80] dBm/Hz
Texas: background noise
Texas: [100 90 80] dBm/Hz
Student Version of MATLAB
Figure 3.5: Noise PSD reference measurements
3.1.9 Noise level accuracy
Figure 3.5 shows the result of noise reference measurements with the output
power set to -20, -10 and 0 dBm which corresponds to -100, -90 and -80 dBm/Hz.
It is clearly visible that the two modems report dierent noise levels. The
background noise from the CO is also present as reference. The same dierence
between the two modem chipsets is clearly visible. The spike at 1.104 MHz in
the measurement made with the modem using the Texas chipset is internally
generated in the modem.
3.1.10 Noise level correction
The solution to the deviation is to adjust the QLN values reported in the DELT
measurement with the measured reference. The average of 10 reference mea-
surements on -80 dBm/Hz is used to generate a PSD correction Mask which will
be applied to all QLN measurements made using modems based on the Texas
chipset.
3.1.11 Measuring the Modem Background Noise
As mentioned in section 2.4, the noise oor is determined by the analogue front
end of the modem and the background noise induced into the local loop. To
be able to determine the internal noise level in the modem a reference mea-
surement was made on the 0-loop with a static attenuation inserted near the
modem, as shown in Figure 3.6. Two attenuation arrangements was used. One
with two 4.7kohm resistors in series and one with two 10kohm resistors. The ex-
pected attenuation for a series resistance, given in Equation (3.8) and measured
attenuation using DELT is listed in table 3.4. The internal noise generated
by the series resistors is irrelevant in this case since it can not be more than
-174 dBm/Hz into a matched load, which is not the case for a DSL modem
receiver with an input impedance of 100 ohm which matches the impedance of
47
DSLAM MODEM
0-loop
10k ohm
10k ohm
Figure 3.6: Modem noise reference measurement set-up
Table 3.4: Calculated and measured attenuation of line noise
Resistance Calculated Measured
ohm IL [dB] LATN [dB]
9.4k 33.6 34.0
20k 40.1 40.5
the twisted pair.
_
A B
C D
_
=
_
1 R
0 1
_
H
IL
=
Z
L
+Z
S
A Z
L
+B +C Z
S
Z
L
+D Z
S
(3.8)
=
Z
L
+Z
S
Z
L
+Z
S
+R
(3.9)
The resulting QLN measurements with the two attenuators inserted is shown
in Figure 3.7. The reason for using two dierent attenuation set-ups was to see if
the measured QLN diered by more than 5 dB between the two measurements,
which it would if the noise measured was originating from the DSLAM. For the
Texas chipset based modem, the measured noise was the same so the internal
noise in the modem exceeds the background noise of -140 dBm above 1.1 MHz.
The modem using the Broadcom chipset shows a slight decrease in noise level,
but not more than 6 dB, therefore it is assumed that the lower curve is the
modem internal noise.
3.1.12 Measuring the line background noise
To estimate the attainable rate on loops where the loop background noise be-
comes dominant over the FEXT noise, measured noise on long loops can be
used to collect noise statistics in a certain area.
The conclusion from the measurements in Section 3.1.11 is that the back-
ground noise can only be measured if it exceeds the internal modem receiver
noise, which when using the Texas chipset is higher than -140 dBm/Hz for fre-
quencies above 1.1 MHz. On long loops background noise determines the noise
oor at higher frequencies while FEXT dominates in the lower frequency band.
48
0 276 552 828 1104 1380 1656 1932 2208
160
150
140
130
120
110
100
Frequency [kHz]
Q
L
N

[
d
B
m
/
H
z
]
Texas: line noise attenuated by 34 dB
Texas: line noise attenuated by 40 dB
Broadcom: line noise attenuated by 34 dB
Broadcom: line noise attenuated by 40 dB
Student Version of MATLAB
Figure 3.7: Modem noise measurements using set-up in Figure 3.6, with correc-
tion for Texas chipset applied as explained in Section 3.1.10
Figure 3.8 shows the measured QLN noise in the receiver using DELT on two
3 km loops. The (electrical) length of the loop was determined by tting a
modelled ETSI PE05 to the measured loop attenuation. The measurement us-
ing the modem with a Texas chipset shows that the noise level above 1 MHz is
dominated by the modem noise, while FEXT dominates the received noise up
to 1 MHz.
The measured noise with the modem using the Broadcom chipset shows that
the noise level follows the FEXT level down to the background noise oor.
Summary
If background and RF noise is to be estimated on long loops where FEXT is
attenuated by the loop itself, it must be based on measurements using modems
with a much lower internal noise than -140 dBm. This disqualies measurements
using the Texas chipset based modems for background noise estimation.
49
0 276 552 828 1104 1380 1656 1932 2208
160
150
140
130
120
110
100
Frequency [kHz]
Q
L
N

[
d
B
m
/
H
z
]
Texas: line noise on a 3110 m loop
Broadcom: line noise on a 3070 m loop
Texas: modem noise
Broadcom: modem noise
Student Version of MATLAB
Figure 3.8: Field measured line noise using DELT on long loops (3 km), with
correction for Texas chipset applied as explained in Section 3.1.10
50
Chapter 4
Channel capacity
estimation
4.1 Fitting the Loop channel model
The main task of the loop capacity estimation is to t the cable model derived
in section 2.2 and 2.3 and the crosstalk model derived in section 2.4.3.
4.1.1 Fitting the cable model
The transfer function of a twisted pair local loop is given by equation (2.30)
where the ABCD parameters is given by a chosen cable model with a certain
cable length. The loop transfer function is then
H
model
(f, d) =
Z
L
+Z
S
A Z
L
+B +C Z
S
Z
L
+D Z
S
(4.1)
where
_
A B
C D
_
=
_
cosh((f)d) Z
0
(f)sinh((f)d)
1
Z
0
(f)
sinh((f)d) cosh((f)d)
_
(4.2)
Z
S
and Z
L
is the source and load impedance, which is chosen to match
Z
0
(f), the frequency dependant characteristic impedance of the cable. Z
0
(f) is
considered constant for frequencies above 300 kHz and is dominated by its real
part. (f) is the frequency dependant propagation constant and d the length
of the cable segment in meters. The secondary cable parameters Z
0
(f) and
(f) is derived from the cable model using Equation (2.17) and (2.14). The
frequency dependant RLCG parameters are given by the chosen cable model.
The electrical length of the cable is determined in a least square sense, with
respect to the length d, by minimizing the expression

i
(H
delt
(if) 20 log
10
(|H
model
(if, d)|) H
offset
)
2
(4.3)
Over a chosen set of frequencies given by i = m..n times subcarrier spacing
f. The coecient H
offset
is a parameter representing a loss factor that is con-
stant over all frequencies that covers dierences in source and load impedance,
as well as any other loss in the loop that is not accounted for by the cable model.
Such as manufacturing variations, contact resistance in cable splices, reection
loss due to impedance mismatch etc.
The results from the reference measurements described in Section 3.1.7 shows
that without the H
offset
parameter, the cable t optimization would end up
51
0 1 2 3 4 5 6 7 8 9 10 11 12
220
200
180
160
140
120
100
80
60
40
20
0
Frequency [MHz]
H
l
o
g
(
f
)

[
d
B
]
ETSI 0.5mm BT (dwug)
ETSI ADSL PE05 BT (996.1)
ETSI ADSL PE05 MAR2 (hand optimized to fit ETSI 0.5mm)
ETSI ADSL PE05 MAR2 (optimized to 996.1 parameters)
Student Version of MATLAB
Figure 4.1: Simulated attenuation of 3000 m cable using 4 dierent models for
0.5 mm cable
with an electrical cable length that compensates for the attenuation oset in-
stead of matching the frequency dependant attenuation of the cable.
The set of frequencies is chosen so that the measurement is not aected
by the transmitter or receiver lters, as well as the PSD masks. H
delt
is the
ADSL2/ADSL2+ Loop diagnostic parameter obtained during Initialisation or
Diagnostics mode. For those frequencies where attenuation is below (50 dB)
H
delt
is given by
H
delt
(if) = 20 log
10
|H
lin
(if)| (4.4)
When attenuation is above 50 dB, the values from H
log
is used instead. In
the implementation the cable model was tted using the measurements from
tone 64 to 256, which corresponds to a frequency range of 276 kHz to 1104 kHz.
4.1.2 Error analysis of the cable model
The accuracy when tting the cable model depends on how well the measured
loop conforms to the chosen model. The chosen approach has been to model
the loop as single cable segment consisting of 0.5 mm plastic insulated twisted
pair cable. The ETSI-PE05 cable model was used, based on the assumption
regarding the topology of the Swedish copper access network presented in [14].
ETSI-PE05 models a 0.5 mm polyethylene twisted copper cable using the BT
model with measured and optimized parameters as specied in [9]. Those cable
parameters has also been converted to the MAR2 model by tting the model
parameters onto the tabulated RLC parameters [15]. Figure 4.1 shows the
transfer function of four cable models over the frequency range used for VDSL2
capacity estimation (0 to 12 MHz).
52
Table 4.1: Result when tting a PE05mar cable to simulated cables of varying
gauge
2000 m cable Fitted Error [dB]
simulated using Length [m] Avg Max
ETSI MAR 0.32 mm 2913.0 9.3 16.7
ETSI MAR 0.40 mm 2391.0 3.3 5.9
ETSI MAR 0.63 mm 1686.9 0.2 1.5
ETSI MAR 0.90 mm 1177.3 0.1 2.3
0 1 2 3 4 5 6 7 8 9 10 11 12
25
20
15
10
5
0
5
Frequency [MHz]
A
v
e
r
a
g
e

E
r
r
o
r

[
d
B
]
ETSIADSLPE032(MAR) 0.32mm
ETSIADSLPE04 (MAR) 0.4mm
ETSIADSLPE063(MAR) 0.63mm
ETSIADSLPE09 (MAR) 0.9mm
Student Version of MATLAB
Figure 4.2: Dierence from 2000 m cable model & tted ETSI-PE05mar model
Dependency between cable model and error
A test was performed where the ETSI-PE05mar cable model was tted to simu-
lated insertion loss from four dierent cable models. The PE05 cable model was
tted to the calculated attenuation of 2000 meter of each model, minimizing
the error over the frequency range corresponding to tone 64 to 256. Then the
attenuation from the tted model was calculated in the frequency range from
2.2 MHz to 12 MHz. This is the frequency range used in the VDSL2 proles
considered for CO deployment. The attenuation in the frequency band outside
the tted region was compared with the simulated cable. The result is shown
in Table 4.1 where the length of the tted PE05 cable and the deviation from
the simulated model is shown. A graph of the deviation is also shown in Figure
4.2 and Figure 4.3
The selection of cable model is fairly critical when if comes to accurate
estimation of the attenuation outside the tted region.
Figure 4.4 shows the correlation between the average error in the region
where the error is minimized (f = [64..256] f) and the average error in the
extrapolated region (1.104MHz < f 12MHz) for the four simulated cable
gauges. The gure shows that there is a strong correlation. Which implies
that the average error between the tted model and the measured attenuation
53
200 300 400 500 600 700 800 900 1000 1100 1200
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
Frequency [kHz]
D
e
v
i
a
t
i
o
n

[
d
B
]
ETSIADSLPE032(MAR) 0.32mm
ETSIADSLPE04 (MAR) 0.4mm
ETSIADSLPE063(MAR) 0.63mm
ETSIADSLPE09 (MAR) 0.9mm
Student Version of MATLAB
Figure 4.3: Dierence from 2000 m cable model & tted ETSI-PE05mar model
could be used to choose a more suitable cable model than assuming that all
loops consist of ETSI PE05 cable. Unfortunately, modelling real telephone
loops is more complex than choosing the exact cable model. Real loops consists
of multiple cable segments each carrying 100s or 1000s of pairs, where each
segment might have a dierent cable characteristic. To be able to properly
model a loop, some knowledge about the loop topology must be known a priori.
The only information about the loop is that it mainly consists of 0.5 mm twisted
pair copper cable.
4.1.3 Further analysis of the model error
The deviation between the simulated cable and the extrapolated optimized
model can be correlated with the deviation within the region where the model
error is minimized. Figure 4.4 shows the correlation between the average error
in the region f = 276kHz to 1104kHz and 1104kHz < f 12MHz for a
PE05mar cable tted onto the four cable simulations listed in table 4.1. The
correlation shows that the best cable model can be selected, based on the aver-
age error within the tted region, which gives the least average error which the
extrapolated region.
Length dependency
The average and maximum error listed in Table 4.1 shows that the error is
larger for thin cables. That is the error is proportional to the electrical length
of the cable, or more directly the attenuation of the cable, which is larger for
longer cables. Figure 4.5 shows the correlation between average error in the
tted model and the length of the simulated cable.
Optimizing the cable model
The MAR model has the benet of maintaining a causal and physically correct
behaviour even if the modelling parameters are changed. One approach was to
54
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
0
2
4
6
8
10
12
1/192
i=64
256
|HLOG
err
(i f)| [dB]
1
/
2
5
2
6

i
=
2
5
7
2
7
8
2

|
H
L
O
G
e
r
r
(
i

f
)
|

[
d
B
]
Student Version of MATLAB
Figure 4.4: Relation between error in tted and extrapolated region
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
1
2
3
4
5
6
7
8
9
10
A
v
e
r
a
g
e

e
r
r
o
r

[
d
B
]
Cable length [m]
Student Version of MATLAB
Figure 4.5: Relation between average error and length for PE05 tted onto
PE032 of increasing length
55
Table 4.2: MAR2 cable model parameters
R
0
DC resistance per km
L

High frequency inductance per km

a Proximity factor for skin eect
C
1MHz
Capacitance at 1 MHz per km
Shunt capacity loss angle
Table 4.3: Result when tting an optimized cable model to simulated cables of
varying gauge
2000 m cable Optimized model Error [dB]
simulated using Length [m] Avg Max
ETSI MAR 0.32 mm 3006.7 2.8 8.6
ETSI MAR 0.40 mm 1934.9 2.5 6.4
ETSI MAR 0.63 mm 2074.7 0.3 0.8
ETSI MAR 0.90 mm 1398.6 0.1 0.6
increase the number of tted parameters to include the cable model as well as
the length of the cable. The adjustable MAR2 model parameters are listed in
table 4.2. The optimized cable model is derived by minimizing the expression

i
(H
delt
(if) 20 log
10
(|H
MAR2
(if, d, R
0
, L

, a, C
1MHz
, )|) H
offset
)
2
(4.5)
Where H
MAR2
(if, d, R
0
, L

, a, C
1MHz
, ) is Equation (4.1) with
=
_
Z
S
(f, L

, R
0
, a, b, c) Y
P
(f, C
1MHz
, ) (4.6)
Z
0
(f) =

Z
S
(f, L

, R
0
, a, b, c)
Y
P
(f, C
1MHz
, )
(4.7)
and where Z
S
and Y
P
is given by Equation (2.52) with b = 2 and c = 2.765.
4.1.4 Results from tting the optimized cable model to
simulations
When testing the optimized cable model using simulated loops the results listed
in Table 4.3 was achieved.
It is clearly visible that the average and maximum error is decreasing, except
for an optimized model tted to a simulated attenuation of 2000 m cable mod-
elled using the ETSI MAR PE04 model. The graphs in Figure 4.6 and Figure
4.7 shows how the model error has decreased in the extrapolated region as well
as the tted region.
4.1.5 Correctness of the optimized model
When the parameters in the MAR2 model is allowed to be set independently
of each other without any constraints, the resulting model parameters has no
physical relevance to real cable parameters. The optimized model has shown
56
0 1 2 3 4 5 6 7 8 9 10 11 12
10
8
6
4
2
0
2
4
6
8
10
Frequency [MHz]
A
v
e
r
a
g
e

E
r
r
o
r

[
d
B
]
ETSIADSLPE032(MAR) 0.32mm
ETSIADSLPE04 (MAR) 0.4mm
ETSIADSLPE063(MAR) 0.63mm
ETSIADSLPE09 (MAR) 0.9mm
Student Version of MATLAB
Figure 4.6: Dierence from 2000 m cable model & tted optimized MAR model
200 300 400 500 600 700 800 900 1000 1100 1200
0.01
0.008
0.006
0.004
0.002
0
0.002
0.004
0.006
0.008
0.01
Frequency [kHz]
D
e
v
i
a
t
i
o
n

[
d
B
]
ETSIADSLPE032(MAR) 0.32mm
ETSIADSLPE04 (MAR) 0.4mm
ETSIADSLPE063(MAR) 0.63mm
ETSIADSLPE09 (MAR) 0.9mm
Student Version of MATLAB
Figure 4.7: Dierence from 2000 m cable model & tted optimized MAR model
57
0 500 1000 1500 2000
0
1
2
3
4
5
6
x 10
8
P
r
o
p
a
g
a
t
i
o
n

v
e
l
o
c
i
t
y

[
m
/
s
]
Frequency [kHz]
Student Version of MATLAB
Figure 4.8: Propagation velocity for dierent parameter optimized MAR2 mod-
els
to be stable, but the resulting parameters is not useful to anything else than
attenuation calculation. Figure 4.8 shows the signal propagation velocity, as a
function of frequency, for the optimized models tted to the 4 simulated cables
of varying gauge. All real twisted pair cables has a propagation velocity about
2 10
8
[m/s], and no signal can never travel faster than 3 10
8
[m/s]. That is
clearly not the case for the optimized models. Propagation velocity is given by
the relation
v(f) =
2 f
{(f)}
[m/s] (4.8)
Where is the frequency dependant propagation constant, per meter.
4.1.6 Importance of cable model optimization
Apart from the lack of relation to real world cable and topology parameters, the
importance of an absolute optimum model t can be questioned. First the DELT
measurement result is aected by how well dierent vendors have implemented
the algorithm. Second the termination impedances may be aected by adaptive
hybrids, which might not be accounted for by the DELT algorithm. Third the
local loop consists of a cascade of cable segments with dierent characteristics.
Any attempt to model the loop as a single uniform segment is by itself incorrect.
But most important, Capacity estimation is in general not dependant of the
accuracy of the estimated cable attenuation. Which we will show later in section
4.3. The computational complexity of tting 6 model parameters compared to
only optimizing the electrical length, shall not be neglected.
4.2 Fitting the noise model
The crosstalk noise is modelled as FEXT originating from disturbers on adjacent
pairs in the same binder, since VDSL2 uses FDM and is spectrally compatible
58
Pair 1
d
Pair 2
Pair 3
Pair N
S
FEXT
(f,d,N,H(f),S
2
(f))
Figure 4.9: Topology used in ANSI FEXT model
with ADSL2/ADSL2+. The FEXT model given in Equation (2.58) is the model
used For simulation and performance calculations. It assumes that all N dis-
turbers shares the same cable binder as the disturbed pair, gure 4.9 shows the
modelled topology.
A real subscriber loop has the topology similar to the one illustrated in
Figure 2.2. In the rst section, all pairs share the same binder. Then the feeder
cable is split into multiple distribution cables going to separate areas and so on.
If the FEXT noise added in section i is denoted by S
FEXT,i
then the sum of
the crosstalk from all 1..N sections is
S
FEXT
(f) =
N

i=1
S
FEXT
(f, d
i
, N
i
, H(d
i
, f), S
2
(f)|H(
i1

k=1
d
k
, f)|
2
) |H(
N

k=i+1
d
k
, f)|
2
=
N

i=1
k
FEXT
(N
i
) f
2
d
i
|H(d
i
, f)|
2
|H(
i1

k=1
d
k
, f)|
2
|H(
N

k=i+1
d
k
, f)|
2
S
2
(f)
=
N

i=1
k
FEXT
(N
i
) d
i
f
2
|H(d, f)|
2
= f
2
|H(d, f)|
2
S
2
(f)

i
k
FEXT
(N
i
) d
i
= f
2
|H(d, f)|
2
S
2
(f) k
sum
(4.9)
Where all sections is assumed to have the same cable characteristics. N
i
is
the number of disturbers in each section and d
i
is the length of each section.
Equation (4.9) shows that, given the same topology and that all disturbers uses
VDSL2 transmission, the FEXT coupling from adjacent pairs to the modelled
pair can be determined by minimizing the expression

i
(QLN
delt
(if) S
FEXT
(if, k
sum
))
2
(4.10)
With regard to k
sum
which is the total FEXT loss from all sections. All
disturbers is assumed to be ADSL2/ADSL2+ disturbers. QLN
delt
is the noise
prole measured during DELT, as described in section 3.1.6.
By tting the noise model within the frequency range 276 kHz to 1.104
MHz both ADSL2 and ADSL2+ disturbers will be accounted for. Figure 4.10
shows the PSD of the various DSL disturbers that can coexist in the same cable
binder. Within the frequency range where the FEXT model is tted to the
59
0 276 552 828 1104 1380 1656 1932 2208
160
140
120
100
80
60
40
20
Frequency [kHz]
P
S
D

[
d
B
m
/
H
z
]
ADSL2+
BRISDN
HDSL
Student Version of MATLAB
Figure 4.10: PSD from various DSL disturbers [9]
measured noise prole, FEXT from ADSL2 and ADSL2+ disturbers will be
the dominating noise source since FEXT from ISDN and HDSL will be much
weaker.
4.2.1 Validity of the Noise Model
When estimating the sensitivity to crosstalk and noise in a given pair the worst
case scenario is used. In this case we assume that all ADSL2/ADSL2+ dis-
turbers is upgraded to VDSL2 and that the crosstalk noise prole will follow
the VDSL2 PSD with the same coupling factor k
FEXT
. This will result in a
capacity estimation using the worst case scenario. In reality not all present
ADSL2/ADSL2+ subscribers will be upgraded to VDSL2. For example, cus-
tomers on long loops ( 3 km) will not benet by upgrading to VDSL2, because
the cable attenuation above 2.2 MHz will be larger than the ratio between max-
imum transmit power and the noise oor, which is about 90 dB for VDSL2.
Therefore not all ADSL2/ADSL2+ disturbers present in the same binder will
become VDSL2 disturbers. On the other hand, the broadband deployment in
a certain area may be far from saturated and in some binders, the number of
disturbers may increase. It is beyond the scope of this thesis to further inves-
tigate the loop topology for DSL deployment, for the FEXT loss we establish
a minimum bound on the estimated capacity based on the current noise prole
on the measured loop.
4.3 Channel capacity estimation
With a suitable cable model tted to the measured transfer function, H(f), and
a FEXT coupling factor k
FEXT
, determined by tting the FEXT model to the
measured noise prole QLN(f), a model for the capacity of the local loop can
be derived. The Shannon limit for the capacity of a communications channel is
given in (2.1). Given a transmit PSD limit and the models derived, the signal
60
to noise ratio S(f)/N(f) is given by
SNR(f) =
|H
model
(f, d)|
2
PSD(f)
S
FEXT
(f)
=
|H
model
(f, d)|
2
PSD(f)
f
2
|H
model
(f, d)|
2
PSD(f) k
sum
=
1
f
2
k
sum
(4.11)
The equation above shows that capacity is not dependant on the cable model.
That is because in the channel model, both signal and noise is equally atten-
uated. The cable model is still needed to calculate the capacity on long loops
where the attenuated signal becomes weaker than the background noise oor,
which denes the maximum reach for a given transmission mode. Equation
(4.11) is then written as
SNR(f) =
1
f
2
k
sum
+

2
|H
model
f,d|
2
PSD
(4.12)
Where
2
is the background noise power, assumed to be -140 dBm/Hz.
4.3.1 Calculating the attainable rate
With the SNR given per Hz, the capacity of the channel can be calculated
using the waterll algorithm described in section 1.6 or by using the method of
calculating the Attainable Data Rate, ATTNDR, given in Equation (3.7). The
later does not take into account the aggregated output power and provides the
maximum unconstrained channel capacity. For realistic capacity estimations,
the power constrained waterll algorithm will be used. When using the waterll
algorithm, the carriers will be loaded in order of best SNR to worst and that is
power conservative.
VDSL2 uses the same modulation technique as ADSL2+ so the same coding
gain and SNR margin will be used when calculating the attainable rate using
the waterll algorithm.
The capacity in bits/Hz as a function of SNR/Hz is then given by
b = log
2
(1 +
SNR(f)
(P
e
)
) (4.13)
Where is the gap given a target bit error rate
(P
e
) = 10
(GAP
QAM
(P
e
)+SNRMGAIN
CODING
)/10
(4.14)
GAP
QAM
(P
e
) gives the required gap from the Shannon capacity for QAM
transmission, given a target bit error rate P
e
. The target bit error rate according
to the VDSL2 standard is 10
7
which corresponds to a Shannon gap of 9.8 dB.
SNRM is the target SNR margin added to the minimum SNR required for
the given target bit error rate, which accounts for non stationary noise conditions
on the line. In the calculations, an SNRM of 6 dB is used.
GAIN
CODING
is the coding gain achieved by the combined Trellis coding
and Reed Solomon forward error correction. In the calculations, a coding gain
of 3.5 dB is used, which is slight conservative.
The cyclic extension, which in VDSL2 has the duration of 5/64 of a symbol,
is accounted for by calculating the data rate using a symbol rate of 4000 sym-
bols per second. The VDSL2 sync word inserted every 257s symbol has to be
adjusted for in the rate calculation.
61
Table 4.4: VDSL2 proles
Prole Max DS Power Index of highest Upper band edge
dBm used subcarrier MHz
8a +17.5 1971 8.5
8b +20.5 1971 8.5
8c +11.5 1971 8.5
8d +14.5 1971 8.5
12a +14.5 2782 12
12b +14.5 2782 12
997
998
DS1 US1
DS1
DS2
DS2
US2
US2
POTS/US0
US1
f
3
f
4
f
2
f
5
f
3
f
4
f
2
f
5
f
1
Figure 4.11: VDSL2 band plans, US bands transmits data upstream and DS
bands transmits data downstream towards the end user
4.3.2 VDSL2 proles
The VDSL2 proles that is considered for use in central oce installations are
listed in Table 4.4. In all proles, the maximum aggregated upstream power
is limited to 14.5 dBm. Depending on regulations, not all proles will be de-
ployable. The dierence between prole 12a and 12b is that in 12b the support
of upstream band US0 is not mandatory. US0 is the upstream band used for
ADSL1, ADSL2 and ADSL2+. With the support of US0 the reach for VDSL2,
using prole 8b for example, will be the same as for ADSL2+. When estimating
the attainable rate, the estimation will be based on the prole resulting in the
highest bi-directional net data rate.
4.3.3 VDSL2 band plans
To be spectrally compatible with other services, VDSL2 must use the same band
plan below 2.208 MHz as ADSL2+, which includes optional support for US0.
Figure 4.11 shows the two band plans dened for VDSL2. Table 4.5 lists the
frequencies for the band edges.
In band plan 997, the total bandwidth allocated for upstream transmission
is larger than the bandwidth allocated for downstream transmission. Band plan
997 is thus more symmetrical since the SNR is lower for higher frequencies due
to the frequency dependant attenuation.
For the Swedish copper network band plan 998 has been chosen, which re-
sults in a more asymmetrical upstream/downstream ratio. When estimating the
attainable rate for VDSL2 on a given loop, the edge frequencies for band plan
998 will be used with an US0 edge frequency compatible with ADSL2+ Annex
62
Table 4.5: VDSL2 band plan edge frequencies
f
0L
f
1
f
2
f
3
f
4
f
5
Downstream Upstream
Band plan MHz MHz MHz MHz MHz MHz MHz MHz
997 25 138/276 3000 5100 7050 12000 4674 7301
998 25 138/276 3750 5200 8500 12000 6774 5201
M, which allocates 32 more carriers to upstream data.
63
Chapter 5
Implementation
5.1 Measurements
5.1.1 DELT measurements
DELT measurements was carried out using the Ericsson PEM Service Cong-
uration Manager client. From within the client, individual DSL lines can be
selected and performance statistics can be gathered such as trained rate, at-
tainable rate, error statistics etc. Both DELT and SELT measurements can be
performed. The results from the DELT measurement was exported to a text
le, using the built in le export function. The exported data was read into
Matlab using a developed Matlab script. All the parameters listed in Table 3.2
is imported.
5.1.2 HLOG/HLIN measurement correction
The reported HLIN and HLOG is adjusted by a gain factor so that the average
attenuation per carrier, calculated using Equation (3.6), matches the reported
LATN parameter for the measured carriers in the downstream direction. An
optional correction mask can be applied prior the gain adjustment, which makes
it possible to calibrate the result of the cascade the transmit lter, receive
lter and signal processing over the 0-loop. In the tests the only correction
applied was the ADSL2+ transmit PSD mask for Texas based chipsets, which
as shown in Figure 3.1 is not properly accounted for in the reported downstream
HLIN/HLOG values.
5.1.3 QLN measurement correction
The downstream QLN values from DELT measurements using Texas based
modems is corrected according to the reference measurements described in Sec-
tion 3.1.10
5.2 Fitting the loop model
5.2.1 The FTW xDSL simulation tool
Parts of the xDSL simulation tool [16], developed at the FTW, Telecommunica-
tions Research Centre Vienna, was used. The calculation of cable insertion loss
was done using the implemented cable models for BT and MAR, using cable
64
parameters presented in ETSI TM6 working group. The function getTwoPort-
Model in [16] was used to calculate the ABCD parameters which then is used
to calculate IL using Equation (2.30) and (2.31).
The data rate in the downstream and upstream direction was calculated
using the function bitloadingRounded ni [16] which is an implementation of
Chows bit loading algorithm, which is described in detail in [17].
5.2.2 Fitting the cable model
To calculate the cable attenuation for various frequencies the parametrized ca-
ble models in the xDSL simulation tool kit was used. The loop is modelled
as a single 0.5 mm PE copper cable. In the lab and eld measurements the
ETSI-ADSL-PE05mar model was used. The cable model was tted to the mea-
surement by minimizing the expression in Equation (4.3) by nding the opti-
mum length and oset. The built in Matlab function fminsearch was used using
a termination tolerance of 10
6
for the distance and oset parameters and a
termination tolerance of 10
12
for the error function.
5.2.3 Fitting the Noise model
The measured and adjusted QLN values was tted to the FEXT model given in
Equation (2.58) using the developed function tFEXT. The input to the function
is a vector containing the center frequencies for the QLN measurements, a vector
containing the measured QLN values, a vector containing the insertion loss for
the cable at the given frequencies, the estimated cable length in meters and a
vector containing the PSD of the disturbers as a function of frequency.
The PSD of the disturbers is assumed to be at. The noise is tted over
the frequencies common to ADSL1/ADSL2/ADSL2+ which is 276 kHz to 1.104
MHz. It is assumed that all disturbers output maximum energy of 20.4 dBm,
which for ADSL2+ corresponds to -42.8 dBm/Hz if all downstream tones is
used (32..512), and for ADSL1/2 corresponds to -40 dBm/Hz, which is the
transmitter output limit.
The coupling constant k
FEXT
is given by the expression
k
FEXT
= 10
1
10N

f
QLN(f)10log10(f
2
d|H(f)|
2
PSD(f))
(5.1)
Where N is the number of measurement values included in the t. Equation
(5.1) basically centres the FEXT model on the measured QLN values. The
calculated coupling constant k
FEXT
together with the distance parameter d is
then used to calculate VDSL2 FEXT based on the VDSL2 transmit PSD using
Equation (2.58), except for the feet to meter conversion factor p.
5.3 Calculating the attainable rate
Based on a selected VDSL2 prole, the attainable rate is calculated using the
bitloadingRounded function in the FTW xDSL Simulator.
The input to the function is listed in table 5.1
The output of the results of Chows rounded bit loading is a vector containing
the number of bits per subcarrier, a vector containing the Energy per subcar-
rier, a vector containing the inverse SNR times the coding gap and a vector
containing the SNR margin for each used subchannel. By summing all bits in
all subcarriers, multiplying with the symbol rate of 4000 baud and adjusting for
the sync word every 257nth symbol the raw data rate in each direction can be
calculated. On top of that, FEC coding overhead has to be taken in account,
but it is not considered in the current implementation of capacity estimation.
65
Table 5.1: Input arguments to bitloadingRounded, from the Matlab source docs
Parameter Description
Hds The channel magnitude response |HLIN(f)|
Nds The noise psd [mW/Hz] 10
QLN(f)/10
E
total
The total energy [mW] normalized for subchannel bandwidth 10
MATPds/10
/f
Gap The eective coding Gap 10
(9.8+SNRMcodingGain)/10
PSDmask The nominal transmit PSD [mW/Hz]
const The constellation sizes used [2..15]
used tones The dmt tones used, depending on prole and RFI notches
66
Chapter 6
Results
6.1 Lab measurements
Telenor has a DSL test lab located in the main oce, where DSLAM equipment
is installed is the basement. From there one multi pair binder containing 5x10
PE05 twisted pairs connects to the modem test lab on the 6Th oor, in the
other end of the building. The length of the cable is about 200 m.
6.1.1 Test set-up
One bundle of 10 pairs out of the 50 pairs binder was connected to an ADSL2+
DSLAM using the Broadcom chipset, no other disturbers was present on the
50 pair binder. In the other end 10 ADSL2+ modems using the Texas chipset
where connected.
6.1.2 QLN measurement
First the QLN in the downstream direction on each one of the 10 pairs was
measured using DELT, all other 9 lines was inactive during each measurement.
The result is shown in Figure 6.1. In the gure, pair number 9 shows a much
higher level of background noise than the other 10 pairs.
Then all 10 lines where activated and the QLN on each line was once again
measured in the downstream direction using DELT. Now measuring the FEXT
coupling from all other active pairs into the measured pair. All lines was con-
gured in ADSL2+ transmission mode with a maximum downstream rate of 24
Mbit, thus enabling all carriers in the downstream direction. The aggregated
output power on all 10 lines was on average 12.5 dBm, min 12.1 dBm, and max
12.9. Thus the output PSD of each downstream transmitter was assumed to
be -50.2 dBm/Hz at, assuming 432 downstream tones is used. The estimated
electrical length of each loop, after tting an ETSI PE05 mar cable is shown
in table 6.1. The oset in dB between the loop model and the measured, and
LATN adjusted, loop attenuation is listed as well.
Also pair number 9 does not stick out as the worst pair regarding FEXT,
even though it showed a much higher level of background noise during the
QLN measurements. The noise level in pair 9 caused by FEXT is higher than
the measured background noise in over all carriers. The reason for the higher
background noise level on pair 9 has not been further investigated, connections
was swapped in both ends to rule out faulty equipment.
67
0 276 552 828 1104 1380 1656 1932 2208
130
125
120
115
110
105
100
95
90
Q
L
N

[
d
B
m
/
H
z
]
Frequency [kHz]
Loop 1
Loop 2
Loop 3
Loop 4
Loop 5
Loop 6
Loop 7
Loop 8
Loop 9
Loop 10
Student Version of MATLAB
Figure 6.1: QLN measurements on 10 pairs in test lab without disturbers
0 276 552 828 1104 1380 1656 1932 2208
130
125
120
115
110
105
100
95
90
Q
L
N

[
d
B
m
/
H
z
]
Frequency [kHz]
FEXT loop 110
FEXT average loop 110
FEXT worst pair: 8
FEXT best pair: 1
Student Version of MATLAB
Figure 6.2: QLN measurements on 10 pairs in test lab with FEXT
68
Table 6.1: Fitted electrical length and HLOG oset from DELT
Loop Length oset
# [m] [dB]
1 227.5 -1.6
2 226.8 -2.6
3 233.4 -2.7
4 196.1 -1.6
5 181.4 -1.8
6 199.9 -2.1
7 177.2 -1.7
8 214.9 -2.5
9 193.3 -2.0
10 195.6 -2.0
6.1.3 VDSL2 estimation
The result of the ADSL2+ DELT measurement for each pair was then used to
estimate the minimum attainable VDSL2 rate, assuming that all 10 disturbers
used VDSL2 transmission. The disturber PSD was calculated using a at level
of -53.7 dBm/Hz, corresponding to a total aggregated output power of 14.5 dBm
over 1536 subcarriers. This corresponds to VDSL2 prole 12a using band plan
998 which was the conguration used on all ports in the DSLAM. In the method
developed for rate estimation in the eld, a much more pessimistic FEXT PSD
is assumed.
6.1.4 VDSL2 measurements
The ADSL2+ DSLAM was disconnected and a VDSL2 DSLAM was connected
to the same binder. 10 VDSL2 modems of the model Shapiron, using a VDSL2
chipset from Inneon, was connected. The trained data rate and max attainable
rate, as well as other performance parameters was extracted. Trained and max
attainable rate was also collected from each VDSL2 modem while the other 9
modems were inactive. The resulting trained bit rate in both scenarios together
with the estimated downstream bit rate based on the ADSL2+ measurements
is shown in Figure 6.3. The results shows a strong correlation between the
estimated downstream rate and the actual trained downstream rate.
6.1.5 Deviations between real and calculated rate
The average dierence between the real and calculated downstream rate is 3.2
Mbit/s. With 4000 symbols per second and 1536 used subcarriers, this corre-
sponds to 0.57 bits of information, or about 1.56 dB in SNR. This is consid-
ered quite low, regarding the possible variation between the measured ADSL2+
FEXT noise and the true VDSL2 FEXT noise.
Unfortunately VDSL2 DELT was not implemented in the lab test equipment.
So when analysing the deviations between the estimated and trained bit rate,
the bit loading was compared. By using an simple network management proto-
col (SNMP) browser, various performance parameters, as well as conguration
settings can be read directly from the DSLAM.
Figure 6.4 shows the dierence in bit loading between the estimation results
and the trained channel for pair 2, the pair with the smallest deviation between
69
1 2 3 4 5 6 7 8 9 10
40
45
50
55
60
65
70
75
80
D
S

r
a
t
e

[
M

b
i
t
/
s
]
Loop #
trained rate, 9 active disturbers
trained rate, no disturbers
estimated rate, 9 active disturbers
Student Version of MATLAB
Figure 6.3: Estimated VDSL2 downstream bit rate and actual trained bit rate
on a 10 pair binder
estimation and true rate and for pair 9 with the largest deviation. The dif-
ference between estimated and trained bit loading in the beginning of the rst
downstream band, DS1, is mainly caused by the transmitter high pass lter.
The dierence in bit loading is then more or less constant from the end of
DS1 and onwards, which on average in DS2 is 2.9 bits/carrier for pair 2 and
3.5 bits/carrier for pair 9. In average the gross data rate on each channel is
14.5% higher than the net data rate, which corresponds to the FEC RS code
(n,k)=(144,128) used, which causes a rate increase of 12.5%. Also the trellis
code uses 1 bit per two 2-D symbols, or 1/2 bit extra per subcarrier.
The actual data rate depends on the net coding gain achieved after channel
training. The conservative coding gain of 3.5 dB used in the estimation resulted
in a calculated data rate which on average is 6% lower than the trained data rate.
The actual coding gain achieved in the tranceivers is proprietary information.
6.2 Field measurements
The estimation method was then tested on real DSL loops in the Telenor access
network. Eight loops was selected, with average attenuation ranging from 6 dB
to 33 dB per subcarrier, representing lengths between 0.5 to 3 km. The results is
shown in Table 6.2 and Table 6.3. Table 6.2 lists the estimated electrical length
of each loop for an ETSI PE05 mar tted model. The oset in dB between the
cable model and the measured loop attenuation is listed as well as the average
error between the tted model and measured attenuation. Also the aggregated
QLN noise power on each line in the ADSL2+ frequency range is listed.
6.2.1 Capacity results
Table 6.3 shows the estimated attainable data rates in both downstream and
upstream when using band plan 998. The estimated capacity is calculated using
70
0 256 512 768 1024 1280 1536 1792 2048 2304 2560
0
5
10
15
b
i
t
s
/
s
u
b
c
a
r
r
i
e
r
subcarrier #
pair: 2
DS trained bitload
DS estimated bitload
0 256 512 768 1024 1280 1536 1792 2048 2304 2560
0
5
10
15
b
i
t
s
/
s
u
b
c
a
r
r
i
e
r
subcarrier #
pair: 9
DS trained bitload
DS estimated bitload
Student Version of MATLAB
Figure 6.4: Estimated downstream bit load vs. trained bit load
two dierent VDSL2 proles. Prole 8b uses frequencies up to 8.5 MHz and a
total aggregated output power of 20.5 dBm in the downstream direction.
Prole 12a uses frequencies up to 12 MHz and a total downstream power
of 14.5 dBm. Prole 8b basically has the same reach performance as ADSL2+
if the upstream band US0 is used. While prole 12b provides a much more
symmetric capacity due to the larger upstream band US2 above 8.5 Mhz
The max value represents the estimated attainable rate if the subscriber is
the only VDSL2 subscriber in the binder, e.g. only the measured FEXT from
existing ADSL2+ disturbers is accounted for. Such scenario is only valid for
early adopters. As the number of VDSL2 users increase, the VDSL2 FEXT will
increase and the attainable rate will be closer to the min value.
6.2.2 Model dependency
For loops longer than 600 meters, FEXT is not the largest limiting factor, which
is also shown in the results as the min and max values do not dier. When FEXT
noise becomes weaker than the background noise, due to loop attenuation, the
attainable rate from that frequency and onwards is entirely determined by the
loop attenuation. As shown in section 4.1.3, the error in the modelled loop
depends on the electrical length of the loop. For a 1300 m long loop, an average
attenuation estimation error of 5.5 dB can be expected. This corresponds to
almost 2 bits/subcarrier. The estimated FEXT noise on Loop #4 hits the noise
oor of -140 dBm/Hz at 3 Mhz. Which means that the SNR, and capacity
of the upper end of DS1 and the whole DS2 band is determined by the loop
attenuation. For the DS2 band this 5.5 dB error corresponds to a deviation in
the maximum downstream rate of 6 Mbit.
71
Table 6.2: Electrical loop length, model oset, average error and total noise
power
loop length oset avg err QLN
# [m] [dB] [dB] [dBm]
1 170 -5.3 0.28 -85.8
2 423 -5.3 0.20 -85.0
3 595 -4.8 0.21 -89.5
4 1339 5.1 0.17 -90.5
5 1283 -2.7 0.26 -92.5
6 1723 -2.0 0.21 -96.4
7 2933 6.8 0.36 -101.8
8 3039 8.2 0.45 -107.5
Table 6.3: Estimated VDSL2 capacity on the 8 loops
8b 12a
loop DS US DS US
# min max min max min max min max
1 74.1 92.5 16.6 21.6 71.0 89.5 45.5 70.1
2 64.5 87.9 14.6 21.6 61.0 84.4 38.2 62.2
3 69.0 80.4 15.4 19.5 65.6 77.0 41.2 49.0
4 45.6 46.1 10.1 10.1 42.2 42.7 10.5 10.5
5 36.2 36.6 7.7 7.7 32.3 32.7 7.7 7.7
6 23.4 23.4 2.2 2.2 20.2 20.2 2.2 2.2
7 13.2 13.2 1.5 1.5 10.7 10.7 1.5 1.5
8 14.1 14.1 1.6 1.6 11.9 11.9 1.6 1.6
72
Chapter 7
Conclusions
The method of using ADSL2+ DELT for loop characterization by tting a
theoretical model of the local loop to estimate signal attenuation and noise
prole has been evaluated and tested. The lab tests has shown that for short
loops, where FEXT is the dominating noise source, VDSL2 capacity estimation
based on ADSL2+ DELT shows a strong correlation to performance results
using real VDSL2 transmission. Field tests has indicated that for loops longer
than 600 m, loop attenuation gradually becomes the limiting factor for capacity.
7.1 Further work
During this project a lot of time was spent on analysing DELT measurements
and studying the ADSL2/ADSL2+ standards to gure out what was expected
from the measurement result. Also time was spent on experimenting with in-
sertion of known loop impairments such as load resistors and analysing the
measurements. Also the lab environment did not have any measurement equip-
ment such as oscilloscopes, spectrum analyzer or network analyser. The only
tools available was a noise injector, waveform generator and a cable simulator
which was not used.
Even if a spectrum analyser and a network analyser would be useful, the
goal was to nd a method to perform loop characterization using the existing
devices in the network. But to be able to qualify the results of DELT, reference
measurements using precision instruments is necessary.
7.1.1 Improving the DELT results
DELT measurements on various test loops shows a high level of precision. Re-
peated measurements during the same conditions show identical results. By
using automated test procedures, multiple reference measurements on modems
from dierent models and batches could be performed to derive a more exact
estimation of the DSLAM-Loop-Modem signal chain and eliminate the eects
of the DSLAM and Modem from the loop measurement. But that requires ref-
erence measurements are made using precision instruments, as discussed in the
previous section.
7.1.2 VDSL2 DELT
DELT is was not implemented in the current version of the rmware in the
VDSL2 equipment used during the tests. Therefore VDSL2 DELT measure-
ments has not been possible. To verify the capacity estimation method for
73
VDSL2, reference measurements using VDSL2 DELT must be performed. The
VDSL2 DSLAM-loop-Modem signal chain must be analysed ujsing DELT and
compared with the ADSL2+ DSLAM-Loop-Modem signal chain. Based upon
the correlation between the ADSL2+ DELT, loop model, and VDSL2 DELT
and VDSL2 line rate, a more correct estimation model can be derived.
7.1.3 Measuring background noise
Multiple VDSL2 DELT measurements in the eld is necessary to determine the
background noise PSD for long loops and medium loops. On those loops, FEXT
noise is attenuated due to the length of the loop and background noise dominates
in higher frequencies. The statistical properties of the assumed background noise
level of -140 dBm/Hz must be veried by measurements of many dierent loop
conditions. It is also necessary to identify RFI and the statistical properties
regarding loop sensitivity to various known radio transmitters operating in the
VDSL2 spectrum.
7.1.4 Estimating the loop topology
The estimation method uses the worst case assumption that all ADSL2+ dis-
turbers become VDSL2 disturbers. That is obviously not the case. As eld
measurements show, loops longer than 1500 meters will not give any increased
data rate with VDSL2 over ADSL2+. Therefore not all ADSL2/ADSL2+ dis-
turbers will become VDSL2 disturbers. As presented in [14], the average loop
length in the Swedish copper network is about 1.5 km. The loop topology will
dier between COs located in densely populated areas and COs in rural areas.
By measuring all loops in the CO, loop length statistics for binders connected
to that CO will give a better estimate on the number of possible VDSL2 dis-
turbers. Also by using the street address to identify co located subscribers,
combined with noise pattern matching, it might be possible to identify the cop-
per pairs that share the same binder.
74
Bibliography
[1] Proakis John G., Salehi Masoud, Communication Systems Engineering.
Prentice Hall, 2nd Edition, 2002.
[2] Starr Thomas, Cio John M., Silverman Peter J. Digital Subscriber Line
Technology. Prentice Hall, 1999.
[3] Chen Walter Y. DSL Simulation Techniques and Standards Development for
Digital Subscriber Line Systems. Mac Millian, 1998
[4] Philip Golden, Herve Dedieu, Krista S. Jacobsen Fundamentals of DSL Tech-
nology Auerbach Publications, 2006
[5] Cheng, David K. Field and Wave Electromagnetics. Addison Wesley, 2nd
Edition, 1989.
[6] van den Brink, Rob F.M. Cable reference models for simulating metallic
access networks. ETSI-TM6 Permanent document TM6(97)02 revision 2,
Verona, Italy, 1997
[7] Musson, Joseph. Maximum Likelihood estimation of the primary parameters
of twisted pair cables. ETSI-TM6 contribution TD8, Madrid, Spain, 1998
[8] Heylen Luc, Musson Joseph. Cable Models predict physically impossible be-
haviour in time domain ETSI-TM6 basis for discussion, Sparnex N.V., 1999
[9] ITU Test procedures for digital subscriber line (DSL) transceivers ITU stan-
dard G.996.1, 2001
[10] ITU Asymmetric digital subscriber line transceivers 2 (ADSL2) ITU stan-
dard G.992.3, 2005
[11] ITU Asymmetric Digital Subscriber Line (ADSL) transceivers - Extended
bandwidth ADSL2 (ADSL2+) ITU standard G.992.5, 2005
[12] ITU Handshake procedures for digital subscriber line (DSL) transceivers
ITU standard G.994.1, 2003
[13] ETSI Transmission and Multiplexing (TM); Access transmission systems
on metallic access cables; Very high speed Digital Subscriber Line (VDSL);
Part 1: Functional requirements, ETSI Technical Specication TS 101 270-1
V2.0.0 (2002-04), 2002
[14] Thomas Magesacher, Per

Odling, Per Ola Borjesson. Capacity of the
Swedish Copper Access Network. Proc. Radio Vetenskap 2005, 2005
[15] Tomas Nordstrom, Steen Trautmann. Proposal for Parametric ADSL Ca-
ble Models. ETSI STC TM6 working group, 2002
75
[16] Tomas Nordstrom, Daniel Bengtsson. FTW xDSL simulation tool Version
3.0beta1, 2003 <http://www.xdsl.ftw.at/xdslsimu/>
[17] Cio John M. EE379C Course notes, Multi-channel Modulation Stanford
University, Stanford, CA. <http://www.stanford.edu/class/ee379c/>
76