SCM (QAM) VDSL
for
EFM Copper 10PASS- TS
Ottawa June 2003
Page 1
Presentation to IEEE 802.3ah EFM TF June 2003
Supporters
!
Page 2
Presentation to IEEE 802.3ah EFM TF June 2003
Introduction
! Enable an Ethernet-like model for Access
– Low cost transceivers through interoperability and
standardized products
– High volume
– Low complexity
– QAM has a great record on all these points
! QAM VDSL has > 2.5 million lines installed
worldwide
– Over 5% are deployed to Single Family Units, some
estimates up to ~10%
– Used for Data and Video and Data applications
! QAM VDSL is deployed in over 20 countries
Page 3
Presentation to IEEE 802.3ah EFM TF June 2003
What is QAM (SCM) VDSL?
! QAM: Quadrature Amplitude Modulation
! Part of the SCM (Single Carrier Modulation)
family, which includes PAM (SHDSL)
! QAM is a generalization of PAM, modulated
to higher frequencies
! QAM/PAM is broadly used in many
communication channels:
– Cable modems
– Ethernet 1000BaseT
– Other DSL: HDSL, SDSL, SHDSL, RADSL
– HPNA
Page 4
Presentation to IEEE 802.3ah EFM TF June 2003
PHY Sources and QAM Facts
! The following chipset vendors have QAM silicon
developed
– Infineon
– Metalink
– ElectriPHY
– Many others have QAM technology
! Single carrier is well known in the industry
– All standard voice band modems
– Cable Modems
– Home PNA 3.0
– Ethernet – 1000 Base-T
! QAM VDSL highly successful with Ethernet
! If QAM is chosen for EFM many companies could rely on
existing SCM/QAM knowledge to step into the market
Page 5
Presentation to IEEE 802.3ah EFM TF June 2003
QAM VDSL has a large partner base
The following companies publicly endorsed QAM at FS-VDSL
•
Actelis
•
GoDigital Networks
•
Stellar One
•
ADC Communications
•
Iskratel
•
Tellabs
•
BATM/Telco Systems
•
MRV Communication
•
Telson
•
Cisco
•
Motorola / Next Level
•
Tut/VideoTele.com
•
Corecess
•
Paradyne
•
UTStarcom
•
ECI/Inovia
•
Salira
•
Wistron Nexus
•
Extreme Networks
•
Siemens
In addition, the following companies have publicly
announced the availability of QAM VDSL systems
Page 6
•
Hitachi Communications •
Huawei Technologies •
Samsung
•
Eastel System Corp.
•
Lucent
•
Tellion Inc.
•
Gigalink CO LTD
•
Locus Corp.
•
ZyXEL
•
Hana System Inc.
•
NeoWave Inc.
•
Hyundai Network Inc
Presentation to IEEE 802.3ah EFM TF June 2003
QAM VDSL has a large deployed base in the Access space –
QAM is the only veteran First Mile VDSL under consideration
Korea:
China:
Canada:
" VDSL deployments by
Bell Canada & MTS
" VDSL deployment by
regional operators
Belgium:
" Belgacom trial
Tender 3k lines
" China Draft VDSL
standard is released
" 200k VDSL lines in
the field, potential
10M VDSL lines
(Olympic Games
2008)
" KT has more than
1.5 million VDSL
users, over 300K
with 25/3 service
" Broadband Internet
subscribers
surpassed 10M in
Nov-2002
USA:
" VDSL service by
Qwest & IOCs
" > 500k homes
passed
Japan:
" Market requires
52MBit VDSL
(500m)
" Pre-standard
deployment of
about 200K lines
" VDSL tenders from
Softbank BB, KDDI
and NTT all
expected in 2003
Hong Kong:
Note: QAM VDSL deployments as LAN
extension not shown
Page 7
Presentation to IEEE 802.3ah EFM TF June 2003
" Deployed already 125k
VDSL lines
(Hutchinson)
" PCCW Hong Kong
Telecom, New World
Telecom, and Wharf
New T & T announced
to deploy VDSL
Taiwan:
" Chung Hwa Telecom
already deployed 5K
VDSL lines and further
plans to deploy another
8K VDSL Lines in 2003
System Provider’s perspective:
Full Service Access QAM VDSL in NAFTA
! Field Deployment
– >140,000 subscribers with pre- & standard Plan 998
!
>70,000 as Single Family Units (SFUs)
– 2, 3, and 4-band solutions
– Asymmetric data rates mostly
– High percentage have video services
– Broad MDU/MTU deployments
! QAM report card
– Low power consumption means smaller cabinets, low heat
dissipation requirements, low OPEX
– Higher density equates to low CAPEX
– Impulse noise immunity of QAM avoids a serious problem in
homes. DMT is an unknown for VDSL & video versus impulse
noise, no matter what a lab test may show
Page 8
Presentation to IEEE 802.3ah EFM TF June 2003
System Provider’s Perspective:
Ethernet over QAM VDSL
!
QAM VDSL 2, 3, 4-band as 10/100BaseT extension
!
MxU environment
!
Requirements
– Performance & Cost
– Stability & Robustness
– Symmetric rates delivery
– Low Power & High Density
!
Experience with QAM
– No DSL experience necessary
– PSD shaping
– Impulse noise
– Advanced blind equalization
– Low power, low cost
! QAM delivered as promised
Page 9
Presentation to IEEE 802.3ah EFM TF June 2003
QAM has lower complexity/cost
! Lower complexity and relative cost of chipset
– Cost is a function of die size, volume, and process
– QAM VDSL wins in all three areas
! Lower power consumption (<1w per port for
singles and multi-port chipsets in 2003)
! Lower complexity of the analog part (even with
implemented digital duplexing)
! EFM has two objectives, based on
SHDSL(PAM) and on VDSL
– Lower complexity is also important for a combined
SHDSL + VDSL EFM chipset
Page 10
Presentation to IEEE 802.3ah EFM TF June 2003
How does the Transceiver cost affect
equipment cost?
Transceiver cost includes everything between tip/ring and
Utopia/MII
! ADSL CPE
– Today, transceiver w/ integrated network processor is >
50% of total Bill of Material cost
! ADSL DSLAM
– Transceiver price is approximately 30-40% of total BOM
! QAM VDSL Ethernet Switch
– Ethernet components are commoditized, VDSL transceiver
costs make up >60% of BOM
! QAM VDSL Ethernet CPE
– Transceiver w/integrated bridge is about 50% of total BOM
cost
! Trend is the same in IP and ATM
Transceiver cost still has a big impact on equipment cost
Page 11
Presentation to IEEE 802.3ah EFM TF June 2003
QAM VDSL Port density chip-sets
! QAM VDSL chip-set are 4th and 5th generation
! QAM VDSL chip vendors today offer various
types of chip sets
– Single port chip set
– Quad port chip set
– Octal port chip set
– Single port in a single package
!
Page 12
Digital pump, AFE, Filterless, Line driver
Presentation to IEEE 802.3ah EFM TF June 2003
Interleaver
Interleaver Framer Modulator
Framer Modulator
RS-Encoder
RS-Encoder
DAC
DAC
Power
Power
Control
Control
Post
Post
Filter
Filter
SRAM
SRAM
DeInterleaver
DeInterleaver
Deframer Demodulator
Demodulator
RS-Decoder
RS-Decoder Deframer
ADC
ADC
Variable
PrePre- Variable
Gain
Gain
Filter
Filter Amplifier
Amplifier
Line
Driver
Transformer
Transformer
UTOPIA // xMII
xMII
UTOPIA
Where DMT and QAM implementations
differ in size/cost?
Analog
Analog Front
Front End
End Control
Control
Timing
recovery
Timing recovery
Controller
Controller ++ Host
Host Interface
Interface
Digital data pump
Analog Front End
! A basic VDSL transceiver has a digital chip, and analog
chip (AFE), a line driver and a transformer
! The dotted segments highlight the functions that differ
the most in size/complexity between QAM and DMT
! The transformer, framer and data interface similar in size
Page 13
Presentation to IEEE 802.3ah EFM TF June 2003
Analog part (AFE and Line Driver)
! Both QAM & DMT use digital duplexing to meet flexible
bandplan requirements
! Basic ADC & DAC requirements are equivalent, except…
– DMT has Higher Peak-to-Average ratio
!
1 more bit is needed to compensate for higher PAR (each ADC bit
equals 6dB)
!
The impact in area and power consumption on ADC is a factor of x2
– Impact on AFE
!
Extra power -> 30%
!
Extra size
!
Advanced AFE design can lower the extra DMT power and size to
about 10 to 15%.
-> 30%
– Impact on Line Driver
Page 14
!
Extra power in the line driver ->40% (= higher power rails)
!
PAR reduction techniques can be used but will have capacity penalty
Presentation to IEEE 802.3ah EFM TF June 2003
Digital part (Data pump)
! ADSL2+ (512 tones) chip vs. a QAM VDSL chip
Same size,
Yet,
QAM modem
smaller than
DMT
ADSL2+
4-band QAM VDSL
(note: a complete chip includes a QAM or DMT specific PMD and a TPS-TC part which is line
code independent. The ADSL2+ chip shown has an ATM-only TPS-TC, the QAM VDSL chip
shown has an ATM/Ethernet Bridge TPS-TC. If we compare the QAM & DMT specific PMD,
the 4-band VDSL QAM is 30 % smaller than the ADSL2+ PMD)
What about 2048/4096 tones vs. 4-band QAM VDSL?
Page 15
Presentation to IEEE 802.3ah EFM TF June 2003
QAM VDSL Interoperability
! QAM VDSL vendors are continuously working
together in order to guarantee full
interoperability between their chip sets
! Ethernet interoperability of QAM VDSL chip
sets already demonstrated July 2002 in
Vancouver
! QAM VDSL chip vendors committed to a
simple definition of QAM VDSL
– See the QAM Spec – compare it to Part 3 of T1.424
for DMT and see the difference
– Guarantee ease of use
– Plug&Play concept
Page 16
Presentation to IEEE 802.3ah EFM TF June 2003
SNR Averaging
! QAM VDSL uses its SNR Averaging capability to get optimal usage
of the VDSL bands, with almost no loss of SNR
! QAM requires smaller constellations (= lower power)
Effective DMT
bandwidth
•
Effective QAM bandwidth
See ref [1]
Minimal SNR level •
Signal level
Noise level
Frequency
Page 17
Presentation to IEEE 802.3ah EFM TF June 2003
DMT does not use the
signal in this area. QAM
provides a higher
bandwidth!
DMT requires high
constellation for the
highest SNR region
Power Consumption
! Fact: QAM VDSL chipsets have small die
sizes and low resolution ADC and DAC
equating to low power consumption
! Fact: QAM vendors now supplying singles
at less than 1 W/port and multi-port at even
lower watts/port for all components needed
in a design
Page 18
Presentation to IEEE 802.3ah EFM TF June 2003
Field Proven Performance: QAM has
same or better performance
Field results from Korea - May 2003
60
50
40
Mbps 30
QAM VDSL
DMT VDSL
20
10
0
Conditions:
300 meter300 meter
DOWN
UP
1000
meter
DOWN
1000
meter UP
•
Real cable 300m, CEPV 0.5 mm, bundle 50 lines, 24 modems in parallel, plan 998, start at 138 KHz
•
Real cable 1000m, CEPV 0.5 mm, bundle 25 lines, 24 modems in parallel, plan 998, start at 138 KHz
Note: DMT system did not provide more than 7Mbps upstream during those trials
Silicon with pricing constraints is very different from a demo silicon
That is why system vendors, having tested both, stand behind QAM
Page 19
Presentation to IEEE 802.3ah EFM TF June 2003
QAM Met the Short Reach Performance
Objective in lab tests… did anyone else?
Reach [ft.]
Optional Testing - 10 M Symmetric @ 24 AWG
3500
3000
2500
2000
1500
1000
500
0
750 m = 2460 ft.
1
2
3
3F
Test #
Page 20
Presentation to IEEE 802.3ah EFM TF June 2003
5
6
Performance at High Bit-Rates
! At high and symmetric data rates as required by EFM, extra
RS coding and low bit constellations (the Alcatel/STM
Gambit) are not applicable as extra BW not available
! These rates are important for MxU deployments worldwide
and are of special interest in the Asia-Pacific region
The following results were tested at Telcordia optional testing:
– QAM: 40/25 Mbps
Page 21
20 self FEXT
1300 ft 26 AWG
– DMT: 39.6/20.1 Mbps 20 self FEXT
1000 ft 24 AWG
Presentation to IEEE 802.3ah EFM TF June 2003
VDSL QAM future:
!QAM VDSL is a vital technology. Advances are
achieved continuously:
– True broadcast P2MP
!
Blind equalization and blind acquisition makes it simple
!
High symbol rate improves U/S bursting
– Vectoring and alien noise cancellation (MIMO)
– Matrix equalization, straightforward on SCM
– Well known and widely used (1G/10G Ethernet and others)
– DSM
!
Compatible for layers 0-3
!
Coordination technologies, similar to IWF and alternative are available
– Combined 10PASS-TS and 2BASS-TL in a single QAM/PAM
SCM EFM PHY
Page 22
Presentation to IEEE 802.3ah EFM TF June 2003
QAM is committed to the EFM vision &
schedule
! QAM VDSL is as simple as Ethernet
! The market has chosen QAM as the de-facto
solution
! Korea and Japan moved to Ethernet over
VDSL for their next generation access
networks. Others are following.
! The vision of Ethernet in the access is
prevailing. The simplicity and performance of
QAM meet that vision
! EFM can move ahead
Page 23
Presentation to IEEE 802.3ah EFM TF June 2003
Additional slides
Page 24
Presentation to IEEE 802.3ah EFM TF June 2003
Page 25
Presentation to IEEE 802.3ah EFM TF June 2003
“S” DMT- 16/1 US PSD in Configurations
8&9
! 2.5 MHz used to
transmit 1 Mbps
! Spectral efficiency of
0.4 bits/Hz
! Optional RS coding
with high overhead
need to be used
! Use of optional mode
should have been
reported!
Page 26
Presentation to IEEE 802.3ah EFM TF June 2003
“I” DMT - 16/1 US PSD configuration 8
! 1.5 MHz to transmit 1 Mbps
! Spectral efficiency of 0.66 bits/Hz
Page 27
Presentation to IEEE 802.3ah EFM TF June 2003
QAM Configuration 8 US1 PSD mask
! 0.7 MHz BW used to
transmit 1 Mbps
! Spectral efficiency of
1.4 bits/Hz
! Mandatory RS
coding used
! Field experience shows that in full binder high spectral
efficiency is
a better approach !
!This is the foundation of DSM !
Page 28
Presentation to IEEE 802.3ah EFM TF June 2003
“S” DMT- 22/3 Configuration 10 US PSD
! 6 MHz BW is used to
transmit 3 Mbps
! Spectral efficiency is 0.5
bit/Hz
! Extra BW is used for
coding using short code
words
! This is an optional mode.
! Use of optional mode
should have been reported!
! Coding gain is
UNRELATED to line code.
QAM can use short code
words as well.
Page 29
Presentation to IEEE 802.3ah EFM TF June 2003
QAM PSD for the Same Configuration
! 1.2 MHz BW used to
transmit 3 Mbps
! Efficiency is 2.5 bits/Hz
!
Page 30
Presentation to IEEE 802.3ah EFM TF June 2003
Out of Band NEXT – What is the Issue ?
! During the testing we have discovered that the high
level of out of band next changes the statistical
behavior of the noise.
! Out of band next is received out of the RECEIVER
in-band.
! It does not affect capacity !
! To demonstrate this issue we have repeated testing
with reduced power level of out of band NEXT
! 3 Test points have been tested:
– Test case 13 (16/1) - 3400 ft. - 550 ft. improvement!
– Test case 14 (16/1) – 3400 ft. – 550 ft. improvement!
– Test case
Improvement!
17 (22/3) - 2850 ft. – 300 ft.
! These results matches DMT performance on these
cases
Page 31
Presentation to IEEE 802.3ah EFM TF June 2003
VDSL NEXT PSD – VTU-C
US Self NEXT FTTCab M1 Unnotched
-80
-100
dBm/Hz
-120
-140
-160
Out Of Band NEXT
-180
-200
Page 32
0
Presentation to IEEE 802.3ah EFM TF June 2003
0.5
1
1.5
Frequency
2
2.5
3
7
x 10
Noise Generated Eliminating OOB NEXT
Effect
Page 33
Presentation to IEEE 802.3ah EFM TF June 2003
Comparative performance of QAM and
DMT
Testing without OOB
4000
3500
3000
2500
2000
1500
1000
500
Metalink
Page 34
ikanos
Presentation to IEEE 802.3ah EFM TF June 2003
st
34
31
28
25
22
19
16
13
10
7
4
1
0
Metalink No OOB
Size comparison: The figure speaks for itself
A QAM VDSL PHY: Dig. pump + AFE +
LD
Page 35
Presentation to IEEE 802.3ah EFM TF June 2003
QAM VDSL Notching and spectral
shaping
! Notching in QAM is performed with an IIR filter.
Different filter order and implementation will lead to
different notch shapes.
Example of an implemented notch
Example of an implementable notch
! QAM deployment so far have shown that QAM notching works
! Olympics results demonstrate that, even with notching, the capacity of
QAM systems is close to theoretical capacity
Page 36
Presentation to IEEE 802.3ah EFM TF June 2003
Blind Acquisition/Handshake
To establish a link:
! QAM VDSL uses Default link parameters
! QAM VDSL includes an embedded
message channel and a state machine.
! DMT, in the contrary, performs a training
procedure of the system, and a Handshake
process, which result in a very long cycle
for link establishment.
Page 37
Presentation to IEEE 802.3ah EFM TF June 2003
T1E1 Olympics:
Line-Code or Implementation Evaluation
! EFM claimed: will give “Due Weight” to T1E1 Olympic results
! T1E1 Olympic compared rate/reach only. Other EFM criteria were
not tested.
! The results demonstrate the effect of implementation on
performance
! The results do not demonstrate the effect of line code on
performance
! Specifically, BTexact test results demonstrate strongly that:
– “M” QAM implementation significantly better than “I” QAM
implementation
– “S” DMT implementation significantly better than “I” DMT
implementation
– “M” QAM and “I” DMT performance is similar
! Different implementations affected strongest at low bit rate
scenarios
Page 38
Presentation to IEEE 802.3ah EFM TF June 2003
“M” QAM vs. “I” DMT performance in
BTexact
! Difference is an average of 3%
QAM vs. IKANOS %, BT.
Positive is DMT better
30%
20%
10%
0%
-10%
-20%
-30%
1
2
3
4
5
6
7
8
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
Test No.
Page 39
Presentation to IEEE 802.3ah EFM TF June 2003
“I” DMT vs. “S” DMT performance in
BTexact
! Difference is an average –416 feet
Ikanos vs. ST Micro, BT. Minus means ST Micro better
0
-100
-200
-300
-400
-500
-600
-700
-800
1
2
3
4
5
6
7
8
9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
Test No.
Page 40
Presentation to IEEE 802.3ah EFM TF June 2003
“M” QAM vs. “I” DMT: Symmetric
service test results
olym pics s ym m e tric rate re s ults
QAM
DMT
2000
1800
1600
1400
Reach
1200
1000
800
600
400
200
10/10Mbps
Page 41
Presentation to IEEE 802.3ah EFM TF June 2003
(7
)1
3/
13
P
1L2
N
0
(8
)1
3/
13
P
1L3
N
(9
1
)1
3/
13
P
2L1
N
(1
3
0)
13
/
1
3
P
3L3
(1
N
1)
1
13
/
1
3
P
4L3
(1
N
2)
3
13
/1
3
P
4L1
N
1
(1
)1
0/
10
P
1L1
N
(2
1
)1
0/
10
P
2L1
N
(3
0
)1
0/
10
P
2L1
N
1
(4
)1
0/
10
P
2L1
N
(5
3
)1
0/
10
P
2L2
N
0
(6
)1
0/
10
P
3L1
N
0
0
Test number
13/13Mbps
How does Implementation Affect
Performance?
! Performance is affected by:
– Spectral utilization (transmit signal PSD)
– Coding Gain
! Both are line-code independent
! In some cases QAM and DMT implementation for
Olympics were very similar
– Particularly, 10/10 and 13/13 cases utilized entire 4band spectrum in all implementations resulting in
similar performance results
! In low bit rate cases, particularly 16/1 and 22/3,
different spectral utilization, probably due to coding
implementations, created performance gaps
! These implementations are applicable to both SCM and
MCM
Page 42
Presentation to IEEE 802.3ah EFM TF June 2003
What was Different Between the
Results?
! The results differ in cases where low speeds are in
use:
– Specifically 16/1 and 22/3 cases
! The difference in implementation in these low bit
rates were:
– Low spectral efficiency of less than 2 bit/Hz (and even 1
bit/Hz)
– Coding gain achieved by high-redundancy RS coding
! Both QAM chip-sets showed:
– Higher spectral efficiency of 2 bits/Hz using QAM-4
constellation
– Mandatory coding using the RS(200,16) code
! Both DMT chip-sets showed:
– Spectral utilization as low as 0.5 bits/Hz
Page 43
– This low spectral utilization, might be a result of using
optional RS coding with higher redundancy, yields
Presentation to IEEE 802.3ah EFM TF June 2003
higher coding gain and additional burst immunity
Applicability to QAM VDSL
! QAM-2 support (as defined in EFM spec)
– Both QAM vendors are ready for re-testing in the
Olympics labs
– Preliminary lab testing shows 300 ft. improvement in
all relevant cases
! Higher redundancy RS coding
– Provides additional coding gain
– Provide immunity against short and frequent impulses
– Such impulses were observed particularly in the
implementation of Out of Band NEXT in noise
generators
– Will be demonstrated off-line by Q3 of this year!
– Implemented and ready for testing by Q4 of this year!
– This will increase performance by additional 300 ft.
Page 44
Presentation to IEEE 802.3ah EFM TF June 2003
The gain of high redundancy RS codes
Page 45
Presentation to IEEE 802.3ah EFM TF June 2003
Olympic Conclusions
! Olympics results highlight the difference of
performance with different implementations
– None of these are line-code dependent
– Low spectral efficiency and high redundancy RS
coding improves performance in low bit rates
scenarios
! Both methods are applicable to both line-codes
! The Information provided by the results is not
sufficient for a line-code decision
– Points differences in implementation which are linecode independent
! EFM Line code decision based only on T1E1
results, is dangerous.
– power consumption ?
– Footprint ?
– Complexity ?
Page 46
– Cost ?
Presentation to IEEE 802.3ah EFM TF June 2003
QAM developments over the last 4 years
! Filter-less – no duplexing filters, flexible band plans
! Multi-Band – From fixed 2 band to flexible 4 band
! Support for band 0
! 12 bit constellation, QAM-4096
! Lower excess bandwidth: 10% only
! Ham radio notching
! Advanced blind equalization
– Simple, fast, automatic and autonomous – Ethernet
like
! Rate adaptive solutions for the field
! Low power, low cost – Ethernet based on QAM
VDSL links provide better price/performance than
ADSL!!!
! Multi-port solutions
Page 47
! Single chip solutions
Presentation to IEEE 802.3ah EFM TF June 2003
Digitally Duplexed QAM
Rext
LINE
RXb_P
Rext
RXa_P
Rext
RXa_N
RXb_N
Rext
line
adaptation
AGC
splitter
hybrid
trafo
TX_P
POCO
(tx filter ?)
LD
TX_N
AFE
! 1st gen. QAM had discrete filters
– Low cost and performed well
! Current gen. QAM from multiple
vendors implement digital duplexing
today.
! Band plans configurable by software
– 998, 997, China, Private, FX …
! Frequency agility is as feasible with QAM
Page 48
as with DMT
Presentation to IEEE 802.3ah EFM TF June 2003
Myth: QAM is vulnerable by RFI ingress
Reality: RFI ingress is tracked in real
time
! When RFI ingress appears, the DFE tracks the RFI
frequency and notches it out
DFE
RFI
s(t)
SAMPLING
1/T’
Pass-band
FILTER
FEEDBACK
FILTER
SAMPLING
1/T
FEED-FORWARD
FILTER F(f)
A/D
NO RFI
S(f)
0
Page 49
SLICER
F(f)
FREQUENCY
f
0
Presentation to IEEE 802.3ah EFM TF June 2003
FOLDED
SPECTRUM
FREQUENCY
f
0
1/T
f
Myth: QAM VDSL operates over short loops only
Reality: QAM VDSL can operate over long loops
! The basic principle of operation is to use US0
– upstream in the VDSL Band 0
– downstream in the VDSL Band D1
! How can a QAM accomplish this?
# Band allocation and symbol rates are
programmable with granularity of 16.875 kHz
and 33.75 kBaud, respectively
# Filterless operation allows PSD scaling
# High constellations (up to 12 and higher)
Page 50
Presentation to IEEE 802.3ah EFM TF June 2003
Detailed analysis: QAM vs. DMT VDSL
complexity
– QAM
!
Using Digital Duplexing
!
Digital portion consists of
– Constellation Mapper
– Transmit Filtering
– Receive Filtering
– DFE
– Three factors define a silicon design. Let’s assign
them variables to compare QAM with DMT
Size of computational units:
X
Size of memory:
Y
Total Die size:
Page 51
Presentation to IEEE 802.3ah EFM TF June 2003
Z
4-band DMT VDSL Digital Complexity
analysis
DMT
!
A DMT modem consists of N x Tone/Constellation Mappers
and Trellis coding that feeds an IFFT. This constitutes of
20% of the modem.
!
The signal passes through bit loading, clipping and PAR
Reduction functions
!
Transmit Filtering/Windowing reduces interference to the
receive path
!
Rx Filtering, Windowing and (possibly) Time domain
equalization (TEQ) is performed on the received symbols
!
FFT and tone equalization is performed according to the bit
loading algorithm.
!
A dedicated DSP and memory performs SNR margin
calculations, bit loading and coefficient calculations for the
equalizers
Size of computational units:
complexity!)
Page 52
Size of memory:
4Y
Presentation to IEEE 802.3ah EFM TF June 2003
3Z
Total Die size:
0.6X (DMT is lower
(But DMT is 3 times the
Complexity/Cost Summary: DMT vs
QAM
! Analog
– AFE and line driver -> bigger for DMT vs QAM
– AFE and line driver -> more power for DMT vs QAM
– AFE and line driver not affected by Moore’s law
! Digital
– DMT has lower computational unit complexity (0.6X)
– DMT has higher memory requirements (4Y)
– DMT Die size is 3 times larger than QAM (3Z)
! What about Moore’s law – won’t it fix things?
– Both QAM and DMT benefit from Moore’s law
!
Memory doubles density every 18 months
!
Computations double density and speed every 18 months
– Redesigns could get a 4x improvement every 18 months
Page 53
– DMT improves slower because its size is limited by
memory,
not
computations
Presentation to
IEEE 802.3ah EFM
TF June
2003
Examine the barrage of DMT myths
! The complexity of DMT VDSL is comparable to SCM VDSL
– Not TRUE
Die Size of MCM 3 times die size of SCM
! DMT VDSL is a straightforward extension of ADSL DMT
– Not TRUE
requirements
VDSL and ADSL have different AFE
! The cost of a DMT VDSL modem is lower than a QAM modem
– Not TRUE
The die size is proportional to cost
2-band QAM VDSL is the same cost as
ADSL
! DMT VDSL will allow better development efficiencies
– Not TRUE
Multi-mode chips (PAM/DMT) exist today
! QAM VDSL is not flexible enough to deal with DSM, M2DSL,
etc…
– Not TRUE
QAM can do digital duplexing and all
required shaping
Page 54
! Moore’s law plays in DMT’s favor
Presentation to IEEE 802.3ah EFM TF June 2003
Not TRUE
Analog part is bigger for DMT and is not
References
Page 55
[1]
Gottfried Ungerboeck, Broadcom corp., “A
spectral efficiency advantages of SCM over MCM”,
document FI-083, ITU, Fiji Island, jan-Feb 2000.
[2]
Cioffi, Dudevoir, Eyuboglu, & Forney, “MMSE
Decision-Feedback Equalizers and Coding - Part I:
Equalization Results,” IEEE Transactions on
Communications, vol. 43, no. 10, October 1995.
[3]
B. Saltzberg, “Comparison of Single-Carrier
and Multitone Digital Modulation for ADSL
Applications,” IEEE Comm Magazine, Nov. ‘98.
[4]
L. Vandendorpe, “Asymptotic Performance of
MMSE MIMO Decision-Feedback Equalization for
Uncoded Single Carrier and Multicarrier
Modulations,” IEEE International Conference on
Communications 1998, Atlanta, GA, June 1998.
Presentation to IEEE 802.3ah EFM TF June 2003