`
`CSCO-1028
`
`Cisco V. TQ Delta, IPR2016-01021
`
`Page 1 of 19
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 1 of 19
`
`

`

`
`
`CSCO-1028
`
`Cisco V. TQ Delta, IPR2016-01021
`
`Page 2 of 19
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 2 of 19
`
`

`

`
`
`
`
`Jfliflflfllflflflmfl Muifluflm
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 3 of 19
`
`

`

`Understanding Digital
`Subscriber Line Technology
`
`‘
`
`Thomas Starr
`
`Senior MTS
`
`Ameritech
`
`John M. Cioffi
`
`Professor of Electrical Engineering
`
`Stanford University
`
`Prentice Hall PTR, Upper Saddle River, NJ 07458
`http:/IWWW.phptr.com
`
`Peter Silverman
`
`Senior Architect — New Business Initiatives
`
`3COM Corporation
`
`
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 4 of 19
`
`

`

`
`
`CSCO-1028
`
`Cisco V. TQ Delta, IPR2016-01021
`
`Page 5 of 19
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 5 of 19
`
`

`

`Contents
`
`Preface
`
`xv
`
`Acknowledgments
`
`About the Authors
`
`xvii
`
`8
`
`1.4.2 Timing
`1.4.3 Channels
`
`9
`10
`
`24
`
`1
`Chapter 1 DSL Fundamentals
`1.1 Alternatives to DSLs: Fiber, Wireless, and Coax
`1.2 Worldwide Extent
`2
`1.3 Voice—Band Modems and DSLs
`1.4 Transmission Modes
`8
`1.4.1 Direction
`
`
`
`1.4.4 Single and Multipoint Topologies
`1.5 DSL Terminology
`12
`1.6 Rate Versus Reach
`12
`1.7 Crosstalk
`13
`
`11
`
`1.8 Enabling and Disabling Forces
`1.9 Applications
`17
`1.10 Evolution of Digital Transmission
`
`16
`
`21
`
`Chapter 2 Types of DSLs
`2.1 DSL Design Margin
`2.2 DSL Precursors
`
`23
`
`23
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 6 of 19
`
`

`

`
`
`CSCO-1028
`
`Cisco V. TQ Delta, IPR2016-01021
`
`Page 7 of 19
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 7 of 19
`
`

`

`Contents
`
`‘
`
`Contents
`
`3.5 Transmission Line Characterization
`
`64
`
`64
`“ABCD” Modeling
`3.5.1
`3.5.2 Transmission Line RLCG Characterization
`
`3.5.3 Characterization of a Bridged-Tap Section
`3.5.4 Loaded Coils — Series Inductance
`75
`
`67
`
`74
`
`75
`3.5.5 Computation of Transfer Function
`3.5.6 Measurements for Computation ofRLCG Parameters
`3.5. 7 Balance —— Metallic and Longitudinal
`84
`Noises
`85
`3.6.1 Crosstalk Noise
`3.6.2 Radio Noise
`
`86
`
`92
`
`98
`
`94
`Impulse Noise
`3.6.3
`97
`Spectral Compatibility
`3.7.1
`Interference Between DSLs and Multiplexing
`3. 7.2
`Se lf-Interference
`99
`3. 7.3 Crosstalk FEXT and NEXT Power Spectral Density Models
`3. 7.4 Emissions from DSLs
`104
`More Two-Port Networks
`
`139
`
`105
`
`3.8.1 Reciprocal and Lossless Two-Port Circuits
`3.8.2 Analog Filter Design and T(s)
`107
`3.8.3 Lossless Realization ofH(s)
`113
`3.8.4 Frequency/ Magnitude Scaling and Frequency Transformations
`3.8.5 Active Filters
`116
`Three—Port Networks for DSLs
`
`106
`
`119
`
`114
`
`3.9.1 POTS Splitters
`3.9.2 Hybrid Circuits
`References
`129
`
`120
`128
`
`Chapter 4 Comparison with Other Media
`4.1 Fiber-to-the-Home
`133
`
`133
`
`4.2 Coax and Hybrid Fiber Coax
`4.3 Wireless Alternatives
`136
`4.4 Satellite Services
`References
`137
`
`137
`
`134
`
`Chapter 5 Transmission Duplexing Methods
`5.1 Four-Wire Duplexing
`
`139
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 8 of 19
`
`

`

`
`
`CSCO-1028
`
`Cisco V. TQ Delta, IPR2016-01021
`
`Page 9 of 19
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 9 of 19
`
`

`

`Contents
`
`Contents
`
`7.2.6 ADSL T1413 DMT
`
`235
`
`236
`
`247
`
`256
`
`265
`270
`
`274
`Interleaving Methods
`7.4.3
`7.4.4 Concatenated Coding and Multilayer Coding
`7.4.5 ADSL Special Case
`277
`7.4.6 CRC Checks
`281
`7.4. 7 Scramblers
`285
`References
`288
`
`277
`
`Initialization, Timing and Performance
`Chapter 8
`8.1
`Initialization Methods
`297
`8.1.1 Activation
`297
`8.1.2 Gain Estimation
`
`299
`
`7.2. 7 Clipping and Scaling (Peak-to-Average Issues)
`7.2.8 Fast Fourier Transforms for DMT
`242
`7.2.9 Multiplexing Methods for Multicarrier Transmission
`7.210 Narrowband Noise Rejection
`251
`Trellis Coding
`256
`7.3.1 Constellation Partitioning and Expansion
`7.3.2 Enumeration of Popular Codes
`262
`7.3.3 Shaping Effects
`262
`7.3.4 Turbo Codes
`263
`Error Control
`264
`7.4.1 Basic Error Control
`7.4.2 Reed-Solomon Codes
`
`344
`
`302
`
`303
`
`8.1.3 Synchronization (Clock, Frame)
`8.1.4 First Channel Identification
`8.1.5 Channel Equalization
`311
`8.1.6 Secondary Channel Identification and Exchange
`Adaptation of Receiver and Transmitter
`314
`8.2.1 Receiver Equalization Updating
`315
`8.2.2 Transmitter Adjustment
`318
`Measurement of Performance
`324
`
`314
`
`3.25
`
`8.3.1 Test Loops and Noise Generation
`8.3.2 Measure of Performance
`333
`337
`Timing Recovery Methods
`337
`8.4.1 Basic PLL Operation
`341
`8.4.2 Open—Loop Timing Recovery
`8.4.3 Decision-Directed Timing Recovery
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 10 of 19
`
`

`

`
`
`CSCO-1028
`
`Cisco V. TQ Delta, IPR2016-01021
`
`Page 11 of 19
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 11 of 19
`
`

`

`Contents
`
`Contents
`
`13.1.2 RFC 1661 — The Point-To-Point Protocol
`13.2 FUNI over ADSL
`388
`13.2.1
`FUNI Frame Structure
`13.2.2
`Reference
`
`Encapsulation
`389
`
`389
`
`388
`
`387
`
`Chapter 14 ADSL in the Context of End-to-End Systems
`14.1 An Overview of a Generic DSL Architecture
`394
`14.1.1
`The Customer’s Premises
`394
`14.1.2
`14.1.3
`
`395
`The DSL Loop
`Termination ofDSL in the Carrier’s Central Office
`or Remote Site
`3.95
`The Carrier’s Back-End Data Network
`
`396
`
`14.1.4
`14.1.5
`
`The Interface to the Service Provider’s Network
`Potential ADSL Services and the Service Requirements
`
`391
`
`398
`398
`
`Specific Architectures for Deploying ADSL in Different
`Business Models
`399
`Several ADSL Architectures
`14.4.1
`14.4.2
`14.4.3
`14.4.4
`14.4.5
`14.4.6
`References
`
`423
`
`402
`
`A Packet—Based Architecture for Small Deployments
`ATM Access Networks
`403
`RFC 1483
`405
`PPP over ATM
`
`407
`
`402
`
`Tunneled Gateway Architecture
`PPP Terminated Aggregation
`410
`
`408
`40.9 -
`
`Chapter 15 Network Architecture and Regulation
`15.1
`Private Line
`411
`15.2
`Circuit Switched
`15.3
`Packet Switched
`ATM
`413
`' 15.4
`15.5
`15.6
`15.7
`
`4\14
`Remote Terminal
`Competitive Data Access Alternatives
`Regulation
`416
`
`414
`
`41 1
`412
`
`411
`
`Chapter 16 Standards
`16.1
`420
`ITU
`16.2 Committee T1
`16.3 ETSI
`
`421
`
`419
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 12 of 19
`
`

`

`
`
`CSCO-1028
`
`Cisco V. TQ Delta, IPR2016-01021
`
`Page 13 of 19
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 13 of 19
`
`

`

`
`
`CSCO-1028
`
`Cisco V. TQ Delta, IPR2016-01021
`
`Page 14 of 19
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 14 of 19
`
`

`

`d communica—
`
`leo) applications,
`luding micropro—
`in the mid-19905
`retwork exceeded
`:nt of DSLs began
`
`1portant, even for
`
`. During the 19805
`phone lines would
`of fiber fever argu-
`
`per transport would
`,d thereby delay the
`
`focused their fiber
`5 sites and to fiber—
`ners. When asked if
`"ADSL is an interim
`
`egarding who would
`at the
`
`:rformance, simplify
`
`to permit the cus—
`
`tomer the freedom to choose among many competitive providers of equipment. System develop-
`ment slowed because equipment designers did not know what features were required, equipment
`vendors did not know which sales channels to use, and telephone company engineers did not
`know who would be responsible for installation and maintenance. For basic rate ISDN in the
`United States, it was resolved that the customer—end transceiver would be owned by the cus-
`tomer. However, in many other countries the opposite resolution was reached. Either model can
`work, but only after the decision is firmly made.
`ADSL encountered a different regulatory quagmire: would the telephone companies offer
`ADSL as a regulated or unregulated service. The regulatory envirdnment in the United States
`made both alternatives equally unattractive, and worse yet was the prospect that the rules could
`change again at any time. Once again, the price paid for stimulating a competitive market was
`delaying the introduction of new services by nearly two years. With a market window of five
`years for some services, the chilling effect of regulation can easily cause innovators to become
`stodgy.
`
`rate per month. Services may be characterized in terms of guaranteed throughput and availabil—
`
`
`
`Applications
`
`/ D
`
`SL Fundamentals
`
`The success of a service (and its underlying technology) depends greatly on its price and
`its relation to available alternatives. Service price, in turn, depends greatly on the cost of equip-
`ment and labor costs for operation. Equipment and operating costs are reduced as the number of
`customers grows. Low-cost service is best achieved by establishing a service that addresses the
`most customers and minimizes additional infrastructure costs via use of existing facilities. For
`DSLs, additional transceiver circuitry that extends loop reach or enables additional applications
`may ultimately enable the reduction of service price by expanding the addressable market. A
`recurring theme in the field of DSLs is that the cost of additional capability in the transceiver
`yields greater savings from reduced operating costs, greater loop reach, or possible additional
`applications.
`
`1-9 Applications
`
`The first step in developing a technology or system is identification of the customers’ needs and
`their implications on functional requirements. End-user demand for products and services is
`driven by saving money, making money, accomplishing necessary tasks, saving time, and some-
`times the possession of a status symbol. TheSe demands are satisfied by applications: hardware
`and software that perform certain tasks for the user. An application is a package of hardware,
`software, and, in some cases, a network service that provides a solution to specific end customer
`needs. A service performs certain tasks, or provides certain capabilities, for which the customer
`(or someone else) pays for on a recurring basis (e.g., per use, or per month). In many cases, a
`Service may be used to support many applications. At the risk of circular reasoning, a service is
`something provided by a service provider, such as a telecommunications carrier company. The
`applications listed in the following tables apply to both public and private network services.
`The end user may be charged for a service per use, per minute of use, per byte, or a flat
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 15 of 19
`
`

`

`Gaussian. but the approximation is good enough for the tutorial purposes of this document.
`
`where T is the symbol period or time it takes to send each constellation point and H is defined
`below.
`When the points in the input signal distribution are not equally likely, as sometimes occurs
`with one—dimensional distribution of the samples entering the line with transmission methods
`like DMT, the entropy generalizes to
`
`Figure 7.21 ANSI T1.413 DMT transmitter.
`
`Clipping and Scaling (Peak-to-Average Issues)
`7.2.7
`Entropy is an information-theorctic quantity that is a generalization of data rate. In DSL,
`designers often use signal constellations where each point is equally likely to occur. If there are
`M equally likely points for each symbol, the entropy and the data rate are related easily by
`
`H log2(M)
`R = —
`T
`T
`
`bits/second
`
`(7 .92)
`
`H =
`
`p. 108297.)
`
`(7.93)
`
`which reduces to log2( M )if all points are equally likely with p]- : l/M.
`If viewed in the time domain as one—dimensional signals, the probability distribution Of
`multichannel signals approaches a Gaussian distribution through the (ab-) use of central-limit
`theorem arguments.12
`
`
`
`12.
`
`Actually, no discrete distribution ever truly becomes Gaussian, and there are always peak limits unlike the
`
`
`
`Chapter 7 - Intersymbol lnterterenee, Equalization, and DMT
`
`Constell
`-ation
`encoder
`and
`gain
`sealer
`
`IIIIIII
`
`ordering
`'fime
`
`0
`Trellis Input
`Dara flame k: 0 m 255
`
`
`
`2’“? a“
`na 09 processing
`
`|III
`
`III|
`
`E
`FEC Output
`Dara Flame
`
`
`
`Sync
`Control
`
`WWW P0”'”-
`
`A
`Mux
`Data Frame
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 16 of 19
`
`

`

`A clip is defined to occur when the transmit signal sample exceeds the maximum imple—
`mented value for the transmitter (often set by a digital-to—analog converter’s maximum value).
`This random event’s probability of course depends on the clip level. For a Gaussian distribution
`and a probability of clip of 10—7, the clip instantaneous power level is 14.2 dB higher than the
`average power of the signal. For a clip probability of 1075, this value reduces to 12.5 dB, and for
`10—3, the value is 10 dB. A reduction of a very large peak—to—average ratio (PAR) to a smaller
`one can reduce the complexity of a transmitter, especially reducing the power consumption of
`linear analog driver circuits.
`
`As an example, let us take an ADSL signal, as per T1413. The probability of a clip is
`about 10‘8 with the specified 15 dB PAR. Thus, at a sample rate of 2.208 MHZ, clips occur then
`at data rate 1078(2208 >< 10°) - [761,-], where [29],}, is the number of bits necessary to convey the size
`and position of the clip to the receiver, typically less than 20 bits. Thus, it would then take about
`0.5 bits/second to eliminate clips with PAR 2 15 dB, a very low data rate loss. This 0.5 bits/sec—
`ond is the entropy of the clip. Equivalently, if the clip occurs, in reality only 0.5 bits/second of
`data rate should be lost.
`'
`
`.
`
`‘Z‘
`
`t
`’
`
`This study of entropy summarizes the motivations of Wulich [49], and Ochiai and Imai
`[50], Davis and Jedwab [51], and Patterson [52], who all investigate the use of various coding
`methods that preclude the occurrence of data sequences that cause high PAR in OFDM transmis—
`sion. Such methods, unfortunately, seem to exhibit higher rate loss than necessary, largely to
`simplify implementation. Further study is necessary before any of these methods are applicable
`to DMT ADSL.
`
`eument.
`'
`-
`-
`5 Peak “mus “nuke
`
`th
`
`e
`
`7.2.7.1 Dynamic Clip Scaling
`Dynamic clip scaling [59] is an early proprietary clip—mitigation method used in some
`.
`.
`.
`.
`.
`.
`.
`.
`DMT DSLs. A Single tone is reserved for indication of a scale factor for the more symbol. This
`
`MW
`:e Equalization, and DMT
`Multichannel Line Codes
`
`Analog processing
`
`1 0f data rate' 1“ DSL’
`
`
`
`Indeed, ADSL modems could increase PAR significantly without noticeable reduction in
`the data rate. Even at a PAR of 10 dB, the loss in data rate, or the entropy of the clip, is still
`below 50 kbps, which is much less than the downstream data rates of 1.5 to 8 Mbps.
`A practical drawback is that to keep the additional data rate reduction small, a large buffer
`(memory) may be necessary, along with the consequent delay increase. If a transmitter first
`needs to compute a preliminary output to detect a clip and then redo this computation with a
`slightly altered bit stream to encode the clip instead of allowing it, the transmitter complexity
`can double. A similar doubling can occur at the receiver in first decoding the information to
`learn of the clip, reconstruct the large sample, and then decode again.
`Fortunately, a number of low—complexity and low-delay methods have been developed and
`are individually discussed in the next few subsections. The two Tellado methods, tone reduction
`[53] and tone injection [54], [55] (parts of which were independently found by Gatherer and P01-
`ley [56] and by Kschischang, Narula, and Eyuboglu [57]) are those that gain the most PAR
`reduction and appear strong candidates to be implemented in DSLs of the future (see Sections
`7.2.7.4 and 7.2.7.5). An upcoming book [66] has a chapter by M. Friese of Dcutsche Telekom
`devoted to PAR reduction methods (see Chapter 4).
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 17 of 19
`
`

`

`
`
`CSCO-1028
`
`Cisco V. TQ Delta, IPR2016-01021
`
`Page 18 of 19
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 18 of 19
`
`

`

`:e, Equalization, and DMT
`
`ecoding and is added
`: of a clip. The four
`3, —2 dB, and —3 dB;
`Usually 0 dB is sent,
`—1, —2 , or —3 is sent.
`will not cause another
`al back to its original
`
`:rease in noise is less
`scurred. This method
`
`:quires the transmitter
`.es combined with the
`
`l are discussed below.
`
`lity, it is very unlikely
`
`tear to have simulta—
`
`, [61], [62], which has
`ted mapping method
`e known transmit sub—
`
`ed) in a corresponding
`n a tone and/or part of
`re receiver then knows
`
`omplexity of the addi-
`ip to the nth power of
`resulting in a few dB
`peaks, i.e., the second
`
`Iuller and Huber [64]
`
`,t separates the DMT/
`phase rotation applied
`11 IFFTs are performed
`LIE added in a way that
`
`ups needs to be carried
`ese investigates a num-
`e PAR, in his work and
`
`
`Multichannel Line Codes
`239
`
`Verbin’s Method Another approach due to Verbin [67] instead makes use of the obser»
`vation that the sum of two cyclically prefixed sequences will produce at the channel output a
`sequence that corresponds to the sum of the two corresponding sets of parallel channels. Verbin
`then selects from up to 7 possible sums, each corresponding to essentially an orthogonal energy—
`preserving transformation, of the even and odd parts of the nominal channel input. The one with
`the smallest peak is transmitted (and again the receiver must be told which corresponding
`receiver transformation to use via a control channel like a single tone or the pilot). This method
`obtains about 2 dB improvement. To avoid complicating gain issues with the FEQ implementa—
`tion, Verbin selects orthogonal transformations that correspond to permuting the order of the
`subchannels at the output, but otherwise do not lead to interference between DFT outputs. This
`restriction is not necessary, but makes the receiver simpler. This method also causes a reduction
`in the probability of a peak roughly as the nth power of the clip probability, where n is the num-
`ber of transforms. However, the occurrence of secondary, tertiary peaks limits gain to about 2 to
`3 dB as Verbin shows [67]. This method is simpler than recomputing an entire IFFT for intro—
`duction of randomness.
`
`signature waveform. Indeed simple analog lowpass filtering can lead to loss of the PAR reduc-
`
`7.2.7.3 Gatherer/Policy Method
`The Gatherer/Polley (GP) method [56] represents a significant improvement on dynamic
`clip scaling and on all the randomized reencoding methods. First, this method does not require
`coordination between transmitter and receiver, a significant benefit becaUSe standardization is
`less relevant in this case, Second, no extra IFFT or transforms are necessary although there is
`hidden complexity in a search procedure.
`The GP method notes that tones are often unused in DMT (intentionally or because DMT
`turns them off). Some energy may be allocated to unused tones as desired as long as spectral
`mask constraints (and any power constraints) are met. A signal may be constructed correspond-
`ing to use of unused tones. Offline optimization of the possible inputs is done to find a time—
`domain signature waveform that has a large peak in it and is otherwise small at other time
`instants. Addition of the signature waveform to the transmitted DMT signal corresponds to
`insertion of energy on unused tones and does not increase intermodulation distortion. When a
`large peak is detected, the signature waveform is added to the transmit signal with negative
`amplitude to the peak to reduce the level to under the desired clip level. The signature waveform
`must be cyclically rotated into the position of the peak, but that corresponds only to a linear—
`phase change on the inputs of the unused tones (meaning it can be ignored by the receiver). Bet-
`ter signature waveforms require more unused tones and allow increasingly lower settings for the
`clip level. Gatherer and Policy have been able to reduce PAR by 3 dB using this method.
`One limitation is the possibility of generating new peaks elsewhere, and so the signature
`waveform may have to be added several times until acceptable peak level is obtained, thus limit-
`ing the reduction in PAR. Another caution is that peaks can be shifted to the receiver if only sub—
`channels that would not make it through the transmission Channel are used to generate the
`
`CSCO-1028
`Cisco v. TQ Delta, IPR2016-01021
`Page 19 of 19
`
`