Chapter 11 – Dynamic Spectrum Mang. (JC)110/08/18
Chapter 11 - DSM (J. Cioffi)
11.1 DSL Unbundling Evolution
11.1.1 Cable Modem Architecture – DSL’s competition
11.1.2 DSL Evolution
11.1.3 The Essential Steps of DSM
11.2 Multiuser Basics
11.2.1 Multiuser channels and rate regions
11.2.2 The matrix channel
11.2.3 GDFE Theory
11.3. Spectral Balancing (Iterative Water-filling)
11.3.1 Yu’s Iterative Water-filling
11.3.2 Examples of Improvements
11.4. Vectoring
11.4.1 Upstream Multi-Access
11.4.2 Downstream Broadcast
11.4.3 Examples
11.5 MIMO Channel Identification
11.5.1 Method of Least Squares
11.5.2 Packetization
11.5.3 Blind Training
11.6 Predictions for the DSL Age
References
11.Annex-- Multiuser Detection
11.A.1 Basic Receiver prerequisites for MUD
11.A.2 Soft-Cancellation
11.A.3 Simplied Cancellation by Grouping of Excess Dimensions
Acknowledgements: G. Ginis, T. Chung, W. Yu, D. Gardan, S. Zeng, C. Aldana, J. Lee, A. Salvekar, J. Lee, W. Rhee, W. Choi, K.W. Cheong, J. Fan, N. Wu, B. Rezvani, Ioannis Kanellakopoulos, S. Shah, Michail Tsatsanis, R. Sonalkar, R. Miller, G. Cherubini, V. Poor, G. Zimmerman, G. Al-Rawi, K. Foster, T. Pollet, C. del-Toso, S. Schelesratte, N. Nazari, M. Alba, K. Kim, K. Hasegawa, N. Mihito, Y. Kim, C. Hansen, J. Chow.
Chapter 11 - DSM (J. Cioffi)
A 500 meter cable of 50 twisted pairs has acapacity of 10 Gbps in each direction to be shared among the customers of the telephone plant, greater than HFC and, yes, even greater than a single fiber shared in the most popular projected fiber-to-the-home architecture known as a passive optical network (PON). Today’s DSL operates at less than 1% of this capacity -- dynamic spectrum management offers the promise of eventually realizing the goal of broadband connection at 100BT-like speeds to every customer of every phone line in the world, thus enabling the broadband age.
Dynamic Spectrum Management (DSM) is the newest of DSL methods that enables an array of highly desirable improvements to DSL service:
- automatic detection and/or prevention of service faults caused by crosstalk
- greater mixture of symmetric and asymmetric services
- higher and more reliable data rates .
DSM can be thought of as an adaptive form of the earlier spectrum management (see Chapter 10).
Figure 1 – DSM Reference Diagram.
Figure 1 shows an early reference diagram for DSM. The DSL interface is presumed to be one of the standardized DSLs in common use. As depicted in Figure 1 and Figure 2, a single service-provider’s DSL maintenance center performing DSM-Data coordination may optionally accept line and crosstalk information in a specified format across an interface DSM-D. The DSL spectrum maintenance center (SMC) can process the information received and provide such processed data results for the use of DSL service providers according to specified criteria and format at interface DSM-S. The SMC performing DSM-Control coordination may also dynamically specify downstream and upstream DSL line spectra and signals of any modems to which it is attached when those modems are fully DSL capable. Thus, the SMC may be implemented within a DSLAM (especially when used in a fiber-fed remote DSL-LT). DSM is adaptive in that the DSL-LT and DSL-NT can react to recommendations provided automatically by the SMC. The DSL service provider may also adaptively react to reported situations in the loop plant by altering loops to which specific DSL modems are attached or by taking other corrective actions. It is also very much possible for significant DSM improvement when the lines operate autonomously (without any directed control from the service provider other than perhaps the data rate desired). This chapter explores the possible improvements as DSL progresses from a highly autonomous state to one in which ultimately signals are co-generated at remote DSL-LT’s of the future.
DSL standards such as ADSL (Chapter 3) and VDSL (Chapter 7) have capably enabled individual transmission links to play near their individual fundamental theoretical limits, igniting a worldwide interest in broadband access. To date, these standards normally operate autonomously without active concern for other lines, whether or not their own performance exceeds or misses the capability demanded by the individual customer associated with that line. High performance of one line could generate large interference into other lines, perhaps preventing the desired reliable transmission of a good data rate on some of those other lines. Spectrum management in Chapter 10 is an early attempt to mitigate such “DSL binder hogging” by any one DSL line through enforced imposition of a fixed set of DSL spectrum masks on each type of DSL modem. These masks compromise different customer’s interests based on presumed fixed channel conditions. The management thus enforced by early spectrum management only best balances the customers’ varied performance in the one fixed situation presumed in the design of the masks. DSM attempts to balance customer interests adaptively as a function of line information supplied by some or all the DSL systems about loop topology and crosstalk activity. Dynamic behavior may occur at several periodic intervals as well as act at several degrees of coordination from autonomous to co-signal generation: the time interval may be per-service order, per-day, per-session, or continuously adaptive.
At time of writing, DSM was in early drafting stages within the T1E1.4 DSL standardization group. This chapter attempts to illustrate the basic concepts and issues in DSM. Section 11.1 address the evolution of DSL unbundling and services, showing that unbundling is inevitably migrating to a packet or “wholesale” level, especially as fiber is increasingly deployed in the feeder portion of the telephone loop plant. Section 11.2 then addresses a cable characterization that can be used to enable DSM, given the evolution described in Section 11.1. Section 11.1 also outlines the fundamentals of multiuser communication theory that apply to DSL. Section 11.3 provides technical description of the improvements of early autonomous DSM that can dramatically improve existing DSL deployments without need for the SMC. Section 11.3 illustrates the significant benefits of a method known as “iterative waterfilling” that may be implemented autonomously and that always improves DSL binder performance if used correctly no matter what the mixture of DSM-practicing or non-practicing modems occurs. Section 11.4 then proceeds to the more sophisticated vectored DSL systems that do require DSLAM-side coordination and enable extremely high data rates and variable mixtures of symmetry in downstream/upstream data rates. Section 11.5 investigates methods for on-line in-service measurement of the critical channel and crosstalk parameters. An annex describes some mutiuser detection methods that can be used in the absence of DSM to limited benefit in all DSL systems.
11.1 DSL Unbundling Evolution
Unbundling is the incumbent local exchange carrier’s (ILEC’s) lease of a telephone line or some part of its bandwidth to a competitive local exchange carrier (CLEC). Unbundling first began in the United States as a consequence of the 1996 Federal Telecommunications Act. Current unbundling practice with DSL service usually allows the CLEC to place modulated signals directly on their leased physical copper-pair phone line, sometimes referred to as the lease of “dark copper.” Such unbundled signals may have services, and consequently spectra, that differ among the various service providers. The difference in spectra can magnify crosstalking incompatibilities caused by electromagnetic leakage between lines existing in close proximity within the same cable. ILECs and CLECs then try to ensure mutual spectral compatibility by standardizing the frequency bands that can be used by various DSL services. This standardization is known generally as spectrum management (see Chapter 10). However, there are many DSL types and bandwidths, and service providers are often competitors, which complicates the process of such spectrum-management standardization. Further, the cooperation and connection between spectrum regulators and DSL standards groups is still in early evolution, so that regulators may allow practices different than those presumed in the process of spectrum-management standardization. The Tauzin- Dingell bill passed by the US House and before the US senate at time of writing would create a structure that allows the highest possible data rates and services technically, although its encouragement of competition and the consequent rate DSL service installations is debated.
In advanced DSL service the location of the line terminal (LT or “central-office side”), as well as network termination (NT or “customer premises side”), can vary. That is, not all LT modems are in the same physical location. Often the location may be an optical network unit or cabinet, where placement and attachment of CLEC equipment may be technically difficult if not altogether physically impossibleeconomically infeasible. The difficulty arises because CLEC fiber access to the ONU may be restrictedthere may be no spare fibers for the CLEC to access the optical network unit (ONU), and/or the ONU may not be large enough to accommodate a shelf/rack for each new CLEC. Placement of such CLEC equipment for dark copper is often called “co-location,” when it is in the central office. Space and facilitation of such central-office co-location for unbundling of the dark copper is mandated by law in the United States. Presently, many ILECs are finding regulator acceptance of their control of all physical-layer signals that emanate from remote terminals (as long as they can prove they provide wholesale packet-level unbundled service fairly), as opposed to emanating from the central-office where the physical-layer unbundling is forced by law. This represents a change in architecture with respect to what many standards groups have presumed in spectrum-management studies. The control of all the physical layer signals by a single service provider allows potential coordination of the transmitted signals in ways that can be beneficial to the achievable data rates, reliability, and complexity of DSL service, such as is now studied in DSM. Even without such coordination, there is much that DSM can offer, as we shall see in Section 11.3.
11.1.1 Cable Modem Architecture – DSL’s competition
Figure 2 illustrates a general architecture common to cable-TV service providers. Of particular interest is that the cable system is operated by a single service provider, and the coaxial-cable bandwidth used is shared by all users. Cable modem technology basically uses time-division multiple access of the different users on a shared coaxial segment in a common up or down frequency band. The up band can be located below 40 MHz, but additional up bands may be appropriated from the existing TV channels for upstream transmission at higher frequencies. At least one downstream TV channel is also appropriated for the downstream cable modem, and shared again in the time domain. Cable systems today are operated by a single operator, and this operator controls all content (for example, which internet service(s) or voice service(s) may be offered, as well as what TV channels are offered). However, the FCC [2] in the USA has opened discussion on whether cable operators will be forced to provide other content. Even if not forced, many cable operators are investigating allowing competitive service suppliers on the same cable, effectively implementing wholesale “unbundling.” Also, as illustrated in Figure 2, as fiber moves into the HFC network, a single fiber will attach to many homes replacing the coax and providing higher bandwidth for all services. That fiber would consequently be shared in the time-domain according to the same conventions as with the coax (just with more channels available for all services) – this is similar to PONs or passive optical networks, sometimes also studied as an alternative by telephone service providers for fiber migration to the customer. A single service provider, perhaps eventually restricted from controlling content, would control that fiber. Various mixes of time-, frequency-, or code-division access will not change this aspect of a single common carrier, with likely multiple services/contents provided on the system. Such a system is now, and will be in the future, a competitor to DSL.
To emphasize and draw a comparison later with DSL, in the cable system, the issue of competition and unbundling is forced to a higher level, which we call here “packet unbundling” by the physical coordinated nature of the shared media. A single common carrier, the cable operator, maintains the physical layer and the consequent bits that flow over that layer. In DSL systems with current unbundling practice, the bits are managed independently by each DSL service provider, and indeed the physical-layer signals may be different, so different that they cause harmful interference to each other. Spectrum management attempts to contain this harmful interference in DSL, while in cable, there is no such spectrum management because all signals provided by a single service provider are necessarily compatible.
Because the physical medium is shared, a MAC (medium access control) is REQUIRED in the cable system to coordinate data to different users. An aspect of having a MAC is that it moves the unbundling problem to a higher layer in the protocol stack. The MAC doesn’t care who’s providing the incoming data – it just routes it to the right customer. A similarity of the DSL and cable system is that the DSL crosstalking has the same electrical interference problem as the shared common media in cable, which is increasingly important at the higher frequencies used by DSLs on shorter lines from LTs, and causes a performance dependence between lines. However, DSL has yet no MAC to accommodate this problem, and DSM can be construed as a first step towards enabling such a MAC. This no-MAC observation provides a market-competition argument to support the eventual DSL-regulatory migration to packet unbundling, which is inevitable anyway if multiple fibers to each home are to be avoided as DSL evolves (such arguments appear in Section 11.1.2). As the cable system evolves to greater bandwidths and more fiber, a single fiber eventually reaches all customers, and its bandwidth is shared among any common customers and content providers.
Figure 2 – Basic Cable modem architecture, HFC (hybrid-fiber coax). (All links shared among the common users to that link).
11.1.2 DSL Evolution
An often-presumed DSL evolution appears in Figure 3(a) with remote-terminal-based DSL. Individual uncoordinated twisted pairs run to each customer. The content of a pair is controlled by the service provider whose modem attaches to that pair in the line terminal (LT). If that LT modem is in a central office, several service providers may compete for the privilege to supply DSL service to that customer as mandated by law. However, the issue is yet formally undecided at the LT outside the central office (be it for ADSL, VDSL or any other DSL), although at least one ILEC in the United States (SBC) has permission to instead “packet unbundle” at a higher digital layer in the protocol stack. Figure 3(b) illustrates how a 2nd service provider would connect, with their own fiber from the central office to the DSLAM presumed if physical-layer unbundling were continued.[1] A 3rd service provider would have their own fiber, and so on, resulting in many fibers to the LT. As this system evolves, the loop plant eventually has many fibers to each customer in FTTH to maintain the present form of physical-layer unbundling. While a multiplicity of fibers connecting to DSLAMs co-located in a central office is perhaps a common expectation, the purpose of the use of fiber is to avoid many parallel wires/paths to a customer. Thus, clearly present physical-layer unbundling leads to a ludicrous technical evolution of multiple fibers to every customer; nonetheless existing DSL spectrum-management decisions (i.e., see [28]) address only this evolution path. Note this evolution is different that the cable system’s evolution discussed in the previous section, which has one fiber shared among many customers, as discussed in the previous subsection.
Figure 3(a) – Mutliple service provider, line-unbundled LT-based DSL.
A clear alternative is to maintain one fiber as in Figure 3(b), but carry the different service providers signals on that same fiber – i.e., packet unbundling or wholesaling. Technically, when one common fiber carrier carries all the signals, necessarily there is a demultiplexer in the LT for all the individual digital signals. If the common carrier must implement this demultiplexer, that carrier might also as well implement the modem – this allows coordination of the lines at the LT, which can lead to enormous gains in data rate as in Sections 10.3 – 10.5, as well as a nearly arbitrary mixture of asymmetric and symmetric services. Spectrum management then becomes more of a multiplexing problem, than one of just fixed worst-case minimization of crosstalk between lines that may be operated by different service providers.
Figure 3(b) – Packet-unbundled DSL evolution.
Figure 4 predicts the timeline of argued evolution of DSL from its present configuration shown as Stage 0 to a likely packet unbundled future. The ILEC could actually be any service provider, and the twisted-pair network in the final step might actually be a private network. Early DSM could have the network maintenance center for DSL collecting performance/line information from the lines and possibly making recommendations to the lines/DSLAMs as to maximum binder-benefical data rates to attempt. Clearly the DSL maintenance center can also provide information to the ILEC service personnel as to potential or identified problems, resulting in either manual or automatic repair/prevention. As DSM progresses and DSL evolves to fiber fed remote or “line” terminals (LT), then highly advanced maintenance can be used remotely or placed directly at the common copper interface to customers in the LT. Signals with such later coordination could ultimately be co-generated to avert crosstalking problems between lines that otherwise dramatically reduce data rates. The data rates listed are representative of what is possible at each stage of evolution with existing modem technologies, but just with increasing management and coordination.