Chapter 18 Architectural Elements and Examples
18.1 Introduction
Network architecture is the logical and physical interconnection of all the elements between a signal’s generation and its termination. Chapter 17 dealt with ways of measuring architecture-related parameters and the needs of various types of services. This chapter will first delineate each of the elements from which an architecture is constructed and then present some examples of high-level HFC distribution networks that have been proposed or constructed. Fiber-deep architectures will be covered in Chapter 19; the relationship between network architecture and service reliability and availability will be dealt with in Chapter 20.
18.2 Architectural Elements
Although network architecture is often thought of in terms of the physical distribution topology, everything from the internal headend (and preheadend) processing through network termination equipment affects how services are delivered. It is important, therefore, to understand the implications of available options at all levels of the network. In this section we discuss only those structures that are part of the cable television network; we do not include information flow outside that system.
18.2.1 Terminal Equipment
The network terminal equipment provides the interface between the user’s end application devices and the shared broadband distribution system. Depending on the service, it may perform format-conversion, testing, or security functions as well as protecting the network from disruption. One example is set-top converters that perform tuning and selective descrambling functions for video services and may also provide menuing or ordering capabilities. In the case of digital video channels, the box also converts the signal to standard analog video format. In two-way systems, the box communicates user’s service requests to the headend. Another example is a telephony point-of-entry network interface device (NID), which converts between digitally modulated RF signals and analog baseband telephone signals for routing within a home. Choices in terminal equipment can have a significant effect on overall cable system design and performance.
In-Home Equipment
Historically, the full cable spectrum (possibly minus some channels blocked by traps) has been distributed throughout subscribers’ homes. In many cases, subscriber-owned television receivers have been connected directly to cable outlets; in others, converters or converter/descramblers have been placed between the drop and the subscriber’s receiver.
This scenario is consistent with U. S. regulations that allow customers to wire their own residences and that require cable operators to deliver all channels in at least the lowest television programming tier in unscrambled, analog broadcast format.
One advantage to delivering the full cable spectrum throughout homes is that any service can be accessed from any outlet by connecting the proper equipment. Second, no home equipment investment is required in advance of service delivery. Finally, the cost and responsibility of equipment powering are borne by subscribers.
Balancing those advantages are the relatively high and variable loss of drop wiring in homes, which makes it more difficult to ensure that terminal equipment receives the proper signal levels. Additionally, home wiring has been a major egress point for cable-carried signals and ingress point for broadband noise, signals, and electrical transients in two-way systems. Although upstream transmitters may be located anywhere within the home drop system, the full bidirection bandwidth must be preserved, and upstream plant is therefore exposed, on average, to more frequent interference. Often cable operators reduce this risk by installing filters to block upstream ingress on portions of the drop system that will not be used by two-way devices. Finally, and depending on the services being offered over the cable system, there may be a duplication of some functionality among different pieces of terminal equipment.
Point-of-Entry (POE) Equipment
Terminating the broadband linear network at a common point exterior to the residence provides a means to isolate two-way cable networks from signals generated or picked up within the home. Any upstream signals to be transmitted into the distribution system can be generated by a common transmitter at the POE device. Similarly, multiple downstream services may share some of the common POE circuitry. For example — the POE bidirectional transceiver may generate and receive standard baseband signals for transmission over twistedpair cables to telephone instruments and drive an Ethernet bus or wireless transceiver serving multiple computers while transmitting and receiving RF signals from the headend.
Where several services are provided and can be coordinated, the use of shared POE equipment can result in both cost savings and improved operation. The addition of automated status monitoring may be justified by improved monitoring capabilities. On the other hand, the cost of equipment for the first service delivered to a home is often more expensive than a single-purpose in-home version. Finally, lacking interindustry standards, use of a multiservice POE device restricts the ability of operators to buy equipment from multiple sources.
There is a long-standing debate regarding the relative advantages of network versus home powering of POE devices. Home powering significantly reduces the power consumption of the distribution network, but it is less reliable. Also, it is much simpler to maintain large shared standby powering systems than individual in-home units. The goal to provide 99.99% availability for at least basic telephone service certainly tends to favor network powering. The proper decision is less obvious for nonessential services.
Shared Terminal Equipment
In some cases, terminal equipment is integrated with multiport taps, so some functions (for instance, power supplies and data transceivers) can be shared among multiple homes. The most common examples for video services are addressable disconnect functions, interdiction, and “whole-house” descrambling systems (these are discussed in Chapter 21).
Two obvious advantages of shared terminal equipment are easier maintenance access and, potentially, improved security. Deployment of such a system, however, requires either that the network provider completely equip the plant with such equipment or that the plant be modified each time a new subscriber is installed. In other words, the operator is forced to choose between a high initial capital cost and an even higher evolutionary upgrade cost accompanied by service interruptions.
As with POE equipment, shared terminal equipment has the further disadvantage of limiting future flexibility — it is hard to upgrade program scrambling security when the descramblers are part of the distribution system. These problems have limited the widespread adoption of shared interdiction technology by the cable television industry. The greatest commercial success with shared equipment has been in multiple dwelling units (MDUs), where the rate of connects and disconnects exceeds the average, signal theft is high, and/or physical access is difficult.
Shared terminal equipment is central to fiber-to-the-curb (FTTC) architectures. In an FTTC architecture, the fiber distribution network extends to curbside shared terminal equipment, whose functionality is shared among a few homes or businesses (typically 6 to 25). Individual homes are fed from the terminal equipment through fibers, twisted-pair lines, or coaxial drop cable. Fiber-deep architectures are treated in depth in Chapter 19.
Hybrid Arrangements
Most commonly, where either POE or shared terminal equipment is used, it is coordinated with some in-home equipment. For example, several manufacturers have announced POE network interfaces for telephony and data that pass the bandwidth containing video signals to set-top descramblers. The upstream bandwidth is divided, with data and telephony signals generated by the interface equipment and signals related to ordering of television signals originating in the set-top box. Filters block only a portion of the upstream bandwidth from the home wiring. Though such arrangements are arguably more expensive than totally integrated solutions, they preserve the network operator’s flexibility to deal with different vendors and formats and work well where new services are to be installed without disrupting older service platforms. Figure 18.1 illustrates terminal equipment location options.
18.2.2 Coaxial Distribution Network
The generic design of coaxial distribution systems was covered in Chapter 11. Within the general framework of amplified, split, and tapped networks, however, there are many architectural variables.
RF Bandwidth
The most basic parameter of any coaxial network is the forward and return bandwidth. Since these are determined primarily by the amplification equipment, some operators are proposing the use of coaxial distribution networks small enough that they can be totally passive between the output of the fiber node and the user. Such networks have the advantage that the division between upstream and downstream bandwidth can be altered at will, as can the upper frequency limit. As a practical matter, however, these freedoms are restricted by the spectrum usage issues discussed in Chapters 9 and 17, and the upper frequency will be constrained by the characteristics of the passive devices used, by the increasing loss of the cables, and by the practical limits on signal levels in the network, as discussed in Chapter 10. The eventual conversion of all video transmission to digital and the cessation of VHF over-air broadcasting, of course, will enable major changes to cable system spectrum allocations.
Coaxial Serving Area Size
The next most basic decision in the design of an HFC network is the size of the area (measured in number of homes passed) to be served by one physically-separate coaxial network. Since the use of the RF bandwidth is shared among all the users in the serving area, this determines the upper limit on bandwidth per home passed.
Since reliability is inversely proportional to the number of series-connected critical components, the required network availability and failure rate for the services to be offered should also be a factor in determining serving area size. The calculation of network reliability and availability is discussed in detail in Chapter 20.
Typically, three to four separate coaxial distribution legs will be served from each fiber-optic node, offering opportunities for subsequent node size reductions. Some operators, in fact, design nodes for subsequent subdivision.
Coaxial Powering
The average reliability of commercial power is often lower than the total end-to-end required network performance. Thus, systems are required to employ a powering strategy that minimizes the effects of power interruptions. Conventional cable television “standby” power supplies typically provide 1–2 hours of operation from self-contained batteries. While this is sufficient for short interruptions, some percentage of interruptions will exceed this length. In systems where customers are affected by more than one such power supply, reliability will further decrease due to the typically noncorrelated nature of those interruptions.
Where the individual coaxial distribution legs are limited to approximately 150 homes passed (assuming typical suburban, single-family home densities), it is practical to power the entire coaxial subnetwork from a single location, even when network-powered POE interface devices are used. This is, in fact, the strategy typically used by U. S. telephone companies who have constructed HFC networks. Single point powering of this size of network, however, generally requires using 90 volts, with maximum currents of the order of 15 amperes. Voltage limitations as well as electrical safety issues are controlled in the United States by the National Electrical Safety Code or local codes, at the option of each community.1
In order to achieve telephone-grade reliability, operators offering primary-line telephone service and high-availability data services have generally equipped these common power supplies with 8-hour battery capacity and/or backup generators. The generators, in turn, can be powered by self-contained fuel supplies or tied to available natural gas mains.
In order to avoid the problems and costs of installing such power supplies at each node, one telco (Southern New England Telephone) proposed powering nodes from the central office. To accomplish that, they developed a hybrid cable containing optical fibers for signals plus copper conductors to carry 440-volt, three phase power. At each node, the power was converted to acceptable voltages for multiplexing onto the coaxial distribution systems.2
Where the current-handling capabilities of components or the voltage drop through distribution cables might otherwise limit application of single point powering, the output from one power supply is sometimes divided among several power distribution cables whose resistance per unit length is less than the distribution cables. These power bypass cables may be routed alongside the signal cables past the first split point, where lower currents can be inserted into each of several network legs. This is illustrated in Figure 18.2. Since these power distribution cables do not carry RF, lower impedances and nonstandard construction are sometimes used to reduce power loss.3
When battery backup supplies are used, major improvements in outage rate and availability can occur if the supplies are equipped with monitoring systems that alert network operators of commercial power outages so that crews with portable generators can provide interim power before the batteries expire.
18.2.3 Fiber Node
Individual fiber nodes provide the interface between the linear fiber trunking system and coaxial distribution legs. At the simplest level, such a node may consist of a single optical receiver whose output is amplified to feed the downstream amplifier and, optionally, an upstream optical transmitter whose input is driven by the output of a combiner whose inputs are the upstream signals from all connected coaxial distribution legs. There are, however, many architectural variations.
Split-Band Multiple Fiber Inputs
Some nodes provide the option for multiple downstream input fibers that are fed to separate optical receivers whose outputs are then combined in a diplex filter. There are basically two advantages: improved signal quality and network flexibility. In the first case, the C/N and distortion of the fiber-optic links can be improved by simply dividing the total signal load between two transmitters. This allows a greater optical modulation index and thus an improved C/N. At the same time, fewer distortion products will be produced. The relative gains in C/N and distortion can be exchanged, within a limited range, by adjusting the optical modulation depth.
A more common reason to use split-band optical inputs is to allow those signals that are common to multiple nodes to be carried on one fiber and those that are node-specific to be carried separately. As will be seen later in the chapter, this can result in overall network flexibility and cost savings. One should be cautioned, however, that a system that uses split-band inputs cannot easily adjust those bands, because the diplex filters in every node would have to be changed.
Redundant Multiple Fiber Inputs
An alternate use for multiple downstream input fibers is redundancy of the fiberoptic trunking network. In nodes equipped with redundant inputs, each fiber is routed to its own full-bandwidth receiver. A switch at the output of the receivers selects the better (or surviving) signal in the case of failure or degradation of one path. If the fibers feeding the node are diversely routed (so that the probability of simultaneous cut is low), then significant gains in reliability can result. As a practical matter, the gain in reliability is limited by the reliability of the switch (and its control circuit), which is an added critical element in the signal path. As Figure 18.3 shows, the configurations are very similar; in fact, some manufacturers offer either as an option in the same node platform.
Redundant Signal Processing
For applications where availability is critical, nodes are available with additional redundant elements. Power supplies, downstream and upstream amplifier modules, and upstream optical transmitters are all available in dual configurations. In the last-mentioned case, dual upstream optical outputs are provided, enabling diversely routed external cables.
Handling of Multiple Coaxial Distribution Legs
Generally, optical nodes are used to feed more than one coaxial distribution leg. Although this is a design convenience that allows for shorter coaxial cascades by placing the feed point near the logical center of the node-serving area, electri- cally separating the coaxial legs also offers opportunities for effective bandwidth expansion and improved upstream performance. One option, described in Chapter 16, is block segment conversion, whereby the upstream signals from different legs are converted to nonoverlapping frequency ranges. Thus the bandwidth per home passed is limited by the coax-serving area rather than by the larger node-serving area. Offsetting the per-subscriber bandwidth gain is the increased performance required of the upstream optical transmitter, which must handle more signals and increased RF bandwidth. In some cases, this will require upgrading from FP to DFB transmitters, at a considerable increase in cost.
An alternate way of achieving the same upstream bandwidth expansion is to use separate upstream transmitters and fibers for each distribution leg. This reduces the performance requirements for each transmitter, but requires more transmitters and fibers. A third alternative is to use separate upstream transmitters operating on different optical wavelengths and to combine their outputs using a wave division multiplexer (WDM). Whether separate fibers or WDM is less expensive will depend on the length of the fiber transmission path and the relative cost of the transmitters and multiplexers.
A final option is first to process the entire upstream spectrum from each coaxial leg by converting it to a high-speed baseband digital signal, to time multiplex these signals, and then to feed them to a high-speed baseband digital optical transmitter. Several manufacturers offer versions of this technique, with either two or four inputs multiplexed to create a single baseband datastream of about 2 Gb/s. These transmitters are also available in ITU-grid DWDM wavelengths, so they can be further multiplexed at the hub before transmission to the headend.
These four options are illustrated in Figure 18.4. Note that optical splitting or use of additional laser transmitters can, in theory, be used with any option to create dual outputs for feeding diversely-routed fibers for higher upstream reliability, though the incremental cost is least for the single-fiber configurations.
18.2.4 Fiber Interconnects
Linear fiber-optic links provide the connectivity between the point where the FDM spectrum is created and its delivery to coaxial distribution legs. Fiber links are also used at higher levels in the network to distribute selected signals of all types, often in digital form. Although fiber-optic links are used among headends and between headends and outside information sources (such as broadcasters and Internet service providers), this section will discuss only those links that form part of the network extending from headends to and from end customers.
Although this book will treat interconnect architectures as if they were all constructed with fiber optics, it should be remembered that amplitude modulated broadband microwave radio technology (AML), as discussed in Chapter 14, is still used for both individual signal and broadband distribution in those cases where it is more cost effective than construction of physical optical cables.
Clearly, however, microwave links can replace only point-to-point fiber links and not more complex fiber topologies. Also, the available microwave spectrum may not permit as many channels as in a fiber-linked system. For instance, the U.S. 13-GHz CARS band will support only 80 analog television channels.
Star
The simplest optical architecture, and the one most commonly deployed for transmitting analog FDM signals from headends or hubs to and from nodes, is the star, defined as separate paths from a common point to multiple termination points. When constructed with dedicated optical fibers, driven by separate transmitters, the star architecture allows complete independence in the signals delivered to each node (as opposed to the most common type of microwave broadband star network, which splits the output of a common transmitter to feed multiple antennas and paths). In fact, many cable systems optically split fiber transmitter outputs to feed two to three nodes, but do that splitting in the headend so as to allow easy separation of nodes in the future when greater network segmentation is appropriate.
Though the failure of an individual fiber feeding a receiving point affects only the subscribers on that node, practical cable routing often results in one cable sheath holding the fibers for many nodes and thus creates a possible single point of failure that is much larger (see Figure 18.5).
Sheath Ring
The exposure to outages caused by fiber cuts or failures can be reduced significantly by sending the signals between the common point and each termination via physically diverse routes. When multiple termination points can be arranged in a ring configuration, considerable savings result from the sharing of common construction costs, as shown in Figure 18.6. This architecture is known as a sheath ring. Note that the diagram shows only a single line in each direction around the loop from the common point to each node, whereas in fact there may be several (at least one upstream and one downstream). Analysis of the figure will show that the cable can be cut at any point without destroying continuity from the common point to each node.
Although the sheath ring architecture preserves the signal independence of the star and adds a considerable degree of protection against fiber cuts, it does so at the expense of a considerable increase in total fiber footage. Depending on the physical layout of the ring, however, the total cable footage may be only slightly greater than for a simple star. Since construction costs typically are much greater than fiber costs, the incremental cost of the installation may be only moderately higher.
Sheath rings are a common architecture among competitive access providers (CAPs) who offer highly reliable data transport to commercial customers. They are also commonly provided between headends and hubs in multiple-level architectures, to protect against potentially very large outages should a major optical trunk cable be cut.
Analog Shared Ring
Where a common set of signals must be sent to multiple nodes, an economical structure is the shared ring, illustrated in Figure 18.7. The common signal spectrum is transmitted both clockwise and counterclockwise from the common point using separate fibers (and often separate transmitters to increase the degree of redundancy). At each node, a portion of the signal is coupled from both rings and fed to separate optical receivers. A redundancy switch, as described in the optical node options, selects the surviving path in the event of a cable cut or transmitter failure. Alternatively, the optical switch can be placed ahead of a single receiver.
Though shared rings can be constructed at 1310-nm wavelength, the maximum available optical budget of approximately 17 dB (see Chapter 12) limits the number of receiving nodes. At 1550 nm, however, fiber losses are less, and optical amplifiers are available that can considerably increase the available loss budget.
A major advantage of the shared ring is obviously a considerable savings in optical fiber usage and number of transmitters, as compared with either a simple star or a sheath ring, while preserving the protection against fiber cuts and transmitter failures. Although its application is limited to distribution of a common signal spectrum, shared analog rings are effective when combined with other structures.
Digital Repeating Ring
A ring structure that is well suited to baseband digital signal distribution is the repeating ring, an example of which is shown in Figure 18.8. As discussed in Chapters 12 and 19, baseband digital optical transmission allows much greater optical budgets (of the order of 30 dB) and very high modulation rates.
In a repeating ring of the form shown, the information destined for all termination points (usually hubs) is time multiplexed into a single datastream, which is transmitted both clockwise and counterclockwise around the ring. At each hub, the signals destined for customers served from that hub are separated from the stream and signals destined for the headend are inserted. Should a cable be cut or any piece of terminal equipment fail, the remaining equipment automatically “loops back” so that it communicates with the common point via the surviving path.
As with the shared ring, an advantage to the architecture is the small number of fibers required while preserving protection against cuts and equipment failures. A limitation is that its application is restricted to the distribution of information that can be converted to the common digital transmission format. Several such formats are available, including proprietary time division multiplexed (TDM) schemes, asynchronous transfer mode (ATM), synchronous optical network (SONET), and gigabit Ethernet. When used to transport NTSC video, digital transmission requires costly signal processing at the point where the signals are converted to RF, because each must be demodulated to a baseband digital stream, converted to an analog baseband signal, remodulated to the desired RF channel, and, finally, combined into the desired FDM spectrum for delivery.
A video transmission variant that is lower in cost than SONET and designed specifically for the cable television market uses pulse code modulation (PCM). Rather than starting with a baseband digital signal, it encodes the modulated IF signal for transmission. Then, at the destination, it is necessary only to reverse the process and convert the reconstituted IF signal to the desired RF output channel.
Depending on the format, digital terminal equipment may take the form of switches, routers, or add-drop multiplexers. Where standards-based formats such as SONET are used, network capacity can be upgraded by simply using terminal equipment that supports higher data rates on the network side. Rates up to 10 Gb/s are commercially deployed.
18.2.5 Switched Video Headend Structures
In traditional entertainment-only cable systems, the headend served the function of receiving signals from various locations and processing them into a common spectrum for distribution throughout the distribution network. As discussed in Chapter 9, considerable complexity is added by the requirement to generate custom spectra for each node containing both regional differentiation and information to be transmitted to and received from individual customers.
An important example that illustrates the architectural trade-off between node size, RF bandwidth, and frequency reuse is the use of headend switching in the offering of near video on demand (NVOD) — typically a selection of movies played on a fixed schedule with frequent start times — or narrowcast (meaning low viewership) video services. The characteristics of both service categories are that the video sources operate on a fixed schedule not controlled by the viewer.
If a common spectrum of signals is distributed throughout a cable system, then the number of program offerings is obviously limited by the spectrum devoted to that service classification and the efficiency with which that spectrum is used.
The usage of video channels, however, is very nonuniform, with the result that some channels are not always being watched in all nodes. For example, if a network is built with 1,000 home nodes, subscriber penetration is 60%, and simultaneous usage (percentage of homes watching television at any one time) is 67%, then 400 viewers will be watching at the peak hour of usage in any one node. Based on current viewing patterns in North America, the major television networks command approximately 50% of the viewing audience. Including those, the total percentage of customers tuned to any of the 20 most popular channels on any cable system is likely to be over 90%. Thus, regardless of the number of channels offered, the maximum number of additional channels that will be watched at any one time in any node is less than 40 and will approach that number only if the viewers of less popular programming all want to view something different from each other. Put another way, any channel whose audience share is less than 0.3% has a less than 50% likelihood of being accessed in any given node at any given time.
These statistics suggest an alternative way to provide channels of limited interest. If viewers’ channel requests could be communicated to the headend, then a video switch and separate set of modulators for each node (or multinode group of customers) could be used to provide less popular channels only as required. If this structure were put in place, it would remove any limitation on the number of program offerings except for the economic limitations on switch size. Such an analog architecture was tested successfully in a U. S. cable system, though it was not widely deployed.4 A digital video variant is now being offered by a different manufacturer. This version has the dual advantages of more efficient use of bandwidth and the ability to utilize the two-way set-top box communications infrastructure already provided for interactive video services, such as VOD.5 The principles are illustrated in Figure 18.9.
The switched NVOD/narrowcast service differs from a true video-on-demand (VOD) service in the assumption that all available NVOD and narrowcast channels are always present at the input of the matrix switch, whereas in the VOD case, channels are generated only by the server in response to (and under control of) individual subscriber requests. In the application described, video is not generated in response to subscriber request but only switched, so it is available to subscribers fed from specific nodes. Thus, the result is switched broadcast, not interactive VOD.
The trade-offs in deploying switched broadcast services are complex.
• Serving Group Size (r in Figure 18.9). Smaller serving groups will each need fewer multiplexers and modulators (n) because the number of switched channels required will be statistically less, though the relationship is not linear. Balancing that, the number of sets of multiplexers and modulators increases linearly with the number of serving groups (m), which is inversely proportional to serving group size.
• Division between common and switched bandwidth. The more channels that are offered through the switched platform (x), the higher the statistical likelihood that a customer on a given serving group will request a switched channel (thereby increasing the number of modulators per serving group), though, again, the relationship is not linear.
• Statistical sharing of modulators among nodes. At the expense of additional RF switching hardware and complexity, frequency agile modulators could be shared among nodes, in recognition of the fact that the number of subscribers requesting switched channels will vary randomly with time. That would require replacing the RF combining networks with switching matrices and managing both the frequencies of the modulators and their assignment to nodes.
• Sharing of channels among subscribers. When more than one subscriber in a given node requests a switched channel, the system can simply supply the program on two channels or, at a slight increase in complexity, can direct the second requestor to the already supplied program. This capability protects the system against overload when an occasional offering on a switched channel is unusually popular.
As an example, consider an NVOD service that offers the top 10 movies, each starting every 15 minutes. Given a 2-hour average movie length, this would require 80 program streams if every event were to be available to every subscriber. At a 10:1 net compression ratio (that is, 10 digitized, compressed movies transmitted in the spectrum generally used by a single analog channel), 8 RF analog channels would be required.
If this offering were made using a switched broadcast architecture, then we need to make some traffic assumptions. Assume that each service group includes 1,000 homes. At a 70% basic service penetration, the group will include 700 video customers. If 50% of those choose to lease a digital set-top converter, the potential market for NVOD services in each node will be 350 customers. Assuming the monthly movie buy rate among those customers is 100% (one movie, on average, per subscriber per month), the daily buy rate will be about 12. If one-third of those daily buys overlap in the busiest evening hour, then four simultaneous streams would be required. A single RF channel with 10-stream capacity would thus have 2.5 times the required capacity in 1/8 the spectrum required without switching.
A major technical issue with switched NVOD is reducing the time between subscriber channel selection and the appearance of video on the TV screen to the point that it is not noticeably slower than selection among nonswitched channels. Whether utilization of a switched architecture for delivery of NVOD and/or narrowcast channels is cost effective when compared with increased RF bandwidth is dependent on many factors, which vary with the system and will change with time as the cost of various technologies matures.
Note that the switched broadcast architecture shown in Figure 18.9 could be combined with VOD services in the same spectral range if some of the inputs to the MPEG MVX were fed by on-demand servers.
18.3 Architectural Examples
Engineering is the art of solving a technical problem for the lowest cost, given a well-defined set of requirements for performance, reliability, scalability, and other factors. Unfortunately for the engineer, the problem is not always well defined, though the budget for a given project may be. In that case, the task is to find innovative ways of meeting the short-term essential goals while planning for ill-defined future expansion. Such is almost always the case when designing the architecture for a cable television system. The many variants now deployed are testimony to the ingenuity of different designers. Thus, this section will concentrate more on the trade-offs among options than on comparing complete architectures.
The basic requirements are the same for any architecture.
• The end-of-line technical quality must meet corporate standards. At the least, it must meet FCC-defined performance levels with a sufficient margin to cover normal operating variations.
• The reliability and availability must be adequate for the services to be offered.
• The bandwidth must be sufficient to deliver all services to all customers. As a practical matter, the available bandwidth will be divided among commonly-delivered signals and those delivered to specific customers on a demand basis.
• The number of potential customers affected by any single network element failure must be limited to those who can be reasonably responded to by the customer service organization.
• The initial capital investment must be as small as possible, consistent with delivering the initial slate of services to the existing, or anticipated, initial number of customers.
• The network must be capable of cost-effective expansion to deliver additional services, to respond to higher utilization of initially defined services, and to respond to greater penetration among homes passed. This is especially true when the operator is not the incumbent operator with an established customer base. This requirement is known as scalability, and the best designs allow scalability that is specific to individual services and specific geographic areas.
18.3.1 Historic Development and Small-System Architectures
One of the earliest HFC architectures to develop was the cable area network (CAN), conceived as a rebuild strategy for older coaxial networks. The principal is illustrated in Figure 18.10. Fiber-optic links from the headend were used to transmit the common spectrum to selected nodes inserted at regular intervals in a former all-coaxial distribution network. This effectively shortened the cascades, which allowed the system to be upgraded to a greater RF bandwidth. Reliability, which was improved by shortening the cascade, was further enhanced by leaving the coaxial network between the nodes in place as a backup in case a fiber strand or transmitter failed.
Though a number of systems were upgraded using the CAN architecture, they were limited by being constrained to a common channel lineup. In fact, the architecture was useful primarily as a mechanism for minimum-cost bandwidth expansion and reliability improvement.
The first step toward truly innovative fiber architectures was similar to the CAN architecture, except that instead of just inserting nodes into an existing amplifier cascade, greater cascade reduction was realized by feeding both directions from the nodes. This was originally known as a fiber backbone and, like CAN, was proposed as a cost-effective system upgrade strategy. It became favored over CAN because, although it lost the redundancy feature, far fewer fiber links were needed for a given cascade reduction and fiber links were perceived to be highly reliable. The critical difference, however, was that the fiber backbone architecture had created the ability to independently program nodes, thus creating a simple star network.
18.3.2 Node Size and Architecture
A basic building block in any HFC architecture is the node, where optical signals are detected and launched into coaxial distribution cables, and the coaxial distribution subnetwork extending from there to customers. An important decision that fundamentally limits per-customer effective bandwidth is the number of potential customers served by each node.
As the cost of fiber cables, transmitters, and receivers has dropped and performance has improved, the smallest economic node sizes have dropped continuously. With sufficiently small nodes, it has become practical to offer bandwidth-intensive subscriber-specific services such as telephony, high-speed modems, and video on demand. RF bandwidths of 870 MHz combined with 500-home nodes are common for new construction, but many designs call for much smaller nodes, with a continuum from there down to fiber-to-the-curb (FTTC) designs that utilize individual drops to customers from the node. Also common are designs in which multiple coaxial distribution legs are fed from a common node location, with the node configured so that it can be divided electrically into separate logical nodes, one for each distribution leg.
Figure 18.11 shows one example of a three-step serving-area division plan that utilizes evolutionary capabilities built into some manufacturers’ nodes. As initially installed, one optical receiver detects the commonly delivered analog signals and the other detects node-specific signals. After RF combining, the output is split to feed all four coaxial output legs, each of which is designed to deliver signals to approximately 125 homes. Each leg consists of an express feeder followed by an amplifier that drives tapped distribution cables. In the upstream direction, all four legs are combined at RF and fed to one input of a shared A/D converter, followed by an upstream transmitter. Thus, the node size is 500 homes in both directions. Three active fibers are required.
At the first level of scaling, a second node-specific downstream receiver is added and the downstream configuration changed to create two 250-home downstream virtual nodes, and a second two-input upstream converter/transmitter module is added (or a four-input A/D used — both are commercially available) to create four 125-home upstream virtual nodes. This asymmetry is consistent with predictions that upstream bandwidth will be saturated before downstream bandwidth. Five fibers are required to support this configuration, unless WDM is used. This is a very cost-effective upgrade, since only modules in the node, and the matching headend equipment, need be changed.
At the third level of evolution, fibers (dashed lines in the figure) are lashed over the express feeder to each amplifier location. Optical receivers are added there for only the node-specific programming, with the power and common signals traveling over the coaxial link. There are several options for upstream node division. In this example, CWDM transmitters are added at each amplifier location and a CWDM multiplexer at the original node location. On the assumption that eight amplifiers are ultimately required to serve the 500-home serving area, the final virtual node size has been reduced to approximately 62 homes. With downstream and upstream CWDM devices at the original node location, as few as three fibers from the hub would be sufficient to support this level of division.
18.3.3 Multilayer Networks
As pointed out earlier, pure star networks require a lot of glass and typically have single points of failure affecting many subscribers because the fibers feeding multiple nodes share the few available physical paths leading away from headends. These problems are proportionately worse as systems get bigger and as nodes get smaller. Given the trend toward regional consolidation and deeper fiber penetration, different architectures are required.
These larger networks include intermediate processing points called hubs. Depending on the specifics of the design, more than one intermediate layer may be specified. The addition of hubs offers some or all of the following advantages, depending on design:
• Shared use of fibers in the headend-to-hub link, thereby reducing the number of fibers required
• Provision of redundant transport between headend and hub, thereby improving reliability and reducing the possibility of outages affecting many subscribers simultaneously
• Use of intermediate signal processing at each hub to realize a net improvement in end-of-line quality
• Distribution of switched-service processing equipment to reduce instantaneous demand on headend-to-hub circuit capacity
There are two divergent opinions on how best to use hubs. One camp prefers to push as much signal-processing equipment out to the hubs as possible. This reduces the bandwidth requirement between headend and hub, because many signals are locally generated at the hub and only the multiplexed network-side signals need be transported back to the headend. For instance, VOD movies may be stored at hubs and streamed from there to specific customers on demand; only the occasional content updates need be sent down from the headend to hub-based servers. Similarly, telephone host digital terminals (HDTs) at hubs can terminate calls from subscribers, and only the multiplexed DSx signals are transported over baseband optical links back to the headend where the switch is located.
A second advantage is potentially improved analog video quality. If such signals are digitized, transported in that form to the hub, and reconverted to analog for transmission over the relatively short distance to customers, a higher end-to-end quality may be realized, especially where the total distance from headend to customer is at the outer limits of analog transport links. If the signals transported from headend to hub are all digital, the distance between them can be virtually unlimited, allowing arbitrarily large networks.
Offsetting these advantages, it is almost always more costly to divide processing among several locations than to do it all in the headend. Second, the more complex the signal processing at each hub, the more likely failures are to occur there. Then system operators are faced with a choice between a longer time to restore service outages and the higher staffing costs required to station personnel at each hub. Finally, the larger and more complex the hub, the higher will be the associated costs for real estate, emergency powering systems, air conditioning, etc.
The alternative multilayer philosophy centralizes as much signal processing as possible at the headend. Hubs are then used primarily or exclusively for optical processing and redundancy switching. For instance, DWDM transport may be used in a redundant sheath ring between headend and hub, with redundancy switching and DWDM demultiplexing there to separate the signals for each served node. Thus, the operator is able to realize reduced fiber usage and redundancy over the greater part of the path between the headend and the subscriber while utilizing a very compact and inexpensive hub. With 16-wave-length multiplexing it is entirely possible to build a hub serving more than 100 nodes in a half-dozen standard equipment racks, with similarly modest requirements for power and air conditioning.
Offsetting these advantages, purely optical nodes, though more efficient than pure star networks, typically require more fibers in the headend-to-hub link than hubs that include more local processing functions. Also, the use of linear optical transport puts an upper limit on headend-to-hub distance and thus on system size. Although optical amplification is possible, each stage of amplification adds noise, and fiber-related cross modulation and second-order distortion increase with distance, as discussed in Chapter 13. Depending on the specific design, a typical upper limit is about 30–40 miles.
These architectural philosophies are not necessarily exclusive. A large system might well have several interconnected major hubs containing much of the signal generation and service-support equipment, each connected through linear optical links to smaller optical hubs that support clusters of nodes.
18.3.4 Distributed Architecture Example
Figure 18.12 is a simplified diagram showing a typical distributed architecture that illustrates the trade-offs possible. To keep the diagram uncluttered, the upstream paths from subscribers are not explicitly shown. It should be understood, however, that they terminate at the hub and are routed to the appropriate service equipment. Also, the optical transmitters feeding nodes are not shown. The distributed architecture is characterized as follows.
• All signals between the headend and hubs are transmitted in baseband, digital form, so distance limitations are removed.
• Each hub creates its own FDM spectrum for local nodes, thereby allowing the lineup to be customized on a regional basis (for example, for public, educational, and governmental access channels and regionalized advertising insertion.)
• Direct subscriber electronic interaction all takes place at hubs.
• Headend functions include “outside world” interfaces, master signal acquisition and manipulation, master content control for interactive video services, and network monitoring functions.
Within that general description, however, many variations are possible. For instance, the diagram shows diverse routing between two points, whereas in fact, a ring configuration is usually used that connects several hubs. In that case, some combination of repeating and tapped ring is employed, depending on the nature of the signals. Also, the diagram shows separate optical transceivers for each service category, though an operator may choose to partition the bandwidth of shared transceivers among several services. As an example, a SONET transceiver may be partitioned to carry both DSx services and gigabit Ethernet services. Finally, although a single master headend is shown, in fact multiple headends may be utilized for redundancy and/or because it is more efficient to centralize the equipment for specific services at different locations.
Analog Video Options
The form of digital transport for analog video channels is not shown in Figure 18.12. In order to avoid visible degradation, however, lossy compression techniques should be avoided. Options include high-bit-rate MPEG2 and proprietary constant-bit-rate schemes based on digitizing the signal at the common intermediate frequency of 41–47 MHz rather than at baseband.
Noninteractive Digital Video Options
The common digital video spectrum will almost always be delivered in multi-program MPEG2 streams to set-top terminals. Multiprogram streams can be created at the headend or the hub, and the individual streams, whether multi-program or individual, can be transmitted via MPEG transport, IP transport, DSx, or other common protocols.
Interactive Digital Video Options
The object of distributed processing for video-on-demand services is to stream individual programs to customers from the hub while managing the local server content from the central headend. As with noninteractive digital video, there are many options for sending content and control information to hubs, because it needn’t be done in real time.
Telephone Options
In the example shown in Figure 18.12, the telephone service is provided over a proprietary constant-bit-rate (CBR) technology, using separate hardware and bandwidth dedicated to this service. With this option, host digital terminals (HDTs) provide the interface between the RF transmissions to and from subscribers and standard multiplexed telephone formats known collectively as digital signals (DSs). A hierarchy of DSs are available, including DS0 (a single digitized phone call), DS1 (1.544 Mb/s, comprising 24 DS0s plus overhead), and DS3 (about 45 Mb/s and comprising 28 DS3s). HDTs usually internally multiplex to the DS1 level, and an external DSx multiplexer combines those into DS3s, which can then be transported over SONET links between hub and headend, where they can be interfaced with a standard Class 5 telephone switch using a standard GR303 interface.
Should the operator offer telephone service using voice-over-IP (VoIP) technology, the telephone equipment at the hub would be eliminated. Instead, a router in the high-speed data equipment at the headend would sort voice from Internet packets and route the voice packets to a gateway device that would convert them to standard DS format for switch interface.
High-Speed Data Options
Each CMTS is capable of a downstream speed of 30–38 Mb/s. Therefore, the network-side interface needs to be capable of at least that rate. Depending on the design, however, multiple CMTSs may share a common chassis, with internal multiplexing to higher rates. In either case, various options are available for connection to an aggregation router, including 100BaseT and gigabit Ethernet. The aggregated hub data traffic may be transported, as suggested by Figure 18.12, by a gigabit Ethernet link but equally over a SONET link or any other link of adequate speed.
If VoIP services are provided over the shared data network, it is vitally important that the entire data network between the CMTS and voice switch interface respect the latency requirements necessary for transparent voice transmission.
Distribution Network Options
Figure 18.12 shows a simple combining of one leg from the common services combiner with the output of a node-specific combiner, which could be used to feed a dedicated transmitter for each node. Although an entirely usable scheme, most headends are much more complex. In general, each interactive service will be configured so that various numbers of nodes can share signals. Additionally, common signals and node-specific signals may travel over separate paths to the nodes, as shown in Figure 18.11. Various options for transmitting signals are discussed in Chapter 13; flexible node combining is discussed in the next section.
18.3.5 Centralized Architecture Example
Figure 18.13 is an example of the optical distribution network in a centralized architecture. As can be seen, it is characterized by:
• Separate processing of commonly distributed downstream signals (or, more accurately, area-specific, since some channels will generally vary from city to city in a metropolitan area), node-specific downstream signals, and upstream signals
• Generation of node-specific signals at the headend level
• Analog, QAM-modulated redundant transport of node-specific signals from headend to optical hub using DWDM to reduce the number of fibers required
• Analog, baseband-modulated redundant transport of upstream signals from hub to headend using DWDM
• An all-optical hub, with no RF or baseband signals except for status monitoring
• Flexibility in scaling of each interactive service within the headend
It is easiest to analyze this network in sections.
Optical Transport Options
The optical transport architecture uses an externally modulated transmitter and a dedicated fiber for the analog-modulated video signals. At the hub, the signal is optically amplified and split 32 times, with one leg feeding each node on a dedicated fiber. Two headend transmitters drive separate fibers in the sheath ring configuration to provide redundancy against both transmitter and optical cable failure. DWDM is not an option for this link because the crosstalk mechanisms would result in interference below system specifications. Directly modulated transmitters are similarly not an option, because more than an octave is carried and self-phase modulation interacting with the fiber would result in serious CSO distortion.
The QAM signals directly modulate DFB transmitters that are combined using 16-wavelength DWDM between headend and hub. There, the combined signal is amplified and the wavelengths then separated, with a second dedicated fiber feeding a separate receiver in each node. Since less than an octave is carried and the output of the QAM detector is filtered before combining with the analog video spectrum, the second-order distortion is not a problem. Crosstalk mechanisms are a factor in determining the end-of-line C/N, but it is possible to achieve a 40-dB end-of-line system specification — adequate for 256 QAM signals.
There are other options that a system engineer should consider in deciding on an architecture for the downstream signals. Here are some examples.
• Common and node-specific signals could have been combined at the headend and used to modulate an externally modulated transmitter for each node. This would require fewer node receivers but more expensive transmitters as well as a dedicated fiber for each node.
• The common signals and node-specific signals could have been detected at the hub, filtered and combined, and then used to modulate a 1,310-nm DFB transmitter for transmission to the node, as shown in Figure 18.14. This would trade the cost of the optical amplifiers at the hub, the second downstream receiver at each node, and one downstream fiber for the cost of one DFB transmitter at the hub and the additional noise penalty from two series-connected transmitter-receiver pairs. It would also allow city-specific channels to be mixed at the hub level.
• The common and node-specific optical signals could have been combined optically at the hub, then transmitted over a single fiber to a single receiver at the node. As discussed in Chapter 13, however, this would restrict the node-specific bandwidth to less than 50 MHz to keep the second-order products below channel 2.
• The last option could be made workable with the full-bandwidth QAM signals if the headend-to-hub link were constructed with dispersion-shifted fiber, and very flat DWDM demuxes were used to reduce the second-order products.
• Although the upstream signals in the example were generated by converting the entire 5- to 41-MHz return band to a digital stream and transmitting as a high-speed baseband signal, the architecture would be similar if they were transmitted as analog signals. The trade-off is cost versus upstream optical budget and signal “durability” relative to degradation in the optical link.
There is no right answer here, just options that should be considered and evaluated against required performance, cost, scalability, and maintainability.
Optical Redundancy Options
Other choices determine the degree of redundancy. Figure 18.15, for instance, shows some alternatives for the analog common signal spectrum that could also be applied to the other signals:
• The expanded redundancy option shows that placing two optical amplifiers ahead of the redundancy switch would protect against failure of any active hub device for the cost of one high-power amplifier for every set of nodes.
• As the highly redundant option shows, using two optical amplifiers in the hub driving separate, diversely routed fibers to each node, with a redundancy switch after the detectors there, would protect against amplifier failure, cable cuts between hub and node, and node receiver failure. For highest redundancy, the two sets of hub equipment could be in separate facilities, offering added protection against a major facility failure (such as a fire). The use of redundant rings between hub and nodes is a feature of the Cox Communications “ring-in-ring” architecture.6
• Rather than two separate headend transmitters, a single transmitter can be used to drive both directions around the headend-to-hub ring, at a considerable savings in cost, as shown in the reduced redundancy option. In the case of externally modulated transmitters, this is particularly easy, since a second, isolated output is “free” because of the way Mach-Zehnder modulators work.
Scalability
Scalability is simply a way of expressing the ability of a system to meet growing demand for existing services or to deliver new services. What that means, on a technical level, is the ability to grow the service- and area-specific ratio of bandwidth to homes. Section 18.3.2 covered options for scaling the node and its serving area. Such scaling can obviously be applied on a node-by-node basis, as required, but it affects all services. Additionally, the downstream virtual node can initially be made larger by splitting the downstream QAM transmitters in the headend to feed multiple nodes.
Services can be scaled independent of the distribution plant scaling in a properly designed headend, however. As Figure 18.16 shows, each service-specific set of equipment is connected in an independent leg between the upstream service splitter and the downstream node combiner. One coax lead from each node’s upstream applications splitter is wired to the input combiner/splitter network for each of the services shown, and one coax lead from the output combiner/splitter network for each service is routed to the node combiner area. The only exception is video-based services, where the input to each service (VOD in the example) comes from the shared set-top box (STB) transceivers. Within this general framework, two parameters can be varied to scale services in a node- and area-specific manner — the bandwidth allocated to the service and the number of nodes included in the service group.
Figure 18.17 shows an example of data scaling. Initially, the signals from four nodes are combined at each cable modem termination system (CMTS) input port. The equipment chosen has eight input ports, so the output is split to feed the same 32 nodes. A single 3.2-MHz, 16-QAM upstream RF channel and 6-MHz, 256-QAM downstream RF channel are used. The figure shows some of the evolutionary steps, which ultimately include elimination of upstream combining, addition of three more frequencies, and a reduction in the number of input ports used. In the ultimate configuration, this example allows upstream data scaling of 16:1 and downstream data scaling of 128:1, independent of any distribution system scaling.
Combining the distribution-scaling example of Section 18.3.2 with the first step in the preceding example of data-specific scaling would allow a total downstream data scalability of 256:1 and upstream data scalability of 64:1 within the same basic HFC architecture. Further assuming four 125-home coaxial legs served out of each node, a data service penetration of 50% of homes passed, and 20% simultaneous usage among subscribers, the system would be able to deliver average data rates of 6 Mb/s downstream to each simultaneous user and to receive 3.2 Mb/s upstream from each simultaneous user, all within the standard DOCSIS 1.0 or 1.1 framework and while utilizing only 24-MHz downstream and 12.8-MHz upstream bandwidth.
18.4 Summary
The architecture of a system determines what services it can deliver. It controls effective bandwidth, reliability, flexibility, and distribution of signal processing. This chapter has described some of the architectural elements and how each is related to essential network characteristics. The examples given illustrate choices made or proposed to meet specific service requirements in specific service areas.
The most common architecture in use today is the single star, with nodes connected directly to a single headend. In large regional systems, formerly independent headends are often linked by either digital or 1,550-nm analog fiber-optic links to a large master headend and become hubs. Single coaxial cable low-split-band plans are used in the vast majority of systems, with individual nodes serving 500-2,000 homes. Coaxial amplifier cascades vary from one to six in typical new upgrades.
It is possible to “push” much signal processing out to hubs in order to increase the efficiency of the headend-to-hub links, or to centralize the processing for easier management at the expense of needing more fiber capacity. In the largest systems, both structures may be used in multitiered structures.
In considering an architecture, initial cost is certainly a factor. But just as important is the ability to scale to meet market demand and opportunities without “stranding” capital and without causing excessive service interruptions to existing customers because of required reconfiguration.
The relationship between architecture, service reliability, and service availability is a major topic in itself. This chapter has been limited to options within conventional HFC networks. Chapter 19 discusses FTTC and FTTH networks. The methodology for estimating reliability-related parameters is given in Chapter 20.
1. National Electrical Safety Code. Published by the Institute of Electrical and Electronics Engineers, New York, and approved by the American National Standards Institute as C2-1997.
2. Tom Osterman, Broadband Powering Issues, Cable-Tec Expo ‘96, Nashville, TN, June 1996, pp. 469–473. SCTE, Extor, PA.
3. Dan Kerr, Power Distribution in a Lifeline Network. Communications Technology, May 1997, pp. 64–70.
4. Leo Hoarty, Interactive Television — Architecture Under Test, Symposium Record. International Television Symposium, Montreux, Switzerland, June 1995, pp. 221–239.
5. Ran Oz, Amir Bassan-Eskenazi, and David Large, Unified Headend Technical Management of Digital Services. 1990 NCTA Technical Papers, NCTA, May 2000, pp. 87–97.
6. Cox Communications Corporate Engineering Staff, Rings, Clamshells and Spurs. CED Magazine, February 1995, pp. 24–34.