Cable Modem Systems

One of the most promising solutions to the demand for reasonably priced high-speed Internet access comes from the cable television world. Cable operators, who sometimes call themselves Multiple Service Operators (MSOs) when they deliver access to things other than TV programming, can devote a small fraction of their bandwidth resources to digital data and collect about as much money from subscribers as they would charge for basic cable service and a couple of premium channels.

The transformation of the cable TV network to a digital data network has been fraught with obstacles, though none so serious that money can't overcome them. First, cable TV distribution was designed to transport analog signals via Frequency Division Multiplexing (FDM). Conversion to digital format is not a problem: There are many ways to modulate a carrier to transport digital material-in fact, part of the problem has been that cable operators have been slow to adopt a standard means of digitizing content.

Second, digital streams tend to be somewhat less hardy than TV signals on long segments of coaxial cable. In most cases, MSOs that want to provide data services have found it necessary to upgrade their physical networks with fiber optic cable at the core , creating hybrid fiber-coax infrastructures .

This fiber build-out, which must be performed system-wide, regardless of the number of subscribers who want data services, has been estimated to cost at least $150 per home along the fiber route. It should be noted, however, that fiber trunks tend to increase the overall bandwidth of cable networks, permitting MSOs to provide substantially more channels than a pure coax system could (see Figure 1).

Figure 1: Cable Architecture. Large cable providers typically have termination systems at distribution hubs, with centralized management and Operational Support Systems (OSS) servers.

The third pothole is the problem of upstream transmission. Cable systems were designed for one-way, downstream signal distribution. Provisions for upstream communication require a separate channel-a stream of data on a cable system is not full duplex, as it is on a phone line.

The tree and branch topology of the cable system complicates the upstream transmission problem by exacerbating "ingress noise." Loose connectors, poor shielding, and similar points of high impedance on the coaxial cable (particularly on the customer premises) develop noise signals from amateur radio transmissions, electric motors, and other sources of electrical impulses.

With the cable system's branching structure, ingress noise is additive-in the upstream direction, the problem gets worse as you approach the head end of the system. On typical cable systems, the low end of the spectrum-between 5MHz and 40MHz-is not used for television signals, primarily because this band is noisy , and tracking down the sources of interference is time consuming.

However, MSOs have chosen to use these frequencies as the upstream channel for data services. Several partial solutions have been deployed. One is to install a high-pass filter on customer lines. This step blocks low-frequency noise, but it requires some sort of "windowing" so that upstream data transmissions won't be filtered out along with the noise.

The principal solution to ingress noise has been the modulation technique cable modems and termination systems employ . While downstream traffic is usually modulated using 64 Quadrature Amplitude Modulation (QAM) or 256 QAM, with 6-bit or 8-bit symbols, respectively, upstream traffic uses Quadrature Phase Shift Keying (QPSK) or 16 QAM, with 2-bit or 4-bit symbols, respectively. (The efficiency of a modulation scheme is higher with greater numbers of bits per symbol, but noise interferes with the resolution of those symbols with more bits per symbol.)

Thus, while a 6MHz TV channel can sustain downstream data throughputs as high as 27Mbits/sec, upstream data throughput, limited to narrower bands between 200KHz and 3.2MHz, typically hits the 320Kbit/sec to 10.24Mbit/sec range.

An added problem for upstream flows is contention resolution. Because there is only one downstream transmitter for the whole cable data system, there is no question of downstream contention. On the upstream path , contention is inevitable. Cable data networks typically solve the problem by assigning time slots at the head end. The cable modem termination system is responsible for dividing the available upstream channels into Time Division Multiplexed (TDM) slots and assigning those slots to endpoints that need to send data.

Some MSOs take a shortcut to deal with the upstream conundrums. They use cable modems that are designed to work with an analog modem and dial in to the cable company's router over a telephone line. This option is sometimes referred to as telephone, or telco, return.

Telco return isn't unthinkable because, first of all, common Internet actions such as Web surfing and file downloading are asymmetrical in the same direction as this technique. For these activities, the ratio of downstream traffic to upstream traffic is often greater than 10 to 1. With telco return, MSOs can basically get into the high-speed data business without upgrading their networks for two-way communications.

Disadvantages for users of telco return systems include the need for the phone line to go online; the delay imposed by modem dialing and mating; and, in some instances, the inability to use the cable capacity for symmetrical sorts of activities, such as Internet gaming, IRC, or videoconferencing. For MSOs, telco return systems are complex to configure, often occupying two router ports for a single customer. The dual pathways also complicate troubleshooting and support.

It's The Architecture

The fact that the cable system medium is shared among many subscribers leads to two principal challenges for system operators and users. The first issue is declining performance as incremental users are added. The second issue is security from the activities of other users.

A typical fiber node-the point where optical transmissions are converted to electrical signals that are distributed over coaxial cable-serves between 500 and 1,000 residences. While not every residence will subscribe to high-speed data services, and not all of those who do subscribe will be online at once, cable subscribers often experience service degradations at peak hours-typically in the early evening.

Cable operators can minimize these degradations by allocating additional channels to the data service and by extending fiber nodes farther toward neighborhood end nodes. Both of these solutions are costly, however, and devoting additional bandwidth to downstream data doesn't help if the bottleneck is upstream. In fact, some cable operators, including Excite@Home, have found it necessary to cap the upstream throughput of users at 128Kbits/sec, ostensibly because of subscriber abuses such as operating Web servers in defiance of the operating agreement subscribers must sign.

The security problem may be invisible to many users, but could have an even greater impact on users than a service slowdown . With some models of cable modems, any subscriber with a protocol analyzer can see all the traffic attached to a particular fiber node. Taking financial data, capturing passwords submitted in the clear, and simply eavesdropping on the content of another user 's downloads are all potential threats that cable systems must take steps to address.

Another potential vulnerability associated with a shared-medium network arises when Windows or Macintosh users set up file-sharing , peer-to-peer networks. All the subscribers on a fiber node will be able to see these shared files, directories, and folders. If they aren't password protected, they can be copied and erased at the whim of your neighbors.

Unlike Digital Subscriber Lines (DSLs), which are not physically shared among subscribers, file-sharing services need not be bound to IP to become visible to others. Thus, a NetBEUI-based peer-to-peer network that is not bound to IP is still relatively secure on a DSL system, but not on a cable modem system. If you really need a home network with cable modem service, you may want to use servers with hearty access control features, such as NetWare, Windows NT/2000, or Unix, or else install a firewall between the cable modem and your home network.

The Data Over Cable Service Interface Specification (DOCSIS), a cable industry standard developed by Cable Labs, defines a Baseline Privacy Interface (BPI) in version 1.0. DOCSIS 1.0-compliant cable modems and termination systems from multiple vendors began shipping in 1999. The BPI calls for 56-bit DES encryption of all cable modem traffic and provides for key exchange based on 768-bit RSA public key encryption. BPI solves the eavesdropping problem, though there remain IP and Address Resolution Protocol (ARP) spoofing issues that could be exploited if cable modem vendors don't protect against them.

DOCSIS also defines Physical and Data-link layer protocols, and version 1.1 supports Data-link layer prioritization, which is an important step toward providing reliable QoS for voice and video data.

The cable industry has threatened for years to compete with the Incumbent Local Exchange Carriers (ILECS) in the voice market-after all, it's the only service besides the LECs to run physical media throughout most parts of North America. Until the cable network was converted to two-way traffic with optical fiber, these threats of competition were utterly hollow. Now, however, the possibility of competition in the local loop grows ever more plausible, especially with AT&T's resources poised to convert the rest of the former TCI network to a modern hybrid fiber-coax infrastructure.

The cable modem industry is maturing rapidly . It has faced numerous obstacles and managed to work its way around most of them. Excite@Home and RoadRunner, the two largest providers of cable modem service, have both attempted to be providers of special content as well as raw high-speed access. Given both AT&T's interest in distancing itself from content and AOL's acquisition of Time Warner (which includes RoadRunner), it appears likely that cable systems will be opening up to numerous ISPs, much as DSL services have.

In the coming struggle against DSL, cable modem systems have several advantages. MSOs have deployed optical fiber closer to subscribers-especially residential subscribers-than the phone companies have. At least in theory, MSOs could have a downstream throughput advantage over DSL, as long as they move aggressively to maintain that advantage as they add new subscribers. Cable providers also have a marketing advantage: Once a system has been upgraded, they can be confident that every subscriber's premises will support a connection.

Telcos, on the other hand, face a number of anomalies in their facilities, including distance from a Central Office (CO) and the presence of undocumented past wiring practices, which may make it impossible to connect some customers.

DSL is not without its own advantages over cable modem systems. The absence of the shared medium is foremost, though this advantage will be somewhat neutralized as DOCSIS-compliant cable modems come to dominate the installed base.

Cable modem advocates sometimes claim that DSL systems have the same shortcomings as cable because DSL service providers will be tempted to oversubscribe the link between the CO and the Internet POP. While it's true that a shortsighted service provider could hurt its customers' performance with an undersized uplink, this potential bottleneck is completely different from a medium with a fixed throughput at the core. In other words, the subscriber part of a cable network is a performance bottleneck, while the subscriber part of a DSL network is not.

DSL is more likely to be available in many business districts because cable build-outs have been primarily restricted to residential neighborhoods. Users with a choice between these two high-speed options are fortunate indeed.

This tutorial, number 140, by Steve Steinke, was published in the March 2000 issue of Network Magazine.