Widening Cable's Upstream Path
By Leslie Ellis
from the May 21, 2001 issue of Broadband Week
The subject of cable's upstream signal path often is likened to the old saw about trying to fit 10 pounds of, shall we say, "stuff" into a five-pound bag. Inherently slender, the 5-40 MHz upstream zone is shared among users; spectrally, it's not exactly a warm, cozy place.
And if that's been true since the early days of two-way signaling--when upstream signals consisted of impulse pay-per-view clicks--then it's even more profound now, as peer-to-peer, telephony and other new applications jostle for upstream access.
Cable's hybrid fiber-coax (HFC) architectures are inherently asymmetrical, with as much as 99 percent of total capacity dedicated to downstream (headend to customer) traffic, and less than 1 percent available for upstream (customer to headend) use. If the downstream ride is a point-to-multi-point architecture, the upstream ride is multipoint-to-point ... with a predictable effect on data passage. Plus, the modulation techniques used to send interactive data upstream were built for sturdiness, not speed.
That's the bad news. The good news: A growing list of techniques, both operational and technological, are or will be available that make the upstream ride future-friendly. Node splitting, for example, halves or quarters the number of subscribing households sharing the upstream ride; better modulation increases the speed of the ride.
There's no easy way to model a worst-case scenario for upstream traffic. Straight division of available capacity by number of users doesn't work, because there are too many variables: Service penetration rates, how many users are simultaneously online, and what those users are doing. An interactive TV request, for example, isn't nearly as needy as a bandwidth-slurping video file transmitted by a cable modem engaged in peer-to-peer computing.
As it turns out, only about 25 MHz of capacity in the 5-40 MHz upstream path is really usable--the zone between 25-40 MHz--because of the spectral rowdiness below 20 MHz.
Cable providers currently use an upstream modulation technique called "QPSK," or "quadrature phase shift key." It's the same technique used in direct broadcast satellite transmissions. It's different, slower and sturdier, from the downstream modulation used for digital traffic ("QAM," or quadrature amplitude modulation). QPSK yields a usable capacity (after overhead) of about 1.2 Mbps/MHz. Multiplied by the 25 MHz of usable upstream bandwidth; the aggregate upstream capacity is about 30 megabits per second.
For now, QPSK is a workable way to shuttle traffic through an environment teeming with interlopers, which are the bursts of noise leaking into the system either from individual customer premises or external sources. Regardless, noise elbows in on the intended signal, then smooshes together with everyone else's noise and intended signal as traffic moves multi-point to point. Everything gets amplified as it moves to the headend.
What's the buzz?
There are two primary noise types that ruffle the upstream path. One is impulse noise, also known as "electrical transient noise." "Ingress noise" is the other. Both sometimes are referred to as "spurious noise," to mean undesirably intrusive noise.
Impulse noise is a form of interference caused by quick jolts of unwanted electrical energy--when electric-powered things, from furnaces to hair dryers, kick on. It briefly slams into the cable plant, crushing any information traveling on the network. Then, as quickly as it careened in, it's gone. The trampled bits usually are not recoverable and need to be re-sent by the cable modem, set-top, phone unit or other device that had been trying to transmit.
Ingress noise happens when unwanted RF signals leak into a section of coaxial cable, then travel along with the intended signal as it traverses the upstream path. Engineers estimate that as much as 70 percent of the noise in the upstream leaks in from individual subscriber homes, for a variety of reasons: Improperly installed F-connectors; cracked or improperly shielded coaxial cable; even bad shielding around the TV's tuner can do it.
Who's the Boss?
It is the cable modem termination system (CMTS, the headend part of cable modem systems) that arbitrates who gets what upstream bandwidth. Current CMTS technology treats every upstream request identically and slots available upstream capacity equally among all.
Example: If 250 people on a 50 percent penetrated, 500-home node need to use the upstream path--including, say, 50 greedies and 200 non-greedies--the CMTS accepts the traffic on a one-by-one basis. After the first set of packets comprising everyone's individual needs is collected, the CMTS starts through the queue again. Whether it's a one-word reply to an instant message or a portion of a video file, it's the same: Get in line. One packet at a time.
On the other hand, if one cable modem customer happens to be the only one online at a given moment in time, that doesn't equate to a 30 Mbps upstream ride. Most MSOs cap the amount of upstream bandwidth to around 384 kbps.
Changemakers
When new gear based on the cable industry's DOCSIS 1.1 standard comes out, it will be technically possible to charge more for prioritized CMTS treatment. Plus, there are two other ways to cajole the upstream into carrying more stuff.
One is to split the node, which halves or quarters the number of homes and businesses sharing an upstream ride. Most current opto-electric node equipment comes with four output legs. Each leg connects to the coaxial portion of the network. That yields two potential splits, say, from 500 passings to 250 passings, then from 250 passings to 125 passings, although the numbers usually are not that exact.
MSOs say it takes about a half-day and around $10,000 to split a node, including labor and equipment (mostly lasers transmitters/receivers).
The second way to massage the upstream is to use what engineers call "higher order modulation, " such as 16-QAM, in devices such as cable modems, interactive digital set-tops and phone gear that rely on the upstream path. While 16-QAM already is aboard most cable modem chipsets shipping now, most MSOs haven't yet activated it. It's faster, but requires a cleaner path.
In the "good" upstream spectral zone, between 25-40 MHz, 16-QAM provides about 62.5 Mbps of aggregate throughput, to QPSK's 30 Mbps. Coupling 16-QAM in the upstream with a single-split of a 500-home node gives some subset of 250 homes-passed, shared access to 62.5 Mbps. Half the homes share twice the speed. That's much better than some subset of 500 homes-passed sharing 30 Mbps of aggregate upstream throughput.
And, better forms of upstream modulation are on the horizon. The CableLabs DOCSIS group is considering two forms of advanced modulation now and plans to make a decision this summer. At one time, this work was unofficially called "DOCSIS 1.2." Then, it was officially changed to "Advanced PHY," for advanced physical layer.
The Advanced PHY contenders: "S-CDMA" (Synchronous-Code Division Multiple Access) modulation, developed by Terayon Corp., and a technique called "FA-TDMA" (Frequency Agile-Time Division Multiple Access) proposed by Broadcom Corp., Texas Instruments, and others.
Boiled way down, both techniques claim to increase upstream throughput dramatically, even in the face of big noise. Terayon says its method is backward-compatible to DOCSIS 1.0 and 1.1, approaches 128-QAM speeds, and does well with noise. Broadcom touts 64-QAM speeds and backward-compatibility. Both claim to upshift and downshift around noise.
Whatever the ultimate decision, advanced PHY will emerge in the 2002 timeframe; moving to 16-QAM likely will happen sooner. While most MSOs say they haven't yet seen traffic patterns that necessitate node-splitting, all are armed to do it now. All together, they go far in assuaging cable's upstream dearth.
|
