Digitizing That Rambunctious Return Path


Some inventions happen by mistake: The guy who set out to make a compound to replace the use of ivory in billiard balls instead came up with the curious substance we now know as Silly Putty.

Other inventions come more painfully: Nikola Tesla struggled for years, enduring scientific scorn from powerful contemporaries (Thomas Edison, for one) before his zealous pursuit of alternating current (AC) power became a still-used reality.

Other inventions make us smack palm to forehead, muttering laments about not conceiving them ourselves: Retractable dog leashes. Picnic backpacks. Luggage with wheels. A 5-42 MHz upstream path, digitized remotely from the headend, with most of its gunk extracted algorithmically.

OK, OK: The eureka of the latter invention wouldn't naturally occur to most people, like you and me, who delight in simple, useful, easy to conjure inventions — like something to thump the excess snow off the encrusted windshield-wiper blade during a big storm.

But to cable technologists, who fret about the scrawny, noisy state of cable's upstream signal path, such an invention is garnering considerable attention. If digitization could do to cable's snarly, 5-42 MHz return path what it's done for the quality of sound and video, it can't be all bad.

In fact, it could be quite good. Extracting upstream signals digitally, and not going through the many analog processes that define today's gear, doesn't just mean a reduction in noise. It also leads to more usable bandwidth, which is important as more and more two-way services require safe, swift upstream passage.

True, cable's suppliers of plant-related equipment have already introduced digitization to the upstream signal path. But so far, their digital stretch reaches only as far as the optical node, and not all the way to the home. An estimated 6 million cable homes are passed by this sort of equipment — not much in plant terms.

This column is not, and never will be, used for purposes of endorsement. But every so often, an invention comes along that seems plausible enough for wider translation. Such is the case with a recently-uncloaked Silicon Valley startup, Pacific Broadband Communications, and a method its 100 engineers devised to make the upstream path a safer, faster place, through remote digitization.

It starts with an Application Specific Integrated Circuit, or "ASIC" (pronounced "a-sick," with a hard "a"). ASICs are chips built for a single, customized purpose. In this case, ASICs are programmed to sit in the headend portion of cable modem systems, called "CMTS," for "cable modem termination system." There, the chips "listen" to the entire, 5-42 MHz spectral chunk, then digitize it.

This "listening" happens across the time and spectral domains — or in real time, and including any noise that temporarily or permanently messes with intended signals. Algorithms (secret number codes) inside the ASIC subtract the noise (called "noise cancellation") and compensate for distortions (called "equalization").

If you had this chip stuck into your forehead (and assuming it worked that way), you could stand at a crowded cocktail party, focus on a person at the other side of the room, and hear precisely what she was saying — minus the clinking of glasses, the clunking of forks and the ridiculous babble of the large man standing next to you. Actually, you could tune up to 16 conversations you wanted to hear, and discard the rest.

In technology terms, noise cancellation and equalization improve the upstream carrier-to-noise (CNR) ratio — tech-speak for a baseline measurement that compares signals sent against the noises they encounter. The improvement is quantified by as much as 7 decibels, or 7 dB. If true, that's pretty huge. It's the difference between, say, 2.5 mbps and 10 mbps, in the same 3.2 MHz upstream chunk.

Translation: Way more throughput, for the same bandwidth.

Plus, because everything happens digitally, this particular ASIC is equipped to automatically handle more, and simultaneous, upstream "channels" — sixteen, as opposed to one in contemporary CMTS gear.

The ability to dynamically assign more upstream channels could, in turn, be helpful with respect to node re-combining. In today's analog cable-modem gear, CMTS ports are often shared among four or so 500-home nodes, at least until subscriber penetrations get high enough to warrant a decoupling.

Node decoupling, as it exists right now, can cause operational angst, not to mention service interruptions. Thus, the ability to handle penetration-related throughput issues by automatic node-realignment is of particular interest to data technologists. Every little bit helps.

Of course, a useful invention — whether accidental, tortured, or obvious — is one thing. Making it work is quite another, particularly in a setting as predictably hostile as is cable's upstream signal path.

But, if the old adage about necessity being the mother of invention is true, then cable's upstream path has never been more acutely needful of some new mothering. And maybe it all starts with a digital cleansing.