The Ultimate Cable Modem

5:40 PM -- DENVER -- CTAM Summit -- The CableLabs Docsis 3.0 specs call for a modem to be capable of bonding, at a minimum, four upstream channels and four downstream channels. There's not much actual upstream channel bonding occurring in the wild yet, but that 4x4 configuration ensures that MSOs can offer burst speeds of more than 100 Mbit/s in both directions.

Not lame. But not where things are heading.

These days, vendors are working on modem configurations that bond eight or 16 downstream channels, while maintaining a four-channel upstream, Carey Ritchey, general manager of Microtune Inc. (Nasdaq: TUNE)'s cable unit, mentioned to me yesterday. At full burst, a 16-channel downstream (using 6MHz-wide channels) would generate something like 640 Mbit/s.

Definitely not lame. But also not where some see this all going... at least among the mad scientists out there.

Ritchey said he's already seen some requests for Docsis modems that can bond 32 downstream channels. My cable math suggests that would put a burst speed in the neighborhood of 1.2 Gbit/s.

"But that's been the extreme," Ritchey says. It's the type of extreme Cisco Systems Inc. (Nasdaq: CSCO) openly conceptualized in January 2008 at the Society of Cable Telecommunications Engineers (SCTE) Conference on Emerging Technologies with a presentation titled "A proposal for Docsis 4.0."

Seems as though that vision has since taken a few steps closer to reality.

Of course, the question is what sort of extreme would require such a massively wide speed pipe. If I had to guess, it would have to be for something along the lines of a QAM-IP gateway that can do everything today's digital set-tops can now, but also help MSOs pursue a unicast video model in a big way. Someday.

Or maybe it's about something completely different. But what? I'll do more asking around here. Got an idea? Please share it on the message board.

— Jeff Baumgartner, Site Editor, Cable Digital News

Page 1 / 3   >   >>
drewcwsj 12/5/2012 | 3:53:24 PM
re: The Ultimate Cable Modem

32 channels is almost 200Mhz of spectrum. RG-6 has about 3Ghz or less of usable spectrum with just a few splitters and not too much length. Broadcast and on-demand video will need several hundred Mhz based on how many channels and at what resolution are delivered. So maybe 10 subscribers can reserve this kind of bandwdith simultaneously. This may help short term as the average subscriber is still mostly ad-hoc for data usage (email, www, low bandwidth streaming and gaming) but I'm betting that by the time this technology is fully baked the usage models will move have moved to high-bandwidth, continuous streaming. The TV is going to be the display for digital content and an on-demand, quality experience is a necessity. Oh and don't forget a whole lot of people come home at 6pm and turn on the TV. Now imagine a few hundred people on a cable loop all starting a 1080P MPEG4 stream at the same time. Bursting doesn't help. You have to get a whole lot more spectrally efficient.

SabrinaChow 12/5/2012 | 3:53:23 PM
re: The Ultimate Cable Modem

Let's be honest, D3 is really just a means to an end.  Once all ROI has been recouped, the cable architecture must change to mimic that of a Verizon optical system.  Now we are talking RFOG, DPON, DWDM type of design with fiber direct to the home.  The HFC design is already half way there.

paolo.franzoi 12/5/2012 | 3:53:23 PM
re: The Ultimate Cable Modem



You left out one parameter before anyone can give an effective answer which is - at what BER?

In the end it is all the same electromagnetic transmission theory from wireless, dsl and fiber - just have to know the target SNR to represent what BER for what bit rate.


PS - Just to add, of course the coding itself can have SNR gain through FEC.

Stevery 12/5/2012 | 3:53:23 PM
re: The Ultimate Cable Modem

You have to get a whole lot more spectrally efficient.

I am a cable ignoramus.  Judging from the article tho:  640Mbps/16 channels = 40 Mpbs/channel stuffed into 6 MHz (implies something like 128QAM).  Is this in the ballpark?

If true, that means about 7 bps/Hz already present, which begs the question:  What is the ballpark S/N for the system design?  I assume you can always demand that the cable be shorter, better or any one of a number of unrealistic business assumptions.


Stevery 12/5/2012 | 3:53:22 PM
re: The Ultimate Cable Modem

You left out one parameter before anyone can give an effective answer which is - at what BER?

Sorry, the use of the term "modem" made me assume that the wired guys are working under Shannon–Hartley, but that's just leftover bias from dialup days.

But you are right:  Is there an assumption of BER, or is it clean enough to avoid FEC?

Edit:  A quick perusal of wikipedia reminds me that I am an information-theory ignoramus as well.

paolo.franzoi 12/5/2012 | 3:53:21 PM
re: The Ultimate Cable Modem


LOL - If we all sat down and worshipped at the feet of Shannon and Nyquist we would all be better off.

Generally most systems talk about 10-7 BER.  I am not so sure about that anymore with HDTV being more like 10Mb/s per stream and 8 - 15 Mb/s downstream to a home.  I would like to see a 10-10 or 10-11 BER.  Maybe 10-12 if there is satellite reception involved in the path.



paolo.franzoi 12/5/2012 | 3:53:20 PM
re: The Ultimate Cable Modem


I get your SDV argument but I think it has trouble with the homes passed bit.  Let me use 2 SD channels + 1 HD channel + 1 Mb/s (to represent HSI and Voice) per home.

I think that equates to about 25 Mb/s per home.  So, 100 homes on a cable segment represent 2.5 Gb/s.  So, it is a scaling issue - which is basically what FiOS has overcome (with a 32 home per segment system).



paolo.franzoi 12/5/2012 | 3:53:20 PM
re: The Ultimate Cable Modem



There are some FEC schemes for which the virtual SNR improvement is better than the lost bandwidth for certain kinds of noise.  That is why I put it in there.  I am not a cable guy but did spend a lot of time in leased line and dial up modems (as well as DSL and Fiber transmission).

You are right about the channel capacity.  Which is why I always wonder about LTE.




opticaljunkie 12/5/2012 | 3:53:20 PM
re: The Ultimate Cable Modem

Spectrum is not a problem on coax. New cable modems use 256qam downstream, aka spectral efficiency of 8bits/s/hz. Right now HDTV (1080i) uses about 20mbps per channel (yes your cable companies use a lot of compression), with additional channels allocated for on demand etc.

So let's be conservative and say you have only 1ghz of useful coaxial bandwdith (instead of 3), and let's say you can only push 64 qam (6 bits/s/hz) over the spectral range, that gives you raw bandwidth of 6gbps. Let's throw in 100mbps internet (both up and down stream), and let's upgrade the hi-def broadcast to 3D 1080i (~80mbps per channel), assume 5 on demand channels and 5 always on channels, and let's add 100mbps for phone & in home video conferencing, we are only looking at total bandwidth of around 1.1gbps, well below the available (and VERY conservatively estimated) spectrum of 6gbps. Even if we assume a monster FEC with 30% overhead to get BER < 10e-12, there's still plenty remains.

See the trick is to use Layer 2 switched video, instead of broadcast every channels you can selectively broadcast the channel requested by user, which saves massively chunk of bandwidth. Hope this helps.

Stevery 12/5/2012 | 3:53:20 PM
re: The Ultimate Cable Modem

OK, now that I have coffee and a couple of minutes: When Seven made his comment about BER because you get a SNR gain from FEC, I realized I have pretty much forgotten what little I know from Shannon.

The original reason this thread captured my curiousity is that drewcwsj made a comment about increasing the channel capacity.  Now, I have some ideas on this (who doesn't), but I was wondering how close to the limit cable systems are, as well as how much margin the system guys want built in.

The limit on channel capacity depends only on the bandwidth and the SNR.  I believe that Shannon tells me BER does not matter:  eg  Any SNR gain by FEC does not affect the capacity (the improved error-rate by SNR gain is lost to the reduced effective bit rate from FEC), but please correct me if I'm out to lunch.

So my question to The Cable Guys should be: 

<li>You have an installed base of cable, and that cable has some distribution of lengths.&nbsp; </li>
<li>You probably design your system for some upper bound of length, based on the revenue collected for the best service.</li>
<li>This length probably implies an SNR for your 6 MHz channel.&nbsp; What is that number?</li>

As I mentioned previously, it looks like they are at 7 bps/hz right now, implying a lower bound of SNR = 20 dB.

There, it's taken me three posts to just ask the question.&nbsp; I suspect there will be a fourth.

Page 1 / 3   >   >>