We were pretty blown away by the idea of using G.hn chips for MDU broadband access links, when we wrote it up in our 27 February edition of Faultine, under the headline ‘Marvell throws G.hn at the last mile for 500 Mbps Korean broadband,’ so we scheduled a call with the company to talk through the possibilities.
So this week Chano Gomez Martinez, described simply as a product manager at Marvell, finally had time to talk us through the implications, of which we think there are many.
Apparently the Marvell G.hn chip was first being shipped a year ago and sampling clients asked whether or not it could be used for broadband access networks. It was never part of the plan, but suddenly it seemed to make sense.
The main focus on G.hn was that it enabled any of three existing lines in the home, coax, twisted pair and powerline, to be used with the same chip and an almost identical MAC. Our main coverage point was ‘would it work’ and then it became ‘will it survive,’ as it had Broadcom, Qualcomm Atheros and a number of other major chip vendors hostile to it.
Had those stories not been the order of the day, along with the rise and rise of the IEEE 1905 specification, that was supposed to blend all of its rivals, HomePlug, WiFi and MoCA into a single network, then we may have noticed that broadband access networks were a probable destination, rather than just in-home applications.
But the story in February was simply that Korea Telecom was signing up to get 500 Mbps to each apartment in a multi-dwelling unit, over existing twisted pair. The idea is to push up to 10 Gbps over EPON to the basement and then split this between 24 twisted pair lines. It calls its solution G.Now and said that it can also bond 2 channels to help extend the reach, and add noise protection using two chips – the Marvell 88LX3142 G.hn digital baseband processor and the 88LX2718 G.hn analog front end.
Other chip vendors supporting G.hn include Sigma Designs and Chinese group Metanoia, while Lantiq, which was one of the original supporters of G.hn seems to have fallen by the wayside and has put all its eggs into the VDSL2 vectoring basket. Which is perhaps why Gomez Martinez has left Lantiq and turned up at Marvell.
The twisted pair PHY on G.hn uses most of the spectrum up to 100 MHz, leaving only a small gap to avoid FM radio interference. So it makes sense that it can outperform VDSL2 which uses up to a maximum of 30 MHz of spectrum on the wire, but usually just 12 MHz or 17 MHz. The problems is that it will attenuate faster than VDSL2.
‘It all depends on distance,’ says Gomez Martinez, ‘if you use up to 30 MHz in VDSL2, you can use it over 600 to 700 meters. Our technology sweet spot is closer to 100 meters to 200 meters.’
He said very clearly at this point that he was not considering central office installations, but if you think about it, this is misleading, because no-one is. G.Fast itself, which is due for standardization by the end of 2014, with chips a year later and services a year after that, works best at 250 meters, and that certainly isn’t considering a central office any more. In fact the central office is moving lock stock and barrel to the cross connect box, and could eventually move, like cable systems to just outside each home.
Gomez Martinez positions the G.Now chips using twisted pair as something that can stave off moving to fiber, which is precisely what G.Fast is targeted at. The difference is that G.Now can give you 500 Mbps now, in an existing chip, and G.Fast can’t realistically bring it for almost three years. You have to ask ‘Who’s in a hurry?’ because fiber has to get closer and closer to the home, and that’s a lot of work and Capex before you need a solution, and in many cases this won’t make much sense any earlier than G.Fast. This is true everywhere that telcos are comfortable with their broadband speeds, but where they are up against strong cable operators with advanced levels of DOCSIS, there is a need to go the extra last mile, as it were.
‘The price on VDSL2 and G.Now is going to be very similar given that we have deployments, and our shipments are rising. But the price of vectoring makes the cost far higher for VDSL2,’ says Gomez Martinez. That surely gives the edge to G.hn in the access network then, since VDSL2 without server side vectoring is likely to give rise to just a 50 Mbps service, while G.Now would be far faster, depending precisely on the distance a given house is from the fiber. The G.Now installation has near end Cross talk cancellation built in, but has no defense against far end Cross talk. Even if you compare top of the line between these two technologies, server based cross talk cancellation at an exchange using VDSL2 and line bonding tops out at 200 Mbps. Even if this was slightly cheaper, the 500 Mbps that Marvell has boasted for MDU broadband delivery is not even its top speed. ‘When we get to within 50 meters from the home, we can get the speed over twisted pair up to 750 Mbps,’ Gomez Martinez said. This probably means aggregate throughput, so could be 150 Mbps upstream and 600 Mbps downstream, but it’s still plenty of bandwidth to usher in 4K OTT streams.
Which is why he is targeting this as an interim fiber technology. About 50% of the cost of taking fiber right to the home, is the last 50 meters, as you come into the property, dig up and fix their garden, and require scheduled visits and user permissions. If you put the cost of laying fiber to the home (different in different countries and topologies) at around $1,000 per home, then this approach saves you $500 for now, and keeps telcos on track to compete with cable. And it’s available now, not in three years’ time.
If you take that calculation a little further, it means that it becomes overnight a ‘no brainer’ for the rising tide of fiber installations, to halve their costs and have the entire industry switch overnight to this kind of hybrid. Anything up to 200 meters from the home is going to be worth considering for a FTTC strategy.
Now we know that the G.Fast standard allows for two profiles, one starting at where VDSL leaves off at 30 MHz and going up to 106 MHz and another at 212 MHz, which attenuates even faster. We suppose that chips will emerge that can do either. But if all the low hanging fruit is already taken in FTTC installations, and remember it has 3 years in which to achieve this, this seriously undermines the economic installation rate for G.Fast, so much so, that it may prevent it being a success, except in that higher 212 MHz profile, which will no doubt come later.
Potentially in one customer signing Marvell has changed the economic profile of one of the most anticipated telco based broadband technologies, and that’s surely significant. ‘Don’t say that,’ said Gomez Martinez, ‘I still have a lot of friends in VDSL.’ Maybe he doesn’t want to say it, but someone has to.
Marvell has multiple customer trials going on in secret, both in parts of Asia Pacific and in its native America for G.Now as an access technology over twisted pair. We would hazard that Verizon would be crazy NOT to have a trial of this. It has pushed it FiOS FTTH system out to within reach of about 18.5 million US homes, and it has more or less stopped there. It’s strategy was considered by investors to be too expensive, and it is preparing for a shift to mobile video service delivery.
But if it could re-enliven that broadband drive, using this type of technology, while saving 50% of the costs of deployment, then it could accelerate its broadband sign ups. The top end of that service is called Quantum and already headlines with an offer of 500 Mbps and upload speeds up to 100 Mbps ‘ which can clearly be hit with G.Now over the last 50 meters.
And the possibilities do not end there. G.hn over coax runs across 90 MHz of coax spectrum, offering a PHY of 900 Mbps and according to Gomez Martinez a real world throughput of 750 Mbps, as well. While this is not something that can be used in cable installations, it can be used in MDUs and wherever existing coax is laid, and it can go up against the existing efforts of MoCA chips from Entropic and Broadcom, as well as a DOCSIS based Ethernet over Coax from Broadcom, and a strange HomePlug AV chip version of Ethernet over Coax from Atheros. This is a very different game with many local technology specialists having to be supported for what are initially smaller direct sales. As a result this has been a slow burn access market, but Marvell already claims to be speaking to one Tier 1 operator in the Asia Pacific about this very subject. This market has very local difficulties, and it depends on where signal amplifiers are placed inside an MDU and whether or not they are bidirectional, which often they are not, which is why it has traditionally been so slow.
Which brings us to the powerline capabilities of G.hn. Could it be used to take broadband access into and around a home, well yes it could, but it is slower than the other existing wires and still requires a termination point inside the home, a Home Gateway, and it has difficulties in MDUs where power suppliers are often shared and interfere with one another. So while it is a nice idea to shove broadband in the basement of an MDU or to within 50 meters and broadband simply arrives in your electricity supply, that is unlikely to happen. Powerline remains diverse and tricky for access, but is a perfect target for in-home distribution, and is a straight fight between Homeplug and G.hn. The more chips G.hn ships, the more of a threat it is here.
‘We could conceivably take a multi-line approach rather like the IEEE 1905 effort. An operator might create a single network from the basement that can use coax, twisted pair or the powerlines, all with an intelligent Ethernet switch at the bottom managing things,’ said Gomez Martinez. ‘Path selection could be carried out using real world info and the system being load balanced, and WiFi Access points being used to distribute inside an MDU apartment.’ But here we can tell he is just speculating to show off the power of his technology, because he confirms there are no such trials in process.
It will be interesting to see if this technology can steer its way through the various bake-offs that it will need to in the near term, in time for it to steal the early G.Fast market, or as Gomez Martinez prefers to call it the FTTC market. We think it will certainly be taken seriously by every Telco out there, and many of their equipment suppliers too.