One of the significant challenges facing network operators today is the high capital cost of deploying next generation broadband network to individual homes or schools. Fiber to the home only makes economic sense for a relatively small percentage of homes or schools. One solution is a novel new approach under development in several jurisdictions around the world is to bundle the cost of next generation broadband Internet with the deployment of solar panels on the owners roof or through the sale of renewable energy to the homeowner. Rather than charging customers directly for the costs of deployment of the high speed broadband network theses costs instead are amortized over several years as a small discount on the customer’s Feed in Tariff (FIT) or renewable energy bill. There are many companies such as Solar City that will fund the entire capital cost of deploying solar panels on the roofs of homes or schools, who in turn make their money from the long term sale of the power from the panels to the electrical grid. In addition there are no Energy Service Companies (ESCOs) and Green Bond Funds that will underwrite the cost of larger installations.
Thursday, April 1, 2010
Must read: The economics of last mile fiber
Fiber-to-the-X: the economics of last-mile fiber
Lately the word "fiber" has started to become ubiquitous in advertisements for broadband. It’s a synonym for the future, for speed and quality. Everybody tries to connect that synonym to their brand, regardless of their actual network design. In the trade press, acronyms like FttX (which stands for Fiber-to-the-X, where X is your favorite letter or word) are used as if all last mile network architectures with optical fiber are more or less equal.
But they're not. Let’s take a look under the hood and analyze the reasons why fiber is chosen as a medium, then look at the topologies, the architectures, the trade-offs, and the inherent path dependencies of a particular deployment method. Fiber-to-the-curb, fiber-to-the-basement, fiber-to-the-home—truly, not all fiber is created equal.
Getting fiber closer to the home
Ninety-nine percent of the Internet's physical distance has been strung with fiber already; just a minor hop, and home and business users can have a fully fiber connection. The obvious question is, why has fiber been rolled out in globe-spanning networks without any public discussion whatsoever, while deploying fiber in the last mile is a huge deal? The answer is two-fold: money, and natural monopolies.
A Utility Infrastructure Law commonly quoted by engineers says, "The closer you get to the home, the more investment is needed, averaged per home connected." This law applies to all parts of the physical network, like water pipes, sewage pipes, and electricity cables. What are the applicable numbers for telecom cables?
A useful division of communication networks is between core networks (deep sea intercontinental, international, or core networks countrywide between exchanges), backhaul and middle-mile networks (from exchanges to local aggregation points), and access networks (from homes to local aggregation points). A quick, back-of-the-envelope calculation based on expert estimates indicates a relative investment level of 1:3:10 for core:middle:access networks, proving the Utility Infrastructure Law.
Incremental versus sunk costs
The last mile access line is the biggest part of the per-home average investment in the network. On a per-home basis the absolute level of investment isn’t that impressive relative to other investments (two iPhones? A 50-inch HDTV? A front door?) The useful life of a fiber access line probably exceeds 40 years, but the multiplier of the total number of homes makes it a sizable overall investment—much larger than investments in other parts of the network. To make matters worse, there is a major difference between core/middle-mile and access networks.
Middle-mile and core networks are generally shared between many customers, so capacity can be reallocated. These shared networks can be incrementally extended and increased in capacity as demand grows. Moore’s law helps as well; the fact that the increase in capacity per dollar invested in transmission electronics doubles every 18 months already covers a large part of the yearly increase in traffic for middle-mile and core networks.
But for a last-mile access line, in contrast, there is only one potential customer: the occupant of the home. The majority of the cash outlay goes to laying the access line, and it's all sunk costs. The investment can only be recovered for a minor part of the project (the last drop), and then only the first time out—you don’t remove the last drop even if the customer quits paying you for services. It is not possible to reallocate that access line investment to another potential customer, so that line either pays for itself or it doesn't. (Note: electronics are excluded in this analysis, because the gear can be deployed and reallocated as penetration grows).
The maintenance costs of a fiber line are very low. Verizon estimates that the difference in maintenance costs between a copper line and a fiber line, expressed in a Net Present Value of all future gains, exceeds $200 per connection.
Utilization is key
The investment in access infrastructure, i.e., stringing or digging new cables, consists mainly of labor, not materials.
The net effect of the dynamic outlined above is that penetration levels in access networks are critical to making money with fiber, as Benoît Feltenrepeatedly demonstrates in his presentations. The level of penetration has such a financial impact on the cost price per active connection that an ISP can't really offset that impact by having a more efficient operation in the rest of the value chain. A 20 percent relative variation in penetration can lead to a difference in cost price per connection of $10 or more per month.
Given the fact that almost all costs in the access network are sunk, it is hard to envision two or more new fiber access networks being deployed in parallel to each home, leading to a stable competitive environment over time. (Unless the ISP’s or network's owners are allowed to divide the market and raise prices to compensate for the underutilization of the networks). If the medium is no longer limited and the access network is the expensive part of the investment, why duplicate the cables? We not do duplicate cables for electricity or other utilities either, for the same reasons.
Note that the current competition between the two wired communication infrastructures to the home—cable and telephone—is a historical accident. Both networks were built for and financed by services that were originally mutually exclusive (telephony and TV). Providers of each these services financed each access network, with relatively high utilization rates. The much later discovery that each network type could unexpectedly deliver packets of information to the home for some new newfangled thing called the Internet, and that users were willing to pay for that new service, was a stroke of luck. But now, with VoIP and IPTV, utilization is a major factor, and the sell-off of older, underutilized parts of the access network and investment in new networks has started.
It remains to be seen what models will emerge where. The European trend is to strive for unbundling and sharing of at least pieces of the new access network (France, Portugal), if not the complete access network (Netherlands, Switzerland). This gets utilization ratios up and average costs down. The US, in contrast, has chosen up to now a competition between networks.
So if fiber-to-the-home is a relatively large investment, most of it up-front with an expected lifetime of many decades, and if utilization ratios are the key to fiber's financial success, then it makes sense to look hard at architectures and path dependencies. The risk of premature technical obsolescence or a lock-in that would require new investments should be reduced, or at least weighed in the decisions.
Topology versus technology
A fully optical path gives so many options in transmission technology for such a low price that we have only just begun to explore the possibilities for home use.
The tendency amongst vendors and operators is to define fiber access network architectures by the technology that is used instead of the possible topology. Active Ethernet technology is typically coupled to a homerun network, and GPON technology is typically coupled to a Passive Optical Network with shared fibers to each block. This is a mistake, as different architectures allow more or less flexibility in topology and therefore have more or less path dependency in the choice of technology. The right method is to analyze the architectures and the topologies that can be supported by each architecture. In each potential "fiber-to-the-X," the main architectural choice is this: a full optical path to each home/apartment, or not?
In Fiber-to-the-Curb (aka VDSL), Fiber-to-the-Basement, or HFC (Hybrid Fiber/Coax) access networks, the length of copper wire or coax cable to the home is shortened, but not to zero. Somewhere before entering your home, the optical signal is transformed to an electrical signal, which is transported by an older medium. In the HFC network, the electrical signals are shared in the last segment, while in FttC or FttB, they're not. So these access networks effectively have the same architecture as before (telephony/DSL and cable), but with shorter distances for the old medium, allowing for higher capacity. The optoelectronic conversion of signals is positioned closer to the home.
In Fiber-to-the-Home access networks, a full optical path is built to the home. The optoelectronic conversion which defines the transmission technology is positioned in the home (or on the outside wall). The three main architectures are deep PON, staged PON, and homerun (or point-to-point). In deep PON (DPON), a single fiber is fed to a block of homes and split optically close to the homes with a passive splitter to as many fibers and homes as you design for, typically a multiple of two in the range of 32 or 64. Each endpoint receives the same light from the source, and sends light back through the same single fiber. Staged PON (SPON) is a variation where a much smaller number of homes (typically eight) are served in the same manner by a single fiber. In the aggregation point a 1:4 splitter is used to get to the 1:32 ratio. A homerun architecture (point-to-point) does not split light at all, at least between a home and the aggregation point. Each home gets one or more dedicated fibers all the way from the aggregation point.
If the networks at home are all made out of copper/coax (or wireless) anyway, does it matter where the optoelectronic conversion is positioned in the network, inside your home or a few hundred meters away? To start with, the assumption that copper and electrical signal networks will stay dominant in the home is questionable. Swisscom already has started to invest in optical plastic fiber networks that can be used inside the home, to accommodate the much higher capacity needed.
Second, a fully optical path gives so many options in transmission technology for such a low price that we have only just begun to explore the possibilities for home use. The universities and researchers have already pushed the envelope very far, and Moore’s law is helping us to get price levels down to a level where anyone can afford the technology. If you have a fully optical path, the only thing you need to change if the need arises is the equipment. So a fully optical path allows for a very fast and selective upgrade (compared with a slow and laborious physical network upgrade if you did not invest in a fully optical path). Imagine what a fix AT&T would be in if a service that would require a full optical path to the home suddenly would become popular. Verizon could respond quickly, but AT&T would not be able to follow.
The best fiber possible does nothing except guide light of all possible wavelengths. Each wavelength you can separate from others is comparable to the complete “ether” available for all wireless communication. As long as you can modulate the light source as you wish and keep the signal intact through the fiber, you can do whatever you want with the light. This fact allows for many variations in technology over the same fiber.
So let’s look at the options and see what optical technologies are already available and practical for everyday use at the home. What are the path dependencies for each architecture?
Active Ethernet is a technology for data transmission which requires a unique fiber or wavelengths (one down, one return), and it's very well known and stable. One of the technology's advantages is the distance you can cover or the speed of modulation you can achieve (or the reduction in light power you need), because there is no optical splitter that reduces the light strength between you and the receiver and shares the capacity with your neighbors.
GPON is currently the most dominant FttH technology, since it's specifically designed for shared architectures. Again, specific wavelengths, one for down and one for the return channel, are used. Through clever time-division-multiplexing between users, a guaranteed capacity and a much higher peak capacity can be delivered. Currently this requires the same brand of equipment both at the aggregation point and the home.
10GPON is one of the more promising next steps in GPON, delivering up to four times as much capacity. However, the limits of the pulse width/distance/light-strength combination can become a limitation: if the splitter ratio of a Deep PON architecture becomes too high there is a risk that the number of photons per pulse is getting dangerously low after being split (bit error rates high). Increasing the light strength at the source is prohibited by safety laws so that is no option. In a Deep PON architecture you are stuck unless you want to run more fibers to the splitter. A Staged PON architecture does give the option of removing the 1:4 split at the aggregation point. A home run architecture has the option of putting splitters in the aggregation point at will.
Linear TV (RF over Fiber) is a method for delivering radio frequency channels over fiber. A wavelength is modulated with these RF signals, broadcasted to each home, and converted to the same signal that an HFC cable network delivers. In fact, the fiber part of the cable network uses this technology to deliver all these channels to the coax segment.
Because coax has its limitations when it comes to transmitting high frequencies (over 1 GHz) over longer distances, the frequency band for linear TV is limited to the standard 50-900 MHz. The optical equipment and the fiber, however, do not have these limitations (5 GHz is easily broadcasted). Recently, equipment has been introduced that superimposes multiple satellite TV signals in the frequency band higher than 1 GHz in the same optical wave. These signals are separated at the home and converted to multiple outputs that can be fed to standard satellite receivers, as if you had 4 satellite dishes. This technology is very useful in dense cities where homeowners hate all the individual dishes on top of the roof or the balcony.
Wave division multiplex (WDM) involves sending multiple colors over one fiber, and the technology was first used to increase the capacity in core networks. One of its uses in FttH networks is to reach faraway homes in a shared architecture. A standard shared topology (PON) runs into problems, like not enough light, at larger distances, and requires a lower split ratio, more fiber, and a higher average costs per customer. WDM has a much larger optical power budget (15db more), allowing for much longer distances.
Applying WDM in an PON architecture requires ripping out the splitters in the field and replacing them with a wavelength filter that separates the colors to individual customers. The filter selects one wavelength per user, making the fiber plant dedicated to WDM. In a peer-to-peer architecture, all you need is patching in the aggregation point.
The options outlined above are just what are currently available and affordable. Most likely others will be developed in the coming years.
One thing we can learn from history is that “things are difficult to predict, especially the future”. Indeed, we often cannot imagine what we possibly could do with new technology.
European regulators like having the option to unbundle physical access lines, with wholesale bandwidth access as the second best option. But the various architectures have different unbundling path dependencies as well.
For fiber-to-the-curb, physical unbundling is more theory than practice. Allowing multiple ISPs with their own hardware inside a small cabinet is a questionable idea to start with, but the backhaul is more of a problem. With something like 800 connections per cabinet, you need a very deeply distributed backhaul network, something not many can afford, especially competitors of incumbents. Regulating backhaul access for competitors is an option, but it's not an easy situation for new entrants.
Unbundling Deep PON by requiring patch cabinets close to homes (aggregating homerun fibers in splitters from different operators) is an option that has not been tested in practice. It's not clear that this overhead and cost is worth the effort, compared to alternatives like homerun.
Upgrading the architecture
In theory and in practice nothing is forever: you can upgrade a fiber-to-the-curb network to full fiber-to-the-home, or you can upgrade an HFC network to full fiber-to-the-home.
The legacy of a fiber-to-the-curb network, however, is a large number of cabinets with active equipment out in the field. This equipment is something that old network engineers abhor, because the point of fiber is to get an outside plant which you can leave alone forever.
A HFC network can be transformed to either a SPON or DPON or even a peer-to-peer network. In fact, some small cable operators who have shareholders that let them keep their cash flow have embarked on a seven-year program to slowly and quietly extend their network fiber up to the home, in order to build a full peer-to-peer network. This is the cheapest way to get the best possible architecture, with approximately a 20 percent lower investment per connection compared to a new build-out.
One thing we can learn from history is that “things are difficult to predict, especially the future.” Indeed, we often cannot imagine what we possibly could do with new technology. Yet it turns out that in ten years, tools and applications can change in massive ways that nobody imagined beforehand. Our appetite for communication bandwidth seems to be unlimited.
Because the rollout of fiber-to-the-home will take more than ten years it makes a lot of sense to think hard about increasing options and reducing path dependencies.
The main issue with halfway options (FttC, FttB and HFC) that do give more bandwidth for data is the time needed to upgrade the network to full optical when the need arises. It may take a decade to do so because it is physical work, home by home. If some other region or country has gone fully optical in the first place, then they have a huge advantage.
A Deep PON architecture creates a serious lock-in. This architecture is the first to hit the wall of limitations of the pulse width/distance/light-strength combination and is hard to unbundle. A Staged PON has more leeway and has less unbundling limitations.
The most expensive but most flexible architecture is peer-to-peer (home run). All technologies can be mixed and matched on a per-customer basis, without modifying the outside plant. Unbundling is relatively easy, and the extra investment is relatively low in dense areas but increases with distance.
Ultimately, it turns out that not all fiber is equal, as we will discover the hard way in the coming years.