residential networking over telephone

Recently, I covered some of the history of Ethernet's tenuous relationship with installed telephone cabling. That article focused on the earlier and more business-oriented products, but many of you probably know that there have been a number of efforts to install IP networking over installed telephone wiring in a residential and SOHO environment. There is a broader category of "computer networking over things you already have in your house," and some products remain pretty popular today, although seemingly less so in the US than in Europe.

The grandparent of these products is probably PhoneNet, a fairly popular product introduced by Farallon in the mid-'80s. At the time, local area networking for microcomputers was far from settled. Just about every vendor had their own proprietary solution, although many of them had shared heritage and resulting similarities. Apple Computer was struggling with the situation just like everyone; in 1983 they introduced an XNS-based network stack for the Lisa called AppleNet and then almost immediately gave up on it [1]. Steve Jobs made the call to adopt IBM's token ring instead, which would have seemed like a pretty safe bet at the time because of IBM's general prominence in the computing industry. Besides, Apple was enjoying a period of warming relations with IBM, part of the 1980s-1990s pattern of Apple and Microsoft alternately courting IBM as their gateway into business computing.

The vision of token ring as the Apple network standard died the way a lot of token ring visions did, to the late delivery and high cost of IBM's design. While Apple was waiting around for token ring to materialize, they sort of stumbled into their own LAN suite, AppleTalk [2]. AppleTalk was basically an expansion of the unusually sophisticated peripheral interconnect used by the Macintosh to longer cable runs. Apple put a lot of software work into it, creating a pretty impressive zero-configuration experience that did a lot to popularize the idea of LANs outside of organizations large enough to have dedicated network administrators. The hardware was a little more, well, weird. In true Apple fashion, AppleTalk launched with a requirement for weird proprietary cables. To be fair, one of the reasons for the system's enduring popularity was its low cost compared to Ethernet or token ring. They weren't price gouging on the cables the way they might seem to today. Still, they were a decided inconvenience, especially when trying to connect machines across more than one room.

One of the great things about AppleTalk, in this context, is that it was very slow. As a result, even though the physical layer was basically RS-422, the electrical requirements for the cabling were pretty relaxed. Apple had already taken advantage of this for cost reduction, using a shared signal ground on the long cables rather than the dedicated differential pairs typical for RS-422. A hobbyist realized that you could push this further, and designed a passive dongle that used telephone wiring as a replacement for Apple's more expensive dongle and cables. He filed a patent and sold it to Farallon, who introduced the product as PhoneNet.

PhoneNet was a big hit. It was cheaper than Apple's solution for the same performance, and even better, because AppleTalk was already a bus topology it could be used directly over the existing parallel-wired telephone cabling in a typical house or small office. For a lot of people with heritage in the Apple tradition of computing, it'll be the first LAN they ever used. Larger offices even used it because of the popularity of Macs in certain industries and the simplicity of patching their existing telephone cables for AppleTalk use; in my teenage years I worked in an office suite in downtown Portland that hadn't seen a remodel for a while and still had telephone jacks labeled "PhoneNet" at the desks.

PhoneNet had one important limitation compared to the network-over-telephone products that would follow: it could not coexist with telephony. Well, it could, in a sense, and was advertised as such. But PhoneNet signaled within the voice band, so it required dedicated telephone pairs. In a lot of installations, it could use the second telephone line that was often wired but not actually used. Still, it was a bust for a lot of residential installs where only one phone line was fully wired and already in use for phone calls.

As we saw in the case of Ethernet, local area networking standards evolved very quickly in the '80s and '90s. IP over Ethernet became by far the dominant standard, so the attention of the industry shifted towards new physical media for Ethernet frames. While 10BASE-T Ethernet operated over category 3 telephone wiring, that was of little benefit in the residential market. Commercial buildings typically had "home run" telephone wiring, in which each office's telephone pair ran directly to a wiring closet. In residential wiring of the era, this method was almost unheard of, and most houses had their telephone jacks wired in parallel along a small number of linear segments (often just one).

This created a cabling situation much like coaxial Ethernet, in which each telephone jack was a "drop" along a linear bus. The problem is that coaxial Ethernet relied on several different installation measures to make this linear bus design practical, and home telephone wiring had none of these advantages. Inconsistently spaced drops, side legs, and a lack of termination meant that reflections were a formidable problem. PhoneNet addressed reflections mainly by operating at a very low speed (allowing reflections to "clear out" between symbols), but such low bitrate did not befit the 1990s.

A promising solution to the reflection problem came from a company called Tut Systems. Tut's history is unfortunately obscure, but they seem to have been involved in what we would now call "last-mile access technologies" since the 1980s. Tut would later be acquired by Motorola, but not before developing a number of telephone-wiring based IP networks under names like HomeWire and LongWire. A particular focus of Tut was multi-family housing, which will become important later.

I'm not even sure when Tut introduced their residential networking product, but it seems like they filed a relevant patent in 1995, so let's say around then. Tut's solution relied on pulse position modulation (PPM), a technique in which data is encoded by the length of the spacing between pulses. The principal advantage of PPM is that it allows a fairly large number of bits to be transmitted per pulse (by using, say, 16 potential pulse positions to encode 4 bits). This allowed reflections to dissipate between pulses, even at relatively high bitrates.

Following a bit of inter-corporate negotiation, the Tut solution became an industry standard under the HomePNA consortium: HomePNA 1.0. HomePNA 1.0 could transmit 1Mbps over residential telephone wiring with up to 25 devices. A few years later, HomePNA 1.0 was supplanted by HomePNA 2.0, which replaced PPM with QAM (a more common technique for high data rates over low bandwidth channels today) and in doing so improved to 10Mbps for potentially thousands of devices.

I sort of questioned writing an article about all of these weird home networking media, because the end-user experience for most of them is pretty much the same. That makes it kind of boring to look at them one by one, as you'll see later. Fortunately, HomePNA has a property that makes it interesting: despite a lot of the marketing talking more about single-family homes, Tut seems to have envisioned HomePNA mainly as a last-mile solution for multi-family housing. That makes HomePNA a bit different than later offerings, landing in a bit of a gray area between the LAN and the access network.

The idea is this: home run wiring is unusual in residential buildings, but in apartment and condo buildings, it is typical for the telephone lines of each unit to terminate in a wiring closet. This yields a sort of hybrid star topology where you have one line to each unit, and multiple jacks in each unit. HomePNA took advantage of this wiring model by offering a product category that is at once bland and rather unusual for this type of media: a hub.

HomePNA hubs are readily available, even today in used form, with 16 or 24 HomePNA interfaces. The idea of a hub can be a little confusing for a shared-bus media like HomePNA, but each interface on these hubs is a completely independent HomePNA network. In an apartment building, you could connect one interface to the telephone line of each apartment, and thus offer high-speed (for the time) internet to each of your tenants using existing infrastructure. A 100Mbps Ethernet port on the hub then connected to whatever upstream access you had available.

The use of the term "hub" is kind of weird, and I do believe that at least in the case of HomePNA 2.0, they were actually switching devices. This leads to some weird labeling like "hub/switch," perhaps a result of the underlying oddity of a multi-port device on a shared-media network that nonetheless performs no routing.

There's another important trait of HomePNA 2.0 that we should discuss, at least an important one to the historical development of home networking. HomePNA 1.0 was designed not to cause problematic interference with telephone calls but still effectively signaled within the voice band. HomePNA 2.0's QAM modulation addressed this problem completely: it signaled between 4MHz and 10MHz, which put it comfortably above not only the voice band but the roughly up-to-1MHz band used by early ADSL. HomePNA could coexist with pretty much anything else that would have been used on a telephone line at the time.

Over time, control of HomePNA shifted away from Tut Systems and towards a competitor called Epigram, who had developed the QAM modulation for HomePNA 2.0. Later part of Broadcom, Epigram also developed a 100Mbps HomePNA 3.0 in 2005. The wind was mostly gone from HomePNA's sails by that point, though, more due to the rise of WiFi than anything else. There was a HomePNA 3.1, which added support for operation over cable TV wiring, but shortly after, in 2009, the HomePNA consortium endorsed the HomeGrid Forum as a successor. A few years later, HomePNA merged into HomeGrid Forum and faded away entirely.

The HomeGrid Forum is the organization behind G.hn, which is to some extent a successor of HomePNA, although it incorporates other precedents as well. G.hn is actually fairly widely used for the near-zero name recognition it enjoys, and I can't help but suspect that that's a result of the rather unergonomic names that ITU standards tend to take on. "G.hn" kind-of-sort-of stands for Gigabit Home Networking, which is at least more memorable than the formal designation G.9960, but still isn't at all distinctive.

G.hn is a pretty interesting standard. It's quite sophisticated, using a complex and modern modulation scheme (OFDM) along with forward error correction. It is capable of up to 2Gbps in its recent versions, and is kind of hard to succinctly discuss because it supports four distinct physical media: telephone, coaxial (TV) cable, powerline, and fiber.

G.hn's flexibility is probably another reason for its low brand recognition, because it looks very different in different applications. Distinct profiles of G.hn involve different band plans and signaling details for each physical media, and it's designed to coexist with other protocols like ADSL when needed.

Unlike HomePNA, multi-family housing is not a major consideration in the design of G.hn and combining multiple networks with a "hub/switch" is unusual. There's a reason: G.hn wasn't designed by access network companies like Tut; it was mostly designed in the television set-top box (STB) industry.

When G.hn hit the market in 2009, cable and satellite TV was rapidly modernizing. The TiVo had established DVRs as nearly the norm, and then pushed consumers further towards the convenience of multi-room DVR systems. Providing multi-room satellite TV is actually surprisingly complex, because STV STBs (say that five times fast) actually reconfigure the LNA in the antenna as part of tuning. STB manufacturers, dominated by EchoStar (at one time part of Hughes and closely linked to the Dish Network), had solved this problem by making multiple STBs in a home communicate with each other. Typically, there is a "main" STB that actually interacts with the antenna and decodes TV channels. Other STBs in the same house use the coaxial cabling to communicate with the main STB, requesting video signals for specific channels.

Multi-room DVR was basically an extension of this same concept. One STB is the actual DVR, and other STBs remote-control it, scheduling recordings and then having the main STB play them back, transmitting the video feed over the in-home coaxial cabling. You can see that this is becoming a lot like HomePNA, repurposing CATV-style or STV-style coaxial cabling as a general-purpose network in which peer devices can communicate with each other.

As STB services have become more sophisticated, "over the top" media services and "triple play" combo packages have become an important and lucrative part of the home communications market. Structurally, these services can feel a little clumsy, with an STB at the television and a cable modem with telephone adapters somewhere else. STBs increasingly rely on internet-based services, so you then connect the STB to your WiFi so it can communicate via the same cabling but a different modem. It's awkward.

G.hn was developed to unify these communications devices, and that's mostly how it's used. Providers like AT&T U-verse build G.hn into their cable television devices so that they can all share a DOCSIS internet connection. There are two basic ways of employing G.hn: first, you can use it to unify devices. The DOCSIS modem for internet service is integrated into the STB, and then G.hn media adapters can provide Ethernet connections wherever there is an existing cable drop. Second, G.hn can also be applied to multi-family housing, by installing a central modem system in the wiring closet and connecting each unit via G.hn. Providers that have adopted G.hn often use both configurations depending on the customer, so you see a lot of STBs these days with G.hn interfaces and extremely flexible configurations that allow them to either act as the upstream internet connection for the G.hn network, or to use a G.hn network that provides internet access from somewhere else. The same STB can thus be installed in either a single-family home or a multi-family unit.

We should take a brief aside here to mention MoCA, the Multimedia over Coax Alliance. MoCA is a somewhat older protocol with a lot of similarities to G.hn. It's used in similar ways, and to some extent the difference between the two just comes down to corporate alliances: AT&T is into G.hn, but Cox, both US satellite TV providers, and Verizon have adopted MoCA, making it overall the more common of the two. I just think it's less interesting. Verizon FiOS prominently uses MoCA to provide IP-based television service to STBs, via an optical network terminal that provides MoCA to the existing CATV wiring.

We've looked at home networking over telephone wiring, and home networking over coaxial cable. What about the electrical wiring? G.hn has a powerline profile, although it doesn't seem to be that widely used. Home powerline networking is much more often associated with HomePlug.

Well, as it happens, HomePlug is sort of dead, the industry organization behind it having wrapped up operations in 2016. That might not be such a big practical problem, though, as HomePlug is closely aligned with related IEEE standards for data over powerline and it's widely used in embedded applications.

As a consumer product, HomePlug will be found in the form of HomePlug AV2. AV2 offers Gigabit-plus data rates over good quality home electrical wiring, and compared to G.hn and MoCA it enjoys the benefit that standalone, consumer adapters are very easy to buy.

HomePlug selects the most complex modulation the wiring can support (typically QAM with a large constellation size) and uses multiple OFDM carriers in the HF band, which it transmits onto the neutral conductor of an outlet. The neutral wiring in the average house is also joined at one location in the service panel, so it provides a convenient shared bus. On the downside, the installation quality of home electrical wiring is variable and the neutral conductor can be noisy, so some people experience very poor performance from HomePlug. Others find it to be great. It really depends on the situation.

That brings us to the modern age: G.hn, MoCA, and HomePlug are all more or less competing standards for data networking using existing household wiring. As a consumer, you're most likely to use G.hn or MoCA if you have an ISP that provides equipment using one of the two. Standalone consumer installations, for people who just want to get Ethernet from one place to another without running cable, usually use HomePlug.

It doesn't really have to be that way, G.hn powerline adapters have come down in price to where they compete pretty directly with HomePlug. Coaxial-cable and telephone-cable based solutions actually don't seem to be that popular with consumers any more, so powerline is the dominant choice. I can take a guess at the reason: electrical wiring can be of questionable quality, but in a lot of houses I see the coaxial and telephone wiring is much worse. Some people have outright removed the telephone wiring from houses, and the coaxial plant has often been through enough rounds of cable and satellite TV installers that it's a bit of a project to sort out which parts are connected. A large number of cheap passive distribution taps, common in cable TV where the signal level from the provider is very high, can be problematic for coaxial G.hn or MoCA. It's usually not hard to fix those problems, but unless an installer from the ISP sorts it out it usually doesn't happen. For the consumer, powerline is what's most likely to work.

And, well, I'm not sure that any consumers care any more. WiFi has gotten so fast that it often beats the data rates achievable by these solutions, and it's often more reliable to boot. HomePlug in particular has a frustrating habit of working perfectly except for when something happens, conditions degrade, the adapters switch modulations, and the connection drops entirely for a few seconds. That's particularly maddening behavior for gamers, who are probably the most likely to care about the potential advantages of these wired solutions over WiFi.

I expect G.hn, MoCA, and HomePlug to stick around. All three have been written into various embedded standards and adopted by ISPs as part of their access network in multi-family or at least as an installation convenience in single-family contexts. But I don't think anyone really cares about them any more, and they'll start to feel as antiquated as HomePNA.

And here's a quick postscript to show how these protocols might adapt to the modern era: remember how I said G.hn can operate over fiber? Cheap fiber, too, the kind of plastic cables used by S/PDIF. The HomePlug Forum is investigating the potential of G.hn over in-home passive optical networks, on the theory that these passive optical networks can be cheaper (due to small conductor size and EMI tolerance) and more flexible (due to the passive bus topology) than copper Ethernet. I wouldn't bet money on it, given the constant improvement of WiFi, but it's possible that G.hn will come back around for "fiber in the home" internet service.

[1] XNS was a LAN suite designed by Xerox in the 1970s. Unusually for the time, it was an openly published standard, so a considerable number of the proprietary LANs of the 1980s were at least partially based on XNS.

[2] The software sophistication of AppleTalk is all the more impressive when you consider that it was basically a rush job. Apple was set to launch LaserWriter, and as I mentioned recently on Mastodon, it was outrageously expensive. LaserWriter was built around the same print engine as the first LaserJet and still cost twice as much, due in good part to its flexible but very demanding PostScript engine. Apple realized it would never sell unless multiple Macintoshes could share it---it cost nearly as much as three Mac 128ks!---so they absolutely needed to have a LAN solution ready. LaserWriter would not wait for IBM to get their token ring shit together. This is a very common story of 1980s computer networks; it's hard to appreciate now how much printer sharing was one of the main motivations for networking computers at all. There's this old historical theory that hasn't held up very well but is appealing in its simplicity, that civilization arises primarily in response to the scarcity of water and thus the need to construct irrigation works. You could say that microcomputer networking arises primarily in response to the scarcity of printers.