T-carrier

Few aspects of commercial telecommunications have quite the allure of the T-carrier. Well, to me, at least, but then I have very specific interests.

T-carrier has this odd, enduring influence on discussion of internet connections. I remember that for years, some video game installers (perhaps those using Gamespy?) used to ask what kind of internet service you had, with T1 as the "highest" option. The Steam Hardware Survey included T1 among the options for a long time. This was odd, in a lot of ways. It set T1 as sort of the "gold standard" in the minds of gamers, but residential internet service over T1 would have been very rare. Besides, even by the standards of the 2000s T1 service was actually pretty slow.

Still, T1 involved a healthy life as an important "common denominator" in internet connectivity. As a regulated telephone service, it was expensive, but available pretty much anywhere. It also provided a very high standard for reliability and latency, beyond many of the last-mile media we use today.

Telephone Carriers

We think of telephone calls as being carried over a pair of wires. In the early days of the telephone system, it wasn't all that far off to imagine a phone call as a single long circuit of two wires that extended all the way from your telephone to the phone you had called. This was the most naive and straightforward version of circuit switching: connections were established by creating a circuit.

The era of this extremely literal form of circuit switching did not last as long as you might think. First, we have to remember that two-wire telephone circuits don't actually work that well. Low install cost and convenience means that they are the norm between a telephone exchange and its local callers, but for long-distance carriage over the phone network, you get far better results by splitting the "talk" and "listen" sides into two separate pairs. This is called a four-wire telephone circuit, and while you will rarely see four-wire service at a customer's premises, almost all connectivity between telephone exchanges (and even in the internals of the telephone exchange itself) has been four-wire since the dawn of long-distance service.

Four-wire circuits only exacerbated an obvious scaling problem: in the long distance network, you have connections called a toll leads between two exchanges. In a very simple case, two towns might have a toll lead between them. For simple four-wire telephone lines, that toll lead needs four wires for each channel. If it has four wires, only one phone call can take place between the towns at a time. If it has eight wires, two telephone calls. This got very expensive very fast, considering that even heavily-built four-crossarm open wire routes might only have a capacity for eight simultaneous calls.

For obvious reasons, research in the telephone industry focused heavily on ways to combine more calls onto fewer wires. Some simple electrical techniques could be used, like phantoms that combined two underlying pairs into a single additional "phantom" pair for a 50% increase in capacity. You could extend this technique to create more channels, with a noticeable loss in quality.

By the 1920s, the Bell System relied on a technique that we would later call frequency division multiplexing (FDM). By modulating a phone call onto a higher-frequency carrier, you can put it over the same wires as other phone calls modulated onto different frequencies. The devices that actually did this combined multiple channels onto a single line, so they were known as channel banks. The actual formats they used over the wire, since they originally consisted of modulation onto a carrier, came to be known themselves as carriers. AT&T identified the carriers they developed with simple sequential letters. In the 1940s, the state of the art was up to J- and K-carrier, which allowed 16 channels on a four-wire circuit (over open wire and multipair cable, respectively). A four-crossarm open-wire circuit, with sixteen pairs, could support 256 unidirectional circuits for 128 channels---or simultaneous phone calls.

FDM carriers reached their apex with the coaxial-cable based L-carrier and and microwave radio TH and TD carriers ¹, which combined channels into groups, groups into supergroups, and supergroups into mastergroups for total capacities that reached into thousands of channels. Such huge FDM groups required very large bandwidths, though, which could not be achieved over copper pairs.

In the 1950s, rapidly developing digital electronics lead to the development of digital carriers. These carriers relied on a technique called Pulse-Code Modulation or PCM. PCM has sort of an odd name due to the history; it dates so far back into the history of digital communications that it was named before the terminology was well-established. "Pulse-code modulation" really just means "quantizing an analog signal to numbers and sending the numbers," which is now intuitive and obvious, but was an exciting new development of the 1940s.

PCM had a lot of potential for carrying phone calls, because the relaxed needs of voice transmission meant that calls could be digitized to fairly low-rate streams (8kHz by 8 bits) and engineers could see that there was a huge variety of possible techniques for combining and transporting digital signals. The best thing, though, was that digital signals could reliably be transmitted as exact copies rather than analog recreations. That meant that PCM telephone calls could pass through a complex network, with many mux/demux steps, without the reduction in quality that analog FDM carriers caused.

Even better, analog channel banks were large systems with a lot of sensitive analog components. They were expensive to build, required controlled environments, and were still subject to drift that required regular maintenance by technicians. Digital technology involved far fewer sensitive analog components, promising cheaper equipment with less maintenance. Digital was clearly the future.

The question, though, was how to best carry these digital signals over the wiring that made up much of the telephone system: copper pairs. In the late 1950s, Bell Laboratories developed T-carrier as the answer.

T-Carrier

T-carrier is a specification for transporting a stream of bits over copper pairs. The plural here is important: because T-carrier supported multiple channels, it was developed as a trunk technology, used for communication between telephone exchanges rather than between a telephone exchange and a customer. So, like other carriers used for trunks, T-carrier was four-wire, requiring two pairs to carry bidirectional signals.

T-carrier operates at 1.544Mbps, and that's about all you can say about it. The logical protocol used over T-carrier, the actual application of those bits, is determined by a separate protocol called the digital signal or DS. You can roughly think of this as a layer model, with DS running on top of T.

Here, we need to address the elephant in the room: the numbers. The numbers follow a convention used throughout AT&T-developed digital standards that is most clearly illustrated with DS. A DS0, by analogy to DS raised to the zeroeth power, is one single telephone channel expressed as a PCM digital signal. Since a telephone call is conveyed as 8-bit samples at 8kHz, a DS0 is 64kbps of data.

A DS1 is a combination of 24 DS0s, for 1.544Mbps.

A DS2 is a combination of 4 DS0s, for 96 channels or 6.312Mbps.

A DS3 is a combination of 7 DS2s, for 672 channels or 44.736Mbps.

Each level of the DS hierarchy is a TDM combination of several instances of the level below. The numbers are kind of odd, though, right? 24, 4, 7, it has the upsetting feeling of a gallon being four quarts each of which is two pints.

The DS system was developed in close parallel with the carriers actually used to convey the signal, so the short explanation for this odd scheme is that a DS1 is the number of DS0s that fit onto a T1 line, and a DS2 is the number of DS1s that fit onto a T2 line. The numbers are thus parallel: DS1 over T1, DS2 over T2, DS3 over T3. The distinction between T and DS is thus not always that important, and the terms do get used interchangeably.

But still, why 24?

Well, it's said that the number of channels on a T1 was just determined empirically. The 64kbps rate of a DS0 was fixed by the 8b x 8kHz digital format. A test T1 installation was built using a typical in-place copper telephone cable, and the number of channels was increased until it no longer functioned reliably. 24 channels was the magic number, the most achieved without unacceptable errors.

T-Carrier Infrastructure

T1 was designed for operation over a "typical" telephone cable trunk. In the 1950s, this meant a twisted-pair telephone cable installed in 6,600 foot sections with a loading coil at the junction of each section. A loading coil was essentially a big inductor hooked up to a telephone line at regular intervals to compensate for the capacitance of the line---long telephone cables, even four-wire, needed loading coils at regular intervals or the higher frequencies of speech would be lost. Loading coils also had disadvantages, though, in that they imposed a pretty sharp maximum frequency cutoff on the line. High-speed digital signaling needed to operate at those high frequencies, so T1 was designed to fit into existing long cables by replacing the loading coils with repeaters.

That means that T1 required a repeater very 6,600 feet. This repeaters were fairly small devices, often enclosed in pressurized cans to keep water out. 6,600 feet might sound pretty frequent, but because of the loading coil (and splice box) requirements trunk lines usually had underground vaults or equipment cabinets at that interval anyway.

Over time, the 6,600 foot interval became increasingly inconvenient. This was especially true as end-users started to order T1 service, requiring that T1 be supported on local loops that were often longer than 6,600 feet. Rather than installing new repeaters out in the field, it became a widespread practice to deliver T1 over a version of DSL called HDSL. HDSL is older and slower than the newer ADSL and VDSL protocols, and requires four wires, but it was fast enough to carry a DS1 signal and could cover a much longer span than traditional T-carrier. HDSL used the voice frequency band and thus could not coexist with voice calls like ADSL or VDSL, but this had the upside that it "fully controlled" the telephone line and could use measures like continuous monitoring (using a mark signal when there was no traffic) to maintain high reliability.

For the era of internet-over-T1, then, it was far more likely that a given customer actually had an HDSL connection that was converted to T1 at the customer premises by a device called a "smart jack." This pattern of the telco providing a regulated T1 service over a different media of their choice, and converting it at the customer premises, is still common today. T1s ordered later on may have actually been delivered via fiber with a similar on-premises media converter, depending on what was most convenient to the telco.

T1 is typically carried over telephone cable with 8P8C modular connectors, much like Ethernet. It requires two pairs, much like the two-line telephone wiring commonly installed in buildings. However, like most digital carriers, T-carrier is more particular about the wiring than conventional telephone. T1 wiring must be twisted-pair, and it is polarity sensitive.

DS1 Protocol

The DS1 protocol defines the carriage of 24 64kbps channels over a T1 interface. This basically amounts to TDM-muxing the 24 channels by looping over them sending one byte at a time, but there are a surprising number of nuances and details.

Early versions of DS1 only actually carried 7 bits for each sample, which was sufficient for a telephone call when companding was used to recover some of the dynamic range. The eighth bit was used for framing. T-carrier is a completely synchronous system, requiring that all of the equipment on the line have perfectly synchronized "frame clocks" to understand what bits belong to which logical channels. The framing mechanism provided a synchronization signal to achieve this perfect coordination. Later improvements in the framing protocol allowed for the use of all eight bits in some or even all of the samples. This gets to be a complicated and tangled story with many caveats, so I am going to leave it out here or this article would get a lot longer and probably contain a bunch more mistakes.

The various combinations of technologies and conventions used at different points get confusing. If you are curious, look into "robbed bit signaling," an artifact of the transition of where framing and control signals in T1 were placed that was, for some reason, a pet topic of one of my college professors. I think we spent more time on robbed-bit signaling than we did on all of MPLS, which is way cooler. Anyway, the point of this is to understand the protocol overhead involved: T1 operates at 1.544Mbps, but at least one 8-bit "sample" must be used for framing purposes, leaving 1.536Mbps of actual payload. The payload may be further reduced by other framing/signaling overhead, depending on exactly how the channel bank is configured. Most of these issue are specific to carrying telephone calls (and their related signaling); "internet service" T1 lines typically used a maximally-efficient configuration.

The Internet

So far we have pretty much only talked about telephone calls, because that's what T-carrier was developed for. By the 1980s, though, the computer industry was producing a lot of new applications for high-speed digital connections. T1 was widely available, and in many cases a tariffed/regulated service, so it was one of the most widely available high-speed data connections. Especially in the very early days of the internet, was often the only option.

Into the 1990s, T1 was one of the dominant options for commercial internet access. It was rarely seen in homes, though, as it was quite expensive. Keep in mind that, from the perspective of the phone company, a T1 line was basically 24 phone lines. They charged accordingly.

To obtain internet service, you would order internet service either from the telco itself or from an independent provider that then ordered a connection from your premises to their point of presence on an open access basis. In this case you were effectively paying two bills, one to the telco for the T1 line and the other to the independent provider for internet connectivity... but the total was still often more affordable than the telco's high rates for internet services.

Because of the popularity of T1 for internet access, routers with T1 interfaces were readily available. Well, really, the wide variety of media used for data connections before Ethernet became such a common standard means that many routers of the era took interchangeable modules, and T1 was one of the modules you could get for them.

In implementation, a T1 line was basically a fast serial line from your equipment to your ISP's equipment. What actually ran over that serial line was up to the ISP, and there were multiple options. The most classical would be frame relay, an X.25-derived protocol mostly associated with ISDN. PPP was also a fairly common option, as with consumer DSL, and more exotic protocols existed for specialized purposes or ISPs that were just a little weird.

When the internet was new, 1.5Mbps T1 was really very fast---the NSFNET backbone was lauded for its speed when it went online as an all-T1 network in 1991. Of course, today, a 1.5Mbps "backbone" internet connection is pretty laughable. Even as the '90s progressed, 1.5Mbps started to feel tight.

One of the things I find odd about the role of T1 in the history of internet access is that the era when a T1 was "blazing fast" was really pretty short. By 2000, when online gaming for example was taking off, both DSL and cable offered significantly better downstream speeds than T1. However, the nature of T1 as a circuit-switched, telephone-engineered TDM protocol made it very reliable and low-latency, properties that most faster internet media performed poorly on (early DSL notoriously so). Multiplayer gaming would likely have been a better experience on T1 than on a DSL connection offering multiples of the theoretical bandwidth.

A faster T1

Of course, there were options to speed up T-carrier. The most obvious is to combine multiple T1 connections, which was indeed a common practice. Later T1 interfaces were often supplied in multiple for that reason. MLPPP is a variant of the PPP protocol intended for combining the bandwidth of multiple links, referred to in the telephone industry as bonding.

But there were also higher levels in the hierarchy. Remember DS2 and DS3? Well, in practice, T2 wasn't really used. It was far more common to bond multiple T1 connections to reach the equivalent speed of a T2. 44.736Mbps T3 did find use, though. The catch is that T3 required specialized cabling (coaxial pairs!) and had a fairly short range, so it was usually not practical between a telephone exchange and a business.

Fortunately, by the time these larger bandwidths were in demand, fiber optic technology had become well-established. The telephone industry primarily used SONET, a fiber media over which Synchronous Digital Hierarchy (SDH) channels were carried. SONET comes in formats identified by OC (Optical Carrier) numbers in a way very similar to T-carrier numbers. An OC-1 is 51.840Mbps, already faster than T3/DS3. So, in practice, DS3 service was pretty much always provided by a media converter from an OC-1 SONET ring. As bandwidth demands further increased, businesses were much more likely to directly order SONET service rather than T-carrier. SONET was available into the multiple Gbps and enjoyed a long life as a popular internet carrier.

Of course, as the internet proliferated, so too did the stack of network media designed specifically for opportunistic packet-switching computer networks. Chief among them was Ethernet. These protocols have now overtaken traditional telephony protocols in most internet applications, so SONET as well is now mostly out of the picture. On the upside, Ethernet networks are generally more cost-effective on a bandwidth basis and allow you to obtain much faster service than you would be able to afford over the telephone network. On the downside, Ethernet networks have no circuit switching or traffic engineering, and so they do not provide the quality of service that the telephone network did. This means more jitter and less predictability of the available bandwidth, an enduring challenge for real-time media applications.

Miscellaneous

One of the confusing things about T1 to modern readers is its interaction with ISDN. ISDN was a later development that introduced a lot more standardization to digital communications over the telephone network, but it incorporated and reused existing digital technology including T-carrier. In the world of ISDN, a "DS0" or single telephone channel is called a basic rate interface (BRI), while the 24-channel T1 bandwidth is called a primary rate interface (PRI). Many T1 connections in the 1990s were actually ISDN PRIs.

The difference between the two is in the fine details: many of the details related to framing and control signals that were, shall we say, "loosey-goosey" with T-carrier are completely standardized under ISDN. An ISDN PRI always consists of 23 bearer channels ("payload" channels) and one control channel, and the framing is always the same. Since there is a dedicated control channel, there's no need to do weird things with the bits in the bearer channels, and so the overhead is standard across all ISDN PRIs.

In practice, the difference between T1 and ISDN PRI was usually just a matter of configuring your router's interface appropriately. Because of the curious details of the regulation and tariff processes, one was sometimes cheaper than the other, and in general the choice of whether to use a "T1" or a "PRI" was often somewhat arbitrary. It's even possible to use some T1 channels in the traditional way and others as ISDN.

While T1 is now mostly forgotten, some parts of its design live on. DSL and T1 have always had a relationship, DSL having originally been developed as basically a "better T-carrier" for ISDN use. In the modern world, DSL pretty much always refers to either ADSL or VDSL, newer protocols designed for consumer service that can coexist with a voice line and provide very high speeds. Many aspects of how DSL works have their roots in T-carrier, including the common use of protocols like ATM (now rare) or PPP (fading out) to encapsulate IP over DSL.

Okay, I realize this article has been sort of dry and factual, but I thought it'd be interesting to share a bit about T-carrier. I think it's something that people my age vaguely remember as important but have never thought about that much. I, personally, am probably just a bit too old to have had much real interaction with T-carrier. When I worked for an MSP for a bit in high school I saw a few customers that still had HDSL-based T1 service, and when I later interned at General Electric we had an OC-3 SONET connection that was in the process of being turned down. Just really catching the trail end... and yet for years later the Steam Hardware Survey was still asking if I had a T1.

Why did T1 get stuck so long in the very specific context of video games? I assume because video game developers frequently had T-carrier connections to their offices and knew that its guaranteed bandwidth provided very good performance for video games. The poor latency of ADSL meant that, despite a theoretical bandwidth several times larger, it was not really a better choice for the specific application of multiplayer games. So the "T1 is god tier" thing hung around for longer than you would have otherwise expected.