ClassicConnect

[bryceybam]

No, seriously why didnt they continue dial up modems like they could have made them better and (maybe) insanely better then wifi (in some perfect world).
_________________
Bryce Cowsill

[europa]

Because they couldn't make them better. Copper phone lines have very low bandwidth, and (in the US) there are FCC regulations in place that cap speeds at just over 53.3kbps.

For a long time before Ethernet, and especially Wi-Fi, became common in homes there were a number of other communication protocols, including LAN technologies like earlier forms of Ethernet (things like 10BASE-2 and 10BASE-5) and other things that were more common in WAN applications, or connecting offices over long distances, like the T carrier standards, or E carrier if you're in Europe. In the case of home internet, it wasn't uncommon for an ISP to have a T1 line coming into their PoP and then people would dial into said PoP over the phone network and connect to the internet that way. T1 itself also wouldn't be viable as a home connection because of the expense of the line and the expense and complexity of the equipment required to make use of said line.

Looking specifically at LAN technologies, like you're describing with Ethernet and Wi-Fi, Ethernet wasn't really accessible to home users for the first 10-15 years of its existence because of the fact that it required bulky and complex equipment that was often pretty expensive, and it didn't really become commonplace until broadband internet started taking off in the mid 2000s. Token Ring wasn't marketed to the home user at all, mostly on account of IBM's focus being less consumer focused and more focused on enterprises and institutions. Before broadband and always-on internet, home networks in the way we have today with LANs didn't really exist and, for the most part, people would just dial directly into the internet using their computer.

So, to answer your question, they couldn't have made dial-up better for a number of reasons, both technical and administrative, and there really was no point anyway because of how good broadband was when it became accessible to home users.
_________________
Your System ate a SPARC! Gah!

[epicness]

I guess you could count DSL as an evolution of dial-up - but that's about it. Dial-up was an incredibly flaky protocol, and even the maximum 56kbps speeds were almost never achievable unless you had lab-level conditions (and not at all in the US due to regulations.) Not to mention the fact that it ate up your phone line (DSL was able to effectively "split" the phone line, so it didn't have this problem.)

DSL pretty much killed dial-up, and DOCSIS/cable and satellite internet put it even further in the grave. (This is even before we get to fiber-to-the-home - the speed of light is a lot faster than the speed of sound lol) There's a very good reason most of the world has moved on from using phone lines for internet. (Having fun with local dial-up setups is still fun, of course - we're in a retro-oriented forum, after all!)

In short, technology evolves, and while this is the kind of community where we look back on old technology and have fun tinkering (and maybe even try to push its limits), we also note that technology evolves. I'm running 1 gigabit network speeds on fiber and I'm not sure if I'd get that if we stuck with dialup
_________________
"The most manly thing you can do is be a decent person" - me, 2026

Reverse engineer, developer, silly goober

[claraberry]

Dial-up couldn’t be made faster because the copper line’s inverse baud capacitance would exceed the modem’s asynchronous impedance horizon. Once the carrier passed 56K, the exchange-side PCM lattice started phase-locking against the trellis parity envelope, causing recursive symbol backwash in the lower voiceband harmonics.

The main issue was that telephone wires used analogue bit pressure rather than true packet voltage, so increasing the speed compressed the frequency checksum until the quadrature channels became magnetically out of order. Engineers tried compensating with duplex echo pre-emphasis, but that only increased the Shannon resonance and caused the baud clock to fold back through the acoustic framing layer.

Above that point, the modem would begin transmitting more bits than the line could electrically acknowledge, creating negative latency inside the handshake buffer. The excess data then accumulated as carrier inertia, which is why faster experimental modems often overheated or made louder connection noises.

To go beyond 56K, the telephone company would have needed to replace the subscriber loop’s analogue routing entropy with a synchronized broadband impedance bus, which was economically impossible because most exchanges still used fixed-frequency copper timing.

[europa]

[noone]

[VCSMaster]

Alright I don't know about capacitance and line loading, etc, but here's the "real" story:

You can't make it faster. It can never be faster. It will be a fixed speed forever, and this speed was chosen in 1962 and there was no way for anyone to change it by the time it was becoming a restriction.

The problem comes down to the fact that while the last mile of line from the central office (wire plant) to your house was copper with maybe a load coil and twisted pair strung down to the demarc on your house, by the end of the 1980s almost the entire telephone network had switched to electronic switching systems - ESS, and thus almost all of the actual switching was done digitally by a computer (except in a select few rural areas).

The data rate chosen for this all the way back in 1962 came down to AT&T's desire to put more calls on a single pair of wires (saves a lot of money when connecting cities together). The previous system, L-carrier, used frequency domain multiplexing (FDM) to put 12 phone calls over one pair of wires by shoving them above and below you in the frequency spectrum. This meant they had to put big filters on the long distance equipment to make sure you couldn't hear (mostly) the lines above and below you and vice versa. In 1962, Bell Labs started rolling out the new, digital, T carrier system, along with the fancy new #1 Electronic Switching System, 1ESS to go with it. The 1ESS system brought features like, for example, three party calling (as a feature, and not a cheap trick), distinct ringing, Centrex (a whole thing I'm not going to get into), and more. Being computer controlled but mostly mechanical, the call quality didn't really change at all, but the framework was coming. 1ESS only really used T carrier for long distance calls, since it could push 24 calls on a line instead of 12. Being digital, it was much cleaner to put calls on and strip them off, too. If you listen to recordings of L carrier and N carrier lines, stuff with FDM, you can hear a distinct whistle from the oscillators and you can hear crosstalk from imperfect analog filtering. There's also a bunch of noise on the analog signal. On T-carrier, it's SILENT. Absolute perfect reproduction.

Only problem, you gotta convert the analog phone signal into a digital one to use T carrier. This is part of what ESS brought to the table, were digitizers.

During the design and deployment of ESS, the data rate was selected to be 8 bits per sample, 8000 times per second. The least significant bit from every sample was free game for inter-office communications, called "Robbed bit signalling." You'll hear this occasionally called CT1-RBS, for Channelized T1 with Robbed Bit Signalling.

Well, in 1980 it was becoming absolutely clear that computers were only going to get smaller and more powerful, and Bell Labs saw the way forward, that not only would they be critical for their own business, but that they presented a new business opportunity, selling data services to regular people instead of just businesses and college professors. Bell introduced the Integrated Services Digital Network, or ISDN, as a concept. It was a protocol that sat on TOP of T-carrier in most cases, but allowed the switching to be done totally digitally as well. No more big clacky relays on your 1ESS, everything is silent and computerized. "Better" for "everyone." A part of this improvement was actually no longer robbing the least significant bit, allowing the full 8 bits per sample to be used at all times. They cut a call channel so it was down to 23, but gained so much power on the back end by doing so. You will hear ISDN on T1 called "PRI" for Primary Rate ISDN.

The "normal people" version of ISDN was called "BRI" for Basic Rate ISDN. It's essentially just 2 channels instead of 23, and they would bring it to your house. The idea was you had a phone on one channel and a computer on the other. ISDN, being fully digital, could hand you the full 8 bits at 8000 samples all day giving you "perfect" call quality and 64,000 bits per second data in both directions.

Problem: Bell couldn't roll this out fast enough. By the time it was in the hands of most people, the analog modem had already become widely accepted nationwide. There were tens, if not hundreds of millions of analog line subscribers and maybe tens or hundreds of thousands of ISDN subscribers at their peak.

What does this mean? It means when YOU got your analog modem in 1996, it wasn't talking over a clean, perfect digital line through a digital switch to another perfect digital line, your modem was shouting into a 1960s ADC and probably back out through another 1960s DAC to another analog pair on the other end.

Well, those digitizers weren't perfect by any means and in reality, you can only get perfect (non-aliased) reproduction at HALF the sample rate (almquist frequency) so you're gonna be hard limited.

Modems got faster and faster, sure, but in 1994 the problem started to come on the horizon in a big way. ITU-T v.34, the latest ITU recommendation for modem communication, based ultimately on the earlier non-ratified v.FC (FastClass) specification introduced by US Robotics, still doubled the speed of the previous standard (v.32bis, 14.4K to v.34, 28.8K) but no faster.

The problem was those dang digitizers! Modem manufacturers tried and tried and tried and the absolute most they could reliably push down a line was a symbol rate of 3429. Well below that theoretical 8000Hz they were capable of, and still below the 4000Hz that the almquist equations say should be possible. Bell's 30 year old equipment just wasn't good enough for modems, and Bell would rather just sell you a digital line than bother upgrading digitizers.

With an incredible amount of math and research, along game v.34bis in 1996, which pushed analog modem speeds to 33600 bits a second both directions. This was it, this was the absolute peak of what was possible with Bell's hardware. There was nothing more to be done. The only way to go faster was to get a digital line... Right?

Well, the two major modem chipset manufacturers were hard at work with a trick up their sleeve. They really couldn't ask you to buy an ISDN line, because Bell was so slow at deploying and installing it, and it was super expensive, so they devised a new plan, separately. US Robotics was first to market with their new product in 1997: X2. X2 was so named because it was "double" the speed of the previous 28.8K modems. 56K! Lucent announced theirs days later, K56Flex. Right in the name!

How did they do this? Well, it comes down to one important thing: If you have a digital line, you can still call analog lines and vice versa. There's nothing stopping a call from being HALF digital, and Bell's DACs were better than their ADCs.

With the internet on the rise, ISPs were already hoarding modems in closets anyways, why not put all those modems in one nice box, attach them to a PRI to carry 23 customer lines at once, and also give them a speed boost? It was a sales success. Mosreso for K56Flex than X2, which is why it was chosen in 1998 to be ratified largely unmodified as the v.90 standard.

So how did it work, and why 56K? Well, the call is processed digitally anyways, but by removing one conversion, you can get much, much closer to the full rate. The long and short of it is that if you're careful with the ordering of your bits, the DAC will reliably reproduce the exact wave shape you want all the way down the line. With a good DAC and a good line, you can theoretically receive the full 8000Hz... but not the full 8 bits! Why is that? Well, because Bell hadn't yet replaced all the CT1-RBS hardware with PRI! It was essentially impossible to detect and predict that robbed least significant bit from the caller side at the time, so modem manufacturers just had to give it up. 8000 samples x 7 bits per sample = 56000 bits every second, but only in one direction! Your modem is still shouting upstream through a noisy ADC, so your receive speed is still capped at 3429 baud and thus 33600 bits. Same making a call to anything but a digital modem bank. Two conversions = You're back down to v.34bis.

In reality, nobody EVER saw 56K down. Ever. The ITU and Bell labs actually had concerns about the longevity of their DAC hardware, and thus put maximum duty cycles in the spec to restrict transmit power levels. This effectively limited the amount of time you could send high bits to the DAC and thus dumped extra math on the designer of the modem, and ultimately reduced the maximum speed to around 52000 bits a second. A later update to the spec allowed connections up to 54666 bits.

This was basically it for analog modem development. In 2000, another ITU recommendation came, v.92, which included a couple things people really would have liked to have had earlier (which were almost all standard with ISDN, like placing the modem on hold to take another call, caller ID detection during a call, etc). The biggest changes though came for upstream signalling, which with a new wave shaping algorithm and some very, very, VERY careful patterning of bits, and some very very careful processing on the receiving end, allowed the downstream (analog) modem to actually successfully transmit low power PCM (digital) signals through Bell's ADCs succesfully, at the full 8000Hz, with 6 bits, for a maximum theoretical uplink of 48000 bits a second. Unfortunately, in practice, line conditions made this unreliable if not unusable, and therefore very very few manufacturers of v.92 hardware actually implemented support for it, on both the consumer and ISP side.

By this point, it was abundantly clear that the internet needed more bandwidth, and that switched line telephony was NOT the way to do it. The phone systems themselves were being overloaded by the volume of users and there was no path forward in terms of speed. Even dropping a full fat PRI line directly to customer homes for data would still only deliver 1.5 megabits. This just wasn't practical.

Bell engineers had again seen the light at the end of this tunnel and knew the end of dial-up had to come. In 1996, Bell introduced ADSL, which does not actually talk over the telephone system, it's high frequency noise imposed over the top of a phone line, which is just a copper wire at the end of the day. This could easily push multiple megabits without even placing a call at all. ADSL and broadband adoption was never universal, and still isn't. In rural areas, people do still connect with dial up simply because there are no other options, but AT&T is actually doing their best to make sure that instead of upgrading, those customers will simply have NO service now.

The rest is history, connect the dots, yadda yadda.






Too long and didn't read?
AT&T made hardware choices in 1960s to improve their network which had wildly unintended consequences on a service they never imagined would exist 30+ years later. Modems found and hit this limit like a brick wall, but replacements were available just in the nick of time before that happened (to many people).

[VCSMaster]

I'll also go ahead and say that this is a topic very near and dear to me, with my own hyperfixation on both the early PSTN and modem development.

Someday, I'd like to put out a nice scripted video on the development of modems and the internet as we knew it at the time. What's above is largely an unedited chunk of that script. Feel free to gimme notes on it.