 |
| [ Home ] [ Table of Contents ] [ About Lee Goeller ] [ Search ] |
Whatever Happened to ISDN?I started writing this in September, 1999, and sent drafts to a number of friends, including Fred Knight at Business Communications Review and Ian Angus at Telemanagement as Christmas presents. DDT, as I called it, seemed too long and esoteric for current publication; I wrote it just to preserve a mood and provide a few laugh. To my surprise, Fred loved it and ran it in the February, 2000, issue of BCR. This was particularly nice in that the February issue was given away at the huge BCR PBX seminar in Crystal City, just south of Washington DC, which I was able to attend. Made me feel important again. The Death Of Digital TelecomVoice-Data ConvergenceBack in 1978, my research study, "The Digital Future of the Telephone Network" [1], discussed the way digital T-carrier, having taken over the world of short-haul telephone transmission on existing copper cables between nearby central offices, was about to take over long-haul toll transmission by riding on fiber optics. As fiber optics replaced obsolescent microwave, the barrier of analog transmission dictated by microwave economics would be removed and digital data from computers and terminals could be mapped directly into the T-carrier bit stream without being converted to a screech that T-carrier's PCM encoding could treat like voice. This would make it possible to dial up end-to-end data connections at 64 kb/s as easily as voice calls because voice and data would both consist of 8-bit bytes. Further, a single voice channel at 64 kb/s was faster than the bandwidth of 12 voice channels in what had originally been a Telpak "A" on analog transmission systems, terminated in 301B (40.8 kb/s) modems at each end (as in the original ARPANET). Noting that at 64 kb/s, an entire novel could be sent in less than a minute, it appeared to me, back when a 9600 b/s modem cost $10,000, that there might be a considerable market for high-speed switched digital connections for the price of a phone call. It also seemed logical that being digital ought to mean more than cost-reduction in the handling of analog voice and voice-like signals. AT&T had been installing digital 4ESS switches in its long distance network as fast as possible since 1975, and during that same interval, several interconnect manufacturers where making PBXs that, like 4ESS, were compatible with T-carrier. All my dream needed was for digital PBXs to: a.) home on an all-digital tandem/toll network and, b.) interface data as well as PCM-encoded voice right at the users' work-space. Much of the business community would then be living in the world of the future with complete convergence of voice and data while using the same facilities to handle both analog and digital fax, and even slow-scan TV. It should be noted that, as late as 1993 or so, there was very little market for residential data. The "Viewdata Revolution," widely discussed [2,3,4] in the early 1980s as a means for residential customers to dial up data bases and obtain all sorts of information ranging from stock quotes and sports scores to the menu of the day at their favorite restaurant, generated no interest in the marketplace.* Prior to 1993, a few of us had modems for sending articles to our publishers or for addressing "bulletin board systems" containing Viewdata type information maintained by valiant local computer freaques, but except for the hackers who took pleasure in cracking government and business data bases, there were few residential data users.
When AT&T cut over its first big "digital island" (on Feb. 10, 1983 [5]), using fiber optics from Washington, DC, to New York City to interconnect 4ESS toll switches on which local central offices (COs) were already homed via T-carrier, some 60% of the telephones in the country could make toll calls that were digital from the output of the originating CO to the input of the CO terminating the call. MCI followed AT&T up the northeast corridor with fiber about a year later [6], replacing many of its analog Danray PBXs which had been serving as toll switches, and in 1987 Sprint, having also gone digital on fiber optics, began selling off its brand new microwave network in a series of ads in industry magazines [7]. It began to look as though the digital future was at hand. My Secret AgendaMy vision for the convergence of voice and data, at least for business users, seemed within reach. However, I also had a secret agenda: I wanted to get rid of "power ringing." The ringing signal used by telephone companies is 86 volts at 20 Hz, about 50 dB louder than the electrical equivalent of voice as carried by the telephone and nearly as big as the voltage you plug your TV or washing machine into. However, it was also about four octaves below the bottom of the voice band so that it didn't have much effect on telephone receivers. Indeed, ringing is so near d.c. that differentiating ringing current from line current after the customer answers is quite difficult and the technical problem of "tripping ringing" is much harder than one would expect. Both the ringing signal and the bell used in the telephone had been invented by Bell's pal, Watson [8], and it seemed to me it was time, after a hundred years or so, to move on to something better. I hated power ringing because, early in my career, I had been a radio technician who occasionally had to handle broadcasts from locations remote from the studios. The telephone company would run a pair of wires from the studio via the central office to the auditorium, church or whatever and then, sometimes just for fun, one of the telephone techs would connect ringing to the broadcast pair. When I arrived at the remote location and tried to hitch up the amplifier for my microphones, I would get a jolt that knocked me flat. Then I would have to listen to a horse-laugh from the telco guys when I asked them, very politely, to please remove the ringing so I could do my job. Later, I saw one opportunity to kill power ringing come to nothing. During my Bell Labs days, one of my tasks had been to make 1ESS, the first of the computer controlled local CO switches, apply ringing to customer telephones. The field trial in Morris, IL.[9], which preceded 1ESS, had used an all-electronic switching matrix which was unable to accommodate the huge size and low frequency of the traditional ringing signal. Tone ringing using tones in the audio band had been developed, along with a tuned electronic amplifier to replace the bell. This ringing signal could be switched to the customer's line through the electronic matrix as easily as the earlier metallic matrices of Step by Step and Crossbar had switched power ringing. Further, the customers loved it. When 1ESS backed off from the all-electronic Morris system and reverted to a metallic switching matrix using small reed switches, tone ringing was no longer needed; traditional power ringing could be switched to the called party's line just as in earlier systems. This meant that 1ESS could replace Crossbar and SXS systems without also having to simultaneously replace existing phones, a massive problem if the total cutover was to be done with no interruption of service. Then, too, Watson's bell, after a hundred years of refinements, was about as low in cost as possible, at a figure the new electronic technology, as of the 1960s, couldn't even approach. There was one good thing: the 1ESS reed-switch matrix was much faster than crossbar, and the computer control was much smarter; thus we were able to simplify trunk circuits enormously by providing ringing from a separate group of circuits. The control would connect ringing to the called party via the switching matrix, monitor it for answer, and then set up a new connection from the called line to the incoming or intra-office trunk to which the calling party was already connected. Because the time a ringing circuit was in use was very brief compared to the length of a conversation, 1ESS needed much less ringing equipment than, say, No. 5 Crossbar which had all the ringing equipment built into each trunk circuit where it did nothing at all after the call was answered [10]. By providing ringing independent of the trunk circuits, we also had the opportunity to provide ringing circuits for different kinds of ringing such as that needed by 4, 8 and 10 party lines. What I wanted to do next was provide a tone ringing circuit so that new telephones with tone ringers could be installed as needed, possibly as a premium feature. Eventually the delighted customers would abandon power ringing which would then fade away. Needless to say, it never happened. Although Touch-Tone, which doubled the cost of the 500 type telephone set, was eagerly embraced, tone ringing was rejected. And then, to add insult to injury, after Carterfone a number of suppliers began to make tone ringers for their phones, but operated them from the standard power ringing signal from the central office! It sounded like progress, but I knew better. Real progress toward the demise of power ringing came in the PBX field when LSI made it economically possible to replace electromechanical switches with electronics starting in 1975. Of that generation, Nortel (then Northern Telecom) came on the market with its SL-1 which recognized the need in a business office for something other than residential telephones with "flash and feature-code" operation for hundreds of unlikely features. To provide the multi-line and multi-featured capability to which customers had become accustomed with 1A2 key telephone systems, Nortel ran a separate digital control channel to its business sets. Although these sets were rung with Morris-style tone ringing switched through the digital matrix, the signaling channel let the switch-hook and push-buttons send information to the PBX, and the PBX operate the intercom buzzer and light and blink lamps on the set. The separate signaling channel was the key to the future [11]. Other PBX manufacturers eventually caught on, and when further advances in LSI continued to make increased chip complexity available at lower cost, the codec for digitizing the voice signal was moved from the line card in the PBX to the multi-button business telephone sets. Then the signaling channel could be multiplexed into the bit stream needed by voice and most PBXs discovered a better way to do tone ringing: use the separate data channel to tell the set to activate appropriately the ringer as well as the buzzer, lamps and other displays, and then send back a digital signal when the customer answered. All of the traditional ring-trip problems of differentiating 20 Hz from d.c. were now gone, but more important, extending the T-carrier digital format all the way to the telephone set made end-to-end circuit-switched high-speed data connections possible [12]. Unfortunately, telephone companies and interconnect companies alike saw this PBX capability as an example of a value-added feature for which the customer should be eager to pay a premium price. They might have gotten away with it had not local area networks (LANs), a result of the same advances in LSI, come on the market at about half the price and twice the convenience of data via the PBX. True, LANs were primarily local, but then, so were the data capabilities of PBXs when the only way they could connect to digital long distance networks was via analog trunks to local analog central offices. Bypassing the local telco was the most direct way for digital PBXs to access digital toll networks, but because that would be a threat the to revenues of local telephone companies, progress was blocked by regulations. During the 1980s and early 1990s, the telephone people pursued ISDN, the Integrated Services Digital Network, at a languid pace [13]. Mostly, they copied and tried to standardize the digital telephone sets available with then-current PBXs, but designed to be used with digital COs. Nortel's DMS-100, a digital local CO switch, came on the market in 1979 [14]. It was bought in quantity by the Bell Operating Companies in the process of being divested from AT&T who feared further anti-trust action if they continued to purchase AT&T telephone equipment from Western Electric. Between legal apprehension and Nortel's local-market head start, AT&T's digital 5ESS, first available in 1982 [15], was at a marketing disadvantage until AT&T split off Western Electric and Bell Labs under the name of Lucent Technologies in 1996. Unfortunately, analog multi-button electronic telephones for Centrex were less expensive than digital sets, and both AT&T/Lucent and Nortel offered them in competition with their own ISDN sets which, as it happens, were not compatible with each other. Needless to say, modemless data designed to run directly on T-carrier never really had a chance. Although the telephone companies showed little interest in business telecom which had a considerable need for data transmission, the failure of Viewdata did not prevent them from continuing to give lip service to ISDN for residential customers even as their technicians were reported to be highly incensed by the very idea. The potential demand of the numerically huge residential market kept them and their new competitors trying. After about a decade, America On Line conducted a disk blitz, "carpet-bombing" everybody who might have a computer with free copies of AOL access software. This made the Viewdata approach stagger upward just in time for the Word Wide Web to tie together AOL and enough other Information Service Providers (ISPs) to become the Internet. With AOL customers able to exchange e-mail with customers on other ISPs, and all ISPs able to access a large number of "Web Sites" provided by government, business, and interested hobbyists, the Internetworked World achieved considerable acceptance in the 1995 time frame. From the mid '80s on, however, people who wanted to take advantage of fast document transmission had turned to fax, finally standardized by international agreement and made affordable by further improvements in LSI [16]. Even today, fax, being circuit switched, is often faster than computerized e-mail, can easily be used by people with no computer skills, and can transmit business graphics as well as legally recognized signatures. And because fax rides on a dial-up phone connection directly to the fax machine of the called party which answers automatically, it doesn't stop off in queue at a dozen store-and-forward nodes like a packet switched document or sit for hours until called for in a mail box on an ISP server where it can be viewed by any hacker who wants to take a look. Fax ran Western Union out of business, and still has more of the point-to-point message traffic than Internet hustlers would like to admit, particularly when their data networks can do a lot more than send pictures of text. "Selling" The InternetAOL, Prodigy and other ISPs with hundreds of FREE files you could download made the Internet take off shortly after 1995; a supply of hype exceeding even the vast number of FREE files got almost everybody who didn't have a computer to buy one, and all to sign up with AOL, Prodigy, CompuServe, or one of many local ISPs to get in on the FREE FILE gravy train. "FREE files" was obviously a better marketing ploy that Viewdata's efforts to sell phone calls. In those mid-90s days, AOL, Prodigy and others charged a minimum fee (about $10) for 5 hours of access to their own servers and the Internet, plus an hourly fee (about $3) for additional use. Then, at the very end of 1996, AOL decided to go "flat rate," "all you can eat" for 19.95 a month [17], possibly to compete with new local ISPs who were already using such pricing "innovations" to seek a competitive edge. Flat rate changed the entire psychology of Internet use; now, instead of minimizing connect time to save money, one had to use as much connect time as possible to get his or her money's worth; because so many were logged on for hours at a time, the number of access lines at ISP Points of Presence (POPs) had to be greatly increased, particularly by AOL. Software was developed to time out and dump users who were logged on but not sending or receiving anything, and entrepreneurs developed software for users to fool the ISPs. Once you got a connection, you had to hold onto it if you wanted it to be there after you finished eating supper, taking a bath, or walking the dog. Once you lost the connection, you were done for the day. Naturally, any ISP that was not flat rate had to join the crowd, and Internet access, if not actual usage, shot straight up. The only problem was that, for residential users to reach the private packet networks that accessed their ISPs which, in turn, could connect them to the Internet, they had to dial up an ISP POP via their local telephone company. Now, the thing that packet networks handle most efficiently is NOTHING. That is, if you are not sending or receiving data, you are not using either the switches or the connecting transmission links of a packet net-work. That leaves the network free to handle packets for others, and is totally unlike a circuit switched telephone network where, once a connection is set up, a certain amount of bandwidth and switching capacity is occupied during the entire conversation, even when nobody is saying anything. Note, however, that the circuit-switched connection is set up only once, and then is taken down at the end of the call. In packet switching, each and every packet, in each direction, requires the equivalent of the complete end-to-end call set-up process. Thus while packet switching only uses the network while something is actually being sent, it demands far more effort on the part of switch control equipment when a message or connection consists of more than a few packets. As the cost of processors and memory, and thus, processing, drops, the cost of call-set-up inefficiency tends to become less important, at least for traditional data transmissions. Parenthetically, it might be of interest here to note that virtual circuits for packet calls, as in frame relay and ATM, are set up only once per connection, as in conventional circuit switching. This reduces the processing load for multi-packet connections while, at least for certain kinds of transmission, it conserves bandwidth almost as well as packet switching. Among other things, the header for virtual circuit packets (or "cells") is much shorter, serving only to identify the connection rather than containing the entire calling and called addresses and various other things required for packet processing. Because most people "surfing the net" send very little information TO their ISPs (a few typed characters and mouse clicks), and only occasionally receive bursts of data (often quite large, but usually of fairly brief duration) FROM their ISPs, packet networks are designed to handle the peak number of packets expected in any one time period rather than the much larger peak number of customers who happen to be logged on. But for each customer logged on, a circuit-switched connection from customer to POP must be set up through the local telephone company, and at the POP, the ISP must provide a modem on a per-circuit basis. After AOL went flat rate, telephone companies across the country were suddenly tied in knots by "all trunks busy" conditions at AOL POPs. AOL had to rush out and rent more telephone lines and buy a modem for each. However, what was needed was not more customer-to-POP circuit switched connections, but a way to extend packet switching all the way to the customer. There was an obvious way to do this: ISDN. Last Chance For Digital PhonesThe ISDN signaling or D channel is capable of handling 16 kb/s, enormous overkill for the control functions required by an ISDN telephone set. Thus D channel protocol (standard X.25) was arranged to allow packet data, up to 9.6 kb/s, to be added to the control data. In the line shelf in the digital CO, both control and data packets use a bus similar to a local area network to allow a number of line cards to send packets to each other or to the switch control equipment. There is no problem in adding a port to this LAN to access another computer, router, or whatever to handle data transmission to remote locations. Indeed, 5ESS, when used as an ACD for directory assistance, uses D-channel packets for operators to interact with the directory data base [18]. When dealing with text only, 9.6 is more than sufficient. For megabyte-file downloads, of course, 9.6 takes forever. But an ISDN line has two B channels in each direction, each running at 64 kb/s. They can be "bonded" together to provide 128 kb/s, and switched like a regular telephone call to suitable ports on the router (which also interfaces the D-channel busses from various line groups) whenever a file download (or, for that matter, upload) is required. Just doing the division, 1 Mbyte, or 8 million bits, can be sent in a little over a minute. This is nowhere near fast enough to send a movie for real-time viewing, but for text files, still pictures in catalogs, etc., it shouldn't be bad, providing ISDN lines are suitably priced. The price trade-off should be between two analog lines, one for the regular home phone and the other for the Internet freaque, vs. one ISDN line where the D channel is used for general ISP access, e-mail, etc. Even if a phone call is using one B-channel, the other is available for data downloads at 64 kb/s, while if the voice phone is not being used, both B channels are available for downloading at 128 (even then, call waiting via the D-channel can identify incoming calls without blasting the data). The main advantage of this kind of connection is that both the telephone company and the ISP do not have to provide ANY per-call hardware for Internet access; no modem at the POP and no long-holding-time switched connection through the local CO and perhaps a tandem switch. If the cost to the customer were competitive with two analog lines, that would be the way to go, assuming the customer had battery back-up on the ISDN terminal adapter. I described this approach on John Dvorak's PBS radio show back in May, 1996, and a year later in Reference [17]. Finally, in 1998, "Always On/Dynamic ISDN" or AO/DI, was announced as soon to be available from Bell South and Southwest Bell in PC Magazine [19]. Since then, I haven't heard a word, and I imagine I never will. Data OverrideNaturally, other ways to handle high-speed downloads were quick to appear. DirecPC, for instance, downloads via satellite at 400 kb/s directly into the customer's PC, analogous to the two B channels of AO/DI, but requires a regular dial-up connection to a POP to perform the function of AO/DI's D channel. Then, too, there are cable modems which use part of the spectrum of local CATV systems as a high-speed LAN to collect packets from and deliver them to residences along the cable run. Even dial-up analog connections have been speeded up with 56 kb/s modems which, after a lengthy battle, have finally standardized their approach (at least I suppose V.90 has prevailed). A 56K modem is asymmetrical: it goes from the user toward the CO at 33 kb/s, but comes back at (almost) 56K. The better download speed comes from something more sophisticated than a modem at the POP. There, analogous equipment using DSP (digital signal processing) maps the bytes coming from the ISP into an optimal T-Carrier bit pattern (that is, without quantizing noise) on a channel in the POP's T-1 access line. This pattern, arriving at the customer's CO, is converted into a modem screech which then proceeds via the local loop to the customer's modem. If the CO switch is analog, as with a 1ESS, the digital-to-screech conversion is made in the channel bank of the T-carrier transmission system; for a digital switch such as a 5ESS or DMS-100, the conversion is made in the switch's line card. This approach provides the customer with a digital picture of an analog impression of a digital signal, but is hardly the use of T-Carrier I would have hoped for. Among other things, if someone dials up a line with a 56K modem on it, POWER RINGING is necessary to tell the modem to answer the call. But more important, Digital Subscriber Lines [20], asymmetrical (ADSL) and otherwise, are suddenly becoming available at a reasonable price, often $70 a month or less. They have the Always On property of AO/DI, but work at much higher speeds than ISDN, downloading at 1.5 to 6.1 Mb/s, and uploading at 500 kb/s or more. Further they don't tie up any switched paths through the CO, ever. They just use the bandwidth on the customer loop above the voice band (between 25 kHz to 1.1 MHz), like the air-space rights in New York City above a 3 story building. At the customer end, there is a "splitter" which delivers the high-frequency band to the user's data interface and the low frequency band to the regular telephone(s). The splitter passes the telephone signal from the CO straight through: d.c. for on-hook/off-hook, 86 volts at 20 Hz for ringing, and the speech band from 300 to 3500 Hz. Thus phone service is isolated from and independent of the high speed data. At the CO, a similar splitter separates the two signals, sending the telephone path to a regular analog port on the voice switch, and the data path to a suitable interface which matches the customer's high-frequency bandwidth to a statistical multiplexer which gathers up a number of such physical circuits and connects them via virtual circuits to various ISPs via a broadband data network. In between customer and CO, one of the better ADSL approaches uses multiple carriers, each with modem-like modulation to carry several bits per baud. If the customer loop is too long so that the high frequencies do not get through, or if an AM radio station or other problem interferes with certain frequencies, the terminal equipment can adjust, making use at a slower speed of the carriers that do get through. Copper pairs to the customer are used not because they are good, but because they are already there, in place, and can continue to provide telephone service essentially unchanged while adding always-on high speed data. This combination seems to me to have the highest probability of success of the various techniques currently available. It will allow almost anybody to use a $2000 computer to emulate a $200 TV set, and to read from the computer screen multi-colored high quality ads comparable to those found in the best slick magazines. How can it lose? But if ADSL (or some other approach) takes over, there will be nothing left on the telephone network but voice-like signals. The digital properties of both switches and transmission facilities will be left largely unused, doing little but reducing the cost of handling analog signals. Further, there will be no motivation to replace analog telephone sets, while fax machines, answering machines, and modems (if customers want direct data connections without going through an ISP) will also remain, all requiring my old nemesis: POWER RINGING. A Sad FinaleOf course, there may not be any voice traffic left for telephone long-distance networks. Voice over Internet Protocol, or VoIP, may have stolen it away at rates heavily subsidized by the Government, venture capitalists, and other parties interested in promoting competition, no matter what the cost. VoIP claims to run at 8000 bits per second when a person is actually talking and, like TASI (Time Assignment Speech Interpolation) on the early Atlantic Cables, sends nothing when a person is not talking. This is a considerable saving over the continuous 64,000 b/s of T-carrier's PCM, or even the 32,000 b/s of ADPCM, but its efficiency has nothing to do with packet switching. VoIP compression comes from the AT&T pavilions at the 1939 World's Fairs: it uses voder technology [21]. The vocoder at the "talk" end examines the analog signal (usually in PCM form) over 10 milli-second intervals, decides how the human speech tract had to be configured to make the sound, and sends that specification as a digital signal. At the "listen" end, the received spec is used to configure an electronic speech tract to regenerate the original sound (the voder operation). AT&T had originally hoped to use this approach to put voice over Trans-Atlantic telegraph cables prior to the first voice cable in 1956. It has been resurrected for bandwidth-starved cell phones as ITU-T International Standard G.729, taking advantage of the huge processing power available in today's inexpensive microchips [22]. It works nicely to tuck voice into spaces between the graphics presentations on which the Internet squanders most of the excessive bandwidth it consumes. To obtain 8000 b/s, each 10 ms speech interval is represented by 80 bits (10 bytes) when a person is actually talking. This is apparently enough to not only preserve the articulation of speech, but also the details that allow the listener to identify the speaker. But the voder approach has some problems. First, the original signal to be encoded has to be accumulated before it can be analyzed. This, of course, takes 10 ms, greater than the round-trip transmission delay at the speed of light between New York and St. Louis, to which must be added the processing time itself. As a result, echo cancelers are required, even for local calls, to solve the traditional echo problem. Second, the 10 byte payload from each interval is appreciably smaller than the header of an IP packet (about 40 bytes), suggesting that some of the claims of efficiency for VoIP packets may be exaggerated. If one speech interval is sent per packet, each packet runs at only 20% efficiency. By sending two or three intervals per packet, doubling or tripling the minimum delay, efficiency goes up to 33% and 43%, respectively, but this delay, plus processing and store-and-forward delay at each node begins to exacerbate the second delay-induced problem, "double-talking." When delay gets to be greater than about half a second, you may start to talk again before my reply reaches you. Third, voder operation is designed to handle voice only; to handle Touch-Tone, Fax, "comfort noise" in the background, and other sounds which a human speech-tract does NOT generate, additional complexity has to be added. Finally, even though 8000 bits per second is far less than PCM T-carrier uses for voice, it is vastly greater than the few keystrokes and mouse clicks that IP expects from a computer user, both per second and per connection; thus VoIP may put more of a traffic load on a data network than its designers expect. Even so, we can expect the intrepid Internet designers to rush in where angels fear to tread and (eventually) deal with these problems, making it possible for the nimble, energetic entrepreneurial spirits of the Internet to eat the lunch of the stodgy, slow telephone companies. Their intention appears to be to make VoIP gateways for long distance calls so that even non-computer people can select an ISP as their long distance carrier. Since ISPs don't have to pay access fees like regular long distance carriers, this free-load is apparently their major competitive advantage and should let them do very well. But from my point of view, they will once again be a failure. They will make just one more excuse for retaining analog telephones and their POWER RINGING in local telephone networks forever. References
[ Top ] [ Table of Contents ] |
|
Copyright 2006 Lee Goeller. All Rights Reserved. |