247 15 ENABLING “SOFT” TECHNOLOGIES 15.1 ALPHABET SOUP Thirty seconds into a review of communication product specifications one encounters a babble of vee-dot (V.) versus MNP variations, and mysterious acronyms such as EC 2 , ETC, and TX-CEL. It is unwise to dismiss this sometimes unintelligible gibberish as an unimportant attempt to break/protect vendor proprietary interests. These alphabetic designators are often surrogates for key communication developments required for the success of wireless data. This chapter is a simplified look at some nonhardware technologies, protocols, and communication standards that drive both network and device development. The focus is on modulation techniques, error detection/correction, and data compression. 15.2 MODULATION 15.2.1 Circuit Switched The modems initially used to transmit data via analog cellular channels were conventional, dial-up wireline devices with CCITT-standardized V series modulation techniques. The most common are listed in Table 15-1. The user either could plug these modems into a cellular phone (with adapter) or use landlines via the usual wall phone jacks. A useful characteristic for cellular was, under adverse conditions, the V.26/27 implementations dropped to half speed. They actually worked well but connected slowly and transmitted data oh-so-slowly. The Wireless Data Handbook, Fourth Edition. James F. DeRose Copyright © 1999 John Wiley & Sons, Inc. ISBNs: 0-471-31651-2 (Hardback); 0-471-22458-8 (Electronic) Between 1976 and 1982 IBMs Gottfried Ungerboeck developed the theory, and received the patent, 1 on what is now known as trellis-coded modulation (TCM). Briefly, TCM introduces redundancy in the information bits to be transmitted. These extra bits are used for detecting, and usually correcting, transmission errors. Of naming interest is the fact that the redundancy is added in such a way that sequential data symbols cannot succeed each other. The possible states can be represented by a transition diagram that (imaginatively) resembles a trellis. The state sequences can be explored recursively using the Viterbi algorithm to predict the most likely sequence. An error that disrupts this sequence can often be corrected by substitution. The TCM goal was to permit 9600-bps transmission over dial lines. Initially ignored, even within IBM, continued development, as well as the evolution of affordable signal processors, proved that TCM was eminently practical. In 1986 the technique was standardized by CCITT as V.32 over the resistance of the U.S. delegates, who stated, the V.32 recommendation is not workable . . . and is needlessly sophisticated for most PC communications. 2 Three years before the formal V.32 standardization, IBM Research saw that the TCM techniques could be pushed to higher user speeds. 3 Thus was born the development efforts that ultimately became V.32 bis . In this implementation, 7-bit symbols are transmitted at 2400 baud to produce a raw transmission speed of 16,800 bps. The user bit rate is specified as 14,400 bps (14% redundancy). This TCM technique was embraced quickly by manufacturers. IBM was beaten to market by the 14,400-bps Codex 2360/2660 in 1984, two years before CCITT endorsed the 9600-bps standard! V.32 bis was standardized by CCITT in early 1991. With relentless reduction in both size and cost, V.32/V.32 bis modems became the favored modulation technique for circuit switched wireline modems by 1992. Since the standard permits fallback to half speed under adverse line conditions (e.g., 9600 bps drops to 4800 bps), this slower speed operation proved useful on cellular. In 1993 multidimensional trellis coding became the core of the V.Fast proposals. In June 1994 they were standardized as V.34 by the International Telecommunications Union (ITU), a United Nations agency that had replaced CCITT. This backward-compatible, variable-speed approach permitted raw data transmission capability of 28.8 kbps and, optionally, 33.6 kbps. Table 15-1 Early wireline modulation techniques CCITT No. User bps HDX/FDX Modulation Technique V.21 300 Both Frequency shift V.22 1,200 FDX Phase shift key (PSK) V.22 bis 2,400 FDX Quadrature amplitude V.26 bis 2,400 HDX Differential quadrature V.27 ter 4,800 HDX Differential 8-PSK 248 ENABLING “SOFT” TECHNOLOGIES Work was already underway on faster 56-kbps modems. Two competing, incompatible technologies were on the table, with all the usual claims for technical superiority (and attendant saber rattling over patent positions): 1. K56 (or K56flex)supported by 3Com, Lucent, Motorola, Rockwell, and others 2. x2supported by Texas Instruments and USRobotics In March 1997, ITU-T began work on harmonizing the two competing proposals. The effort was probably aided by the acquisition of USRobotics by 3Com. On February 6, 1998, agreement was reached on the V.90 56K modem standard, which was expected to be formally ratified in September 1998. These sharp bit rate improvements in wireline modems did not directly transfer to the more hostile wireless environment. Until recently, the best cellular modems were all based on V.32 bis , offering the potential for speeds up to 14,400 bps (Rockwell permits its cellular modems to operate in V.34 mode, but with poor results). Typically wireless cellular modem speeds lag their wireline counterparts by a very great deal, as shown in Figure 15-1. Figure 15-1 Circuit switched wireline vs. wiereless modem rates. 15.2 MODULATION 249 15.2.2 Packet Switched 15.2.2.1 Narrow-Band Low-Speed Devices Most packet switched data radio devices have a history of digital modulation techniques with arcane names such as modified duo-binary (MDIs MMP), noncoherent baseband FSK (Motorolas MDC4800), or binary digital baseband filtered FM modified GMSK (Ericssons Mobitex). The frequency shift keying (FSK) variant was in widespread use in wireless devices at the start of the 1980s. It was easy to generate and detect and permitted much lower cost devices. The problem with FSK is that it is extravagant in its use of bandwidth. It was quite clear that Motorola would be unable to get much beyond 4800 bps using this modulation technique. At the close of the 1980s Motorola, keying off wireline progress, developed an enriched TCM technique to improve both the speed and correction power of RD-LAP. Here 8 bits per baud, not 7, are sent at a raw transmission rate of 19,200 bps. Two of the 8 bits are redundant, producing a user yield of 14,400 bps, comparable to V.32 bis . One of the psychological differences is that Motorola began to claim the raw bit rate, not the effective user rate, for RD-LAP. All competitors would be forced to follow this imprecise convention or thought to be bit rate laggards. CDPD chose not to use TCM, using Gaussian minimum shift keying (GMSK) instead. The initial IBM CelluPlan design goal was 16,800 bps. With Motorolas announcement of RD-LAP in late 1991, that rate would no longer suffice. A variety of modulation techniques were considered that, in future implementations, might yield far higher bit rates in the 30-kHz-wide cellular channel. With the participation of PCSI as the modem/base station designer, the alternatives were narrowed to three with the grossly simplified selection criteria of Table 15-2. GMSK has the weakest spectral efficiency, and there were early concerns about transmission power control. But GMSK variations were employed in both Mobitex and the European GSM system; it had good product cost numbers and could be driven to match RD-LAPs speed. Reed-Solomon coding permitted Motorola equivalency in correction efficiency: 19,200 bps, which yields 14,400 user bps. Table 15-2 CDPD modulation alternatives GMSK 4-FSK DQPSK Radio complexity Good: standard FM Costlier Baseband sophistication Good Costlier Good Spectral efficiency Worst Very good Best Eb/no performance Static Good Worst Best Rayleigh Good Worst Good Process power/memory Good Harder Harder 250 ENABLING “SOFT” TECHNOLOGIES 15.2.2.2 Narrow-Band Medium-Speed Devices In September 1991 4 Motorola announced a new digital wireless technology, then called MIRS. This combination of techniques, employed by Nextel among others, employs M16QAM 5 modulation and achieves a raw bit rate of 64 kbps in a 25-kHz-wide channel. Nextel states that no other modulation technology transmits as much information in a narrow band channel. 6 Quadrature amplitude modulation (QAM) is a solution to the phase jitter problem seen in PSK. PSK, in itself, was designed to squeeze more bits into a given bandwidth. The first commercial QAM modulation appeared in V.29 leased-line modems and was limited to 9600 bps. QAM, as employed in iDEN for Nextel, uses this high bit rate and time division to carry digitized voice. If Nextel chooses to offer a packet data solution, it should be possible to send information across multiple time slots, trading off voice for data. This could provide attractive narrow-band channel rates. As of the fourth quarter of 1998, this narrow-band bit rate has not been exploited for any data solutions. 15.2.2.3 Wide-Band High-Speed Devices An alternative to narrow-band schemes is the use of spread spectrum, a term often used interchangeably with CDMA. Spread spectrum was originally developed to foil jamming of military communications. In this approach the data stream is imprinted onto a spreading signal that has a bandwidth much greater than the data rate. It seems counterintuitive that in a limited spectrum world such a design requirement could lead to a viable, spectrally efficient solution. But it is possible for a spread spectrum receiver to operate when its input signal is literally buried in noise. As long as the received power can be equalized, hundreds of interfering signals can operate in the same spectrum. If all transmitters in a mobile system used fixed transmit power, as they do in ARDIS, BSWD, and CDPD, the power equalization requirement could not be met. Devices close to a base station would swamp the signals from more distant units, a problem known as the nearfar effect. This transmit power control problem has gradually become tractable, as is evident in the (now) successful Qualcomm CDMA techniques employed by many carriers for digital voice. Qualcomm is not the only user of CDMA. Metricoms Ricochet system was a user from the outset. The positioning of many low-power nodes helps to provide a solution to the nearfar problem. In the initial design, device modems separated by distances less than ~600 meters can transfer data to each other without passing through a receiver/relay node. In this special case, rates up to 40,000 bps are achievable. Typical speeds through the nodes ranged from 9600 to 28,800 bps. In March 1997 Metricom announced a new technology to sharply increase data speeds. 7 The new approach, code named Autobahn, utilizes two unlicensed spectrum bands: 900 MHz and 2.4 GHz. The top speed was to be 85 kbps, available in 1998. In addition, broadening the bands still further (Starlite) offered a potential speed of 256 kbpswhich would be really moving! In September 1998 Metricom announced the demonstration of what is now called Ricochet II, at speeds which will be comparable to a wired 56K modem. 8 Clearly, there has been some trouble along the way. It is also a bit late: The new service will begin deployment in mid-1999. 15.2 MODULATION 251 15.3 ERROR DETECTION: V.42/V.42 FAST AND MNP4/10 In 1988, after bruising political battles, CCITT recommended V.42 as its error control standard. The competitive approaches were a protocol called LAP-M and Microcoms MNP4. LAP-M is a variant of the HDLC protocol devised for ISDN: LAP-D. It is a full-duplex protocol and was backed by U.S. modem manufacturer Hayes as well as AT&T and British Telecom. 9 MNP4 is a half-duplex technique developed by U.S. modem manufacturer Microcom and backed by the Belgian, French, and Italian postal, telephone, and telegraph (PTTs) providers. 10 Both approaches are speed independent. One does not have to have V.32/V.32 bis /V.34 for V.42. In fact, virtually all initial implementations were on the older, slower V.22 modulation standard. Both competing error detection techniques use CRC with ARQ retransmission. Technical differences are minor; both protocols provide virtually identical performance. The camel compromise: V.42 requires modems to use LAP-M and support a secondary mode of operation that uses MNP4. The theory was that enhancements would flow to LAP-M, gradually retiring MNP4. Microcom did not permit that to happen. In 1991 Microcom used V.42 as a base to extend error detection with a series of MNP levels, some of which were useful for cellular data. The core of the approach is to depend upon the 16-bit CRC for the detection of an error in a packet that does not exceed 256 octets. With MNP4 the size of this packet varies with transmission quality: the better the line (reflecting the original design), the larger the packet. Continuous ARQ can be used to retransmit faulty packets. Collectively, the MNP Cellular features are called MNP10. This is a proprietary, two-sided protocol. That is, both the remote user and the host (or cellular modem pool) must have MNP10 to gain any benefit. When used with cellular, which is detectable by the modem, the key differences with the MNP4 versions are: 1. MNP4 operating on wireline uses large packets at the beginning of a session; if there are too many errors, the packet size is reduced until the error rate reaches an acceptable level. MNP10 does the opposite: It begins with small packets and continues to increase packet size in the absence of errors. 2. The initial MNP10 connection is made at 1200 bps. 11 Bit rates are reduced in the presence of noise; when the noise eases, the bit rates are advanced (a function not specified in V.42). These speed shifts can be dramatic: 300 bps under adverse conditions to 4000 bps in a clean environment. 12 3. The modem begins transmission at low modulation amplitudes. If channel background noise is detected, the receiver can ask the sender to increase signal power. In the presence of distortion the receiver can ask the sender to decrease signal power. This philosophy tends to make the protocol perform less well if the target is moving since it often detects problems and slows the bit rate just as the target leaves the trouble area. In 1991 Microcom tests on low-bit-rate modems, executed along Bostons Route 252 ENABLING “SOFT” TECHNOLOGIES 128 in high-signal-strength areas, revealed that bit rates for comparable messages on stationary targets was 1.41.9 times higher than for moving targets. Further, MNP10 sees cellular events as damaging noise. The modems lose synchronization and must retrain after each event. This can have punishing impact on throughput since retrain cycles can vary between 6 and 25 seconds. Meanwhile, the original V.42 LAP-M proponent was not idle. AT&T Paradyne continued its work on modem improvements for cellular. In 1993 it introduced its own proprietary, two-sided modem called ETC: Enhanced Throughput Cellular. Confusing everyone, AT&T Paradyne advertisements sometimes make claims 13 for ETC error-correcting algorithms. These claims are advertising puffery. ETC is an error detection protocol that employs 14 V.42cell: essentially V.42, Appendix III modifications to improve reliability with potentially high bit rates under adverse transmission conditions. However, ETC is modulation dependent and uses only V.32, V.32 bis , and V.34. In 1997 Paradyne introduced extensions called ETC2 Quick Connect. 15 Paradyne claims that ETC2 is more reliable, causing V.42 error detection to be invoked less often: the lowest rate of ETC2 . . . (4800 bps) operates 4 dB better than the lowest rate of other cellular protocols. 16 15.4 DATA COMPRESSION: V.42 BIS AND MNP5/7 In a reprise of the V.42 battles, CCITT was faced with two competing proposals for modem data compression in 1988. The first was the extant MNP5 developed by Microcom. This Microcom approach used a real-time adaptive algorithm to achieve a nominal doubling of throughput. That is, under the right conditions a 2400-bps modem would be able to make an effective data transfer rate of 4800 bps. Fully aware of its modest compression achievement, Microcom delivered MNP7 in the second half of 1988. This technique used Huffman coding (few bits for frequent English language letters such as E, T, A; more bits for infrequently appearing letters such as Q and Z) with a predictive algorithm. The result: ~3-to-1 improvement under good conditions. ACTs CommPressor alternative based on Lempel Ziv techniques, which could yield 4-to-1 improvements, was selected as the V.42 bis standard in September 1989. Microcom continues to license and build its MNP alternatives but admits that V.42 bis will increasingly become the preferred method of data compression. 17 The actual compression achieved in practice varies widely with the application. The algorithm works by recognizing repeated patterns in data and substituting shorter symbols for them. The more repetition a file has, the greater the compression. But purely random data contains no patterns at all and cannot be compressed. Precompressed information, such as ZIP files, do not benefit from V.42 bis either. Further, data files that have been encrypted through a randomization process will also show little reduction in size because the data has had identifiable patterns removed. This suggests that CDPD, for example, might not benefit from data compression. A data compression summary 18 is shown in Figure 15-2. These are average ratios. A 15.4 DATA COMPRESSION: V.42 BIS and MNP5/7 253 compression ratio of 2.0 indicates that the data can be compressed by a factor of 2 and transmitted in half the time needed to transmit it uncompressed. 15.5 ON-GOING ENHANCEMENTS TO ALPHABET SOUP MODEMS ETC addressed some MNP10 weaknesses, including the ability to operate (with degradation) if only the mobile user has an ETC modem. It still treats cellular events as noise. It was quickly tested at Bell Atlantic Mobile, Ameritech, and Southwestern Bell 19 and is now available through most modem pools. As might be expected, AT&T Paradyne claimed markedly superior performance to MNP10: ETC blows MNP10 out of the water 20 and V.42cell outperforms . . . MNP10 by a significant margin. 21 Microcom disputed all the test results. However, disquieting murmurs about the effectiveness of MNP10 had long been extant. In 1991 low-bit-rate field tests were performed in Los Angeles with Microcoms Microporte 1042 modem. The customers summary was 22 : The performance of MNP10 . . . may be only marginally better than the performance of other protocols . . . (and) appears to be the same as other protocols in practical applications. In 1994 the results of Network Worlds tests of high-bit-rate modems were bluntly summarized 23 : The results of this trial cast little doubt that ETC modems outperform MNP10 modems. The ETC supporting modems generally connected faster and at higher initial rates than MNP10 modems. When it comes to transmitting over the cellular network, the ETC modems are far superior to the MNP10 devices aggregate throughput. Celeritas Technologies developed, and patented, spectral shaping firmware called TX-CEL (throughput accelerator). The technology is essentially protocol independent and will work with either MNP10 or ETC variants. The licensing has been quite successful, including the gateway products of Primary Access. 24 To improve MNP10 performance, in November 1994 Rockwell announced a chipset called MNP 10EC. The new functions include signal conditioning technology and specific algorithms to deal with cellular impairments such as hand-offs, dropouts, interference, fading, echoes, and audio distortion. By April 1995 more than Figure 15-2 Data compression summary. 254 ENABLING “SOFT” TECHNOLOGIES 100 modem companies had pledged to deploy cellular modems with the MNP10EC technology. 25 The competition between these two approaches has forced continuing improvements in data performance over circuit switched cellularclearly with more emphasis on cellular problems. Motorola, with its CELLect line of modems, used V.32/v.32 bis modulation, V.42 LAP-M error control, and V.42 bis data compression. (Motorola also had problems with ad writers who extolled the virtues of V.42 bis error correction. 26 ) From this base Motorola developed Enhanced Cellular Control (EC 2 ) to optimize cellular data communications by managing the hand-off between one cell and another, noise distortion, and other cellular functions. 27 But Motorola could not compete: With slumping sales, questionable profitability and increasing price competition, Motorolas (October 1997) decision to sell off its consumer modem business 28 was not unexpected. Another casualty was AirTrue from Air Communications. Introduced in late 1994, AirTrue was a cellular-side-only protocol optimized to work with any V.42 modem at the host end. AirTrue was confined to V.32 bis . At this speed AirTrue attacked transceiver noise/distortion and attempted to interpret cellular events as something other than just noise. AirTrue performed extensive testing on both connectivity and interoperability. Connectivity was defined 29 as the ratio of successful first try connects to total numbers called (discounting modem busy and ring-no-answers). Summary results are shown in Table 15-3. But in spite of the boldest assertions that AirTrue offers a level of functionality that its competitors cannot match, 30 the Air Communicator was not a success. Today Air Communications concentrates on GSM products. The new battle is being fought by Paradyne ETC2 31 : No other cellular protocol supports the high rates . . . Rockwell . . . MNP10EC . . . reliability is poor. No other cellular protocol has a quick connect mode. No other cellular protocol is widely licensed, etc. The fights are ugly, but the products are improving and the prices are dropping. What more can we ask? REFERENCES 1. Fractional Tap-Spacing Equalizer and Consequences for Clock Recovery in Data Modems, IEEE Transactions on Communications , Vol. COM-24, 1976, pp. 856864; Table 15-3 Cellular modem connectivity comparisons Category Connectivity (%) Modems with interfaces 2050 EC 2 solutions 3845 ETC direct connect solutions 4049 AirTrue 9096 REFERENCES 255 Method and Arrangement for Coding Binary Signals and Modulating a Carrier Signal, U.S. Patent No. 4,077,021, Feb. 1978; Channel Coding with Multilevel/Phase Signals, IEEE Transactions on Information Theory , Vol. IT-28, 1982, pp. 5567. 2. PC Week , 1-19-88. 3. Proposal for a 14,400 Bits per Second Modem for Use on 4-Wire Telephone Circuits, IBM Europe, Contribution to CCITT Study Group XVII, No. 100, Feb. 1983. 4. SMR and Data , 9-23-91. 5. D. Kurt, Motorola press release, 6-17-96. 6. iDEN Technology Overview, http://www.nextel.com/information/technology/ supporttech. shtml. 7. Metricom press release, 3-31-97. 8. Metricom press release, 9-23-98. 9. PC Week , 5-10-88. 10. Info World , 5-30-88. 11. http://csps1.corp.mot.com/GSS/CSG/Help/DataTutorial/page 04c.html. 12. The Cellular Handbook , Microcom publication 10K-RES-6/90. 13. Example: AT&T Paradyne Access the Globe series. May, 1995 marketing brochures. 14. Mobile Data Report , 2-15-93. 15. Paradyne White Paper, 7-24-97 (http://135.90.25.1/technologies/etc_whitepaper.html). 16. AT&T Paradyne White Paper, ETC2 Quick Connect, 5-9-98. 17. Microcom, Microcom Networking Protocol, 1990 ed., p. 3. 18. Byte Magazine , Nov. 1990, p. 360. 19. Mobile Data Report , 2-15-93. 20. AT&T Paradyne, Mobile Data Report , 7-5-93. 21. R. Scott, Paradyne Senior Technical Staff, who says he was aghast by how much better V.42cell stacked up against MNP10. MNP10 was a terrible performer. 22. JFD Associates, private communication, field test results, Sept. 2426, 1991. 23. A. Iellimo, Network World , 9-5-94, pp. 43-50. 24. Edge On & About AT&T , 6-5-95. 25. M2 presswire, 4-21-95. 26. Motorola CELLect PCMCIA Faxmodem, form BIM-424. 27. Motorola promotional material 1622-0394-01. 28. PC Week Online , 10-7-97. 29. K. Kernahan and K. Surace, Connectivity Test Method for Cellular Data, 3-22-95, Air Communications, Inc., Newsgroups: comp.std.wireless, Path: aircom1@aol.com (AIRCOM1). 30. http://www2.eccosys.co.jp:12345/~brett/wbp/air.html. 31. Paradyne White Paper, 7-24-97 (http://135.90.25.1/technologies/etc_whitepaper.html). 256 ENABLING “SOFT” TECHNOLOGIES . environment. Until recently, the best cellular modems were all based on V.32 bis , offering the potential for speeds up to 14,400 bps (Rockwell permits its cellular modems to operate in V.34 mode,. 85 kbps, available in 1998. In addition, broadening the bands still further (Starlite) offered a potential speed of 256 kbpswhich would be really moving! In September 1998 Metricom announced the demonstration. employs 14 V.42cell: essentially V.42, Appendix III modifications to improve reliability with potentially high bit rates under adverse transmission conditions. However, ETC is modulation dependent