CA1207389A - Packetized ensemble modem - Google Patents

Abstract

ABSTRACT OF THE INVENTION

A high speed digital data modem particularly suited for use on a dial up telephone line is disclosed.
For the transmit ensemble, the telephone passband is divided into sixty-four sub-bands each with a carrier located approximately in the center of each sub-band.
Each carrier is amplitude and phase modulated in order to encode five (5) bits. One carrier is used as a refer-ence signal for phase and amplitude. The modulated carriers can be changed in data content every epoch.
By use of packetization of data, individual amplitude correction, and individual phase correction for each carrier, the high speed modem may achieve up to 12000 bps over a dial up line with a simultaneous 300 bps reverse channel.

Description

~ 2~ ~ 3 ~` ~

PACKETI ZED ENSEMBLE MODEM

BACKGROUND OF THE INVENTION

Field of the Invention The invention relates to the field of data communications and inds particular application as a high ~peed, full duplex, modem.

DescriPtion of the l?riox Art While at one time telephone lines only carried voice communications, for many years now it has been common or digital data to be conveyed over the tele-phone network. ~owever, this clevelopment ha~ not been without ~ignificant technical limitations and obstacle~.
For example, the bandwidth of ~I typical "dial up" tele-phone line is only appxoximately 3 kilohertz (kHz~ which ~exves as an upper limit to the~ data transfer rate upon the line. Fur~her, impairments or performance limita-tions of the dial ~p tel~phone netwoxk have ~everely ~0 limited the ability of the network to reliably transfer digital data at high 6peeds. For exh~ple:
Telephone lines ~uffer from frequency dis-tortion or attenuation of the high and low frPguencies acr~ss t~e available 3 kHz bandwidth.
~ Thexe is c~mmonly phase distortion or a difference in ~he time delay for each freguency compo-ne~t across ~he available 3 kHz bandwidth.
Frequently a hetxodyne off~et ~ay be ~ncoun-tered which ~hifts ~he received compo~ent fr~quencies with respect to the transmitted component fres~lencies;

~,~

.

73~

this results from freguency variations between the tele-phone company'c hetrodyne oscillators.
Amplitude distorion is common and often caused by non-linear c~mplification of the telephone companyls A/D converters.
c Impulse noise is common and often results from line "hits" (i.e. lightning) or dial "clicks" from other telephone circuits.
Cross-talk is not a rare occurrence; cross-talk is the "leaking" of voices of tones from one line to another.
Phase jumps ~i.e. the instantaneous changes of time delay or phase) of the carrier are common.
~mplitude jumps (i.e~ instantaneous changes in amplitude generally as a result of alternate micro-wave link switch overs) are all too common.
Gaussian noise i~ an ever present impair-ment that plagues all electrical systems.
2V Echo ~uppressor are installed by the tele-phone company to permit very long distance voice commu-nication, ~ut they must be disc~led for lon~ di~tance d~ta two way communication.
Typically, high speed modems have ~uffered 2S from ~ome hybrid loss or the undesired return of a por-tion of ~he transmitted ~iynal into the local receiv~r channel.
Echo on telephone lines is the typical oc-currence of return of the transmitted signal back into 3~ the receiver channel usually on long distance communica~
tions.
Satellite delay is a further line imp~ir-~ent caused by ~he delay encountered by the distances travelled by telephone ~ignal~ when beamed to geostc~ble ~ar~h ~atellites.

In addition to the foregoing, several other factors must be kept in mind when considering digital data telecommunications at, say, 9600 bits per second (bps). First, achieving reliable 9600 bps over a dial up line is virtually unheard of. To obtain reliable 9600 bps communication over a telephone line, condi-tioning is often a necessity; a conditioned line is one for which the user pays a premium and is assured by the telephone company that the line is of lower (or custom-i~ed) noise characteristics.
Conditioning is tailored to the high speed modem manufacturer's specifications, and conditioning cannot obviously be obtained for a dial up line (that would mean conditionin~ all lines). Second, 9600 bps full dup:Lex opera~ion can only be obtained with modems of the prior art when utilizing two lines (i.e. four wires), certainly not on a single channel and never on dial up lines. Third, modems ~f the prior art which do manage to achieYe 9600 bps on conditioned lines do not ~racefully degrade their perfonmance in the presence of impairments. That is, a conditioned line does not guar-antee zero impairments; it only statistically reduces the probability of impairments. Yet, when 9600 bps modems of the prior art encounter noise, as they inevi-tably will, they typically reduce the transmitted data rate to 7200 bps, 4800 bps, 2400 bps, 1200 bps, and so on until reliable communication i5 re-established.
Often, though, impainments of the telephone line are limited to particular frequency ~andwidths, and thus, reducing the net data transmission r~te usually by a factor of two or four i~ a needless and uneconomic waste of ~he available telephone bandwidth.
One high ~peed digital modem of the prior art is the SM9600 Super MOdem manufactured by Gandalf Data, Inc. The SM9600 is nominally a full duplex, 9600 bps * - Trade Mark 7,~

modem which can operate over a dial up line. However, the SM9600 data sheets recommend conditioning ~f the line. In the presence o~ impairments, ~he SM9600 will "gearshift" or drop back its transmitted data rate to 4800 bps or 2400 bps. Further, the*SM9600 cannot oper-ate full duplex at 9600 bps over s sirAgle channel but must allocate a portion of the available spectrum for a reverse channel if a single channel is all that is avail-able or utilize a ~econd line (i.e., a total of fourwires) for full duplex operation. Further~ still, the SM9600 cannot cope with ~ingle or multiple extraneous tones within ~he passband.
The most common class of 9600 bps digital data modems of the prior art is available from AT&T
(Model 2096A). This class o modem is extremely sensi-tive to impulse noise (i.e., dial clicks and lightning strikes3. lt will tolerate but 5 Hz frequency hetrodyne offset, a~d it requires use of a conditioned line for 9600 bps. Further, this class of modem is extremely ~ensitive to cross-talk, and the unit does not have any capability for error suppression.
U. S. patent 3,706,9:29 to Robinson et al.
discloses a combined modem and vocorder pipeline pro-cessor. In Robins~n, the modem function is implementedusing 16 frequency division multiplexed channels, data being carried on each channel by means of phase ~hift keyed modulation of each carrier. Robinson, however, requires a four wire circuit or a completely s~parate line for the reverse channel in order to achieve full duplex operation, and conditioning of the line is a necessity to achieve high data rates.
U. S. patent 4,206,320 to Keasler et al. dis-closes a high 6peed modem suitable for operating with a 6witched network. However, Keasler require~ a completely ~eparate reverse channel to obtain full duplex operation.

* - Trade Mark `` i ~2~3~9 Further, while utilizing 32 carriers to convey the in-formation in a frequency division multiplexed manner, Keasler employs the inefficient mechanism of forming a I'hole" ~r delay at the beginning and end of each modula tion sub-period in order to minimize ~he effects of intersymbol distortion. The shortened modulation period can cause undesired crosstalk between adjacent channels.
Several other aspects tend to render all high ~peed digital data modems of the prior art, including those referenced hereinabove, obsolete. No modems of the prior art are capable of functior.ing in both syn-chronous and asynchronous modes. Sy~chronous transmis-~ion is the old standard form of data communication over a telephone line. However/ asynchronous transmis-~ion in the form of pacXetization vastly improves the error perormance of ~he modem. In packet transmission data is ~ormed into blocks (say, from 0 to 256 charac-ters each) which are ~ent as ~elf contained packets.
~ach packet contains housekeeping information for fram-ing, routing, error detection, etc. This housekeeping information is placed into packet header and trai.ling fields. During the last ew years there has been an unpre~edented, rapid adoption of an international stan-dard for packet ~witched communication -I~ternational Standards Organization X.25. X.25 is a multilevel pro-tocol with only the lower levels which deal with the transmission of the data itself now unambiguously de-fined. There is a major advantage in having a high ~peed modem that is capable of X.25 protocol compati-bility ~or obvious internet communication. There is a further 6ignificant advantage in having a modem ~hat can communicate over internets utilizing pro~ocols other than X.25 (each computer company devised its own pxoto-col prior to X.25~. There are ~imply no high 6peeddigital data modems of the prior art which are 7~

compatible with packet ~witched networks (not to mention any that are compatible with multiple packet ~witching internet protocols) and which operate in both synchro-nous and asynchronou~ modes concurrently through ~ignalmultiplexing.
In addition to the foregoing, modems of the prior art suffer $rom the inability to isolate or other-wise pinpoint the source of poor error performance to ei~her the modem Dr the communication medium. ~igh ~peed modems of the prior art commonly calculate a com-posite data error rate which aggregates causes origi~
nating in ~he modem itself and ~he causes originating from ~he telephone cir~uit. Often the most vexing prob-lem to the data communication manager is assigning blamecorrectly for poor communication performance so that the problem can be fixed. It is ùsual for the tele-phone company to immediately deny that any transient ~ault is due to it~ ~quipm~nt. By their nature, tran-sient faults tend to disappear in time and l~ter grossmeasurements often in fact find nothing wrong with the telephone circuit. Further, even if the circuit impair-ment persists, measuxements made by the telephone com-pany are often incomplete with respect to the range o possible impairments. There are simply no modems of the prior art which can ~eparate modem performance errors from telephone circuit impairments. Moreover, there are simply no modems of ~he prior art which can separate modem perormance errors from telephone circuit impairments. ~oreover, there are simply no modems o~
the prior art which permit the modem user to completely characterize the telephone circuit so ~hat the user can adequately direct xepairs to the telephone circuit by ~he telephone company.
.

~ 7~

SUMMARY OF THE INVENTION

It is thus an object of the present inventi~n to provide a digital high speed data modem ~hat can operate in full duplex mode on a dial up telephone line at data rates in excess of 9600 bps.
It is another object of the present invention to provide a digital high speed data modem that can operate in the ~ynchronous, asynchronous, and packet modes.
It is yet ano~her object of the present inven-tion to provide a digital high speed data m~dem that can operate on telephone lines exhibiting higher phase error and requency attenuation roll-off than permitted by modems of the prior art.
It is another object of th~ present invention to provide a digital high speed data modem that can operate in full duplex mo~e in excess of 9600 bps on a dial up telephone line and whlc:h can be manufactured at a lower cost to performance rat:io compared with that of prior art modems.
It is yet another ob,ject of ~he present inven-tion to pr~vide a digitai high ~peed data modem that can opera~e with multiple higher level protocols as commonly u~ed with packet 6witched networ~s.
It i~. another object of thP present invention to provide a digital high speed data modem which reducPs its transmitted data rate in small increments in response to telephone line impairments.
It is yet another object of the present inven tion to provide a digital high speed data modem which exhibit~ ~ignificantly lower error rates than modems of the prior art while operating at 9600 bp~.
It is yet another object of the pre6en~ inven-tion to allow flexibility of data input format and the ~Z~7~

intermixing of a multiplicity of sources of asynchronous, synchronous, and packet format data in an essentially error-free manner.
It is yet another object of the present invention to permit local or remote diagnostics to be performed in an unambiguous rnanner in order to pinpoint errors by characteri~ing the channel's noise, frequency attenuation, phase shift, and frequency offset solely in terms ~f the corruption by the transmission medium and not as an arti-fact of the modem's data characteristics.
It is yet another object of the present inventionto determine whether the telephone line or the modem is at fault when errors occur and to unambiguously inform a re-mote site as to the specifics of the problems in measured quantitive terms with both data and voice backup.
Summary of the Invention In accordance with an aspect of the invention there is provided in a modem utilized for transmitting source digital information and for receiving destination digital information over a communication medium utilizing a plurality of carriers located within the passband of the communication medium, the improvernent comprising means for measuring the transmission charac~eristics o said communi cation medium constructed rom means for transmitting said source digital information on selected ones of said car-riers at a pxedetermined amplitude level and for a pre-determined period, means for analyzing signal corruption caused by communication medium impairments to received carriers carrying said destination digital information by 3Q comparing received data patterns with known good data pat-erns, and means for avoiding use of at least a selected one of said carriers for transmission of said source digital information in the event said analyzing means indicates said one o said carriers is subject to impair-ments to said communication medium affecting transmissionof data.

7~
- 8a -In a preferred embodiment of the present inven~ion, an ensemble of sixty-four (64) orthogonally modulated carriers is digitally generated. Each such carrier or tone is individually modulated to contain five ~5) bits. One (1) carrier within the ensemble of sixty-four (6~) serves as a pilot tone for precise coordination between the transmitter and receîver sections of the modem of ~he present invention. ~his tone maintains timing and frequency calibration or "alignment" irrespective of the telephone network's carrier heterodyne errors or changes in transmission path length. The transmitter and LeCeiver portions of the modem of the present invention operate in conjunction with one another in a coordinated manner to deduce real time information on the performance of the telephone channel being used. All significant signal parameters are measured and corrective signals are returned to the originating modem on a simultaneous reverse channel. Carrier ~ 7 ~ ~

frequencies, generally located at ~he end of the usable spectrum ~but not always), which are impaired from any number of causes are removed from the ensemble. In the modem of the present invention, the spectrum ~pacing is minimized so as to permit the simultaneous transmission of both 9600 bps or greater in one direction ~as high as 12000 bps in a preferred embodiment3 together with 300 bps in the reverse direction on a single, two wire, dial up telephone line.
Several levels of cl~se interaction and coor-dination ~re important between the receiver and the transmitter sections of ~he modem of the present inven-tion in order to permit maximum error-free data transfer.
Data packets are exchanged between the transmitting and receiving sections and use interleaved cyclic redundancy checking (~RCs). Repeat transmission is used upon the detection o errors. This permits the error-free convey-ance of data/ housekeeping information ~e.g. ordering of packets), and diagnostic signals. This arrangement permits adaptation of the modulation of the carriers in order to achieve data transmission at ~he maximum possi-ble effective error-~ree rate for the cha~lel. In the event of impairment, the actual data ~hroughput capacity ~5 gradually decreases by dropping out defective individual ~arriers until a new e~uilibrium point of error-free transfer is reached. It is ~hus clear that unlike modems of the prior art which "gearshift" down the transmitted data rate by a factor of two or four in the presence of impairments to the telephone line, the m~dem of ~he present inventi~n reduces its throughput data rate by only approximately l/64~h of its capacity per ~tep.
Typically, 6ingle step reductions in data ~hroughput will be all that is necessary to surmount most i~pair ments.

:~Z~3519 The present invention also includes the use of a pulsed pilot tone to effect correction of telephone line amplitude variations, amplitude hi~s, frequency offsets, phase drift, and phase hits. Frequency distor-tion is essentially removed in ~he modem of the present invention by modulation of the amplitude of the local oscillator equivalent circuit for each of the sixty-four (64) tones in the ensemble. Phase distortion and channel cross-talk are removed by processing calculations which are based upon initial transmission of a ~nown test pattern.
An important advance over the prior art of the modem of the present invention is packetization of all data and housekeeping information. The modem of the present invention handles all data on a packetized basis and permits intermixing of a multiplicity of data streams. Each data stream may operate at almost any speed or protocol combination by virtue of packetization permitting a degree of fle~ibility never before achieved.
In a preferred embodiment the duration of each frequency ensemble is 2/75th of a second. In this period are transmitted up to 320 bits as an ensemble packet which contains both data and error detection means. This permits closed loop feedback between the transmitting and receiving modems to be based upon new error free intelligence.
The packetizing and extensive test signal arrangements in the modem of ~he present invention also permit rPady use of packets containing solely test data.
These packets are useful in order to provide remote diagnostics and to pinpoint transmission faults. Th~
modem of the present invention measures amplitude, noise, phase delay, and frequency offset for each of a large ensemble of frequencies; these measured parameters are characteri~tics solely of the communication circuit and rf ~
~ ~7;3~

not of the modem. Such measurements are stored over a ~eries of periods to provide operating st~tistics to a remote modem or diagnostic c~nter in a form that unam-biguously defines transmission line problems. To aidin the process of rapid line fault diagnosis, a 6eparate telephone line can be made available that automatically connects to a remote data center. This ~eparate circuit forms the transmission path for the readvut of ~tatis-tics, and it also forms a voice intercom permitting theoperating personnel at the modem site to communicate wi~h the csnnected modem or to a remote diagnostic center.
In a p~eferred embodiment of the present inven-tion, sine and cosine vectors for each of 6ixty-four (64) separate frequencies are generated for each carri~r.
Each is derived digitally. The symbol transmission period is 2/75th of a second, and each symbol conveys five (5) ~its using thirty-two (32) combinationfi of phase and amplitude modulation. The definition of the phase and amplitude combinations is called a "constella-tion" and in the present invention is selected as a function of the characteristics of the real time impair-ments encountered on the telephone line.
The preerred embodiment is described in terms of sinusoidal carrier frequencies. ~owever, the inven-tion is n~t so limited in theory; other orthogonal wave shapes could be utilized in lieu of conventional sinu-soids. Specifically, pseudo-random but orthogonal noise ~treams could be used instead. Each pseudo-random wave-form would be ~tored in a read only memory (~OM). Thewaveform would be one epoch in length and could be ampli-tude modulated. Separation of the waveforms occurs by multiplication by an identical ~et of waveforms in the receiver by use of noise as a ~pread spectrum transmis-~ion concept. The nature of the detection process inthis modem of ~he present invention in which each chan-73~9 lX

nel is anticipating a particular waveform over a rela-tively long epoch period tends to lend itself to the transmission of a simultaneo~ ensemble of pseudo noise channels. This arrangement is particularly fîtting to insure privacy.
It is thus an advantage of the present inven-tion to provide a digital data modem capable of reliably operating in full duplex modem over a dial up telephone line at data rates in excess of 9600 bps by allowing closest spacing between adjacent carriers without fre-guency-to~frequency crosstalk permitting more data chan-nels within the telephone passband.
It is an~ther ~dvantage of the present inven-tion that data throughput degradation in the presenceof telephone line impairment is gradual and tailored to the particular form of line impairment.
It is another advantage of the present inven-tion to be able to utilize a dial up telephone line with ~ignificantly higher phase error and fre~uency attenuation rolloff than previously possible.
It is yet another advantage of the present invention to utilize asynchronous data transmission and reception in the form of packetization of data and house~
keeping ~unction.
It is yet another advantage of the present invention to permit the concurrent multiplexing of a plurality of si~nals, both asynchronous and synchronous.
It is yet another advantage of the present invention to permit operation using a wide range of data communications protocols.
These and other objects and advantages of the present invention will become appar~nt by referring to the drawing figures in conjunction with ~he description of a preferred embodiment.

~%~7.~8~ ), BRIEF I)ESCRIPTIOM OF_THE DRAWINGS

Fig. 1 is a high-level block diag~am of the high ~peed modem of the present invention.
Fig. 2 i~ a pict~rial represent~tion of the transmit and receive ensembles of the present invention.
Fig. 3 is a functional block diagram of the ~ignal generator/hybrid of the high speed modem of the present invention.
Fig. 4 is a functional block diagr~m of the signal extractor of the high speed modem of the ~resent invention.
Fig. 5 is a functional block diagram of the vector de~iner of the hi~l speed modem o ~he present invention.
Fig. 6 is a functional block diagram of the crosstalk reducing circuit of the high speed modem of the present invention.
Fig. 7 i~ a functional block diagram of ~he reference corrector/generator of the high speed modem o~ the present invention.
Fig. 8 i~ a functional block diagram of the co~tellator o~ the high speed modem of the present in~ention.
Fig. ~ is a functional block diagram of the loading docks and the ~hipping department of the present invention.
Fig. lOA is a pictorial representation of the bits transmitted for each of the sixty-fcur 164) carriers for two time epochs of the present invention.
FigO lOB is a pictorial representation ~f the bits transmitted in a reverse cha~nel for each of nine (9~ freguencies for two time epo~hs of the present inven-tio~.

~ ' ~%~7~g Fig. llA is a pictorial representation illus-trating binary si~nals as a function of time for two epochs to transmit a 640 bit packe~ according to a pre-fe~red embodiment of ~he present invention.
Fig. llB is a pictorial representation illus-trating ~he housekeeping and data assignments for a h~pothetical packet according to a preferred embodiment of the pre~ent invention.
Fig. llC is a pictorial representation illus-trating the m~nber of bits in each field comprising an exemplary packet according to a preferred em~odiment of the present invention.

DESCRIPTION OF A PREFERRED EMBODIMENT

Reerring ko Fig. 1, a preferred embodiment of the high ~peed modem of the present invention i~
shown by ~he general reference numeral 10. High speed modem 10 performs both transmit~ing and receiving func-tions over a telephone line 12 in conjunction with a similar high ~peed modem 10' (not shown~ located remotely at the opposite end of telephone line 12. Telephone line 12, typically a 300 to 3100 Hz dial up telephone channel, co~nects to a ~ignal generator/hybrid 14 adapted to separate the tran~mit path from the receive path. Also contained within ~ignal generatorfhybxld 14 is the circuitry for generation of the transmitted 6ignal described in detail hereinbelow in ~onnection with Fig.
3.
Signals received by ~ignal generator/hybrid 14 are pa~ed onto a 6ignal extractor 16. Within signal extractor 16 each component o the undesired tra~mitted signal is removed from the complex received waveform.
The output of signal extractor 16 pa6~es onto a vector definer 18. The vector definer 18 computes the two ~ 3 ~7~ '?

values of ~he received ~inusoidal signal. These ~wo values are referred to hereafter as the x and y vector components and interchangeable referr~d to as the sine and cosine channels, respectively. The output of the vector definer 18 branches into two paths. One path leads to a constellator 20 which matches the received x and y vectors to the closest fitting probably expected signal combination. The other path from the vector definer 18 leads to a reference corrector 22 where the frP~uency, time, and amplitude characteristics of the reference signal are extracted and ~ompared with its expected value and corrections made accordlngly.
The output of ~he constellator 20 proceeds to a diagnostics analyzer 24 which extracts performance information necessary for the coordination of the sending and receiving modems and for diagnostics purposes.
Dia~nostics analyzer 24 performs such functions as deter-mining whether any frequency channels are exhibiting excessive noise levels. Diagnostics anal~zer 24 also generates statistical estimators of individual channel phase delay, amplitude distortion, noise, etc. The in~ormati.on from the diagnostics analyzer ~4 is formed into p~ckets and sent to a ~hipping department 26. The ~hipping department 26 is responsible for the organiza-tion of information into packets for the purpose ~f transmission and or ~he disassembly of packet~ i!l the case of reception of information. Packet switching and the packetization of information itself is well known in the prior art of data communications. The techniques of packet switching will not be described herein other than to ~ote that most information within the high speed modem 10 of the present inv~ntion is transferred in the form of 6tandard ~ize packets ~f information each con-taining a header together wi~h housekeeping data whichfacilitates information transfer between a ~ending high lZ~7;3'~

speed modem 10 and a remcte receiving high speed modem 10'. Internal interlaced error correction ~ymbols permit the detection of error with a high degree of certainty.
Errors are generally corrected by the repeat transmissiPn of defectively receivPd packets. Packets between the shipping department 26 and the outside world (i.e. a computer system to which high speed modem 10 is con-nected) are sent and received by one or more loading docks 28. Each loading dock 28 is tailored, generally by software commandt to match the precise interface requirement of the data user's equipment. For example, many data terminals utilize a standard RS-232-C connector and voltage levels. A loading dock 28 would be used to connect to such an RS~232-C interface. The loading docks 28 also remove and add housekeeping data from ~he packets ~o that the output serial data stream appears to the user as input to the transmitter with the visi-bility of added information reguired by the packet switching process itself.

Ensemble DescriPtion Referring to Fig. 2, a diagrammatic xepresen-tation i~ shown of the transmit and recelve channels that fonm the transmit and receive ensembles for ~he hi~h speed modem 10 of the present inventi~n. High ~peed modem 10 simultaneously transmits an ensemble of freguencies 30 and receives an ensemble of fre~uencies 32. ~igh ~peed modem 10 is designed in a preferred embodiment to opexate with either four-wire telephone ~ircuits (i.e. ~wo lines) or two-wire telephone d rcuits (i.e. a ~ingle line). With a four-wire circuit co~nec-tion (not illustrated), the full ~pectrum from 300 H~
to 3 ~Hz i~ available in both directions simultaneously.
In the two-wlre case (that shown in Fig. 2), 6eparation i! lZ~7~3~9 between the transmit ensemble 30 and the receive ensemble 32 is imperfect. Freguency division is thus used to separate the two directions. Fig. 2 shows the transmit ensemble 30 assembled from sixty-four (64) separate frequencies occupying the band from approximately 600 Hz to 3kHz. Receive ensemble 32 is depict~d in Fig. 2 to be of proportionately narrower bandwidth occupying only the band form approximately 300 Hz to 5Q0 Hz.
This particular choice of frequency allocation capital-izes upon the statistical reality that heavy data flows will generally occur in one direction with a lesser flow in the reverse direction. The receive channel or receive e~semble 32 is utilized to convey user data as well as carrying acknowledgements that transmitted packets were in fact correctly received. If no acknsw-ledgement is received, then the previously transmitted packet ensemble is retransmitted.
In Fig. 2, a plurality of vertical lines 34 ~0 ~eparate each ensemble into time increments of 1/37.5th of a ~econd, known hereafter as "epochs". Each epoch or epoch period is used to transmit a different coded ensemble of freguencies wherein ~ach tone or frequency is phase and amplitude encoded lto convey a multiplicity of bits. In a preferred embodiment, each tone is encoded wi~h five (5~ bits of information; this creates thirty-two (32) discrete comhinations of phases and ~mplitudes (i.e. the well known "constellation"). The composite sixty-four (64) channel capacity comprising transmit ensemble 30 permits conveying up to 320 bits in an epoch or l/37.5th of a ~econd. This is the equivalent of 12~00 bps. However, in a preferred embodiment channel #32 of txansmit ensemble 30 is utilized solely for refer ence purposes. That is, channel ~32 cyclically transmi~s at full inten~ity for one epoch, off for two epochs, on for one, of for two, etc. The amplitude ~f channel 12~7~3~9 ~8 #32 establishes an amplitude reference. The beginning and end of transmissions of channel #32 precisely establish the time boundaries for an epoch. Further, the frequency of ~hannel #32 as received by a remote high ~peed modem lO' provides reference information utilized to correct heterodyne frequency offset that often occurs in passage through the telephone cvmpany's carrier system.
~ ~econd reference signal 35 is used in an identical manner in the return receive ensemble 32. In the case of using high speed modem 10 with a four-wire circuit connectio~, ~he return channel occupies the full 500 Hz to 3kHz bandwidth since it would be other-wise unused. A four-wire connection thus permits nomi-nally 1200 bps in both directions simultaneously, less the capacity used for reference channels and internal housekeeping functions. With a good quality dial up telephone line, high speed modem 10 o the presen~ inven-tion ~hould achieve on the order of 11000 bps. However,if the line quality did not so permit (i.e. in the pre-sence of impairments), certain of the impaired channels would be cropped lowering net data throughput by a factor o 1~64th per channel dropped. This "gracefulli degrada-tion of the net data throughput for the high speed modem10 of the present invention is to be contrasted with the drastic "gearshifting" arrangement utilized on modems of the prior art where minor impairments can reduce the net data throughput by 50% or more.

Si~nal Generator~Hybrld Referring to Fig. 3, a detailed description of ~he ~ignal generator/hybrid i5 ~hown. A~ briefly described above in connection with Fig. 1, ~ignal gener~
ator/hybrid 14 generates the transmit en~emble 30 con-6isting of 6equential digital values each digital value i., ", ' ' ~Z~73~9 -of which represents an analog signal value identical to the signal value components of a bank of eguivalent analog oscillator circuits. All of the tones comprising transmit ensemble 30 are derived within the signal gener-ator/hybrid 14 rom a co~non crystal oscillator 36.
Although high speed modem 10 is actually two modems, a receive modem and a transmit modem, crystal oscillator 36 provides mas~er timing information for both funrtions.
1~ Thus, both modems maintain an integral timing relation-ship. With few exceptions, th~ absolute timing diffex-ence between the transmit and receive modems is a mea-surement of the long distance telephone network propaga-tion delay which tends to stay fixed for the duration of each teleph~ne call.
Crystal oscillator 36 connected to a timing signal generator 38 which in conjunction with a counter 39 provides a plurality o ~iming signals 40 used throughout high ~peed modem 10. The following tech-nique described to generate digi.tal values of tones is~imilar to the technique used for generating an ensemble of tones in ~tate of ~he art electronic musical instru-ments~ Freguencies 1 through 64 as shown are generated in an interleaved manner by u~ed of a time shared read only memory (ROM)42, an adder 44, and a sine/cosine table ROM 46~ The output of ROM 42 is a ~eries of digi-tal values. Each value may correspond to an angular increment used to yenerate pulse code modulati~n (PCM~
values of an ensemble o~ frequencies. As in standard PCM practice, a ~ampling interval is used which is at a rate at least twice the frequency being ~ampled, or ~he (Nyguist rate). At si~ty-four (64) discrete locations within RO~ 42 are 6tored a plurality of incrementing value~ 43 corresponding to the ~ixty-four (64~ frequen-cies to be derived from timing ~ignal~ 40. Each incre-menting value 43 i~ added to the past sum in an adder 7~3~3g

2~

44 which acts as an integrator. The lower order hits from adder 44 form input addresses 45 to the sine/cosine table ROM 46. The pulses which constitute timing signals 40 are counted in an address counter 39. As the value within address counter 39 is se~uentially incremented, a different location of ROM 42 is sequentially addressed producing each of the sixty-four (643 tones or carriers.
The ~ine/cosine ROM 46 contains values corres-ponding to the sine and the next sequential address, ~le cosine. Signals within high speed modem 10 are transmitted by generating two vector components. The fir~t vector component is a sine value ~8, and the second vector component is a cosine value 50. Data which is to be transmitted enters at point 52 and is organized into groups of five (5) bits each. Each ~ive (5) bit gxoup represents a particular combinakion of vector coordinates~ With five l~5) bits thirty-two ~32 separate values or composite vectors are required for transmission. As the ideal choice of angle and magni-tude for each of these vectors depends somewhat upon the nature of the communication line used, the choice may be parameterized ~nd select~ed from a read-only-memory as reguired. This is shown schematically in Fig. 3 as a switch 54. Switch 54 chooses one of a plur ality of 5 bit input x,y vector amplitude ~OM's 56.
The output of each vector amplitude or constellation ~OM 56 is received by a multiplier 62~ Within multiplier 62, x value 58 is multiplied by sine value 48 from the sine/cosine RO~ *6. Shortly thereafter, the y value 60 from vector R~M 56 is multiplied within multiplier 62 by co~ine value 50 from the sine/cosine ROM 46 (x ~alue only illustrated in Fig. 3). This process is repeated for the next freguency modulated by the next fi~e (53

3~ bits to be transmitted. This continues until all 3~0 bits have be@n sent using all ~ixty-four (64) freguen-~ ~2~73~

cies. In a preferred embodiment of the present inven-tion, the per ~ample point time is very fast. During the ~ingle epoch period each frequency will be ~ampled 5 512 times. This corresponds to 19200 samples per fre-quency per ~econd.
Diyital output S4 of the multiplier 62 i5 received by a digital to analog (D/A) converter 66.
D~A converter 66 converts digital signals 64 into analog l~ values 68. A low pass filter 70 removes unwanted higher fxeguency products and the filtered analog signal 72 is then sent to hybrid 74. ~ybrid 74 has the function of transmitting the analog ~ignal over telephone line 12.
In addition to the transmitting function, hybrid 74 also separates a received signal 76 from telephone line 12 in the ca~e of a ~wo-wire circuit. Hybrid 74 is a bridge balancing device resulting in a difference in the adjustment to hybrid 74 as diferent ~lephone line impedances are encountered. A rough first order correc-tion to hybrid 74 is provided by a digital fiignal 78via a D/A converter 8Q. The analog output 82 of D/~
~onverter 80 adjusts the gain o~ hybrid 74 in order that the transmitted ~ignal have minimum cross talk with the receive channel. It E;hould be noted that the 25 primary channel ~eparation mechanism is frequency divi-sion between ~he transmit channel and the receive chan-nel as the two frequency bands do no~ overlap. Adjust-ment to hybrid 74 is used only for second order correc-tion to further reduce reception of spurious energy.

Signal Extractor Referring to Fig. 4, a detailed ~unctional ~chematic of the si~nal extractor 16 is shown. ~eceived signals 76 enter ~ignal extractor 16 and are fir~t fil-tered by a band pass filter 84 designed to remove extra-`~ ~L2~

neous frequencies outside of the useable spectral passband. An output signal 86 of band pass filter 84 is multiplied in a multiplier 88 by a multiplier signal 90 form a D/A converter 92. D/A converter 92 6erves as an analog multiplier to grossly adjust the amplitude of the overall received channel signal 86. After gain adjus~ment in multipli~r 88, a recei~e signal 94 is passed onto a local oscillator multiplier 96. The ampli-tude value of multiplier signal ~0 is controlled by D/Aconverter 92 by a peak detector 93 which insures that the input signal to D/A converter 92 does not ~xceed the linear ceiling of the multiplier 8B.
Into the local oscillator multiplier 96 are passed a plurality of tones 98 which compri e a 6eries of freguencies similar to tho6e transmitted but of con-stant amplitude. Con~tant amplitude is achieved for ~he product Gf the received signal 86 impaired by losses on the telephone line 12 by ad~usting the gain of each local oscillator tone 99 individlually in a channel yain control amplifier lO0. Unique t:o the present invention, information relating to the individual channel gain characteristics for each of ~he ~ixty-four ~64) carriers is stored in a channel gain random access memory ~R~M) 102. The infonmation deposited within RAM 1~2 stored at a plurality of separate addresses 104 is derived from indi~idual gain measurements for each of the car-riers transmitting ~nown amplitudes during dial up con-nection initiation. After initiatisn the received values are quantized to one of ~everal known ~mplitude values insuring that the gain in precisely maintained~ Each channel ~ain adjustment constant is called up ~ynchro-nously from R~M 102 and used to instantaneously adjust the gain of the gain control amplifier 100 processing the local oscillator 6ignals. This is eguivalent to modifying the stren~th of the received ~ignal 94.

` ~2'~7~3~91 It is common telephone line impairment that a frequency offset exis*s between transmitted and rece.ived siynals caused by the telephone carrier system. Signal extractor 16 accounts for this hetrodyne offset error of the telephone system by introducing an an offset control ~i~nal 106 into the incremental adder 44. As described above in connection with signal generator/hy-brid 14, ROM 42 contains the values to be selected for the local oscillator freguencies 1 through 64. These values axe selected to generate freguencies that are interleaved between ~he assigned transmitting frequen-cies. The frequencies selected from ROM 42 pass onto the incremental adder 44. A second common digital value, i.e. an offset control signal, is added to ~ach adder circuit 44 in order to constitute a frequency ~ffset ~ontrol. The o~fset control signa~ is derived from the reference corrector/generator 22. The ~alue o the offset control ~ignal is precisely determined so that the output of the sine/cosine ROM 46 (i.e. the sine value 48 and the cosine value 50) exactly match those of the transmitted ensemble 30. This i~ actually achieved by varying the reference frequency (i.e. fre-quency #32 ) until it achieves its exact anticipated valueO In a preferred embodiment, this scheme permits correction o~ heterodyne off~ets a~ great as plus or minus 18 ~, larger than will ever be encountered in a telephone system. This large range of heterody~e cor-rection is useful o~ ~ingle sideband radio channels where larger freguency offsets can occur. It is also useful in situations employing radio channels where moving ~tations cause a doppler shift.

~L2~7~3~9 Vector Definer Referring to Fig. 5, a detailed functional schematic of the vector definer 18 is shown. The vector definer lB is used to achieve matched frame coherent detection. An input signal 108 to vector definer 18 is received from ~he ~ignal extractor 16. Each sample has already been multiplied by the local oscillator/multi-plier 96 and thus represents each different freguency to be integrated. Integration is readily accompli~hed sin~e each signal 108 is a digital value of the esti-mated value. Input signal 108 is applied to a plurality of sample integratoxs 110. Sample integrators llO
actually are constxucted from individual sine integra-tors 112 and cosine integrators 114 for each discretecarrier or frequency (i.e. 64 in the case of a preferred embodiment). Naturally, there is much background noise in the signal applied to each sample integrator 110 as the signal 108 contains energy from irrelevant ~ha~nels as well as the specific channel desired. Fundamental to this form of detection process is the premise that the desired signal energies will add within sine inte-grators 112 and cosine integrators 114 with each ~epar-ate integrated addition while the unwanted signals ~at least with respect to a particular integrator) occur orthogonally. That is, the unwanted signals 60metimes add, and, just as often on a statistical basis, subtract from the values in ~ine integrators 112 and cosine inte-grators 11~. It i~ the u~e of a large number of samples which permits the desired signal to be extracted from the noise. After the integration process, the output signals from integrators 112 and 114 pass through a plurality of buffers 116~
As briefly discussed above, the refer~nce channel Si.e. frequency ~32 in a preferred embodiment3 (l ) ~Z~738~

is pulsed at full amplitude for one epoch and at zero amplitude for two epochs. The edges of this pulsing of the reference channel provide the basic timing informa-tion needed by the sample integrators 110; that is,when to start and when to dump. However, there i~ a second order difficulty that needs to be resolved that relates to timing, or more properly, phase distortion.
In the high speed modem 10 of the present invention, each epoch is approximately 26.7 milli~econds in dura-tion. Telephone line phase delay is commonly on the order of 2 milliseconds or more. Further, this phase delay is significantly worse near the edges of the 3 kHz pass band. Thus, if each integrator llO (i.e. ~ine integrators 112 and cosine integrators 114) ~tarted at the 6ame time, the channels located at the pass band edges would ge~erally exhibit maximum phase shift error.
It is clear that the value within some integrators 110 ~ould be as much as 10% in error ~i~e. such integra~ors 110 would contain as much as 10% of their values from energy from the previous epoch period). Such phase distortion is unacceptable in auch a high performance modem as the one described her~in.
The approach utilizecl to correct for this 2~ phase distortion which v~ries across the 3 kHz pass band is to use a measurement o:E the phase delay of each of the-6ixty-four (64) carrier frequencies with respect to the reference channel as a pha6e correction factor.
This permits a ~orrection factor to be generated separ-ately for each of the ~ixty-four (64) carriers as if each ~nple integrator 110 had been started and stopped preci~ely at the temporal boundaries of each epoch.
Thu~, a portion of vector definer 18 functions as a phase distortion corrector a~ noted in Fig. 5. The pha~e distortion correction proce6s begins by passing the output of each bufer 116 through a delay line 118 2~7~o~1~9 ') in order to preserve the previous value from integrators 112 and 114. The actual values from integrators 112 and 114 are individually multiplied in a plurality Qf multipliers 120 by a plurality of individual correction constants 1~2. The value of each correction constant 122 (i.e. Kl through K64 in Fig. 5) is determined auto-matically at the time that telephone cQnnection is first established or at the time of a "hot" restart (e.g. a restart after the error rate has exceeded an acceptable limit). In a preferred embodiment of the high speed modem 10 of t~e present invention, the technigue used to measure the pass band phase distortion is to turn on every third carrier for an epoch and then to measure ~5 the energy contained in ~he following epoch. The energy "hangirlg over" duriny the ollowing epoch is a measure-ment of (i.e. directly proportional to~ the phase delay relative to the reference channel used for the epoch timing.
Therefore, the output of multipliers 120 in a plurality of subtractors 124 frr~m the values received from buffers 116 ~uitable delaye!d by delay line 118 from the previous epoch. In a preferred embodiment of the present invention, delay line 118 is actually imple-mented as random ac~ess memory; i.e. values are actually stored in the random access memory and recalled a~ the proper time which effectively works ~he function of a classic delay line. The phase distortion correction process takes place on a per epoch basis as shown.
Only approximately 4800 corrections per second are re-guired for acceptable accuracy.
A phase corrected x value 126 and a phase ~orrected y value 128 are ~ued to produce polar coordi-nate values for further processing. The magnitude of ~he radius vector i~ computed by a squaring operation 130, a 6ummation of the squares 132, and a sguare roct ~ ~y 7~

operation 134 performed upon the sum of the squares of the x and y components. These operations in practice may be performed by computation or by a use of a table 6tored in a look-up ROM. Similarly, the angle of the radius vector i5 determined by calculating the arctangent of the x value divided by the y v~lue. The division of the x component by the y component is performed in an operation 136 and the arctangent is determined by an operation 138. Both the division operation and the arctangent computation may ~e accomplished by storing expected value~ in a look-up table ROM or by computation.

Crosstalk_Reducer Circuit Referring to Fig. 6, an optional crosstalk reducer circuit is shown by the general reference nu-meral 140. Crosstalk reducer circuit 140 provides a ~econd level of correction for signals that "leak" ~rom one frequency carrier into an adjacent freguency carrier (it is possible for crosstalk to occur beyond adjacent channels, but adjacent channel crosstalk is the major component). Of couxse, the individual carrier band centers are spaced apart by approximately 37.5 ~z in order to minimize channel-to-channel crosstalk. In Fig. 6, three representative frequencies each having 2S two vectors (i.e. xll, yll, xl~, yl2, x13, yl3) repre-sentative of the full ~et of sixty-four (64) freguency carriers. Each vector is stored in a RAM buffer 142.
This value is multiplied by a correction coefficient k ~tored in a ~AM buffer 144. The actual correction takes place in a plurality of adders 146. Circuit 140 thus provides a de~ree of crosstalk reduction accom21ished in a ma~ner conceptually similar to the phase correction circuitry utilized in the vector definer 18. That is, at the time of initial telephone connection, every third ~ZC?7~

frequency channel is turned on at full amplitude for one epoch. The magnitude of energy that i~ detected in otherwise quiescent adjacent channels is proportional to the adjacent channel crosstalk and used as a reference.
This information is then stoxed in RAM buffers 144 and used to form a correction constant k which is multiplied by the output of the phase distortion corxection cir-cuitry of vector definer 18 and then subtracted (by negative addition) therefrom. This process is repeated for each of the sixtyOfour (64) channels until the RAM
has stored for a particular t lephone connection the magnitude of adjacent channel crosstalk from any channel into an adjacent channel.
Crosstalk reducer circuitry 140 is placed after the phase distortion correction circuitry operating upon corrected x values 126 and cor~ected y values 128 and before the reckangular to pol~r conversion circuitry of vector definer 18.
Each of the sixty-foux (64) carriers conveyed as two vectors (xl, yl, x2, y2,..., x64, y~4) forms an input to the crosstalk reducing eircuit to produce the corresponding output vectors xl', yl', x2', y2',....
x64', y64'.

Reference Corrector/Generator Referring to Fig. 7, a reference corrector/gen-erator 22 for ~he high ~peed modem 10 of the present invention is ~hown by the general reference ~umber 22.
Reference corrector/generator 2~ uses incoming timing infonmation ~o that ~ample integrators 110 of vector definer 18 start integration and reset precisely a~ the boundaries of epochs. The basic input to reference corrector/generator 22 is a re~erence signal 202 ~i.e.
caxrier frequency 32 in a preferred embodiment)O From ~7.3~

a random start position, the energy in the reference ~ignal 202 is measured with respect to three consecu-tive epochs. If properly timed, the central epoch should S contain all of the energy; the first (or early) epoch should contain zero energy; and the last (or3 late epoch ~hould contain zero energy. ~ modulo-3 counter 204 ~equentially opens an AND gate 206, an AND gate 208, and an AND gate 210. When AND gate 208, 208, ~nd 210 are "on" as det~rmined by modulo 3 counter 204, then enexgy will be accumulated within an early gate 212, a late gate 214, and a normal gate 216, respectively. If input signal 202 is pxoperly timed with respect to modulo-3 counter 204, then all energy will reside within normal gate 21~. The output from the normal gate inter-~al forms a timing signal 217 used by the vector definer 18 to mark the start and end of epoch integration periods.

~ threshold detector 218 and a threshold deter 220 store the energy receiv2d in the early gate 212 and the late gate 214. If ei~her value exceeds a predeter-mined threshold value, then a very coaxse adjustment is made by resetting a common counting chain 222. Howevçr, if high speed modem 10 has been in recent ~ynchronism or if ~he energy in ~he eaxly gate 212 and the late gate 214 is ~ufficiently small, then only a gradual phase shift correctiun is required. This is effected as shown in Fig. 8 by corrective input to the countîng chain 222 by a voltage controlled oscillator 22~. Note that as not all of the processing takes place within the digital domain, a D/A converter 226 receives a digi-tal input from a ~ubtractor X28 for feedback correction to the 06cillation frequency of voltage controlled osril-lator 224 in a ~anner which produce~ a long time constant correction not ~ensitive to minor hort term perturba-tions~

lZS~7.~39 Constellator Referring to Fi~. 6, a detailed functionalschematic of the c~nstellator circuit 20 is shown. It is the function of the constellat~r 20 $o analyze the phase and amplitude values obtained from vector definer 18 and to match said phase and amplitude values (i.e.
the defined vectors) with the closest values stored in a ROM (i.e. constellation vector~) which correspond to specific data patterns. This is a closest-fit, table look-up process. More precisely, phase inf~rmation in a preferred embodiment of the present invention is de-coded by determining the difference in phase from one epoch t ~he next. Utilizing the technique of differen-tial phase shift.modulation removes the necessity ofmaintaining a very long term pxe!cise and absolute phase relationship for each of the sixty-four (64) channels.
In Fig. 8, a phase sic~al 252 for a particular epoch is o~tained from vector definer 18. The phase signal value from the previous epoch is skored in a time delay ~AM 254. The difference in phase between the present epoch ad the previous epoch is calculated in a subtractor 256. A difference ~ignal 248 is then passed ~nto a phase drift corrector 260. Phase drift corrector 260 is a subtractor which corrects ~he differ-ence si~nal 258 by a small amvunt with respect to a phase correction signal (not sh~wn) ~btain~d from the reference correctorfgenerator 22 in order to account for any ~econd order drifts not previously removed.
A corrected phase value X64 is thereafter passed along with an amplitude value 266 derived from vector definer 18 to a constellator ROM 268. A compari-son of th~ paired values wi~hin the ROM 268 is thereafter undertaken with tha derived amplitude value 266 and 35 corrected phase value 264. ~ closest phase angle 270 . .

1~7~

and a closest amplitude value 272 as stored within ROM
268 form a unique vector which corresponds to a uni~ue data pattern. In a preferred embodiment of the high speed modem 10, five (5) bits are encoded in each epoch of each data channel. Thus, there are thirty-two (32) pairs of amplitude and phase values or thirty-two (32) vectors stored within ROM 268. Simple algorithms per-form the matching and output a data pattern. Clearly, if none of the stored vector pairs matches the received and deined vectors or if the received vector value is too far away from the stored values, an error will re-sult. Short of an error, high speed modem 10 of the present invention has the facility to monitor the error angle or phase diference between the stored angles and the xeceived angle. The same is true for the difference between the stored amplitudes and the received amplitude.
This information is of interest for diagnostic analysis of the performance of high speed modem 10 and the analy-2t) sis of the characteristi~s of the telephone connection.Referring to Fiy. 8, closest phase angle 270 and mea-~uxed (or received) phase angle 264 are received by a phase subtractor 274. A phase error signal 276 is ~here-after available ~or error monitoring. A high level of phase errvr provides inormation that one or more chan-nels has a hig~ probability of error and should conse-quently be ignored. This logic is provided in a micro-processor (not shown) which ~ends pac~ets to the con-nected modem 10' containing the individual channel per-3~, formance characteristics. Similarly, ~he closest ampli-tude 272 and the derived amplitude value 266 are passed onto an amplitude ~ubtractor 278. ~he amplitude differ-ence i6 calculated and an amplitude error ~ignal 280 is thereafter available to permit modification of t~e indi-~idual channel gain values 104.

u~ ....

Packet Transmission Referring to Fig. 9, transmitting modem 10and its receiving counterpart modem 10' are illustrated exchanging inormation in the form of a packet 300 and a packet 300' (or ease of illustration, prime notation as used herein refers to structures wi~hin modem lO' corresponding to equivalent structures within moden~
10). Both housekeeping information and user data are transmitted. Housekeeping includes information such as ~he definition of the unusable chann~l frequencies and the failure of error detection checksums. Error detec-tion checksums (e.g. cyclic redundancy checks3 are used to :insure essentially ~rror-free overall data txansmi~-sion. Additional housekeeping information i~ neededwhen multiplexing signals rom a plurality of data sources connected to the plurality o separate loading dock~ shown ~y tha general reference numeral 28. Eacb loading dock 28 has a unique address value, and in a preferred embodiment o~ the pre~ent invention, loading docks are numbered from 0 to lS (established ~y a four bit address discussed hereafter). In Fig. 9, a specific loadin~ dock 0 (LD0) is shown by the reerence numeral 302. Similarly, specific loading docks 1 and 2 (LD1 ~5 and LD2) are shown by ~he reference numerals 304 and 306, respectively. ~ach loadiny dock ~8 within modem 10 presént~ the illusion that it is physically connected (i.e. a "virtual" coNnection) with a correspondin~ load-ing dock 28' located wi~hin modem lO'. That is, loading dock 302 operates as if connected wi~h loading dock 302'. Similarly, loading docks 304 and 306 operate as if physically connected wi~h loading docks 304' and 306', respectively. Internally ~ddres~ed pac~ets within modem 10 are directed from virtual loading dock 302 t~
virtual loading dock 302'. Pack~tization of data allows ~z(~7~ 3 3~

the efficient transmission of a multiplicity of data sources where ea1h source may have a different data rate from moment to moment. Further packetization of data permits effective data transmission even with a communications channel which is subject to the time varying uncorrectable perturbations.
Referring to Fig. lOA, an instantaneous snap shot of two ~eparate epoch periods of the transmit chan-nel 30 are illustxated. Information form modem 10 tomodem 10' is conveyed in the form of ensembles of sixty-four ~64) freguencies. As explained briefly above, each frequency carrier i~ modulated (phase and amplitude~
to thirty-two (32) states i~ order to acsommodate five (5) bits. ~ach epoch therefore transmits sixty-four (64) carriers of five ~5) bits each for a total of 320 bits.
In Fig. lOA, frequency 32 is shown blackened to denote that in a preferred embodiment of the present inv2ntion that is not used for data but for synchroniza-tion purposes. For purposes of illustration, freguency 37 is al~o shown blackened, bec~use it has been dropped as a data channel due to crosstallk from an adjacent telephone wire cable. Further, Xrequencies ~3 and 64 2S are also shown blackened because due to excessive noise ~located.at the passband edge~) they too ha~e been dropped from the ensemble for ~arrying data.
Referring to Fig. lOB, the reverse ensemble 32 i~ illustrated. In a preferred embodiment of the present i~vention, nine ~9) c~rrier~ generally l~cated-in the 300 ~z to ~00 Hz band comprise ~he reverse channel. In ~he reverse cha~nel 32 as in the transmit ensemble 30, each carrier freguency conveys five (5) bits. In reverse channel 32 as illustrated in Fig.
10~, fre~uency 73 is ~hown blackened or unavailable for ~Z~73~

data transmission, because fregu~ncy 73 is utilized for reference purposes.
Referring to Fig. llA, a serial representation of the output data ~tream from the ensembles of Fig.
lOA is illustrated. An ensemble 330 and an ensemble 331 are transmitted within epoch i and epoch i+1, re-spectively. The concatenation of ensembles 330 and 331 comprises a single packet 334 which contains a total of 640 bits in a preferred embodiment. The longer packet length tends to reduce the loss in net data rate caused by overhead functions associated with each packet.
Referring to Fig. llB, the internal structure of the 640 bit packet of Fig. ~lA is shown. The first data ~roup is a four (4~ bit address 346 uniguely as-signed in a preferred embodiment to a loading dock with active traf~ic. In Fig. llB, the address of loading dock 302 (i.e. OOOQ~ would be within the first address field 346. After the loading dock address 346 wi~hin packet 334 is found a field 348 which correspQnds to the number of four bit groups or "nibbles" contained within the data portion of the packet which are to be xeceived by loading dock '302. The unigue addresses 346 of all loading docks with active traffic are suc-cessively followed by fields 348 denoting the number offour bit nibbles to be found within the actual data field ~described hereafter). All fields 348 contains eight (8) bits in a preferred embodiment ~ the modem 10 of ~he present invention (eight bits permit a maximum 3~ of 256 four bit nibbles or 1024 bits of data ~hipped per d w k; as a whole packet can only be 640 bits, an eight (8) bit field 348 is clearly sufficient)~ A ter-minator 350 denotes ~he end of the successively alter-~ating loading dock addxesses 3~6 and nibble fields 348. Terminator 350 contains four bits which consti~ute the addre~ of a non~existent loading dock thus inorming 3~,9 the receiving modem lO1 of the end of the stream of loading docks 28 having information to be shipped.
Following terminator 350 is a two bit modulo-4 ~eguence number 352 (i.e. 0, l, 2, or 3 ) . This is fol-lowed by another two bit field 354 reserved for the ~eguence number of the last correctly received packet.
In operation, each packet is assigned its own sequence or serial number 352. Seguence numbers 352 keep incre-menting by one or each packet shipped in a modulo-4 pattern. As field 354 contains the two bit ~equence number of the last correctly received packet~ the con-tents of field 354 serves as an acknowledgement. If field 354 contains a value ~ the ~e~uence num-ber of the last packet transmitted, transmitting modemlO must retransmit the last packet.
Following field 354 are a plurality of data fields 356. Each data field 356 corresponds to a par-ticular lo~ding dock addre~s 346 and contain~ exactly the number of nibbles of data within field 348. An impor~ant aspect o ~he present invention over prior art modems is tha~ separate headers are not required to 6eparate data fields 3S6. This is because modems lO
and lO' are perfectly ~ynchronized to tra~smit packets 2~ precisely at epoch boundaries, and each data field 356 i~ distinguishable from adjacent fields becau~e the header information of the packet (i.e. fields 346 and 348) informs the receiving modem lO' of exactly how many data bits ~o expect. A further important aspect of the modem of the present invention over the prior art is the ability to ~ultiplex a ~eries of data inputs ~i.e. from di~ferent l~ading docks 28) with a minimum of visible delay time for ea~h multiplexed channel.
This result is achieved because a separate packe~ is not required or each data source in the modem ~f the present invention and inormation may be shipped when l '` ~2(~73~ ~

available from ea~h loading dock rather than waiting for buffers to ~ill.
There are two further fields within a standard packet of the high speed modem 10 of the present inven-tion. Following the data fields 356 is a cyclic redundancy check 358(CRC). A 16 bit or 24 bit CRC is sufficient to insure adequate error-free performance o the modem 10. Lastly, a filler field 360 is appended to the CRC 358 in order that packet 334 have a total of 340 bits.
Referring to Fig. llC, a pi~torial represen-tation is shown for a hypothetical packet according to a preferred embodiment of the present invention. The first ield is a loading dock address field 346 of four

(4) bits. ~he nex~ ~ield is the eight (8) bit field 348 defining the number of nibbles of data shipped by the previously noted loading dock. In Fig. llC, fields 346 and 348 are repeated; in this ~xample, snly two loading ~ocks ~8 have active dat:a to be shipped. A
four (4) bit field 350 follows field 348 defining the termination of the last field 348. ~ield 350 is fol-lowed by seguence and acknowledc~ement fields 352 and 354 respectively (each ~wo [2~ bits). Next in order withi~ packet 334 are two data fields 356, the firs~
has five (5) nibbles of data of four (4) ~its each (i.e.
twenty [20] bits total from ~he first loading dock send-ing data) and the second has six (6) nibbles of data of four (4) bits ~ach (i.e. ~went~-four [24] bit~ total from the ~econd loading dock sending data). A twenty-four (24~ bit CRC 358 allows the data fields 356, and, lastly, a 540 bit field 360 fills out the 640 packet.
Note that i~ the arrange~ent of packet 334 as illustrated in Figs. llA, 11B, and llC, a packet is ~hipped every other epoch and that partially ~illed packet~ are ~hipped. This assures minimum pipe filling ) ~
~2~7~

delay times and removes the effects of any errors which would require retransmission had they occurred at the end of the usable data section of the packet. This arrangement ~implifies the use of high speed modem 10 when telephone circuit 12 is ~ubject to a number of defective freguency channels. When a frequency channel is rendered unusable by noise, or otherwise, a single requency carrier is dropped conse~uently reducing the net data rate ive (5) bits per epoch or ten (10) bits per packet.

Remote Dia~nostics It is one of the significant advantages of high speed modem 10 of the present inventlon to be able to completely characterize the telephone channel to a degree hitherto practicably i~pos~ible and to efecti~e-ly correct each of the multip:Licity of impairments so measured. This information i~3 generated within constel-lato~ 20, particularly by means of the phase ~ubtractor ~ignal signal 274 and the amplitude ~ubtractor signal 27~. This inormation may be readily acces~ed by means of the concept of the virtual loading docks 28 de~cribed hereinabove. Virtual loading dock 302 (LD0) has the capability t~ ~end packets to its remote corresponding loading dock '302 (LD0') in order to exchange perform-ance data of one modem 10 relative to another modem 10' .
Thus each modem 10 (or 10') of the present invention can e~change a complete 6et of measurements with another similar modem. For each of the freguencies used as a carriers, individual mea~urement~ exi~t defin-ing amplitude, phase delay, noi6e, and freguency offset.
Thi~ arrangement is both useful to exchange perormance information between ~ending and receiving modems and ~2~73~

with a remote ~hird modem. For this purpose, an inde-pendent diagnostic channel (not ~hown in the drawing figures) can be used. This independent diagnostic chan-S nel comprises a ~eparate telephone circuit and may usea conventional 300 bps frequency shift modem of the prior art such as an AT&T 103A. The AT&T 103A class of modem employs autoanswer and autodial wherein an incoming ~all from a remote diaqnostic center c~n be connected to read out the packets from the virtual loading dock shipping diagnostic data. Since a dial up connection exists during the period that diagnostic information is being exchanged, the same extra telephone circuit may be shared and also used as a voice communication circuit in order to aid in servicing the telephone line or modem 10 of ~he present invention. ~ote that the use of such an optional dia~nostic telephone circuit in no way is reguired to achieve the basic objective of full duplex operation at data rates in excess of 9600 bps over a two wire dial up telephone circuit; it is nonetheless an attractive advantage of the high ~peed modem 10 of the present invention.
While for the sake of clarity and ease of description, a specific embodiment of the present inven-tion has been described hereinabove, the scope of thepresent invention is intended to ~e measured by the claims as set forth hereafter. ~t is clear that those ~killed in the art to which the present invention applies will ~e able to practice the present invention wi~h variations in structure from the preferred embodiment including the use of one or ~ore micropxocessors to implement the logical processes described herein. ~ll ~uch equivalent variations within the scope of ~he fol-lowing claims are intended to be part of the present i~vention.

Claims (10)

What I Claim Is:

1. A modem for transmitting source information and for receiving destination information in packetized format by means of an ensemble of frequencies through a communication medium having a passband comprising:
a transmit modulator means adapted to impress signals containing said source information upon said communication medium and constructed from means for generating a plurality of first digital values directly representative of an ensemble of carriers whose frequencies are within said passband, means for generating second digital values from said first digital values directly representative of modulating said carriers with first unencoded ortho-gonal signals and encoding said carriers with said in-formation in the form of discrete amplitude and phase deviations, said deviations defining digital data in packets upon said selected ones of said carriers, means for accurately controlling the duration of said modulating of said carriers for discrete epoch periods, means for converting said second digital values into analog signals for impression of said analog signals upon said communication medium as modulated carriers at said frequencies of said ensemble of carriers;
a receive demodulator means adapted to extract said destination information from signals impressed upon said communication medium and constructed from means for generating a plurality of third digital values directly representative of analog signals received through said communication medium, said third digital values being developed by multiplying signals representative of said modulated carriers impressed upon said communication medium by a plurality of signals representative of second orthogonal signals of known amplitude and phase, means for integrating said third digital values precisely over time to determine frequency and phase of unencoded orthogonal signals substantially identical to said first orthogonal signals and generating therefrom said second orthogonal signals, means for comparing said third digital signals with fourth digital signals, said fourth digital signals representative of patterns of possible encoded data conveyed by said selected ones of said carriers to decode said destination information, and means for detecting packets of said data which have been corrupted in transmission; and means for measuring the transmission character-istics of said communication medium for each of said transmitted carriers constructed from means for transmitting selected ones of said carriers at a predetermined amplitude level and for a predetermined period, means for analyzing signal corruption caused by communication medium impairments to received carrier carrying said destination information by comparing re-ceived data patterns with known good data patterns, and means for avoiding use of at least a selected one of said carriers for transmission of data in the event said analyzing means indicates said selected one of said carriers is subject to impairments to said com-munication medium affecting transmission of data.

2. The packetized ensemble modem of claim 1, wherein said communication medium is a bandlimited tele-phone line subject to amplitude and phase distortion.

3. The packetized ensemble modem of claim 1, wherein said first unencoded orthogonal signals are a pair of sinusoids that are 90 degrees out of phase.

4. The packetized ensemble modem of claim 1, further comprising:
means for correcting amplitude and phase dis-tortion for each of the received carriers with respect to a reference carrier wherein said reference carrier is one of the plurality of discrete transmitted carriers.

5. The packetized ensemble modem of claim 1 wherein said carriers are divided into a first group of carriers forming a high speed channel and a second group of carriers forming a low speed reverse channel.

6. The packetized ensemble modem of claim 1 wherein a plurality of input data signals from a plur-ality of data input sources are multiplexed over a plur-ality of epoch periods to form error-protected individual units of source information.

7. The packetized ensemble modem of claim 1, wherein said first group of carriers forming a high speed channel comprises sixty-four (64) carriers, said second group of carriers forming a low speed reverse channel comprises nine (g) carriers, and said error-protected data packets each con-sists of five (5) bits of encoded data per epoch period on each carrier.

8. In a modem utilized for transmitting source digital information and for receiving destination digital information over a communication medium utilizing a plurality of carriers located within the passband of the communication medium, the improvement comprising:
means for measuring the transmission character-istics of said communication medium constructed from means for transmitting said source digital information on selected ones of said carriers at a pre-determined amplitude level and for a predetermined period, means for analyzing signal corruption caused by communication medium impairments to received carriers carrying said destination digital information by com-paring received data patterns with known good data pat-terns, and means for avoiding use of at least a selected one of said carriers for transmission of said source digital information in the event said analyzing means indicates said one of said carriers is subject to impair-ments to said communication medium affecting transmission of data.

9. The packetized ensemble modem of claim 1, further comprising:
means for measuring the phase, amplitude, and noise characteristics of the transmitted carriers cor-rupted by a first communication medium as received by said receive demodulator, means for statistically averaging said mea-sured phase, amplitude, and noise characteristics for each of said transmitted carriers over a plurality of epoch periods, and means for packetizing information constituting said statistically averaged phase, amplitude, and noise characteristics for each of said transmitted carriers and transmitting said packetized information to a remote site over a second communication medium for diagnostic purposes.

10. The packetized ensemble modem of claim 1, wherein said first communication medium is utilized for the transmission of error-protected data packets, and said second communication medium is utilized for the transmission of said statistically averaged diagnostic information and voice traffic.