On Distributed Communications Series

VIII. The Multiplexing Station

V. Detailed System Configuration

We shall now examine the detail of the proposed Multiplexing Station and explain the mechanisms used to perform the operations described in the previous sections. While the number of subscribers that can be handled by any single Multiplexing Station is not fixed, it is basically a function of cost. In this section the discussion is limited to a single Multiplexing Station having 1024 separate input and output subscriber lines. It is further assumed that this "central office" has been designed to possess a much higher peak-occupancy safety factor than used in conventional telephone systems. In civilian practice--and all too often in the military--the assumption is made that only a very small proportion of the users will ever simultaneously demand service. However, in the military network, we must be more conservative--otherwise we invite overload during critical times. Therefore, from the standpoint of traditional practice, this system will appear to be grossly over-conservative in the value of peak-service capability.

Figure 6 is a block flow diagram of the major units comprising the Multiplexing Station. Each block is designated by letters for ease of identification.

Table 1 lists these major blocks and the approximate number of "equivalent flip-flops" necessary to perform the desired function. We may roughly assume about four transistors being required per equivalent flip-flop. The values shown in parentheses are for those units for which an inordinate amount of design would be required to make an accurate determination: the estimates shown are felt to be adequate approximations, inasmuch as we are interested only in gross orders of magnitude--not in specific details. The column labeled Heads indicates the number of read/write heads utilized on the common drum by each of the examined units.

Gross Description:

Unit A, FROM Subscriber TO Drum Unit

Unit A includes the input terminating equipment from 1024 separate subscriber lines. Strobed timing signals sample each of the incoming 1024 wave trains at a point near the center of each pulse interval time. The binary state ("0" or "1") for each subscriber's line is stored in a separate flip-flop. Each such flip-flop is unloaded in synchronism via a buffer flip-flop onto a common drum store controlled by a timing track on the drum. The output of Unit A feeds Unit B.

Unit B, FROM Subscriber VIA Drum Unit

Unit B examines the entire Message Block that has been stored on the drum during previous revolutions. Unit B contains filtering means which allow only active Message Blocks to be transmitted. Active Message Blocks are defined as those Message Blocks containing data or signaling--and not all zeros. The output of Unit B feeds Unit C.

Unit C, FROM Subscriber Crypto Unit

Unit C is a cryptographic subassembly unit having access to stored cryptographic keys on the common drum. This unit applies those cryptographic transformations necessary for processing active fully-formed Message Blocks for transmission into the system. Only when complete and valid Message Blocks are ready for transmission is the crypto storage in Unit C incremented to the next portion of key to be used for each line.

Unit D, Outgoing Line to Buffer Unit

Unit D provides the speed transformation means for the Message Blocks from Unit C. Message Blocks are held until Unit J is ready to accept Message Blocks for further outgoing processing.

Unit J, Central Processor

Unit J is the heart of the Multiplexing Station and contains a high-speed random access memory. This memory is organized in the fashion of a small general-purpose computer for performing complex special operations, as well as for providing the timing flexibility needed between the high-speed synchronous timing of the network block, via the Switching Node links, and the local drum timing. This allows flexibility of the timing necessary for high-speed burst transmission. Units J and D work in conjunction with one another when transmitting the outgoing Message Blocks via Unit F.

Unit F, Output Link Unit

Unit F performs the parallel-to-series buffering into the Input Link Unit, H.

Unit G, Alternate Link Selector and Failure Mode Multiplexer Unit

Unit G is a switch to establish connection to alternate Switching Nodes in the event of a link or Switching Node failure. Unit G also connects the input from the Switching Node or nodes into Unit H.

Unit H, Input Link Unit

Unit H provides the link-by-link encryption devices used by the adjacent Switching Node and provides the link error detection encoding function. Unit H is almost identical in operation to a similar unit contained in the Switching Node and is described in ODC-VII.

Unit I, Translator (Address Table Look-Up) Unit

Unit I is used only when setting up new calls. Unit I converts the incoming called number into the latest correct address to be used for future traffic (Message Blocks). Thus, all Message Blocks are directly addressed to whichever line is presently assigned to the called party. Unit I works in conjunction with, and is controlled by, Unit J which provides control signals to allow the peripheral units of the system to operate in a semi-autonomous fashion. Such semi-autonomous peripheral units include Unit K.

Unit K, Trace and Trouble Display Unit

Unit K contains a 12-digit numeric printout device. This unit prints a tape similar to that used in a conventional electric adding machine, except that the tape material is translucent and preprinted. When an obviously incorrect Message Block or a Message Block "in trouble" (too old) is detected, it is transmitted via Unit J to Unit K. Here the trace portion of the Message Block, together with the to/from addresses, is printed on the adding machine tape. Use of translucent tape allows stacking individual defective trace patterns over one another so that a common fault-pattern can be easily detected, permitting the maintenance man to quickly isolate the source of trouble to within a particular Switching Node.

Unit L, TO Subscriber VIA Drum Unit

If the incoming Message Block is correctly received (passes all tests) it is transmitted via Unit J to Unit E. Units E and L are similar to Units B and D, their output counterparts. The output of Unit I is used by Unit N.

Unit N, TO Subscriber FROM Drum Unit

Unit N provides outputs to the 1024 subscribers at their exact synchronous data rates, with all housekeeping data stripped off.

Unit O, Common TO/FROM Subscriber Controls Unit

Unit 0 maintains an examination of the status of the subscribers and checks for proper operation of the line units. The use or non-use of each line is noted and is passed along to Unit P.

Unit P, Priority Control Console

Unit P performs the priority control functions useful in preventing overloads.[1] Unit P contains information as to which lines are actually in use transmitting traffic, and permits insertion of a variable control over the allocation of the communication resource. Unit P normally would be mounted adjacent to Unit Q.

Unit Q, Intercept Position and Change Translator Table Input Unit

Unit Q is a manual operator intercept position and is the point where manual changes are entered onto the Translator Table, whenever it is necessary to change the address of a subscriber line.

Unit R, High-Speed Drop Point Input/Output Unit

Unit R is a buffer unit designed to process a very-high-speed local subscriber party line directly into the central processor, bypassing the common magnetic drum. Unit R provides the buffering equipment to permit several party line users to connect into the Multiplexing Station. While such users would each operate at very-high-speed output data rates, their expected low duty cycles (operating in a highly intermittent fashion) would seem to permit satisfactory party line operation.

Unit S, Common Timing Sampling and Individual Input Phase Selector Control Unit

Unit S provides the means to test all incoming subscriber lines against a four-phase clock to permit selection of that clock which falls closest to the center of the incoming pulse train, insuring that the incoming lines are strobed at or near the center of their individual input pulse train waveforms regardless of the total amount of phase shift on the lines to remote devices. Unit S measures and compensates for variable phase delay between the transmitted waveform to subscribers and the waveform based upon reconstructed timing from the subscriber back into Unit A of the Multiplexing Station.

Unit T, Power Supply Unit

Unit T provides all proper regulated voltages for operating the individual transistorized units of the Multiplexing Station.

Unit U, Emergency Power Source Unit

Unit U is a standby power source and provides primary power in the event of power failure. The total amount of power required for transistorized equipment is invariably low; probably more power is used by the magnetic drum than all the remaining electronic equipment. Thus, a few kilo-watts of emergency power will suffice. Batteries or perhaps a fuel cell (if economically feasible) would also provide a suitable backup power source. Even a simple gasoline engine and alternator of the type selling for a few hundred dollars would do as a double backup.

Unit V, Emergency Partial Power Supply Unit

In the event of a failure of the primary power supply, Unit V provides power to the most critical units and lines of the Multiplexing Station. During such a failure, it may be possible to interlace-multiplex important subscribers signals using neither the drum nor most of the other units comprising the system. Such a simple time-division multiplexed signal could be transmitted via the Link Units to an adjacent Multiplexing Station for necessary processing. Such a fallback procedure would only be used as a last-ditch emergency measure. The chief objection to this procedure is that only a small portion of the subscribers can be handled and even these would temporarily lose normal end-to-end crypto facilities. However, since link-by-link encryption would still be available on the signals for subscribers feeding to the remote Multiplexing Station for processing, this may be sufficient to safely permit highly classified conversations in emergencies.

Unit W, Common Drum Unit

A storage of nearly 13 million bits is required in the system, most of it provided by Unit W, a single large drum having four hundred heads. Timing tracks of the common drum will establish the timing base for all locally connected subscribers. Although a drum is tentatively suggested as the heart of Unit W, the computer state-of-the art has advanced sufficiently so that the alternative use of high-capacity acoustical delay lines would eliminate the dependence upon a single, large, rotating component. (A good part of the power required by the entire system is consumed by the drum motor and many circuit elements are necessary to compensate for the pulse jitter caused by the drum.) Splitting the drum into a series of smaller Bernoulli discs would simplify fracturing the system into many identical units, a more desirable alternative for continuity of service in spite of the increase in the probability of individual subunit failures.

Unit X, Zero and Signaling Detector

Unit X, in conjunction with Unit A, performs the high-speed scanning operation to examine all input subscriber lines and ignore those that had nothing to say (all zeros) during the last Message Block frame time (about 1/20 sec in the case of 19.2-kilobit/sec digital voice). Unit X also performs the operation of spotting and separating the signaling information from active data transmitted.

Detailed Description

We have briefly reviewed the key operations performed within each of the major blocks of the system, and shall now examine the operations and constructions of each unit in further detail. While a full description of each unit is desirable, such a writeup would become voluminous. Therefore, separate block diagrams are provided for only the least conventional of the major units and those most difficult to visualize. The reader is left to his own devices to fill in the unspecified minutiae.

Operation of FROM Subscriber TO Drum Unit, A

At the extreme left of this drum are the 1024 separate lines each feeding its respective line input unit. Each bank of 32 line units is time-shared by a separate 32-input switch concentrating the sampled waveform onto a single output channel. All 1024 separate line input units have been previously sampled in Unit S by a common time-sampling circuit. Unit S examines each line every few seconds, searching for a slow drift of the arriving wave-form from the subscriber. This long interval between samples is felt to be permissible because of the expected slow-drift-rate nature of the phase shift variation in the input lines from the subscribers. By using Unit S, we may choose one out of the four timing strobe phases per pulse. The specific phase chosen is that which falls closest to the center of the incoming pulse train. A temporary storage flip-flop in each line input unit permits an offset of the strobe point necessary for rephasing to match the timing constraint of the drum. Each bank of 32 lines concentrated by the 32-input switches is connected to a single drum amplifier. The output of this amplifier drives write heads on the drum connected in series to a small bank of magnetic cores. These cores are used to indicate whether signaling or data is being transmitted (by means which shall be described later). The output of each drum amplifier is distributed by a three-way switch to permit sequentially writing on three drum heads. This permits inputs to be written on one bank of heads while the second bank is being read out. Unit A thus provides the multiplexing equipment to concentrate 1024 separate units onto 96 different drum bands.

Composition of Inputs

Table II shows an assumed composition of 1024 inputs used in designing the described Multiplexing Station. This particular system can handle, for example, 128 lines, each operating at 19.2 kilobits/sec, which might be digital telephones; plus, 128 lines at 9600 bits/sec, which could be high-speed facsimile; plus, 256 lines of 2400-bit/sec circuits, which might be vocoder type digital voice or digital data; and lastly, the system will also simultaneously handle 512 lines, each operating at 600 bits/sec, which could be used for high-speed synchronous data, and/or output of conventional teletype equipments. Table II also indicates that if the drum size is chosen so as to complete a single revolution to store 1024 bits, comprising a Message Block of which 866 bits are active information, at 19,200 bits/sec, then two drum revolutions will be required for the 9600-bit/sec channels and 32 revolutions for the lowspeed 600-bit/sec data. The last column of Table II shows that the number of drum amplifiers needed can be reduced by using a single drum amplifier together with an additional layer of switching to further time-share-concentrate the lower-bit-rate channels. Thus, instead of 32 drum amplifiers being required in the arrangement of Fig. 7, only eight drum amplifiers need be used by this additional time sharing. (The component count of Unit A was based on 32 drum-amplifier channels, while many of the subsequently considered units have been limited to the assumption of only ten separate channels.)

It is to be noted that under the assumptions of Table II, the peak input data rate is in excess of 4.5 million bits, about three times greater than the allowable peak output rate over a single 1.5-megabit/sec link.

The ratio between peak input- and output-limited rates is extremely conservative, as the probability of exceeding this output value, even when the bulk of the lines are in use, is statistically a rare event. Even in the event a peak load is experienced in which the number of input devices have an accumulative output data rate in excess of the allowable output rate, overloads will be handled by automatic priority assignment equipment built into the system (Unit P).

Input Strobing

Figure 8 details what happens to the subscriber's input waveform arriving at the individual line units. P1 through P4 are possible input strobing positions. The incoming time wave, regardless of its relative phase to the local timing system, is sampled near the center of each pulse position and a new wave is created at the input rate in synchronism with the drum timing.

Figure 9 shows the line input unit in more detail; its operations are self-evident.

Figure 10 shows a sketch of a 32-input/l-output switch, in order to provide an approximation of the number of components that will be required to implement gates of this type.

Figure 11 is a brief sketch of the l-input/32-output switch, also providing a rough estimate of the number of parts needed.

The Crypto Scheme

A possible end-to-end cryptographic secrecy system that might be used in the system is shown in Fig. 12,[2] illustrating the principle of the proposed crypto scheme. Essentially a newly modified key base is created for each new call. (The actual method of developing this key base when setting up each new call is fully described in ODC-IX).

The system is primed each day or so with a set of start keys assigned to pairs of stations. These are modified after each call. The key base sequence used for each call is "logically-added" to the incoming and outgoing text. The output of this "logical-add" circuit is the encrypted transmitted text. The text is also combined with the last section of the key streams and used as part of the generator function for creating the next portion of the key stream. Thus, the modified key is a function both of the initial key base and of the text being transmitted. This, in essence, is a combination of the auto-key principle and the pseudo-random stream approach.[3]

The receiving decryption equipment is identical with that used in transmission. In order to obtain error-free decryption, we have purposely made it mandatory that no errors appear in the series of Message Blocks received by the receiving Multiplexing Station. It will be remembered that one of the most important fundamental properties of this proposed network is its extremely low error rate. Thus, this cryptographic procedure appears to be usable in the proposed system while not appropriate to conventional communications systems.

Figure 13 exhibits the receiving decryption apparatus in more detail. The round ring bands at the right of the illustration are intended to represent drum bands on the common magnetic drum of the Multiplexing Station. The third drum band would contain the incoming encrypted text, timed out to pass by in synchronism with the clear text of the last Message Block received. The box within the dotted lines represents the circuit which modifies the next key block. Six separate drum bands are required for these cryptographic operations because of our desire never to read and to write on the same drum band at the same time. (This would have required accurately spaced read and write heads.) Thus, we write on three heads while simultaneously reading three heads. At the end of a drum band revolution, we switch and read the three previously written drum bands, etc.

Operation of FROM Subscriber VIA Drum Unit, B

Figure 14 shows in more detail the operation of the From Subscriber Via Drum Unit. The output from the 96 read-write heads from the subscriber inputs are selected by switches; read is by drum-read amplifiers and logical addition is performed by the crypto key. Here, the signals from various subscribers' lines are read off the common drum in synchronism with the crypto key to form outgoing encrypted text.

Operation of FROM Subscriber Crypto Unit, C

Thirty-two read-write heads shown in the read position at the extreme left of Fig. 15 feed ten head switches. The outputs of these switches are connected to drum-read amplifiers. These circuits provide access to previous key information stored in the drum. This allows the incoming text to be combined in a key modifier circuit within Unit C, the output of which is sent to drum-write amplifiers which, in turn, lay down the newly modified key on the drum slot reserved for the next Message Block key. Only ten separate drum-read amplifiers and modification circuits are necessary, as has been mentioned, as the drum revolves many times to process the low-data-rate channels, permitting time-sharing of the read amplifiers to cover many separate drum tracks.

Operation of Buffer Units, D, E; and Input/Output Link Units, F, H

Figures 16, 17, and 18 are essentially the identical circuits of the Buffer Units and the Input and Output Link Units contained within the Switching Nodes. As these units have already been described in detail in ODC-VII, no more will be said about them here.

Operation of Alternate Link Selector and Failure Mode Multiplexer Unit, G

The Alternate Link Selector and Failure Mode Multiplexer is an emergency-use-only unit providing some residual capabilities in the event of failure of the entire Multiplexing Station or of the links to the Switching Node from the Multiplexing Station. Unit G provides an emergency residual capacity, but only at a price of a loss of traffic handling and security capability. In Fig. 19, the output of the Link Unit is connected to a mechanical three-position switch connecting to alternate Switching Nodes. In the event of the failure of one of the outgoing links, a new manual connection can be selected. The output stream from the input link unit, which we have assumed to operate at a data rate of 1.5 megabits/sec, is divided by two to produce a 720-kilobit/sec stream. This continuous pulse pattern is used to create new timing in lieu of the timing derived from the drum (in the event of a failure in a drum or similar major component).

Since four samples are required per pulse received from a subscriber at the line unit, the number of channels that may be handled by such a procedure is greatly reduced. For example, instead of handling 1024 channels, capacity is reduced to about 172 channels, of which 128 lines may operate at 600 bits/sec, 32 at 2400 bits/sec, 8 at 9600 bits/sec, and 4 at 19.2 kilobits/sec.

Table III shows the series of standard 75(2n)-bit/sec digital communications speeds.

Fig. 19--Alternate Link Selector and Failure Mode Multiplexer Unit

In retrospect, it appears that we would be better off to standardize data rates for the high-speed Switching Node links at integral speeds, rather than the tentative 1.54-megabit/sec speed mentioned in other volumes in this series. The nearest two such speeds would be 1.2288 megabits/sec and 2.4576 megabits/sec. If we should so standardize, again in retrospect, we could probably simplify the equipment within the Switching Node by limiting the choice of data rates to these mentioned here. Such a standardization would have simplified the design of the emergency Alternate Link Selector and Failure Mode Multiplexer Unit. Better yet, it is appropriate to reconsider the entire system design and the potential savings if the system were to operate only at these fixed speeds, rather than allow for the infinite range of speeds that has been suggested in the previous volumes in this series.

Operation of Input Link Unit, H

As Unit H is identical with that used in the Switching Node, it will not be described in detail here other than to include the circuit of Fig. 17 (p. 54).

Operation of Translator (Address Table Look-Up) Unit, I

Figure 20 is a schematic view of the Translator Table mechanism. Two tracks of the common drum unit are reserved to store alternate "telephone" numbers for all network users. This unit is fundamentally a simple table look-up device. When setting up a new call, this table is examined to convert the called number into the actual office number and line number currently being used by the called number. (The system also has facilities for inserting alternate numbers if the called person is not at his designated extension.)

Housekeeping data is also stored to: indicate when the alternate number may be used; indicate whether the alternate number is a local or a remote station; indicate whether the actual number is classified; and determine the level of urgency of an incoming call before transferring it to an alternate number.

Information updating the Translator Table is stored in a temporary flip-flop register and read out into its proper place on the drum by the equipment shown in Fig. 20.

Operations Performed in the Process of Setting up a Call

Figure 21 shows some major operations performed when setting up a new call between two network users. The sequence of operations starts from the calling subscriber, passes through the calling terminal station to the called terminal station, and eventually reaches the called subscriber. These operations are self-explanatory.

Creating Signaling Tones by the Use of a Repeating Digital Sequence

Inasmuch as the system is all-digital, it is desirable to have simple devices to convert digital streams into signals which can be expeditiously analyzed by inexpensive end-terminal devices (such as telephones). Below is a suggested procedure.

All signals are in the form of a continuous repetitive digital waveform, selected so that a Fourier analysis of the tone will result in a small set of fixed frequencies of known amplitude relationships. For example, a 2400-bit/sec square wave tone passed through a band pass filter will produce a strong fundamental 'tone at 2400 cycles, with tones one-third as strong at 3 x 2400 and one-fifth as strong at 5 x 2400 cycles. A series of five digital patterns are shown in Fig. 22, each of which can be transmitted on an all-digital line and be readily distinguished and interpreted by a simple band pass filter bank. Simple analog detectors, therefore, are capable of analyzing a digital signaling signal into signals suitable for operating switches to insert ringing signals, busy signals, etc., at the end-device.

High-Speed-Core Housekeeping Operations Upon Message Blocks Newly Arriving at the Terminal Line Distributor

Figure 23 is a simplified flow chart of operations performed within the high-speed core, such as unpacking newly arriving Message Blocks. The details of the operation can be easily gleaned from the flow chart.

Operations of Trace and Trouble Display Unit, K

Each Message Block reserves sufficient housekeeping space to record the last four links it has traversed in its travels. Since there are a maximum of eight links emerging from each Switching Node, only three bits per link, or a total of 12 bits, provide sufficient in-message storage to unambiguously describe the last four links taken by each Message Block arriving at the Multiplexing Station.

The output of the trace signature can be graphically displayed by using a simple 10-column numeral-only printer, such as is found in an inexpensive adding machine; such output is illustrated in Fig. 24. In the event of trouble, an output strip is printed containing the housekeeping information relative to the last Message Block recorded as being in trouble (or in the event of a broken sequence, the last Message Block correctly received). Memory capacity for storing the last Message Block already exists by virtue of the non-destructive nature of the drum memory. Possible causes of trouble might include: maximum handover number exceeded; a reversed leader or broken crypto chain; or no such number existing. As each trouble message will be printed out on a small piece of translucent paper, it is possible to stack them in a small pile against a light source to show the paths used by in-trouble Message Blocks and by correctly received blocks. Interpretation and analysis is left to human intervention. Figure 25 shows a symmetrical network of redundancy level three, illustrating how a path through the network can be described by a simple series of three-bit numbers.

Figure 26 shows the path coverage for a system limited to a four-link trace coverage. The X's indicate the appearance of broken links in an examination of a stack of translucent vellums; the broken links stand out since they are not being used. The area outside the dotted line is that area in which the trace method would not be effective for the particular Multiplexing Station shown in the center of the illustration (large dot).

Figure 27 shows the case of two defective Switching Nodes. This conclusion can be readily determined from the pattern since there are no visible links emerging from either of the two known Nodes.

Figure 28 shows the coverage of several overlapping trace zones, each having a span of four.

The trace scheme is intended only as a last-resort measure when all other system trouble safeguards fail. Error messages are expected to be few and far between. However, some method, such as the trace, is required to tie down stations causing intermittent network interference. The innate perversity of inanimate objects requires a safeguard of this sort, in spite of our lack of expectation of the ailment for which the remedy is proposed.

Operation of TO Subscriber FROM Drum Unit, N

Unit N distributes the decrypted Message Blocks from the drum into a bank of 1024 separate output flip-flops connected to each of the 1024 subscribers. Here, as in the input, a multi-input gate can be used to time-share ten separate drum amplifier. channels. Switches assign each of the drum bands to its proper output line channel. The output data stream at the output of this unit is a synchronous binary stream to each of the 1024 separate subscribers (see Fig. 29).

Operation of the Priority Control Console, Unit P

Figure 30 shows the manual operator position at the Multiplexing Station, the priority control console, whose operation in analog circuit terms is described in DDC-IV.

We are not suggesting the construction of the priority control console in the analog-model manner described. However, all the digital signals needed to indicate the number of devices actively transmitting bits at any one instant which are necessary to the proposed control scheme, are available at the Multiplexing Station with Unit X, the Zero and Signaling Detector Unit.

This priority console controls the allocation of the communications resource among the 1024 users--depending upon their importance and data loading demands. Also seen in Fig. 30, at the operator position, are two telephones used as order wire terminations. The only human operator function that is absolutely necessary in this system is the insertion of information regarding change of address of subscribers and keys; these operations are described in detail in the next section.

Operation of Intercept Position and Change Translator Table Input Unit, Q

Each subscriber has a card upon which his latest line and station number is recorded. This is shown on the top of Fig. 30. Figure 31 shows the subscribers' cards at the operator's position in more detail: the top of each card contains the line code number, below which is the name of the subscriber. The following safeguards are included to prevent imposters from calling in and, posing as a legitimate subscriber, reporting that they are now operating at a different location. To circumvent this potential system weakness is a procedure whereby each subscriber is given a plastic card containing a series of verification numbers. Each verification number is covered with a separate gummed tab. A similar list of verification numbers is on the subscriber's card at the operator's position, and are similarly covered by removable tabs. Thus, if a subscriber moves to a different city, he calls his original Multiplexing Station and requests that the Station change his alternative number to correspond to that of his new nearest Multiplexing Station. The operator at the original Multiplexing Station will then peel off the tab and ask the subscriber for his verification number. If these two numbers agree, then the operator at the original Multiplexing Station is assured that the change is bona fide. Each subscriber is expected to guard his own card, but, in any case, accidental examination or perusal of the card will not reveal the next authentication number unless the covering tab has been removed. This makes it exceedingly difficult for an enemy agent to determine verification numbers which could be used at a later date.

This method for changing individual subscribers' addresses is to be distinguished from the method used to insert the changes of tie-in points for entire Multiplexing Stations which might, for example, comprise an entire large military unit. Changes of the tie-in point of Multiplexing Stations is entirely automatic using the self adaptive means described in ODC-I.

Thus, it might not be necessary to physically man the Multiplexing Station a full 24 hours per day. Perhaps a few hours per day will suffice to allow changing the cryptographic key and making line assignments and priority allocation adjustments.

Operation of Common Drum Unit, W

The drum is common to many of the units of the system. A conventional drum, about 18" in diameter, 21" in length, rotating at about 2200 rpm, is proposed. As it does not appear necessary to include any short registers on the drum (i.e., writing and reading on the same track simultaneously) only one head per track is required and interhead gap spacing problems are avoided. The only thing unusual about this drum is that it is necessary to synchronize the speed of the drum to a crystal clock, due to the synchronous nature of the input/output devices used. Complete synchronization down to an actual bit or fraction of a bit is not necessary, provided that long-term drift of the drum speed is prevented by a locked-in servo. If the servo-system can keep the drum angular position to within about

Processing Drum Calculation

  1. Reserve about 158 bits for header and housekeeping functions and allow 32 Message Blocks or channels to be recorded per drum band.
  2. The drum will be designed to be used with input data rates in the 75(2n) series with minimum intermediate buffering.
  3. The same drum will be used for all channels, both input and output.
  4. The raw Message Block length is 1024 bits, the active portion, therefore, being (1024-158) = 866 bits.
  5. The drum will contain a synchronous timing track(s) prerecorded on the drum containing 32 channels, times 1024 bits, plus 1 bit, equaling 32,769 bits.
  6. Synchronism of drum rotation will be maintained by a servo-system to within
  7. Bit-packing spacing shall be about 600 bits/linear inch on the drum or disc surface (or whatever values fall within conservative practice for future 18" diameter drums or discs).
  8. The drum shall contain about 400 read/write heads, fixed position.
  9. Clock rate shall be 307.2 kc.
  10. Each track shall be about 0.03 in. wide and each track shall have an approximate 0.05-in. centerto-center track spacing (implying a +20" length drum).
  11. No short registers are needed; therefore, precise head alignment is not needed.
  12. The drum need only operate in a room-temperature environment.
  13. The drum shall reach full speed from a cold start within 10 mm.
  14. Modest vibration requirements only need be considered.
  15. Life of the drum bearings and motor shall be in excess of five years.
  16. The read/write heads shall be designed to operate in conjunction with head-switching circuits.
  17. The drum shall normally be operated 24 hours per day, 7 days per week, and shall only be turned off during power or system failures and perhaps at maintenance periods.
  18. Multiple discs or acoustic delay lines may be used in lieu of a magnetic drum if all other specifications can be met.

Operation of High-Speed Drop Point Input/Output Unit, R

Over and above the 1024 separate input/output lines handling synchronous continuous data streams of up to 19.2 kilobits/sec, the Multiplexing Station network can also be designed to handle, on a time-shared basis, those users who make intermittent, but very-high-data-rate demands upon the communications resource (on the order of 1.5 million bits/sec). An example of such an operation might be a computer dumping the contents of its memory into a remote companion unit. In order to reduce the number of high-data-rate feeder lines, a party-line configuration is suggested. Figure 32 shows a "party line," operating at a high data rate, connected between two Multiplexing Stations which are in turn tied into two separate Switching Nodes. While only one of the Multiplexing Stations need be operative at any one time, two provide some measure of backup in the event of a line break. These very-high-data-rate channels can be readily multiplexed by a simple time-sharing of the high-data-rate links between Switching Nodes. For example, every third bit would be reserved for this function. Of course, the data rate of such links would have to be raised from about 1.5 to 4.5 million bits/sec, but the marginal cost appears low.

A more awkward problem is that of the propagation velocity of the very-high-data-rate links. The problem is that each tandemly connected station cannot be sure that at the same instant he activates the line, it will not also become occupied by someone at the far end because of propagation time delay). Whether this will be a major problem or not is unclear at this time. If it is, the difficulty may be circumvented by the expedient of providing one Message Block of storage delay for all traffic passing through each potential local drop point on the way to the end Multiplexing Station. If we examine the input and output streams into and out of the delay, we determine whether we would interrupt an on-going stream if we were to transmit at that instant. Thus, a drop-point station need never worry about "clobbering" traffic in process. Each station on-line would, of course, also utilize error detection and correction equipment in order to permit retransmission of Message Blocks in the event of inevitable link errors. Digital computers designed for on-line operation with an interrupt channel would appear to match the requirement of repeat transmissions with relatively little difficulty.

Calculation of the Number of Drum Bands for Translator Table

Per Multiplexing Station:

L = number of lines = 1024

D = number of octal digits/address = 6

A = number of alternative addresses/line = 1

B = number of bits/digit = 3

M = number of marker bits/line = 6

N = number of bits/drum band = (1024) (32)

R = number of drum bands.

Thus,

Read/Write Head Switching Detail

Figure 33 is a schematic sketch of the detail whereby an entire bank of 32 heads are simultaneously switched from the read to the write mode. Figure 33 is included to obtain a feeling for the complexity of the circuitry involved in preparing the parts breakdown.

Operation of Zero and Signaling Detector, Unit X

Figure 34 is a simplified view of the Zero and Signaling Detector Unit. The heart of this unit comprises a bank of eight core planes, each containing 32 x 32 bits. The bi-stable cores used are initially set to their "0" state. If a single bit other than "0" is found in any bit position of any input line, the core corresponding to that particular line is set to "1". At the end of the duration of each Message Block interval, each core is read out. This provides a rapid determination as to whether nothing, valid information, or signaling has occurred during the last interval. If all zeros were noted on a line, no outgoing Message Blocks were created and no further processing is indicated for this line. Thus, the first function performed is that of a fast filter to focus attention upon those channels that may be transmitting signaling or valid information.

The first procedure only filters out those lines transmitting all zeros. Detection of signaling information is somewhat more complicated. In this system, signaling always takes a pattern in which each bit is exactly similar to the polarity of every eighth preceding bit. In order to recognize the entire category of such signals easily, we separately connect each of the eight core planes to its assigned input line. As there are 866 bits of active information per Message Block, 112 input bits would be examined per core per line. If there had been even a single "1" found in the 112 bits, then this core would be set to "1". The entire core is now read out at the end of each Message Block time. If some of the bits are noted to be zeros and others ones, the probability is high that signaling information is actually being transmitted. The chance of an error in which signaling is mistaken for data is less than 2-112 or about one chance in 1030. Of course, there is a high probability that the signaling will start in the middle of the Message Block. But even with 19.2-kilobit/sec channels, the push buttons used for signaling would stay depressed for a longer period than the Message Block time--hence, there appears to be no problem from this cause. The final extraction of valid signaling dataa relatively rare event--requires post-filtering operations performed by the Central Processor, Unit J.

Operation of the Modify Key Box, Unit Y

It is necessary to change the cryptographic key of the Multiplexing Station at periodic intervals. The equipment necessary to perform this function is provided by Unit Y. In establishing the parameters of the system the following assumptions are made:

  1. About 2000 bits appears sufficient per send-key base and receive-key base between any two Multiplexing Stations.
  2. Allow capacity to handle about 1024 different addressable Multiplexing Stations.
  3. The key shall be contained on a magnetic tape which has been recorded at 2400 bits/sec.
  4. A reasonable tape speed is about 7.5 in./sec for this bit rate.
  5. Time to load tape into Multiplexing Station,

  6. Amount of tape required

  7. A small reel containing 600 ft of tape is one possibility. However, it would be preferable to split the key into two pieces, generated at two different sources: two reels, each holding 300 ft, is thus suggested.
  8. A 300-ft roll of 0.5 mill mylar tape occupies a standard three-inch tape reel.
  9. After a short synchronization period, the tape drive motor servo-system must lock into within a peak drift of about
  10. Assume that the capstan motor of the tape playback unit operates at about 1800 rpm at 60 cps.
  11. 7.5 inches of tape pass by the heads each second, equivalent to 0.00313 inches of tape passing by in one bit-time (7.5/12400).
  12. Maximum drift equals 0.00313 in. x 0.25 bit, 0.00075", implying a tight synchronization loop.
  13. Assume that a simple synchronization system containing a prerecorded timing track can maintain tape speed to within
  14. The maximum differential synchronous motor lag angle at 60 cps, before synchronization is broken, is

  15. The simple servo-system configuration shown in Fig. 35 should suffice. This servo-system utilizes the heterodyned signal between the system's 4800-bit/sec rate and a separate 4740 cycle time signal recording on a two-track magnetic tape, to produce a 60-cycle motor control signal. Speed differences are amplified by a factor of about 80, times the actual frequency difference.

Operation of Remote Devices at Subscriber Stations, Unit Z

While not truly an integral unit of the system, this unit provides the method whereby voice or other analog signals are converted into a digital stream. Figure 36 shows a typical telephone proposed for use with this system. From the outside, it looks very much like a conventional touchtone dialing telephone, but there are some major differences; for instance, the output is a four-wire 19,200-bit/sec binary stream. It will be recalled that all input devices feeding the system derive their timing from an incoming binary stream originating from the magnetic drum of the Multiplexing Station. Both signaling and data are a binary bit stream synchronized to this timing pattern.

The signaling waveforms are chosen so that they are easily detected by the Multiplexing Station. One of the key ways of possibly generating the binary signaling stream at the telephone includes the use of a small counter with changeable feed-back connections. Only four flip-flop stages are required to generate any of the desired digits.

We have seen how a binary stream used for signaling can be easily generated with a few transistors. Next, let us consider the mechanism that turns an audio signal into a digital stream, and vice versa. Figure 37 shows the block diagram of the proposed telephone containing a miniaturized analog/digital converter together with the push-button signaling previously described. Starting at the microphone in the extreme right of the illustration, the audio signal from the microphone is amplified and its dynamic range compressed. This audio signal output is applied to a simple analog/digital converter. The output of the analog/digital converter is transmitted only in the absence of a squelch signal which indicates that the user is talking on his telephone. This output digital signal is then applied to the outgoing line amplifier. To the extreme left of Fig. 37 is the input data stream arriving from the Multiplexing Station at 19,200 bits/sec. This signal is amplified, and the timing period extracted. This fixed-frequency signal is used to time the outgoing digital stream.

Incoming binary data is converted by a digital analog converter circuit; the output is amplified and sent to a small earphone/speaker unit.

At the bottom of Fig. 38 is the counter unit with push-button keys used to establish the binary signaling pattern. The equipment required appears to be relatively simple, composed primarily of transistors and similar small electronic components. The present state of the microminiaturization art appears sufficiently well developed so that the entire unit might even be built within the base of the telephone. It appears that a digital telephone need not be large or expensive (if considered on an overall system basis.). With such digital telephones, additional signaling information may be transmitted even when the receiver is off-hook, permitting the addition of signaling information at times not now conveniently possible. This simplifies the handling of such new services as being able to set up a secure conference call and adding new parties even after the conversation is in progress.

It is necessary to consider methods for signaling information input transducers other than the digital telephone described. Figure 39 shows a modem suitable for start-stop teletype devices containing a signaling attachment. The signaling method is exactly the same as that used in the digital telephones. (It is to be remembered that teletypes operate in this system by modulating a 600-bit/sec square wave pattern.) This tie-in unit is a "black box" connecting the teletype to the line feeding the Multiplexing Station. This box, about the size of a cigar box, contains push buttons to select the called party, a small speaker for signaling and warning, and a high-current teletype line driver. Such a modem provides the interconnection between today's standard input output devices and the potential network of the future.


[1] See ODC-IV.

[2] It should be emphasized that the entire discussion on crypto is merely for illustrative purposes, included only to round out the system description. If, and when, it becomes necessary to precisely determine a form of cryptography for use in conjunction with this system, it will be done by the proper agencies. Such decisions reside outside the purview of responsibility, interest and information available to this writer (see ODC-IX).

[3] Shannon, C. E., "Communications Theory of Secrecy Systems," Bell System Tech. J., November 1949, pp. 656-715.


Contents
Previous chapter
Next chapter