On Distributed Communications Series

VII. Tentative Engineering Specifications and Preliminary Design for a High-Data-Rate Distributed Network Switching Node

III. Internode Timing Considerations

Time Required by Node to Process Message Block

Following is the estimate of total time required per Message Block:

This value assumes that the rewrite cycle of subroutine OUTSTOR is conducted in overlap with other non-core-memory operations

At a 1.54-million bit/sec rate, one Message Block is equal to approximately 0.000667 sec, or 667

= 95.3

Effects of Delay Times in Message Propagation and Error Control

In the distributed network, a store-and-forward technique is used which places rigorous demands upon the efficacy of the link error detection and correction method employed. Previous digital communications systems using error detection and repeat transmission dealt with data rates so slow that the transmission time between Switching Nodes was always much less than a fractional Message Block unit time.

As we move toward 1.5-megabit/sec (and faster) links, we encounter a new and generally unappreciated problem, because of the relatively large amount of time that elapses before the return of information that a particular Message Block has been incorrectly received.

Since the probability that a received Message Block has been received in error in even a relatively noisy link is much less than the probability that the Message Block has been correctly received, we can make a tentative assumption that a Message Block has been correctly received and proceed to transmit the next sequential Message Block. At transmitting Station A, it is then necessary that we keep sufficient "carbon copies" of the most recently transmitted messages for backup purposes, in case we later fail to receive confirmation that the transmitted message has been correctly received at receiving station B. Upon reaching such a finding, we prefer to use an alternate link for retransmission, because we cannot be sure whether we have encountered a random error or a long fade (see ODC-VI).

We shall now consider separately what happens if there is a failure on the transmission link, and/or a failure in the return link conveying the "return receipt."

Failure of the Transmission Link and Not the Return Link

If the transmission link from A to B fails while B to A does not, there is no problem: Both A and B know exactly what Message Block has been corrupted, so B ignores the Message Block. A possesses a clean carbon copy and sends the affected Message Block out over an alternate link.

Failure of the Return Link Only

If the transmission link from B to A fails, then a clear ambiguity exists. A does not know whether B has correctly received the message or not. If B thinks that he has received the Message Block correctly, as indicated by a successful block parity check, he relays the message on to C. As A does not know whether his message was received by B, he is forced to resend the message out over another link.

If the probability of a failure is assumed to be one Message Block in one hundred, then the additional traffic created by this repeat transmission process is only a small percentage of the total volume. Since the original preparation of this Memorandum, it has become apparent (see ODC-VI) that the raw expected error rate will probably be less than one Message Block per thousand. Thus, the additional traffic created will be negligible--an increase of less than one per cent in the total volume.

The appearance of duplicate Message Blocks calls for the Multiplexing Station to have the capability to detect and reject such occasional duplicates. This is not difficult, because each Message Block already contains a sequence number needed for the end-to-end cryptographic reception equipment. Both copies would bear the same sequence number (see ODC-IX).

Magnitude of the Delay

A convenient unit of time is the "Message Block." In the system being examined, it is the time required to transmit a 1024-bit message of 1.54 megabits/sec.

The abscissa of Fig. 11 shows a series of Message Blocks, A101, A102, A103, A104, ..., sequentially transmitted by Switching Node A. A data rate of 1,500,000 bits/sec, and a Message Block length of 1024 bits is assured. Thus, each Message Block requires about 2/3 ms. The ordinate represents distance in miles between Switching Node A and receiving Switching Node B. Two separate ordinate scales are used: one corresponds to a velocity of propagation of light, 186,000 mi/sec; the second to half that value, 93,000 mi/sec.

In the specific illustration, Switching Node B is shown to be at a distance of 103 mi from A. Velocity of propagation of 186,000 mi/sec from A to B and B to A is assumed. The lines slanting to the right represent the time delay of a bit traveling from left to right, or from A to B relative to A. The lines to the left represent the time delay of a bit traveling from right to left, or from B to A.

It can be seen that bit #1024 of Message Block A101, sent by A, cannot be received by B and returned to arrive at A until about 1 ms after transmission by A. Thus, if A101 was defectively received, Station A would not know about it until the moment it was in the middle of transmitting A103.

No Message Blocks are lost, due to the use of carbon copies, but the need for several Message Blocks of such carbon copy storage can be seen.

Links longer than about 150 mi would either have to operate at data rates less than 1.54-million bits or else more carbon copy storage would be needed. The additional storage required at this data rate is 1024 bits of storage per each additional 150 mi desired. Table VII indicates the data-rate/distance tradeoff if no additional carbon copy memory is used. A mixture of these two separate alternatives can also be used.

Use of Satellite Links in the Distributed Network

We shall now briefly consider the problems that might occur if satellite communications links are used in the system. We shall separately consider low-altitude non-synchronous orbit satellites, and high-altitude synchronous satellites. The reader can interpolate between these two examples.

Non-Synchronous Satellites

Fundamentally, the Link Unit described is designed for a non-time-variant communications link. As a non-synchronous communications satellite link changes its length and spatial position with time, a variable time delay is inserted into the communications link which must be considered in order to allow the Link Unit to operate in conjunction with non-synchronous satellites.

Assuming that there are two stations, A and B, at the ends of a satellite communications link, a shift in the radio carrier frequency will be noted as the satellite comes into range and then passes out of range, due to doppler shift. As the satellite comes into range, its frequency will be higher than its transmitted frequency; as it moves beyond the mid-range point, the frequency will decrease by a similar amount.

We shall assume a satellite with a velocity of about 18,000 mi/sec; the plane of the earth flat; the carrier frequency at 4000 megacycles; the satellite flying at an altitude that is low in comparison to the distance between A and B; and that it passes directly over A and B (see Fig. 12). The satellite velocity is

One-way doppler offset frequency, satellite to A

Since this is the same order of magnitude as the bandwidth required to transmit at a 150,000-bit/sec rate, this carrier frequency doppler offset would appear to be manageable with a conventional automatic frequency-control circuit at a slight loss of signal-to-noise ratio.

A 1024-bit Message Block at 150,000 bits/sec will take

for transmission. During this time the satellite will have moved,

Each bit will arrive with shorter-than-transmitted spacing by a factor of

Since there are 1024 bits in a Message Block, the bit timing will be off by 1024 x 2.69 x 10-5 = 2.76x 10-2 bits per Message Block.

Thus, it will take about

If the usable satellite range between A and B is about 3000 mi, the satellite link will be in range for about

If we modify our model by letting the satellite fly higher and extend the range beyond the A-to-B base line by R, then at the extremities of range, the path-length can exceed the path-distance D, between A, by the factor 2R; if R = 1000 mi, D = 3000 mi.

The path-lengths in terms of message length then become:

Therefore, it is necessary to be careful that the error detection feedback digits are returned and assigned to the correct "carbon copy" Message Block. This can be done by examining the handover number of the confirmation signal of the arriving Message Block Sequence Number[2] and subtracting the presently transmitted Message Block Sequence Number. For example, in Fig. 13, Switching Node A is designated the master timing station and is transmitting Message Block A101; some time later Switching Node B receives this same Message Block (A101) and synchronizes B's local timing of messages to A. However, the messages back to A contain the confirmation of correct receipt of Message A101 at B. The Message Block serial number of the present Message Block being received at A, less the value of the instantaneous Message Block Sequence Number A101 being transmitted, tells us which "carbon copy" contains the correct receipt.

This operation is automatically performed once each Message Block under the subroutine which determines the carbon copy number.

Non-synchronous satellites transmitting broadband signals require that the ground stations take advantage of the signal strength gain afforded by directional antennas. The antenna must follow the path of the satellite across the sky. Generally, the ground station has three options for tracking the satellite:

  1. A scanning antenna can be used to seek and lock onto the satellite;
  2. A local computer can be used to predict the future orbit of the satellite and point the ground antenna accordingly;
  3. Remote control centers can compute the pointing coordinates for a large number of satellites and ground stations and transmit the coordinates for immediate or future use. If immediate use is necessary, a separate non-satellite narrow-band communications link to the ground station is necessary. But, if coordinate storage is allowed at each antenna directional control unit, the upcoming coordinates may be transmitted over the satellite link itself for later use.
Let us consider how this last approach might work. We would like to transmit azimuth angle, elevation angle, doppler offset frequency, and perhaps some crypto information. Using the previous assumptions:
  1. The highest expected slew rate is roughly about 360
  2. Assume a 1
  3. Maintain a steering accuracy on an order of magnitude of 0.1
  4. Steering signals need not be transmitted much more often than about six times per second at accuracies on the order of one part in 3600;
  5. Accuracies of one part in 4096 require the transmission of 12 binary bits;
  6. Azimuth and elevation data to the remote antenna require the transmission of (12 + 12) bits x 6 samples per sec, or about 144 bits/sec (doppler offset information can be included at the price of a few more bits);
  7. Assuming that each Message Block can hold in excess of 800 bits of information, the transmission of merely two Message Blocks would provide for about 11 minutes of tracking time- -longer than the time the satellite would probably be within range;
  8. Thus, it appears that a little storage at the ground station might provide a possible convenient way of coping with the antenna pointing problem.
Thus, not encountering any new problems that are not being handled by the proposed system, we shall next consider possible problems that might arise when using synchronous satellite links.

Synchronous Satellites

Because of the "fixed" spatial positions of a synchronous satellite, we may omit the variable length link problems and consider the synchronous satellite link as being a long but fixed link.

Suppose that A and B are 8000 mi apart and the satellite is at a 24,000-mi altitude. The round-trip length, therefore, is equal to

If one additional carbon copy per 150-mi path separation is required at 1,540,000 bits/sec, then the number of carbon copies that must be supplied in a synchronous satellite system will be:

Timing Rrequirement for Very-High-Data-Rate Time Division Multiplexed Links

Although the Switching Node is basically designed to operate at link data rates up to 1.54-million bits/sec, the mini-cost microwave transmission links operate at about three times this data rate (see ODC-VI).

One simple scheme that can be used to multiplex three separate bi-directional channels on a link operating at three times the 1.54-megabit rate (4.62 megabits/sec) is by use of simple time division assignment. Three sequential time slots can be used. The first time slot might carry traffic from A to B, and the second from A to C. By making the widths of the three slots slightly different, information is available to allow positive identification of the time-divided channels. In the "grasshoppering" arrangement of Fig. 15 there is no increase in the number of times new timing must be derived, as compared to that which would be required by a separate single link from A to C. The only new requirement is that the system be able to accommodate links whose length is that of A to C, instead of A to B or B to C. Although this illustration is drawn on the basis of three time slots, it is clear that many more channels can be handled in a similar manner.

[1] See ODC-IV, p. 12.

[2] Note that this is a link sequence number, not the end-to-end number previously described.

Previous chapter
Next chapter