This invention relates to an apparatus for filling a buffer with data and more particularly to such apparatus as employed in a data communication system. Of interest, is copending application serial no.

130,379 filed December 8, 1987 entitled "Modified Statistical Coding of Digital Signals", in the name of N. J. Fedele assigned to the assignee of the present invention and U.S. Patents 4,691,233 and 4,700,226 both in the name of A. A. Acampora and U.S. Patent No. 4,706,260 entitled "DPCM System with Rate-of-Fill Control of Buffer Occupancy" in the name of Fedele et al.

In data communication systems of limited bandwidth, differential pulse code modulation (DPCM) is widely used for data compression. The DPCM system is often used in the transmission of television pictures, especially in teleconferencing applications employing satellite transmission systems or other transmission mediums which have relatively narrow bandwidths. While the present invention is especially suitable with a DPCM system, it may be utilized with any data communication system for which maximum efficiency of transmission and cost effectiveness of hardware is desirable.

Typically in a DPCM system, a transmitter includes an analog-to-digital converter for finely quantizing a source of analog television information signals. The finely quantized signal is applied to a subtractor which subtracts predicted delayed information picture signals which may employ either temporal or spatial filtering techniques. The subtracted signal is coarsely quantized to convert numerous luminance signal values of a television picture, for example, to relatively few signal values. Generally, for television transmission, a coarse quantizer may, for example, have 32 output values as compared to 256 values for the finely quantized signal.

In the interest of transmission efficiency, this kind of system tends to encode redundant picture

information, (that is, information that does not change from one picture frame to another in temporal systems or that does not change from line to line in spatial systems) with zero run length codes. The changing picture portions producing non-zero values are usually encoded first with amplitude codes. Typically, a statistical encoder used in these systems assigns variable length codewords to the different run length and amplitude values of the coarsely quantized signal. The shorter length codewords are assigned to those signal values which occur more frequently and the greater length codewords are assigned to those signal values which occur less frequently.

Statistical encoding schemes employing Huffman or some other arrangement such as, for example, disclosed in aforementioned copending application serial no. 130,379 produce variable-length codewords. Variable-length encoders for such codes are disclosed, for example, in the aforementioned patents in the name of Acampora and in U.S. Patents Nos. 4,093,962 in the name of Ishiguro; 4,569,058 in the name of Grallert; 4,302,775 in the name of Widergren et al.; and 4,488,175 in the name of Netravali, among others.

Normally in these systems, the variable-length codewords are applied to a rate buffer, typically a first-in/first-out (or FIFO), which has a relatively large memory supplying output at a fixed data rate for transmission over a channel. The memory normally is made sufficiently large so that as the variable length codewords are filled into the memory, the constant data rate at its output will neither empty the buffer nor permit the buffer to overfill. However, systems are available to provide more accurate control over the buffer fill state. Such. a control system is disclosed, by way of example, in the aforementioned Acampora patents and in the Ishiguro patent 4,093,962.

Rate buffers employ a memory comprising a plurality of fixed-width N-bit wide locations. The variable-length codewords are each clocked in parallel into

a separate location. Therefore, a codeword that represents a more frequently occurring signal value (which may have a length of say two to five bits, for example), will occupy a buffer memory location capable of holding as many as N-bits (for example, twenty bits). Each codeword is clocked into separate location at a separate clock cycle. The next codeword, which may be thirteen bits long, is thus clocked into the next location. Since codewords are assigned to a given location based on order of occurrence and not on len the locations necessarily need to have capacity for the maximum length word. The words are applied to the buffer in parallel and therefore are each clocked in a given single clock pulse. Typically, rate buffers may include a parallel-to-serial converter at the output thereof for converting the parallel bits to a stream of serial bits for transmission over a channel. The use of excess capacity in the rate buffer to acco odate potential codewords of a maximum length regardless of whether in fact such codewords of maximum length statistically occur less frequently then the codewords of shorter length is a necessary condition to the hardware of such a system.

In some cases to more efficiently pack the rate buffer, a parallel-to-serial converter is used prior to the rate buffer to generate a serial stream of bits to fill the buffer. Such a system is disclosed by way of example in

U.S. Patent 3,761,613 issued to Limb. However, the problem with this kind of a system is that each bit of an encoded codeword needs to be clocked at a separate clock pulse into the buffer memory. To clock each bit into memory separately is time inefficient. For example, if a fifteen bit codeword is to be entered into memory, fifteen separate clock pulses are needed. While the memory may have greater packing density, this kind of a system is slow and not viable for current teleconferencing television transmitting systems. The greater the intensity of the motion in the picture, the slower the information is entered, which could eventually impede the fixed-rate transmission process.

Where variable-length codewords are entered into the buffer memory in parallel, buffer memory space is wasted and thus memories much larger than necessary are required for the amount of data being stored to accomodate the occasional long codewords. As data rates increase, the buffer memory is similarly required to have increased capacity to accomodate the much greater rate of fill of the buffer memory. However, the addition of more memory adds greater complexity to the system, which increases the system cost. As known, to provide an inverse decoding operation, the receiver is required to duplicate the memory of the transmitter. Therefore, all hardware added to the transmitter in the name of increased memory capacity is necessarily added to the receiver as well, further increasing the cost of the system hardware.

An apparatus according to the present invention efficiently packs sequentially occurring variable-length codeword data in an input signal into memory means, for subsequent transmission over a channel. The memory means includes a plurality of data storage locations each having an N-bit storage capacity. The apparatus comprises data grouping means responsive to the variable length codeword signal applied as an input thereto for organizing the variable length codewords into N-bit-long data groups. Each group contains codeword data in an ordinal position according to a first given sequence of occurrence of the data of that data group. The group is outputted from the grouping means in a second given sequence for storage in a memory means storage location in a third given sequence. Each group is stored in a separate, corresponding location whereby each location containing data is filled to capacity with a single clock pulse. In the drawing: FIGURE 1 is a block diagram of a portion of a DPCM communication system according to one embodiment of the present invention;

FIGURE 2 is a more detailed block diagram of the data grouping system of FIGURE 1.

FIGURE 3 is a block diagram of the control of FIGURE 2;

FIGURE 4 is a schematic block diagram of the switches and certain of the latches of FIGURE 2; FIGURE 5 is a block diagram showing in more detail the rate buffer of FIGURE 1;

FIGURE 6 is a timing diagram useful for explaining the principles of the present invention; and FIGURE 7 is a state diagram illustrating the creation of the N-bit groups of data.

In FIGURE 1, system 10 includes a transmitter 12 which transmits over a serial channel 16 to a receiver 14. Only so much of the transmitter 12 and receiver 14 are illustrated as being necessary for understanding the present invention. The present embodiment is described in terms of a DPCM transmission system by way of example. It should be understood that the present invention may be employed in any kind of data transmission system where variable-length codewords are generated. The transmitter 12 includes a variable length encoder 18 which produces variable-length codewords having a maximum length K on conductor 20 and a codeword length signal on conductor 22. The variable length codewords on conductor 20 each have a given length (number of bits) which may vary randomly to a maximum of K. In this embodiment K is assumed to have a value of eighteen. The signal on conductor 22 may have a length L which is assumed in this embodiment to be five bits and represents the bit length of a corresponding codeword on conductor 20. The E clock is a clock signal transmitted on conductor 24 comprising a series of pulses each representing a pixel or codeword on conductor 20.

Encoder 18 may be of conventional design, such encoders being described in the prior-mentioned patents noted above in the introductory portion. Preferably, encoder 18 may be constructed as disclosed in the aforementioned copending patent application in the name of Fedele entitled "Modified Statistical Coding for Digital

Signals". However, what is important about the encoder 18 with respect to the present invention is that a first signal on conductor 20 has a maximum codeword length of K bits, that a second signal of L-bit length on conductor 22 represents the number of.bits in the corresponding codeword on conductor 20 and that a clock signal represents the occurrence of each codeword. Design of hardware to generate the above kind of data signals on conductors 20, 22 and 24 is within ordinary skill in the encoder art. For example, the encoder may include a first ROM look-up table for the codewords and a second ROM look-up table containing the codeword lengths of such codeword. The corresponding codeword and length data are accessed by a given address. The signals on conductors 20, 22 and 24 are applied to data grouping system 25 which is described in more detail in FIGURE 2. System 25 in accordance with the present invention constructs from the information on conductors 20, 22 and 24 N-bit data groups on output conductor 26. The groups of data on conductor 26 always are fixed-length N-bits. A clock pulse, F clock, on conductor 28 represents the generation of a group of N-bitε on conductor 26 and clocks each N-bit data group into rate buffer 30. Rate buffer 30 is of conventional design sometimes referred to as a firεt-in/first-out (or FIFO) based memory system and is illustrated ^' in FIGURE 5 for purposes of explaining the signficance of the data grouping system 25. The rate buffer 30, as will be shown, has a plurality of data storage locations typically by way of example, 16K words (or 40K bytes capacity) for a 384K bits/sec transmission system. In this example each location contains N slots for storing N-bits of data. N, in this case, represents twenty data bits. However for other implementations, the number of bits on conductor 26 in a group of data bits and the bit length of a given location in a buffer 30 may differ from twenty in accordance with that implementation.

Each group of N-bitε is sequentially entered into a different location in buffer 30 in a known way by an F

clock pulse on conductor 28. The rate buffer includes a parallel-to-serial converter near its output, which converter converts the parallel bits stored in each of the locations into a serial stream for transmission over the channel 16. The serially transmitted bits are received by rate buffer 32 at the receiver and the encoded signals are decoded by decoder 34 as part of the receiver 14. The inverse DPCM operation is performed by other well known apparatus not shown in receiver 14. It will be recognized that system 25 of the present invention takes variable-length codewords and arranges the variable length codewords into fixed-data- length groups of N-bits each in real-time, this arrangement being done based on the sequence of occurrence of the variable-length codewords. Each such N-bit group is stored in a different buffer 30 location to completely fill that location to capacity. Each location of the buffer 30 is sequentially filled with the exact number of bits necessary to fill that location. Thus, the rate buffer is efficiently utilized so that the memory locations thereof containing data are filled to a maximum while maintaining the correct data placement within the transmitted data stream. This provides for efficient utilization of memory and is cost effective in terms of hardware for a given data rate.

In FIGURE 2, system 25 includes a data shifter 200. Shifter 200 shifts input signals on input conductor 20 from the variable length encoder 18 (FIGURE 1) a maximum of N-l positions wherein it will be recalled N represents the number of bits in each group to be created by system

25. However, conductor 20, as mentioned above, provides K- bits in parallel to the shifter 200, K in this example being eighteen. The shifter 200 shifts the incoming bits received in parallel on K input lines for each codeword an amount such that at least N output lines of shifter 200 on conductor 202 eventually will receive bits in parallel from the sequential occurring codewords on conductor 20. Should the cumulative successive codewords on conductor 20 which

contain at least N-bits contain bits in excess of N, _F , the excess may be up to K-l bits. As εtatiεtically frequently would occur, those N _E excess bits are shifted to appear on output conductor 204. Statistically, as many as K-l bits may appear on conductor 204 as N L„a bits. The reaεon for K-l bits on conductor 204 is that if nineteen bits (N-l) are shifted to parallel lines of conductor 202 and the next occurring codeword has K bits (18), then the twentieth bit (one of the K bits) is processed on conductor 202 and the remaining bits of the N_ group (K-l) are processed on each of the parallel lines of conductor 204.

By way of example, if a seven-bit codeword X _x is initially received by shifter 200, the bits of that codeword each appear at a different shifter 200 output 1-7 in ordinal position of occurrence starting from position 1. If the next codeword X ₂ is twelve bits long, it is shifted seven places so that its bits appear on outputs 8-19. If the next codeword X ₃ is three bits long, it is εhifted nineteen placeε εo aε to appear at outputs N through N +2 (20-22). Thus, N-bits (20) are applied to conductor 202 and two bits in excesε of N, N„, are applied to conductor 204. Upon receipt of the next codeword X ₄ shifter 200 shifts that number of places from the unεhifted zero position as indicated by the number of bits in excess of N, N _£ , or two places. Assume word X ₄ has four bits. These bits are shifted to shifter outputs 3-7, skipping positions 1-2 for reasons to be explained. Assuming the next codeword X ₅ has five bits, these five bits are εhifted seven places so they appear on outputs 8-12 and so on. Each time N + N _£ bits are εhifted, the shifter shifts the next occurring codeword N _E places from the zero shift position, that is, _E places from the shifter beginning position. It should be understood that the term "conductor" employed herein is intended to include multiple conductor lines for carrying signals in parallel as known in this art. The shifter 200 shiftε the input εignal from conductor 20 to thoεe output lines on conductors 202 and 204 according to a control signal received on conductor AN _X from translator 208.

It should be understood that the shifter 200 does not store the data bits that are received, but serves only to shift postionε of the received bits to direct the bits to different output slots. It is possible for shifter 200 to shift the poεition of the incoming codewords as many as N-l shifts or nineteen places in this example.

Control 206 generates the control signal that is applied to tranεlator 208. Translator 208 in turn generates the shifter control signals produced on conductor AN _! which cause shifter 200 to shift as described above upon the receipt of each codeword in sequence.

FIGURE 3 shows control 206 in greater detail. In FIGURE 3, control 206 receives the codeword length signals on conductor 22, each signal having the same L-bit length. The codewords on conductor 22 indicate the length of each word (number of bits). In this example, L is five bits. The L-bit codeword length signal is applied to adder 308, which adds a subtracted codeword signal on conductor 310 to the codeword length signal received on conductor 22. The cumulative length of the subtracted codeword on conductor 310 is added to the codeword length on conductor 22 by adder 308. The added signalε are clocked into register 312 at each encoder E clock pulse on conductor 24, each E clock pulse representing a codeword. The output of register 312 is L-bit long words which represent the cumulative length values or number of bits in the codewords on conductors 22 and bits on conductor 310. The output of register 312 is applied to conductor 314, to comparator 316 and to subtractor 318. A second input of comparator 316 is on conductor 320 which receives a εignal from a source having a value representing N-bits. Comparator 316 compares the totalled length of bits value of the signal on conductor 314 to the fixed number of bits N on conductor 320. If the number of bits represented by the signal L on conductor 314 is N-bits or greater, then the output of the comparator 316 is a logical high, i.e., a one. This produces a signal referred hereinafter as an F clock on

-

conductor 322. The F clock is applied to buffer 30 (Figure 1) via conductor 28 and to latch 224, Figure 2, delayed one half cycle by delay 207. The F clock signal is applied directly without delay to ROM translators 208, 211 and ^' 212. The logical one level of the F clock signal manifests that a group of at least N-bits has been received. This is significant, as will be explained. If the number of bits at the output of register 312 on conductor 314, FIGURE 3, is less than N, that is from zero to nineteen bits in this embodiment, then the output of comparator 316 on conductor 322 is a logical zero.

The F clock signal on conductor 322 is also applied as an input to MUX 324. The MUX 324, depending on the logical level of the signal on conductor 322, applies either a zero to its output conductor 326 in the case of a logical zero on conductor 322 or a signal having a numerical value N for a logical one. Asεume that the first codeword X _x applied on conductor 22 has seven bits and no prior words have been received, then the output of register 312 is numerical zero at time T ₀ , Figure 6. This numerical zero value is applied to comparator 316 and compared to the number N. Because the output of register 312 is less than N at thiε time, the output of comparator 316 on conductor 322 is a logical low. A logical low causes a numerical zero to be applied by the MUX 324 to output conductor 326. Subtractor 318 subtracts this numerical zero from the numerical zero on conductor 314 applying the difference, which is also a numerical zero to conductors 310 and 310'. This repreεentε an address that is applied to translators 208, 211 and 212 in the data grouping system 25 of FIGURE 2. At the next clock pulse at time Tj , Figure 6, the number seven bits is indicative of codeword X _x having clocked into register 312, applying that value of εeven on conductor 314. This value is less than N so the difference signal on conductors 310 and 310' is seven less zero, or seven.

The seven value on conductor 310 is added to the next received codeword length signal on conductor 22 by adder 308. Assuming that the next codeword X ₂ has twelve bits, those twelve bits are added to the seven bitε on conductor 310 to form a word group having a nineteen bit length. Those nineteen bits are stored in register 312 at the next clock at time T ₂ . After time T ₂ , those nineteen bits appear on conductor 314 and the comparator 316 output is again low because the nineteen bits are still less than N on conductor 320. MUX 324 applies a numerical zero on conductor 326 to subtractor 318 which again subtracts numerical zero from the signal on conductor 314 and produces on conductor 310 a signal representing nineteen consecutive bits. This signal is applied via conductor 310' to the translators 208, 211 and 212 and to adder 308 via conductor 310.

Adder 308 adds the next incoming codeword X ₃ , which is assumed to have three bits, to the nineteen bitε to form a twenty-two bit group which is clocked into register 312 at the next clock pulse at time T ₃ . Conductor 314 at this time has the number twenty-two thereon to be applied to comparator 316. Since N (here assumed twenty) is smaller than twenty-two, the F clock signal on conductor 322 goes high. The high on conductor 322 causes MUX 324 to apply N to the output 326. The N value twenty is subtracted ¹ from the twenty-two bit signal on conductor 314 to produce the number two on conductor 310. This number two is now the address to translators 208, 211 and 212 FIGURE 2. This number two is supplied to adder 308 and is added to the number specifying the length of the next codeword X ₄ on conductor 22, assuming X ₄ to be a four bit-long word, the sum signal six is clocked into register 312 from adder 308 output. This process .continues with the comparator producing a numerical zero on conductor 322 until the number accumulated in register 312 is equal to N or more.

The conductor 310' connected to conductor 310 is the output of control 206. This output is supplied to

conductor 310 which carries a control signal representing the resulting number after subtraction by subtractor 318. Whenever the sum of the codeword length appearing on conductor 22 at the input of control 206 added to the numerical value of the number of bits on conductor 310 is leεε than N, (twenty in this case), then the output control signal on conductor 310' represents the accumulated lengths of the codeword received at conductor 22 and the addition of N _E representative of extra bitε of the prior codewordε, if any. When a codeword length iε received which causes the accumulated codeword group length to be N-bits or greater, then the output signal on conductor 310* represents N _E , the difference of the accumulated bit lengths less N. The signal on conductor 310', FIGURE 2, addresseε the ROM translators 208, 211 and 212 in combination with the F clock signal on conductor 322.

In FIGURE 2, the ROM translator 208 in response to the signalε received thereto on conductors 310' and 322 translateε the number of bitε repreεented ^' by the εignalε into a shift value for causing the shifter 200 to shift a number of places represented by the addresε on conductors 310' and 322. Thus, for example, asεume conductor 310' has the addresε seven on it between time periods T _x and T ₂ . Conductor 322 has a low F clock. These signals are applied to translator 208 (FIG. 2) as a combined address which decodes this address into signals on translator outputs 1, 2, 3 and 4. Shifter 200 in responεe to the tranεlator 208 outputε εhiftε itε inputε seven places from the zero position. The translator 208 outputs 1, 2 and 3 each represent by way of example a three bit code which iε capable of identifying eight εtateε only εix of which are uεed. The outputε 1, 2 and 3 are then capable of describing a total eighteen different states; and the fourth output represents another state. Therefore, the four ROM tranεlator 208 outputε represent nineteen different εtateε and can εhift the shifter 200 output a total of nineteen different positions, i.e. N-l.

Returning to the example diεcussed above, assuming the first address on conductors 310' and 322 represent a codeword having a seven bit length, the translator output 1 on conductor AN _α FIGURE 6 causes the shifter to shift six places and output 2 causeε the shifter to shift one place for a total shift of seven places. At time Ti the next codeword X ₂ received by shifter 200 will be applied to the shifter positions beginning with position 8. Assuming the next codeword X ₂ has twelve bits in it, these twelve bits are each shifted seven places at time _α . So far all of the shifted outputs of shifter 200 are applied to conductor 202 which will be recalled contain N lines each connected to a different shifter 200 output. Assume the next codeword X ₃ iε three bits long and the total number of bits received by control 206 is 22. The subtracted signal on conductor 310' is the number of bits in excess of N, N„, e.g., twenty-two minus N, or two bitε. Two appearing on conductor 310' and the logical one appearing on conductor 322 together causes the translator 208 to generate a shift signal on output 1 causing shifter 200 to shift X ₃ two places from the zero position. Assuming the next codeword X ₄ received has a four bit length, then that four bit codeword will be shifted N- two places from the zero position and the codeword X ₄ will occupy positions 3-6 of the shifter 200 output. This way the shifter applies to conductor 202 the first N-bits of the sequentially received bits of a group of codewords and applies the N„ £_ bits in excess of N of that codeword group to conductor 204, FIG. 2. After at least N-bits are received, the next received bits are shifted N„ places from the shifter N position. This process continues for each group of at least N-bits received by control 206 and shifter 200.

The ROM translator addresses applied from control 206 are delayed one clock pulse from the time the initial codeword is applied to shifter 200 and to the input of latches Bj and B ₂ by the action of register 312 of control

206. The latcheε B _x and B ₂ are clocked by the E clock εignal on conductor 24'.

Since shifter 200 is not a storage device, the bits applied to it are merely shifted in position to the different lines in parallel on conductors 202 and 204. The bits applied to conductor 202 are applied from shifter 200 through switch 210 to latch Bj . Latch B ₁ is where twenty-bit lengths are accumulated for transfer to output latch 224, the accumulation being done by a process involving switches 210 and 218 that will be subsequently described in detail. The bits applied to conductor 204 are applied directly to latch B ₂ . Latch B ₂ provides temporary storage of the N _£ bitε in excess of those needed to complete filling places in latch Bi , which N_ bitε can then be entered into latch Bj after emptying itε contentε to output latch 224. Thiε latter creation iε accompliεhed using switch 218 and will also be described in further detail later on.

Switch 210 serves to couple those outputs of shifter 200 on which codewords are then being applied or are anticipated in the succeeding clock cycle. Switch 210 also serves to decouple those shifter 200 outputs from latch B ₂ which received codewordε in prior clock cycleε until at leaεt N-bitε have been latched into latcheε Bj and B ₂ . Thus, for example, in FIGURE 6, switch 210, at time Tj , opens or decouples the first seven lines of conductor 202 on which seven bits have previouεly been applied to the latch B _α at time T _x . The remaining εwitcheε are closed. The reason for the decoupling at time T _x is as follows: If the shifter 200 were to remain coupled at those first seven outputs to latch B _t , upon the occurrence of the next E clock pulse at time Tj , the next codeword X ₂ , for example, the twelve bit codeword is clocked into latch B _x and the seven bit codeword in latch Bj will be latched out. However, the data in shifter 200 will be otherwise latched into latch B . Since the shifter 200 would have no bits at this time in the first seven outputs, those first seven bits would be lost. Therefore switch 210 decouples those

firεt seven positionε of latch B _x from εhifter 200. However, the problem remainε to retain thoεe firεt εeven bitε in latch B _x which are latched out when the next twelve bits are loaded into latch B _x upon the occurrence of the next clock pulse at time T ₂ . The solution iε to return thoεe εeven bitε to latch Bj via εwitch 218, which will be discussed below.

The way the switching action is obtained is described below in connection with FIGURE 4. Generally, ROM translator 211 receives the addresε produced by control 206 on conductors 310' and 322. The F clock signal on conductor 322 at time T _α iε low and tranεlator 211 outputε 1, 2 and 3 decode the code applied thereto on conductor 322 into a switch 211 control signal on conductor AN ₂ . That signal produced by tranεlator 211 opens the first seven positions of switch 210 at time T _x leaving the remaining thirteen positions closed. The twelve bit codeword X ₂ is placed in ordinal position at latch B _α at time .T ₂ by the shifted position of shifter 200 via the closed contacts of switch 210.

However, at time Tj , the twelve bit word X ₂ is not latched into latch & _x but is at the latch Bi input as shown by state 2, FIGURE 7. In FIGURE 7, an arrow in the circles indicates data is latched, a hyphen indicates data ready to be latched at the next clock. The value twelve descriptive of X ₂ being a twelve bit codeword added to the value seven descriptive of the previous seven bit codeword produces an addresε from control 206 εpecifying a seven place shift in εhifter 200. That address applied to translator 211 opens the first seven positionε of switch 210, FIGURE 6, and the remaining positions remain closed. The twelve bitε are latched into latch Bi by the E clock pulεe at time T ₂ . What happens with the first seven bits previously latched out of latch Bj iε that they are recirculated to the input of latch B _x via εwitch 218, when latched out at time T _α . At time T ₂ , these first seven bits are also relatched back into latch Bi for a total of nineteen bits. These conditions are illustrated by the

state diagram, FIGURE 7, at εtateε 1 and 2. In εtate 1, word Xj is at the input of latch Bjli (Xj-B _j l). At state 2 ₂ word X _α is latched into latch B _x at time T _x (X _1→ Bi) and word X ₂ iε at the latch B _t input (X ₂ -B] _. I). The outputε of latch B _x are applied to conductor

222 and to latch B _x input conductor 216 via output switch 218 and conductor 220. The latch B _x outputε are also applied to output latch 224. Latch B _x has N outputs which are applied in parallel on N lines of conductor 222. However, when the bits in latch Bi are lesε than N, the F clock iε low and latch 224 iε inactive.

The N _E bits in excess of N of a given set of codewords which form the first N-bits processed through shifter 200 are procesεed on conductor 204 to latch B ₂ . The output of latch B ₂ iε applied to K-l lineε on conductor 226 which applieε the latch B ₂ bitε to the latch _x inputε in ordinal poεition through εwitch 218 to conductor 220. At this time, the bitε in latch B _x are latched into latch 224 becauεe the F clock iε high. Switch 218 εerveε to recirculate the bitε in latch B ₂ in the firεt N _E latch poεitionε back to latch Bx at conductor 216 in the same relative position those data bits were originally stored in latch B ₂ from shifter 200. Switch 218 is controlled by ROM translator 212. Tranεlator 212 haε εix output conductors each representing seven states which apply a control signal on conductor AN ₃ . Tranεlator 212 tranεlateε the addreεε on conductorε 310' and 322 into a control εignal on conductor AN ₃ for operating switch 218.

As mentioned above, when the first seven bitε have been latched into latch B _{l t} the next twelve bitε have been shifted seven places by shifter 200 through εwitch 210 to the latch Bi input. When those 12 bits are latched into latch B _x at time T ₂ , the addresε applied to the ROM tranεlatorε 211 and 212 reεpectively open the firεt seven positionε of switch 210 and close the first seven positionε of εwitch 218. Aε a reεult, the codeword X ₂ 12 bitε is latched into latch Bj and the codeword X _α which was previouεly in latch B _x iε recirculated through the closed

first seven positions of switch 218 back to the first εeven places of latch B ₂ . The open state of the first seven positions of switch 210 preclude losing those first seven bits owing to the first seven positions of shifter 200 having no data bits. At time T ₂ latch B _x now contains nineteen bits.

When the next codeword X ₃ is received by the system at time T ₂ , asεuming X ₃ is three bits, shifter 200 is shifted nineteen positionε. One bit of that three bit codeword is applied to conductor 202 at the Nth position and the remaining two bits are applied on conductor 204 at the N+l and N+2 shifter positions. One bit on conductor 202 is applied by switch 210 to the twentieth input of latch B _α and the two bits on conductor 204 are applied. directly to latch B ₂ . See state 3, FIGURE 7. The bits at the input of latch B _α are referred to as codeword X ₃ ' and the bits at the input to latch B ₂ as codeword X ₃ ". The three bitε are loaded into latches Bi and B ₂ at the next occurring E clock pulse at time T ₃ . Latch B _x now has N-bits loaded therein and latch B ₂ has two bits (N _E ). The F clock pulse is delayed one half a clock cycle, FIGURE 6, by delay circuit 207, FIGURE 2, and applied to the output latch 224. Latch 224 outputs the twenty bits in latch B _α to the data conductor 26 by the delayed F clock pulse. The two bits in latch B ₂ are recirculated by switch 218 onto conductor 220 to the first two positionε of latch ^* Bj. For thiε reaεon the εhifter 200 in reεponse to the addresε on control output 310' applied to translator 208, has shifted two places from the zero position so that the next received codeword is shifted in ordinal position two places down from the zero position. This places the next codeword X , FIGURE 6, which is assumed by way of example to be four bits long, into the third through seventh input positionε of latch Bi . Therefore at time T ₄ the latch B ₂ two bit long codeword iε outputted through εwitch 218 to the latch B _% input, switch 210 is open in the first two positions and shifter 200 has been shifted by the ROM translator control signal in responεe to the two bit

addreεε applied to tranεlator 208 from control 206. The. εwitch 210 first two poεitionε are open so as to preclude the latch B _x from losing the first 2 bits of data. This process is continued for all subsequent occurring codewords.

In FIGURE 4, ^" one example of switches 210 and 218 is illustrated employing tristate devices labeled TS. Each line from translator 211 is applied to a separate tristate device which controls the input of data into a corresponding respective input Ij -I _N of latch _x . The tristate devices are either closed (off) or open (on).

Devices 402 through 402 are respectively controlled by tranεlator 212. Triεtate deviceε 402-402n control the outputε 01-ON of latch B _x . Tristate devices 404 through 404 respectively control the outputs 01-OK-l of latch B ₂ . By way of example, line Ei of conductor 202 is coupled to input l _t of latch B _x by conductor JI through tristate device 400. Conductor E ₂ is applied as an input to input I ₂ of latch B _x via conductor J ₂ through tristate device 401. ROM tranεlator 211 controls device 400 via conductor Pj whereas tristate device 401 is controlled via conductor P ₂ and so forth. The output 01 of latch B _x corresponding to input II is applied via tristate device 402 to conductor Gl and itε triεtate device 402 iε controlled by tranεlator 212 via line Cl. The conductor Gl is connected to conductor Jl so that the data bit in latch B _α at position 1 is recirculated via triεtate device 402. to input II back into latch B _x . When a εignal iε εo circulated, tristate device 400 is open (off). Similarly, each of the tristate devices at the outputε of latch Bi recirculate the data bits back to the B _x latch inputs corresponding to those outputs. The input tristate devices 400-400 are open (off) corresponding to those bitε being recirculated aε shown in the diagram of FIGURE 6. Assuming that bits N _E in excess of N have been applied to latch B ₂ then those bitε are inputted into latch B upon their occurrence. Upon the detection of at leaεt N-bits by the control 206, FIGURE 2, at time T ₃ , latch 224

outputs the N-bits of latch _x to the output data conductor 26 by a one half clock delay after time T ₃ . This is shown by the transition from state 4 to state 5, FIGURE 7. Tristate devices 402-402 are all off preventing the bits in latch Bi from being recirculated back to the latch _x input. Tristate devices 404 and 404i are closed (on) and all the remaining tristate devices corresponding to latch B ₂ are open (off).

Thus, at the next clock pulse, T ₄ , the two bits in latch B ₂ are circulated via tristate devices 404 and 404j via conductors HI and H2 to conductors Jl and J2 at the input of latch B _x applying those two bits respectively to inputs II and 12 of that latch. Recall at thiε time the εhifter has shifted two positions and therefore incoming codewords shifted two positions can be latched into latch Bj at the third and et seg. positions of latch B _: . Since the next codeword X ₄ is four bitε then those four bits will occupy the next four positions of latch B ₃ by way of the next four triεtate devices of switch 218 which will be closed (on). In this way, the tristate devices are switched on and off to load the latch Bj . When the bitε in latch B ₂ are recirculated into latch B _α to organize the next group of N bitε a twenty bit codeword iε latched into latch 224 at time T ₃ delay. In FIGURE 5, the F clock, which is delayed one half clock cycle, iε applied to a write regiεter 504 which addresses one of the locations 500 - 500 _n of ROM 502. ROM 502 has 16K locations (40K byte capacity) each location having a capacity of N bits (2.5 bytes). Each group of N-bits applied to ROM 502 completely fills one of the locations 500 - 500 . The write regiεter 504 automatically sequences to each of the locations in a known way to write the next succeeding group of N-bitε into a successive location. The read register 506 reads the information from successive locations in sequence in a known way. The read register 506 causes the twenty bit word groups in ROM 502 to be outputted to an output bus 510. The twenty bit words are applied to a parallel-to-serial converter 511 which

convertε those successive twenty bit parallel bits inputs into a serial stream on conductor 512.

Because of the shift position of shifter 200, a portion of codeword X ₃ representing one bit having a reference designation X ₃ ' is at the input of latch B _x and a second portion of the same codeword X ₃ having a reference deεcription X ₃ " iε at the input of latch B ₂ . Upon the occurrence of the next clock pulεe at time T ₃ codeword portion X ₃ * iε latched into latch B and iε outputted at the delay of a one half clock pulεe at T ₃ delay. At the time T ₃ delay, the codewords X _x , X ₂ and codeword portion X ₃ ' are outputted in a single clock pulse. However, at time T ₃ as mentioned codeword portion X ₃ " is latched into latch B ₂ and codeword X ₄ is applied to the input of latch B _x via the shift position of shifter 200 FIGURE 2. Upon the occurrence of a clock pulse at time T , the codeword portion X ₃ " is latched into latch B _x from latch B ₂ via the tristate devices of switch 218 FIGURE 4. The codeword X ₄ is also latched into latch B _x and the codeword X ₅ is applied to the input of latch B _x waiting for the next input latch clock pulse. At time T ₅ the E clock pulse latches the codewords X ₃ " and X ₄ from B _x back into latch B via switch 218, and the codeword X ₅ is latched into latch B _x in the ordinal position as received by the syεtem relative to the codeword portion X ₃ " and codeword X ₄ .

At thiε time, aε shown in FIGURE 6, asεuming the codeword X ₃ " portion repreεents 2 bits and codewords X ₄ and X ₅ repreεent four and five bits, respectively, then there are a total of eleven bits loaded into latch B with codeword X ₆ waiting to be loaded into latch Bi . This process continues and repetitively outputs groups of N-bit words into the memory of rate buffer 30.

In FIGURE 1, a signal source of video signals is assumed encoded by encoder 18. However, it should be understood that other types of signals other than video may be grouped by grouping syεtem 25. In addition, multiplex εwitcheε may be provided (not εhown) εyεtem 25 for multiplexing different video εignals and grouping those

different video signals into N-bit groups into different rate buffers (not shown). This utilizes syεtem 25 for grouping the data of different video signals or other kinds of signals into different outputs. While tristate device switcheε are illuεtrated other typeε of devices may be used in accordance with a given implementation.