Encoder/decoder buffer control for variable bit-rate channel

<u>Technical Field</u>

0001This invention is related to video image processing and, more particularly, to avoiding encoder and decoder buffer overflow and underflow when the image is transmitted over a variable bit-rate channel.

<u>Background of the Invention</u>

0002ISO MPEG standards group is in the process of defining a Real Time Interface (RTI) specification for vide encoders/decoders which constrains the real-time byte delivery schedule of transport bit-streams to real decoders. A real time decoder model includes three (3) buffers for each elementary video bit-stream: the transport buffer, the multiplexing buffer and an elementary bit-stream buffer. These buffers are intended to be sized to take into account the fact that bytes in the elementary bit-stream will arrive at a decoder with some jitter. When the video bit-stream has been transmitted over an Asynchronous Transport Mode (ATM) network, the jitter is referred to as Cell Delay Variation (CDV).

0003Transform coding techniques and other systems for compressing video motion pictures are well known. Specifically, in our article entitled "Constraints on Variable Bit-Rate Video for ATM Networks", <u>IEEE</u><u>Transactions</u><u>on</u><u>Circuits</u><u>and Systems</u><u>for</u><u>Video</u><u>Technology</u>, Volume 2, Number 4, pages 361-372, December 1992, we indicated that, assuming no delay jitter, the encoder could prevent decoder buffer overflow and underflow by computing the actual decoder buffer fullness. (Also see our U.S. patent 5,159,447 issued October 27, 1992.)

0004Additionally, our standards contribution entitled "Timing Recovery for Variable Bit-Rate Video on ATM Networks", MPEG92/396, AVC-315, Paris France, July 3, 1992, we disclosed a decoder which, in the presence of jitter, prevets decoder buffer overflow and underflow by incorporating additional delay and additional decoder buffer capacity. (Also see our U.S. patent 5,287,182 issued February 15, 1994.) Given guaranteed bounds on jitter, decoder buffer overflow and underflow could be eliminated by chosing sufficiently large buffers and sufficiently large jitter delay. This solution can be thought of as having the decoder use its available buffer capacity to shift the so-called Video Buffer Verifier (VBV) to ensure that there is no buffer overflow or underflow. In practice, the decoder buffer may not have enough capacity to absorb both the VBV fluctuations and the jitter. In addition, it incurs additional decoding delay, and is practical only for the elementary bit-stream decoder buffer.

<u>Summary of the Invention</u>

0005The problems of encoder/decoder buffer overflow and underflow encountered when employing actually variable or effectively variable bit-rate channels for communicating encoded video images are overcome by generating information as to the cell delay variation (CDV) at the decoder which includes possible additional decoder buffer capacity that can be employed to accommodate the CDV. This information is transmitted to the encoder which then has the option of adjusting its parameters to avoid encode/decoder buffer overflow and underflow.

0006Advantages of this solution are that it alleviates the need for the decoder to be designed with enough buffer capacity to prevent overflow and underflow given the maximum possible jitter and, also minimizes the decoding delay.

<u>Brief Description of the Drawing</u>

0007<ul id="ul0001" list-style="none" compact="compact"><li>FIG. 1 shows, in simplified block diagram form, a multiplexer unit incorporating aspects of the invention, a variable bit-rate channel and a demultiplexer unit;</li><li>FIG. 2 shows, in simplified block diagram form, details of the multiplexer unit of FIG. 1;</li><li>FIG. 3 shows, in simplified block diagram form, details of the demultiplexer unit of FIG. 1;</li><li>FIG. 4 is a flow chart illustrating the operation of the compute jitter unit of FIG. 3;</li><li>FIG. 5 is a flow chart illustrating the video encoder buffer control when an effectively variable bit-rate channel is constrained; and</li><li>FIG. 6 shows, in simplified block diagram form, details of the video display unit of FIG. 3;</li></ul>

<u>Detailed Description</u>

0008Shown in FIG. 1, in simplified block diagram form, are multipler unit 100, variable bit-rate channel 101 and demultiplexer unit 102, incorporating aspects of the invention. An original video input signal, VIDIN, that includes frames which contain at least one image representation, is supplied as an input of multiplexer 100. Such video signals are well known in the art. Also supplied to multiplexer unit 100 is an corresponding audio signal and J_dec. As described below, Jdec is representative of the cell delay variation at a video decoder in demultiplexer unit 102 and is employed by a video encoder in multiplexer 100 to adjust its parameters to avoid encoder/decoder buffer overflow and underflow. Multiplexer 100 supplies as an output bit-stream CODETR which is transmitted across variable bit-rate channel 101 to demultiplexer unit 102. As described below, the bit-stream supplied as an output from multiplexer 100 also includes audio and other information necessary to properly decode the video and corresponding audio signals at demultiplexer unit 5 102. However, in the discussion below CODETR is assumed to be an encoded version of VIDIN.

0009It is assumed that the variable channel 102 will be able to transmit at a desired bit-rate R_i. In many applications, however, the variable bit-rate channel 101 will be unaware whether or not it can transmit at the desired bit-rate R_i, as in an ethemet network or on an asynchronous transfer mode (ATM) network. In an ATM network, there may be a policing function or usage parameter control (UPC) function that is used to constrain the number of bits entering variable bit-rate channel 101. An aspect of this invention,is to mimic the UPC function at the channel rate control device 208 (FIG. 2) so as to prevent the UPC function in variable rate channel 101 from reducing the transmitted bit-rate below the desired bit-rate R_i. If the variable bit-rate channel 101 is capable of providing a determination as to whether or not it can transmit at the desired bit-rate R_i, then one skilled in the art can use the techniques disclosed in our U.S. patent 5,159,447, cited above, to make a request of variable bit-rate channel 101 and receive a determination from channel 101.

0010In some implementations, variable bit-rate channel 101 itself might perform the removal of the bits, so that its output may not equal its input. In such implementations, Systems encoder, Enc_CR inserter and J_dec extractor 210 (FIG. 2) may only mark bits for deletion and the actual deletion of some or all bits is performed by channel 101 if it determines that the requested bit-rate cannot be made available. Alternatively, variable bit-rate channel 101 can determine on its own which bits should be dropped from the format of bit-stream CODEIM. The bit-rates R_i that can be supplied by channel rate control 208 are not random, but are constrained in accordance with a predetermined function. Such channel constraint functions are well known, e.g., the leaky bucket function. Because of the constrained nature of variable bit-rate channel 101, the transmission at a particular bit-rate R_i can serve to impose limitations on the bit-rate that may be available to variable bit-rate channel 101.

0011It is noted that for brevity and clarity of exposition, only a single channel of video and its corresponding audio are illustrated in the multiplexer unit 100 of FIG. 2 and the demultiplexer unit 102 of FIG. 3. In practice, it will be apparent that a plurality of channels of video and corresponding audio will be employed to be multiplexed for transmission in multiplexer unit 100 and a similar plurality of channels of video and corresponding audio will be obtained via demultiplexer 102.

0012Specifically, multiplexer unit 100, shown in FIG. 2 in simplified form, includes in the video channel insert PTS, DTS unit 201 , video encoder 202 and video data buffer 203. The audio channel includes insert PTS unit 204, audio encoder 205 and audio data buffer 206. Data from video data buffer 203 and audio data buffer 206 are supplied to systems encoder, Enc_CR inserter and J_dec extractor 210. Systems time clock generator 212 includes oscillator 214 and output counter 215. The output of counter 215 is the encoder time clock Enc_TC value which is supplied to insert PTS, STS unit 201, insert PTS unit 204 and systems encoder, Enc_CR inserter and J_dec extractor 210. It is noted that according to the MPEG specifications, the frequency for oscillator 214 shall be 90 kHz ± 50 ppm. Additionally, the frequency drift of oscillator 214 shall not exceed 250 microhertz/second. In this embodiment, counter 215 supplies a 33-bit Enc_TC value as an output that is incremented once for each pulse supplied from oscillator 214.

0013The video input data is first supplied to insert PTS, DTS unit 201 which inserts Presentation Time Stamps (PTS) on some or all of the image representations, i.e., picture frames. If a PTS is inserted on a non-Bidirectionally Predicted Image Representation (nonB-picture), then a Decoding Time Stamp (DTS) must also be inserted. For these image representations, PTS exceeds DTS (in picture periods) by one plus the number of bidirectional image representations to be encoded. The resulting "time stamped" video data then passes to video encoder 202 where it is encoded in known fashion. In turn, the encoded video data is supplied to video data buffer 203 to await transmission via systems encoder, Enc_CR inserter and J_dec extractor 210.

0014The input audio is in the form of so-called Audio Access Units (AAU), which are supplied to insert PTS unit 204 where PTS values are inserted on some or all of the AAUs. Thereafter, the AAUs are supplied to audio encoder 205 where they are digitally encoded in well-known fashion and, thereafter, supplied to audio data buffer 206 to await transmission via systems encoder, Enc_CR inserter and Jdec extractor 210.

0015The PTS/DTSs are employed to control the decoding and displaying of the audio and video data in demultiplexer unit 102 (FIG. 3). The decoding and displaying in response to the PTS/DTSs are such as to maintain synchronism and to avoid overflow and underflow of data buffers in demultiplexer unit 102. Typically, the PTS/DTS values inserted by units 201 and 204 are equal to some constant value plus the instantaneous Enc_TC values from systems clock generator 212.

0016Systems encoder, Enc_CR inserter and J_dec extractor 210 forms packs and packets of data and supplies them asynchronously to a data communications channel. Packs include packets from a plurality of video and audio encoders (not shown). Pack headers also contain encoder clock reference (Enc_CR) values that are the instantaneous values of Enc_TC from systems time clock generator 212. According to MPEG specifications, packs must be transmitted at least every 0.7 seconds. Thus, the Enc_CR values are transmitted at the rate of a few a second. Additionally, the MPEG specifications assume that the channel data rate is constant for the duration of the pack.

0017The packets in a pack contain video or audio data with the PTS/DTS values, if any, moved to the packet header. Hence, for situations of more than one PTS/DTS per packet, all except the first are discarded.

0018R_i is a representation of the number of bits being transmitted to the data channel during the coding of image representation (picture) i which is generated by channel rate control 208 for a current image representation being transmitted, as described below, and supplied to systems encoder, Enc_CR inserter and J_dec extractor 210. Systems encoder, Enc_CR inserter and J_dec extractor 210 then chooses the actual values according to prescribed principles which are known Additionally, as described below, channel rate control 208 and encoder rate control 209 are employed in response to J_dec supplied from demultiplexer 102 and E_i supplied from video encoder 202 to control the parameters of video encoder 202.

0019FIG. 3 shows, in simplified block diagram form, details of demultiplexer unit 102. Specifically, shown are system decoder, Enc_{_}CR extractor and J_dec inserter 301, which supplies video information to be decoded to video data buffer 302 where it is stored awaiting display. Video data from buffer 302 is supplied to video display control 303 and, eventually, to video decoder 304. Similarly, audio data is supplied from systems decoder, Enc_CR extractor and J_dec inserter 301 to audio data buffer 305 where it is stored awaiting display. Audio data from buffer 305 is supplied to audio display control 306 and, eventually, to audio decoder 307. Systems decoder, Enc_CR extractor and J_dec inserter 301 also detects the received system clock reference (Enc_CR) values and passes them to systems time clock generator 308 along with an enable signal which indicates when there is a valid Enc_CR value being received and a start up value.

0020Systems time clock generator 308 is a phase locked loop including phase comparator 310, filter 311, voltage controlled oscillator (VCO) 312 and output counter 314. The enable signal from systems decoder, Enc_CR extractor and J_dec inserter 301 is supplied to enable filter 311 when a valid Enc_CR value is received. The start up value is supplied to set counter 314 to the value of the initial Enc_CR value being received so that the initial audio or video data in the packets of the initial received pack may properly be displayed. The Dec TC output from counter 314 is compared in phase comparator 310 to the received Enc_CR value. Filter 311 smooths the differences Enc_CR _Dec_TC and generates a control voltage which is used to control the frequency of VCO 312.

0021In the absence of transmission jitter delay, and assuming that the center frequency of VCO 312 is close to that of the oscillator 314 in multiplexer unit 100 (FIG. 2), the control signal output of filter 312 will eventually stabilize to a small, nearly constant value. If the frequency of oscillator 214 (FIG. 2) drifts slightly upward, then the Enc_CR values will increase slightly, producing a corresponding increase in the control voltage being supplied to VCO 312 which, in turn, increases its frequency. This, in turn, causes the Dec_TC values being supplied as an output from systems time clock generator 308 to increase slightly, thus tracking the values of the received Enc_CR. Similarly, for a decrease in frequency of oscillator 214, the Enc_CR and Dec_TC values in demultiplexer unit 102 track each other by also decreasing.

0022The amount of smoothing provided by filter 311, i.e., the time duration over which averaging occurs of the Enc_CR_Dec_TC difference signal, determines how fast VCO 312 can respond. A small amount of smoothing by filter 311 can cause a rapid alignment of Dec_TC and Enc_CR. However, if audio and video clocks were also derived from VCO 312, such a rapid alignment could be detrimental to the audio and video quality. In such an instance, significantly more smoothing by filter 311 may be necessary.

0023The gain of filter 311, or equivalently the input sensitivity of VCO 312, also has an effect on the frequency stability of the output pulses being supplied to counter 314. If the gain is large, then a small difference between Enc_CR and STC will cause a large frequency shift in the output of VCO 312. If the gain is too large, then VCO 312 may never stabilize. If the gain is too small, stable operation could occur even if Enc_CR and Dec_TC were not close to each other.

0024The audio and video decoder timing is controlled via Presentation/Decode Time Stamps (PTS/DTS) that are, as indicated above, included in the respective audio or video data. Assuming PTSs occur on every image representation, the DTS (PTS, if no DTS) of the oldest image representation stored in video data buffer 302 passes to video display control 203. Video display control 303 waits until the Dec_TC value increases to the value of the oldest DTS. It then extracts the coded video data for the corresponding image representation from video data buffer 302 and passes it to video decoder 304 for decoding. For each video image representation, input to decoder 304, the image representation is supplied as the video output for display. The video output may be the same image representation as the input (e.g., for B-pictures), or it may be a stored previously decoded picture (e.g., for nonB-pictures).

0025If an image representation does not have a DTS, then video display control 303 computes an extrapolated DTS value by simply adding the nominal picture duration to the DTS of the previous image representation. It is assumed that video decoder 304 always completes decoding each image representation before the next image representation is due to be decoded. When the oldest received (or extrapolated) audio PTS equals Dec_TC, then a next audio access unit in audio data buffer 305 is supplied via audio display control 306 to audio decoder 207. The decoded audio is immediately passed as the audio output corresponding to the video output to be displayed.

0026Details of the video display control 303 is shown in FIG. 6 and described below.

0027If there is substantial transmission jitter delay, then demultiplexer 102 may not operate correctly. This results because filter 311 of systems time clock generator 208 must average over a much longer time interval than in a jitter free situation. However, even if this is the case, a more serious problem is that the data arrival times may be considerably different than was assumed in multiplexer 100 (FIG. 2). When such an occurrence happens, there is a possibility of overflow or underflow of the video and/or audio data buffers in demultiplexer 102.

0028Overflow of the data buffers can be readily avoided by increasing the buffer sizes beyond what was assumed by the encoder 202 in multiplexer 100. Underflow of the data buffers can be alleviated by the addition of a "Jitter Delay" D_j, which causes an extra accumulation of data in the data buffers prior to decoding. The additional stored data provides insurance against underflow. If bounds on jitter delay can be guaranteed, the possibility of underflow of the data buffers can be substantially eliminated by choosing sufficiently large buffers and a sufficiently large value for Dj.

0029In real world situations, the jitter may exceed the capabilities of decoder buffer 302. In such cases, encoder 202 (FIG. 2) inmultiplexer 100 can handle the effects of the jitter if it knows what it is. To this end, compute jitter unit 317 (FIG. 3) computes an estimate of the peak J_max using as an input the discrepancy between the received Enc_CR values and the decoder Dec_ TC values. The difference value Enc_CR_Dec_TC from phase comparator 310 is supplied to compute jitter unit 317.

0030An example of jitter computation is shown if the flow chart in FIG. 4. The jitter computation in begun via start step 401. Then, step 402 initializes J_max to a nominal jitter value, perhaps a value larger than any expected value. Then, step 403 reads Enc_CR_Dec_TC from comparator 310. Step 405 tests to determine if J &lt; J_max -Δ, where Δ is some safety factor. If the test result in step 405 is NO control is transferred to step 406. If the test result in step 405 is YES, J &lt; _max -Δ, which indicates that J_max is too large. Then,step 407 reduces J_max,for example, by Δ/2 and control is transferred to step 406. Step 406 tests to determine if J_max &lt; J. If the test result in step 406 is YES, J_max &lt; J which indicates that J_max is too small. Then, step 408 increases the value of J_max is increased, for example, to the value of J. Thereafter, step 409 outputs the value of J_max as J_dec which is supplied to system decoder, Enc CR extractor and J_dec inserter 301 (FIG. 3) and control is returned to step 403. In turn, system decoder, Enc_CR extractor and J_dec inserter 301 supplies J_dec to channel 101.

0031The entire process from the initial supplying as an output of a frame of encoded bit-stream CODEIM from encoder buffer 203 (FIG. 2) until the corresponding version of that frame in CODETR', which is received by decoder buffer 302 (FIG. 3) and is supplied to video decoder 303 takes LT seconds, exclusive of any delay in variable bit-rate channel 101. This time period of LT seconds, where T is the time period of one unencoded video frame and L is a system delay parameter not less than one (1) and not necessarily an integer, is arbitrarily selected by the implementor. After the first bit of bit-stream CODETR is received by decoder buffer 302, exactly LT seconds are waited by decoder buffer 302 before it delivers the first frame to video decoder 304. For clarity of exposition, the embodiments of the invention presented herein will be described as if L were an integer. Embodiments of the invention in which L is not an integer will be readily apparent to those skilled in the art.

0032Encoder buffer 203 and decoder buffer 302 each have predetermined fixed maximum sizes, B<maths id="math0001" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>e</mtext></mrow><mrow><mtext>max</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0001.tif" /></maths> and BB<maths id="math0002" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d</mtext></mrow><mrow><mtext>max</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0002.tif" /></maths> respectively, which are known to video encoder 202, channel rate control 208 and encoder rate control 209 a priori. In addition, the value of the delay factor L is known a priori by video encoder 202. channel rate control 208, encoder rate control 209 and decoder buffer 302. In some situations, L is transmitted directly to demultiplexer 102. in bit stream CODETR, and the extraction of this delay value is described below in relation to video display control unit 203 (FIG. 2).

0033Any method of encoding images that allows the number of bits produced as the encoded representation of each image to be specified may be employed by video encoder 202. Such methods are well known in the art. Encoder rate control 209 supplies to video encoder 202 signal RANGE. Signal RANGE is a representation of the range of acceptable numbers of bits that may be produced when encoding each frame of original video signal VIDIN into bit-stream CODEIM. In this example, the range constraining the number of bits that video encoder 202 may produce, is determined so that neither encoder buffer 203 nor decoder buffer 302 will overflow or underflow. The method of determining this range will be described below. Video encoder 202 encodes each frame of signal VIDIN and, thereby, produces a portion of bit-stream CODEIM that represents frame i and actually contains E_i bits in a frame period [(i-1)T,iT]. E_i is a number that falls within the range given by the encoder rate control 209 in signal RANGE for frame i. The frame number index i designates each frame with respect to a predetermined starting frame. The E_i bits, which are a portion of bit-stream CODEIM, are supplied as an input to encoder buffer 203 and stored therein until it is time for them to be transmitted.

0034A selection of a set of estimated channel bit-rates R_i, R_i+1, . . ., R_i+L-1 by channel rate control 208 is controlled by the need to prevent encoder buffer 203 and decoder buffer 302 from overflowing or underflowing. This selection is also controlled by the bit-rate constraint on variable bit-rate channel 101. This bit-rate constraint on variable bit-rate channel 101 may be time-varying. This selection is controlled also by the estimate of the received jitter, J_dec, received by channel rate controller from compute jitter unit 217 (FIG. 2). Additional factors that may be considered in the selection of the estimated channel bit-rates are the cost of transmitting each bit and the fact that for some channel constraints (for example, the well known leaky bucket constraint) the channel bit-rate could be conserved at an earlier time so as to produce a less stringent channel rate constraint later. This might be desirable if a frame of original video signal VIDIN that has been encoded using intraframe techniques will need to be transmitted soon.

0035Encoder rate control 209 receives as inputs the actual bit-rate R_i and the estimated potential future channel bit-rate values for the next L - 1 frames R_i+1, . . . , R_i+L-1 that are selected, in a manner to be described, from channel rate control 208. Encoder rate control 209 also receives as an input the estimate of received jitter, J_dec, from compute jitter unit 217 (FIG. 2). Encoder rate control 209 also receives as an input the actual number of bits E_i that are supplied as an output by video encoder 202 for each frame period i. The range for E_i that is supplied as an output by encoder rate control 209 is determined, in a manner to be described, such that neither encoder buffer 203 nor decoder buffer 302 will overflow or underflow.

0036Channel rate control 208 determines the actual current channel bit-rate R_i for channel 101 and the number of bits E_i that is actually supplied by video encoder 202 as an output in the current frame period. An estimated channel bit-rate R_i+L, which describes the number of bits that are desired to be transmitted across channel 101 in the frame period L frames hence, is selected by channel rate control 208. Each of these selected channel bit-rates are, as noted above, supplied as an input to encoder rate control 209 and to systems encoder, Enc_CR inserter and J_dec extractor 210.

0037The following theoretical exposition should prove useful to one skilled in the art in the understanding of the detailed operation of channel rate control 208 and encoder rate control 209, particularly when variable bit-rate channel 101 is constrained, in an exemplary fashion, by the leaky bucket constraint. E(t) is defined to be the number of bits (or bytes or packets) output by the encoder at time t. The instantaneous bit-rate at any given time of variable bit-rate channel 101, R(t), is variable. B^e (t) and B^d(e) (t) are the instantaneous fullnesses of encoder and decoder buffers 203 and 302, respectively, as computed by the encoder, assuming no jitter delay. Encoder buffer 203 and decoder buffer 302 each have predetermined fixed maximum sizes, B<maths id="math0003" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>e</mtext></mrow><mrow><mtext>max</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0003.tif" /></maths> and BB<maths id="math0004" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d</mtext></mrow><mrow><mtext>max</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0004.tif" /></maths>, respectively. Given BB<maths id="math0005" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>e</mtext></mrow><mrow><mtext>max</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0005.tif" /></maths>, multiplexer 100 is designed to ensure that encoder buffer 203 never overflows, i.e.,<maths id="math0006" num="(1)"><math display="block"><mrow><msup><mrow><mtext>0 ≤ B</mtext></mrow><mrow><mtext>e</mtext></mrow></msup><msubsup><mrow><mtext>(t) ≤ B</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>e</mtext></mrow></msubsup><mtext> ∀t.</mtext></mrow></math><img file="EP0723345A2_D0006.tif" /></maths> Conditions which are required to be placed on bit-rates E_i and R_i to ensure that decoder buffer 302 never overflows or underflows, i.e.,<maths id="math0007" num="(2)"><math display="block"><mrow><msup><mrow><mtext>0 ≤ B</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msup><msubsup><mrow><mtext> (t) ≤ B</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>d</mtext></mrow></msubsup><mtext> ∀t,</mtext></mrow></math><img file="EP0723345A2_D0007.tif" /></maths> are described herein.

0038The problem is made discrete by defining E_i(i = 1,2,...) to be the number of bits in the interval [(i - 1)T,iT], where T is the duration of one unencoded frame of unencoded video signal VIDIN. Therefore,<maths id="math0008" num=""><img file="EP0723345A2_D0008.tif" /></maths> Similarly, R_i is the number of bits that are transmitted over variable bit-rate channel 101 during the ith frame period,<maths id="math0009" num=""><img file="EP0723345A2_D0009.tif" /></maths> which is the actual bit-rate granted by variable bit-rate channel 101.

0039In this description, it has been assumed that the discretization of the transmission bit-rate over variable bit-rate channel 101 is identical to that of the encoding bit-rate E_i. Other discretizations will be apparent to those skilled in the art.

0040Encoder buffer 203 receives bits at bit-rate E(t) from video encoder 202 and supplies as an output bits of encoded video bit-stream CODEIM at a bit-rate R(t). Therefore, assuming that encoder buffer 203 and decoder buffer 302 are empty prior to start-up at time t = 0<maths id="math0010" num=""><img file="EP0723345A2_D0010.tif" /></maths> the encoder buffer fullness after encoding frame i is<maths id="math0011" num=""><img file="EP0723345A2_D0011.tif" /></maths> This can be written explicitly as<maths id="math0012" num=""><img file="EP0723345A2_D0012.tif" /></maths> or recursively as<maths id="math0013" num="(8)"><math display="block"><mrow><msubsup><mrow><mtext>BT = B</mtext></mrow><mrow><mtext>i-1</mtext></mrow><mrow><mtext>e</mtext></mrow></msubsup><msub><mrow><mtext> + E</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msub><mrow><mtext> - R</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><mtext>.</mtext></mrow></math><img file="EP0723345A2_D0013.tif" /></maths>

0041After demultiplexer 102 begins to receive data in decoder buffer 302, it waits LT seconds before starting to decode, as described above. Again, it is assumes for clarity purposes that L is an integer, although this is not necessary.

0042A new time index τ, is defined for demultiplexer 102. τ is zero when decoding begins. Therefore,<maths id="math0014" num="(9)"><math display="block"><mrow><mtext>t = τ + LT + channel.delay</mtext></mrow></math><img file="EP0723345A2_D0014.tif" /></maths> where channel. delay is the transmission delay time across variable bit-rate channel 101. Multiplexer unit 100 can calculate the initial fullness of decoder buffer 302, B^d(e)(0) (when τ = 0), if L is predetermined or transmitted explicitly as a parameter to demultiplexer 102. The initial fullness of decoder buffer 302 is given by<maths id="math0015" num=""><img file="EP0723345A2_D0015.tif" /></maths> The fullness of decoder buffer 302 at time τ = iT can be determined by<maths id="math0016" num="(11)"><math display="block"><mrow><msubsup><mrow><mtext>B</mtext></mrow><mrow><mtext>i</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msubsup><mrow><mtext> = B</mtext></mrow><mrow><mtext>i-1</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msub><mrow><mtext> + R</mtext></mrow><mrow><mtext>L+i</mtext></mrow></msub><msub><mrow><mtext> - E</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><mtext>.</mtext></mrow></math><img file="EP0723345A2_D0016.tif" /></maths><maths id="math0017" num=""><img file="EP0723345A2_D0017.tif" /></maths>

0043In the arrangement disclosed in our prior U.S. patent 5,159,447 issued October 27, 1992, the encoder ensured correct operation of the decoder buffers, assuming no delay jitter. However, if delay jitter is present, the encoder cannot know the exact fullness of the decoder buffer. Here, we extend the buffer dynamic equations to include the possibility of short-term jitter, where data arrive either early or late to the decoder. The maximum delay jitter is defined to be J_max. We assume the decoder clock is stable and accurate. We assume throughout that the jitter is much less than the buffering delay LT.

0044Because of arrival jitter, the encoder cannot know exactly the fullness of the decoder buffer. The encoder can, however, compute a nominal buffer fullness that corresponds to the case when there is no jitter. The actual buffer fullness will differ from this because of the jitter.

0045Let B<maths id="math0018" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d(d)</mtext></mrow><mrow><mtext>i</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0018.tif" /></maths> be the actual decoder buffer fullness at time i and B<maths id="math0019" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d(e)</mtext></mrow><mrow><mtext>i</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0019.tif" /></maths> be the nominal buffer fullness as computed by the encoder, assuming no jitter. Let R_L+i be the number of bits which arrive at the decoder buffer during the i-th interval, and let E_i be the number of bits that leave the decoder buffer during the i-th interval.

0046In general, the encoder computes B<maths id="math0020" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d(e)</mtext></mrow><mrow><mtext>i</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0020.tif" /></maths> as described in Equation (11). However, if data arrive too late, the decoder buffer may be less full than the encoder expected, in which case, J_max R_L+i less data will have been received by the decoder buffer. So the actual decoder buffer fullness, B<maths id="math0021" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d(d)</mtext></mrow><mrow><mtext>i</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0021.tif" /></maths>, will differ from that expected by the encoder, B<maths id="math0022" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d(e)</mtext></mrow><mrow><mtext>i</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0022.tif" /></maths>:<maths id="math0023" num="(13)"><math display="block"><mrow><msubsup><mrow><mtext>B</mtext></mrow><mrow><mtext>i</mtext></mrow><mrow><mtext>d(d)</mtext></mrow></msubsup><msubsup><mrow><mtext> ≥B</mtext></mrow><mrow><mtext>i</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msub><mrow><mtext>-J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>L+i.</mtext></mrow></msub></mrow></math><img file="EP0723345A2_D0023.tif" /></maths>

0047Similarly, if data arrives too early, the decoder buffer may be fuller than expected. In this case, the fill-rate of the decoder buffer is now R_L+i+1.<maths id="math0024" num="(14)"><math display="block"><mrow><msubsup><mrow><mtext>B</mtext></mrow><mrow><mtext>i</mtext></mrow><mrow><mtext>d(d)</mtext></mrow></msubsup><msubsup><mrow><mtext>≤B</mtext></mrow><mrow><mtext>i</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msub><mrow><mtext> + J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>L+i+1.</mtext></mrow></msub></mrow></math><img file="EP0723345A2_D0024.tif" /></maths>

0048Equations (13) and (14) can be combined to obtain bounds on the actual decoder buffer fullness given the decoder buffer fullness computed at the encoder and the actual maximum jitter.

0049If the encoder knows the correct value of J_max, it can ensure that the decoder's buffers will not underflow or overflow by satisfying the following Equation:<maths id="math0025" num="(15)"><math display="block"><mrow><msub><mrow><mtext>J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>L+1</mtext></mrow></msub><msubsup><mrow><mtext>≤B</mtext></mrow><mrow><mtext>i</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msubsup><mrow><mtext>≤B</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>d</mtext></mrow></msubsup><msub><mrow><mtext>-J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>L+i+1.</mtext></mrow></msub></mrow></math><img file="EP0723345A2_D0025.tif" /></maths>

0050If the future transmission bit-rate R_L+i+1 is unknown at the encoding time i, an upper bound can be used in its place. The more accurately R_L+i+1 is known, the more fluctuations can be allowed in the encoded bit-rate and, hence, better video quality can be produced.

0051In a real system, the encoder may not know the correct value of J_max. The decoder must compute and send this information to the encoder. It is possible that the decoder may not observe (and hence be able to transmit) the maximum jitter to the encoder until after it occurs. The decoder may also not detect that jitter is occurring in the unlikely event that all packets with Enc_CRs are jittered identically. Furthermore, there is some transmission delay between the decoder and the encoder. Therefore, decoder buffer overflow or underflow could occur before the encoder receives the correct value of J_max.

0052One solution is to have the decoder start with a large estimate of jitter and reduce it over time. In addition, if the encoder does adjust its coding parameters based upon available jitter information, the probability of overflow or underflow is minimized. Furthermore, if the decoder can estimate the maximum jitter exactly, the encoder can restrict the interval during which overflow and decoder problems can occur to the transmission delay interval.

0053The preceding analysis assumed that the decoder clock was stable and accurate. However, even if the decoder clock is stable with the correct frequency, it could have a constant offset of as much as J_max. For example, in the pathological case that all packets containing encoder clock references, Enc_CR, are all delayed by J_max, the decoder's time clock, D_TC, would be slow by J_max. If the clock is slow, the decoder would remove data from its buffer later than expected by the encoder, and if the clock is slow, data would be removed too soon. If this is taken into account, Equation (15) becomes<maths id="math0026" num="(16)"><math display="block"><mrow><msub><mrow><mtext>2J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>L+i</mtext></mrow></msub><msubsup><mrow><mtext>≤B</mtext></mrow><mrow><mtext>i</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msub><mrow><mtext>≤B</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msubsup><mrow><mtext>-2J</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>d</mtext></mrow></msubsup><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>L+i+1.</mtext></mrow></msub></mrow></math><img file="EP0723345A2_D0026.tif" /></maths> These differences could be accommodated by using an overestimate of the jitter value in the encoder.

0054Conditions necessary to prevent underflow and overflow of encoder buffer 203 and decoder buffer 302 are presented for a general variable bit-rate channel. To prevent encoder buffer overflow and underflow, it is required from Equation (1) and Equation (8) that<maths id="math0027" num="(17)"><math display="block"><mrow><msubsup><mrow><mtext>0 ≤ B</mtext></mrow><mrow><mtext>i-1</mtext></mrow><mrow><mtext>e</mtext></mrow></msubsup><msub><mrow><mtext> + E</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msub><mrow><mtext> - R</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msubsup><mrow><mtext> ≤ B</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>e</mtext></mrow></msubsup></mrow></math><img file="EP0723345A2_D0027.tif" /></maths><maths id="math0028" num="(18)"><math display="block"><mrow><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msubsup><mrow><mtext> - B</mtext></mrow><mrow><mtext>i-1</mtext></mrow><mrow><mtext>e</mtext></mrow></msubsup><msubsup><mrow><mtext> ≤ B</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>e</mtext></mrow></msubsup><msub><mrow><mtext> + R</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msubsup><mrow><mtext> - B</mtext></mrow><mrow><mtext>i-1</mtext></mrow><mrow><mtext>e</mtext></mrow></msubsup></mrow></math><img file="EP0723345A2_D0028.tif" /></maths> which represents a constraint on the number of bits E_i that may be supplied for each encoded frame of bit-stream CODEIM for a given channel bit-rate R_i granted by variable bit-rate channel 101. For example, when variable bit-rate channel 101 has a constant rate, multiplexer 100 can prevent encoder buffer 203 from overflowing or underflowing by varying the quality of the coding employed by video encoder 202. If multiplexer 100 determines that encoder buffer 203 is approaching fullness, the bit-rate of bit-stream CODEIM being supplied as an input by video encoder 202 to encoder buffer 203 is reduced. Such a reduction in the bit-rate of bit-stream CODEIM may be achieved by reducing the quality of the encoding performed by video encoder 202. One method of reducing the encoding quality is to employ a coarser quantization step size. Such methods of reducing encoding quality are well known in the art. Conversely, if encoder buffer 203 threatens to underflow, video encoder 202 can generate bit-stream CODEIM with a higher bit-rate, either by increasing the quality of encoding performed by video encoder 202, or by supplying as an output stuffing bits that are consistent with the coding syntax and can be discarded by video decoder 304.

0055To prevent decoder buffer 302 from overflowing or underflowing the following constraints on the bit-rate of video encoder 202 for a given channel bit-rate can be imposed:<maths id="math0029" num="(19)"><math display="block"><mrow><msub><mrow><mtext>J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>i+L</mtext></mrow></msub><msubsup><mrow><mtext>0 ≤ B</mtext></mrow><mrow><mtext>i-1</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msub><mrow><mtext>+R</mtext></mrow><mrow><mtext>i+L</mtext></mrow></msub><msub><mrow><mtext> - E</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msubsup><mrow><mtext> ≤ B</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>d</mtext></mrow></msubsup><msub><mrow><mtext>-J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>1+L+1</mtext></mrow></msub></mrow></math><img file="EP0723345A2_D0029.tif" /></maths><maths id="math0030" num="(20)"><math display="block"><mrow><msub><mrow><mtext>-J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>i+L</mtext></mrow></msub><msub><mrow><mtext> + R</mtext></mrow><mrow><mtext>i+L</mtext></mrow></msub><msubsup><mrow><mtext> + B</mtext></mrow><mrow><mtext>i-1</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><mspace linebreak="newline" /><msub><mrow><mtext>≥ E</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><msub><mrow><mtext> ≥ R</mtext></mrow><mrow><mtext>i+L</mtext></mrow></msub><msubsup><mrow><mtext> + B</mtext></mrow><mrow><mtext>i-1</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msubsup><mrow><mtext> - B</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>d</mtext></mrow></msubsup><msub><mrow><mtext>+J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>i+L+1</mtext></mrow></msub><mtext>.</mtext></mrow></math><img file="EP0723345A2_D0030.tif" /></maths>

0056Alternatively, the number of bits per frame can be unconstrained thereby yielding the following constraint on the channel bit-rate R_i :<maths id="math0031" num=""><img file="EP0723345A2_D0031.tif" /></maths> which is the same as, for i > L<maths id="math0032" num=""><img file="EP0723345A2_D0032.tif" /></maths> where E_i-L - BB<maths id="math0033" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d(e)</mtext></mrow><mrow><mtext>i-L-1</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0033.tif" /></maths> ≤ (1 -J_max) R_i specifies the condition in which underflow of decoder buffer 302 occurs and<maths id="math0034" num=""><img file="EP0723345A2_D0034.tif" /></maths> specifies the condition in which overflow of decoder buffer 302 occurs. Therefore, there is a restriction on the channel bit-rate R_i that is dependent upon the output supplied by video encoder 202 L frames ago.

0057The right side of Equation (22) can be used to obtain an additional constraint on R_i :<maths id="math0035" num="(23)"><math display="block"><mrow><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>i+1</mtext></mrow></msub><msub><mrow><mtext>J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>≤E</mtext></mrow><mrow><mtext>i-L</mtext></mrow></msub><msubsup><mrow><mtext>+[B</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>d</mtext></mrow></msubsup><msubsup><mrow><mtext>-B</mtext></mrow><mrow><mtext>i-L-1</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msub><mrow><mtext>]-(1-J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>)R</mtext></mrow><mrow><mtext>i</mtext></mrow></msub></mrow></math><img file="EP0723345A2_D0035.tif" /></maths> or<maths id="math0036" num=""><img file="EP0723345A2_D0036.tif" /></maths>

0058A counter (not shown) indicating the fullness of an imaginary buffer is maintained by variable bit-rate channel 101 when it is constrained by a leaky-bucket function. R_i bits for frame period i are supplied as an input to the imaginary buffer (henceforth called the "bucket" herein). The bit-rate at which bits are supplied as an output from the bucket is <maths id="math0037" num=""><math display="inline"><mrow><mover accent="true"><mrow><mtext>R</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0723345A2_D0037.tif" /></maths> bits per frame period. The bucket size is N_max. The instantaneous bucket fullness after frame i has been supplied to variable bit-rate channel 101 is<maths id="math0038" num=""><img file="EP0723345A2_D0038.tif" /></maths> To ensure that high-priority are not dropped, the bit-rate R_i must be such that the bucket never overflows, i.e., N_i ≤ N_max ∀_i, or<maths id="math0039" num=""><img file="EP0723345A2_D0039.tif" /></maths> Equation (26) defines the leaky-bucket constraint on the bit-rate that can be supplied as an input to variable bit-rate channel 101.

0059However, the leaky-bucket constraint is not the only constraint. In fact, preventing overflow of decoder buffer 302 can impose a stronger constraint on the bit-rate that can be supplied as an input to variable bit-rate channel 101. In particular, the right side of the decoder bit-rate constraint specified in Equation (22) may be stricter than the leaky-bucket bit-rate constraint of Equation (26).

0060The upper bound on R_i in the right side of Equation (22) depends on R_i+1, which may not be known before estimating R_i. However, an upper bound on R_i+1 can be obtained from the leaky bucket constraint:<maths id="math0040" num="(27)"><math display="block"><mrow><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>i+1</mtext></mrow></msub><msub><mrow><mtext>≤N</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>-N</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><mtext>+</mtext><mover accent="true"><mrow><mtext>R</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0723345A2_D0040.tif" /></maths> or<maths id="math0041" num="(28)"><math display="block"><mrow><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>i+1</mtext></mrow></msub><msub><mrow><mtext>≤N</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>-N</mtext></mrow><mrow><mtext>i-1</mtext></mrow></msub><msub><mrow><mtext>-R</mtext></mrow><mrow><mtext>1</mtext></mrow></msub><mtext>+2</mtext><mover accent="true"><mrow><mtext>R</mtext></mrow><mo>¯</mo></mover><msubsup><mrow><mtext>+R</mtext></mrow><mrow><mtext>i+1</mtext></mrow><mrow><mtext>UB</mtext></mrow></msubsup><mtext>.</mtext></mrow></math><img file="EP0723345A2_D0041.tif" /></maths>

0061Substituting into the right side of (22), R_i can be chosen to satisfy<maths id="math0042" num="(29)"><math display="block"><mrow><msub><mrow><mtext>(1-J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>)R</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msub><mrow><mtext>≤E</mtext></mrow><mrow><mtext>i-L</mtext></mrow></msub><msubsup><mrow><mtext>+B</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>d</mtext></mrow></msubsup><msubsup><mrow><mtext>-B</mtext></mrow><mrow><mtext>i-L-1</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msub><mrow><mtext>-J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>(N</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>-N</mtext></mrow><mrow><mtext>i-1</mtext></mrow></msub><msub><mrow><mtext>-R</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><mtext>+2</mtext><mover accent="true"><mrow><mtext>R</mtext></mrow><mo>¯</mo></mover><mtext>)</mtext></mrow></math><img file="EP0723345A2_D0042.tif" /></maths><maths id="math0043" num="(30)"><math display="block"><mrow><msub><mrow><mtext>(1-2J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>)R</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msub><mrow><mtext>≤E</mtext></mrow><mrow><mtext>i-L</mtext></mrow></msub><msubsup><mrow><mtext>+B</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>d</mtext></mrow></msubsup><msubsup><mrow><mtext>-B</mtext></mrow><mrow><mtext>i-L-1</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msub><mrow><mtext>-J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>(N</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>-N</mtext></mrow><mrow><mtext>i-1</mtext></mrow></msub><mtext> +2</mtext><mover accent="true"><mrow><mtext>R</mtext></mrow><mo>¯</mo></mover><mtext>)</mtext></mrow></math><img file="EP0723345A2_D0043.tif" /></maths> And so decoder buffer overflow and underflow can be avoided, when the jitter is less than J_max, if<maths id="math0044" num=""><img file="EP0723345A2_D0044.tif" /></maths> Note that Equation (31) assumes J_max &lt;1/2 seconds, which should always be the case in practice.

0062The principles of buffer control for exemplary multiplexer 100 when variable bit-rate channel 101 having a leaky bucket bit-rate constraint are illustrated, in flow chart form, in FIG. 5. Accordingly, the routine is entered via step 500 upon the arrival of the first frame of original video signal VIDIN. Step 501 performs variable initialization at time t = 0 prior to encoding frame i = 1. <maths id="math0045" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>e</mtext></mrow><mrow><mtext>i</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0045.tif" /></maths> representing the fullness of encoder buffer 203 at time t = iT, <maths id="math0046" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d(e)</mtext></mrow><mrow><mtext>-L</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0046.tif" /></maths> representing the nominal fullness of decoder buffer 302 as computed by the encoder, assuming no jitter, at time τ = iT, and N_i representing the fullness of the leaky bucket at time t = iT are all initialized to zero (0). In step 502, estimates of bit-rates available from variable bit-rate channel 101 for present frame i of original video signal VIDIN and for the next L future frames of original video signal VIDIN are determined. Also, upper bounds on future bit-rates are determined. Additionally, leaky bucket fullnesses and nominal fullnesses of decoder buffer 302 for the same next L future frames are estimated. For estimating channel bit-rates inequalities Equation (22) and Equation (26) are employed, where the estimate J_dec replaces the actual jitter value J_max. For k ≤ 0, E_k = 0. Leaky bucket fullnesses and fullnesses of decoder buffer 302 are determined from, respectively, Equation (25) and Equation (12). Rewriting them, for j = i,i+1,..., i+L-1 yields<maths id="math0047" num=""><img file="EP0723345A2_D0047.tif" /></maths> where the left inequality of Equation (32), (E_j-L -B <maths id="math0048" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d(e)</mtext></mrow><mrow><mtext>j-L-1</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0048.tif" /></maths>)/(1-J_dec)≤R_j,<maths id="math0049" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow /><mrow /></mfrac></mrow></math><img file="EP0723345A2_D0049.tif" /></maths> indicates an underflow condition for decoder buffer 302 and the right inequality of Equation (32),<maths id="math0050" num=""><img file="EP0723345A2_D0050.tif" /></maths> indicates an overflow condition for decoder buffer 302,<maths id="math0051" num="(33)"><math display="block"><mrow><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>j</mtext></mrow></msub><msub><mrow><mtext> ≤ N</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext> - N</mtext></mrow><mrow><mtext>j-1</mtext></mrow></msub><mtext> + </mtext><mover accent="true"><mrow><mtext>R</mtext></mrow><mo>¯</mo></mover><mtext>,</mtext></mrow></math><img file="EP0723345A2_D0051.tif" /></maths><maths id="math0052" num=""><img file="EP0723345A2_D0052.tif" /></maths><maths id="math0053" num=""><math display="block"><mrow><msub><mrow><mtext>N</mtext></mrow><mrow><mtext>j</mtext></mrow></msub><msub><mrow><mtext> = N</mtext></mrow><mrow><mtext>j-1</mtext></mrow></msub><msub><mrow><mtext> + R</mtext></mrow><mrow><mtext>j</mtext></mrow></msub><mtext> - </mtext><mover accent="true"><mrow><mtext>R</mtext></mrow><mo>¯</mo></mover><mtext>,</mtext></mrow></math><img file="EP0723345A2_D0053.tif" /></maths> and<maths id="math0054" num=""><math display="block"><mrow><msubsup><mrow><mtext>B</mtext></mrow><mrow><mtext>j-L</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msubsup><mrow><mtext> = B</mtext></mrow><mrow><mtext>j-L-1</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msub><mrow><mtext> + R</mtext></mrow><mrow><mtext>j</mtext></mrow></msub><msub><mrow><mtext> - E</mtext></mrow><mrow><mtext>j-L</mtext></mrow></msub><mtext>.</mtext></mrow></math><img file="EP0723345A2_D0054.tif" /></maths> By substituting the above noted j values in Equation (36) the nominal fullness values of step 502 are obtained.

0063In most cases, for j &lt; i+L-1 we can simplify and reuse previous estimates, in which case evaluations need be made only for j = i+L-1. However, as shown below, B<maths id="math0055" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d(e)</mtext></mrow><mrow><mtext>i-L+1</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0055.tif" /></maths> can change and, therefore, reevaluation of all estimates can be desirable. In general, a value of R_j ≥ 0 that is equal to whichever upper bound of Equation (32), Equation (33) and Equation (34) is smaller is a good choice. However, if it is know that a frame with a large number of bits is imminent, smaller values may be desirable. Smaller values of R_j may be also be desirable if the channel were a video disk so as to employ as few bits as possible. When smaller values of R_j are chosen, they are chosen to be larger than the lower bound of Equation (32) to prevent decoder buffer underflow.

0064For i ≤ L, no frames are yet being decoded by video decoder 304, and decoder buffer 302 is merely filling. In general, the sum of R₁,. . ., R_L should be chosen to exceed the expected bit-rate of the first few frames of bit-stream CODELM in order to avoid underflow of decoder buffer 302.

0065Step 504 estimates an upper bound (UB) on R_i+L by using the leaky bucket constraint specified in Equation (33)<maths id="math0056" num="(37)"><math display="block"><mrow><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>i+L</mtext></mrow></msub><msub><mrow><mtext> ≤ UB R</mtext></mrow><mrow><mtext>i+L</mtext></mrow></msub><msub><mrow><mtext> = N</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext> - N</mtext></mrow><mrow><mtext>i+L-1</mtext></mrow></msub><mtext> + </mtext><mover accent="true"><mrow><mtext>R</mtext></mrow><mo>¯</mo></mover></mrow></math><img file="EP0723345A2_D0056.tif" /></maths> In step 505 an upper bound on E_i is determined by encoder rate control 209 in. This upper bound is determined by employing constraints on the overflow of encoder buffer 203 from inequality Equation (18) and underflow of decoder buffer 302 from inequality Equation (19)<maths id="math0057" num="(38)"><math display="block"><mrow><msub><mrow><mtext>E</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msubsup><mrow><mtext> ≤ B</mtext></mrow><mrow><mtext>max</mtext></mrow><mrow><mtext>e</mtext></mrow></msubsup><msub><mrow><mtext> + R</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msubsup><mrow><mtext> - B</mtext></mrow><mrow><mtext>i-1</mtext></mrow><mrow><mtext>e</mtext></mrow></msubsup></mrow></math><img file="EP0723345A2_D0057.tif" /></maths><maths id="math0058" num="(39)"><math display="block"><mrow><msub><mrow><mtext>E</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msub><mrow><mtext> ≤ UBR</mtext></mrow><mrow><mtext>i+L</mtext></mrow></msub><msubsup><mrow><mtext> + B</mtext></mrow><mrow><mtext>i-1</mtext></mrow><mrow><mtext>d(e)</mtext></mrow></msubsup><msub><mrow><mtext>-J</mtext></mrow><mrow><mtext>max</mtext></mrow></msub><msub><mrow><mtext>UBR</mtext></mrow><mrow><mtext>i+L</mtext></mrow></msub><mtext>.</mtext></mrow></math><img file="EP0723345A2_D0058.tif" /></maths> The minimum of these two upper bounds on E_i is supplied as an output by encoder rate control 209 as signal RANGE to video encoder 202.

0066Step 506 begins encoding of frame i by video encoder 202 subject to the bounds on E_i and the transmission of a frame from encoder buffer 203 at bit-rate R_i. In step 507 the encoding of frame i by video encoder 202 and transmission through variable bit-rate channel 101 of the frame supplied from encoder buffer 203 are completed. Video encoder 202 then supplies the actual value of E_i as an input to channel bit-rate control 208 and encoder rate control 209. The actual value of E_i is used, in conditional branch point 508, to evaluate inequality Equation (17) and, thereby, determine if underflow of encoder buffer 203 occurred during the coding of frame i. If the test result in step 508 is NO, control is passed to step 510. If the test result in step 508 is YES, channel rate control 208 and channel interface 80 terminate transmission early, and control is passed to step 509 which calculates the new lower value of R_i as<maths id="math0059" num="(40)"><math display="block"><mrow><msub><mrow><mtext>R</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><msubsup><mrow><mtext> = B</mtext></mrow><mrow><mtext>i-1</mtext></mrow><mrow><mtext>e</mtext></mrow></msubsup><msub><mrow><mtext> + E</mtext></mrow><mrow><mtext>i</mtext></mrow></msub><mtext>.</mtext></mrow></math><img file="EP0723345A2_D0059.tif" /></maths> Thereafter, control is passed to step 510. Step 510 uses actual values of E_i and R_i to compute actual values of B<maths id="math0060" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>e</mtext></mrow><mrow><mtext>i</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0060.tif" /></maths>,N_i and B<maths id="math0061" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d(e)</mtext></mrow><mrow><mtext>i-L</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0061.tif" /></maths> using Equations (8), (35) and (36), respectively. In step 511 i is incremented in preparation for encoding the next frame. If step 509 was not executed, then all values previously estimated in step 502 are still valid, and for the next frame only an estimate for the latest time interval need be evaluated. If step 509 was executed, then B<maths id="math0062" num=""><math display="inline"><mrow><mfrac linethickness="0"><mrow><mtext>d(e)</mtext></mrow><mrow><mtext>i-L</mtext></mrow></mfrac></mrow></math><img file="EP0723345A2_D0062.tif" /></maths> has changed, and all values previously estimated in step 502 must be reevaluated.

0067FIG. 6 shows, in simplified block diagram form, details of a display control unit which may be employed for either video display control 303 or audio display control 306. It is noted that for audio display control 306 only PTS values are employed. Specifically shown is depacketizer and DTS (PTS) extractor 601 which strips off packet headers and detects the presence of DTS and/or PTS. (The first packet received is required to contain a PTS.) If PTS/DTS is detected (set DTS=PTS if no DTS), then switch 602 is set to the A position via a signal from depacketizer and DTS (PTS) extractor 601, and the DTS value is supplied to the A input of switch 602. In turn, the supplied DTS value is supplied as an output by switch 602 as DTS 1 to DTS (PTS) register 603.

0068After stripping off packet headers, depacketizer and DTS (PTS) extractor 601 passes the remaining packet data upon request to presentation unit (PU) detector 607. For video display control 303, a PU is an image representation (i.e., picture). For audio display control 306, a PU is an Audio Access Unit (AAU). At the start of every first PU (and no others), presentation unit detector 607 outputs a clock pulse that clocks DTS (PTS) register 603, thus causing its input DTS1 to be read, stored and then supplied as an output as DTS2 to subtractor 604 and adder 606. Then, presentation unit detector 607 waits. Dec_TC values are supplied to subtractor 604, where they are compared with the stored DTS2 values. The resulting difference Dec_TC-DTS2 is supplied to zero detector 605. When Dec_TC increases to a value equal to DTS2, enable the next PU signal is supplied as an output to presentation unit detector 607 and the set B input to switch 602. This allows the next PU to be passed to the decoder 304. For video display control 303, upon receiving this enable next PU signal presentation unit detector 304 transfers the data for the next PU to video decoder 304. Similarly, for audio display control 306, the enable next PU signal causes presentation unit detector 607 to transfer the data for the next PU to audio decoder 307. As indicated, the enable next PU signal also causes switch 602 to be set to the B position. If during the output of the PU depacketizer and DTS (PTS) extractor 601 detects a PTS/DTS, it sets switch 602 to the A position as before. However, if no PTS/DTS is detected, switch 602 remains in the B position. At the end of a PU data transfer by presentation unit detector 607, it outputs a clock pulse to the clock input of DTS (PTS) register 603, as before. If a new DTS value has arrived, it is read, as before, from the A input of switch 602 and supplied as a DTS1 value to register 603 and appears as a DTS output value from register 603. However, if no new DTS value has arrived, then switch 602 still remains in the B position. In this situation, an extrapolated DTS value is supplied from adder 606 and is supplied as a new DTS1 value to register 603 and appears at an output thereof as DTS2. This extrapolated DTS value is obtained by adding via adder 606 a constant nominal picture duration value in units of Dec_TC to the previous DTS2 value. For example, if the nominal picture frequency is 29.97 Hz and the Dec_TC frequency is 90 kHz, then the constant is 3003. Presentation unit detector 607 then waits until Dec_TC increases to this new value of DTS2, whereupon it receives an enable next PU signal (pulse) from zero detector 605, thus repeating the data output cycle.

0069The above-described arrangements are, of course, merely illustrative of the applications of the principles of the invention. Other arrangements may be devised by those skilled in the art without departing with the spirit and scope of the invention.