EP2060074A1

EP2060074A1 - Method and device for adapting a scalable data stream, corresponding computer program product and network equipment

Info

Publication number: EP2060074A1
Application number: EP07823837A
Authority: EP
Inventors: Gérard Babonneau; Emmanuel Mory
Original assignee: France Telecom SA
Current assignee: Orange SA
Priority date: 2006-09-15
Filing date: 2007-09-14
Publication date: 2009-05-20
Also published as: US20100005184A1; WO2008032001A1; US8392595B2

Abstract

Method and device for adapting a scalable data stream, corresponding computer program product and network equipment. The invention relates to a method of adapting a scalable data stream organized into blocks of data units, each block comprising at least one base data unit and at least one enhancement data unit, making it possible to define a plurality of levels of quality and throughput (N1 to N3) according to the number and type of data units used, each data unit being initially ranked into an initial level selected from among said plurality of levels. According to the invention, the method implements a regulating mechanism (8) such that, if a data unit is re-ranked into a re-ranking level of lower priority than the initial level of said data unit, all the data units which depend on the decoding of said re-ranked data unit are also re-ranked into re-ranking levels where all the data units indispensable for their decoding are accessible.

Description

Method and apparatus for adapting a scalable data stream, computer program product and corresponding network equipment

1. DOMAIN OF THE INVENTION

The field of the invention is that of the processing of scalable data streams, also called hierarchical data streams, provided to users by real-time transport over a communication network.

More specifically, the invention relates to a method of adapting a scalable data stream to the characteristics of a network and / or users.

The invention applies in particular, but not exclusively, to the adaptation of a scalable video stream (for example an MPEG4-svc stream) to the characteristics of the link of the user and / or of his terminal.

The MPEG4-SVC technique is notably presented in the documents:

- JSVM 2.0: Joint Video Team (JVT) of ISO / IEC MPEG & ITU-T VCEG, N7084, Scalable Video Model Joint (JSVM) 2.0 Reference Encoding Algorithm Description, April 2005, Busan (Julien Reichel, Heiko Schwarz, Mathias Wien) ; and

- WD 2: J. Reichel, H. Schwarz, M. Wien - Scalable Video Coding - Working Draft 2 - ISO / IEC JTC1 / SC29 / WG11, W7084, April 2005, Busan.

2. PRIOR ART 2.1 Scalable video systems (scalable)

Currently, most video encoders generate a single compressed stream corresponding to the entirety of an encoded sequence. Each client wishing to use the compressed file for decoding and visualization must download (or "stream") the entire compressed file. However, in a heterogeneous system (for example the Internet), all the clients do not have the same type of data access: the bandwidth, the processing capacities, the screens of the different customers can be very different (for example, on an Internet network, one of the customers will be able to have an ADSL speed of 1024 kb / s and a powerful microcomputer (PC) while the other will only benefit from a modem access and a PDA ). An obvious solution to this problem is to generate several compressed streams corresponding to different bit rates / resolutions of the video sequence: it is the simulcast. However, this solution is sub-optimal in terms of efficiency, and requires knowing in advance the characteristics of all potential customers. More recently, scalable (or scalable) video coding algorithms have been introduced, that is to say with adaptable quality and variable spatio-temporal resolution, for which the coder generates a compressed stream in several layers, each of its layers being nested in the upper level layer. These algorithms are now being adopted as an amendment to MPEG4-AVC (hereinafter referred to as SVC).

Such encoders are very useful for all applications for which the generation of a single compressed stream, organized into several scalability layers, can serve several clients of different characteristics, for example:

VOD ("Video On Demand" for "video-on-demand"), accessible to UMTS-type radiocommunication terminals ("Universal Mobile Telecommunication

Service "for" universal mobile telecommunication service "), PCs or television terminals with ADSL access, etc., session mobility (for example resumption on a PDA of a video session started on a television, or, restarted on a General Packet Radio Service (GPRS) mobile for "general packet radio service" of a session started on UMTS); session continuity (in a context of sharing bandwidth with a new application); definition, in which a single video encoding should be able to serve both standard definition SD clients and clients with a high-definition HD terminal videoconferencing, in which a single encoding must meet the needs of customers with UMTS access and Internet access, etc. 2.2 MPEG-4 SVC The JSVM MPEG model is described in the already cited JSVM 2.0 document. It is based on a scalable encoder strongly oriented to AVC type solutions, whose schematic structure of a corresponding encoder is presented in Figure 1. It is a pyramidal structure. The video input components 10 undergo dyadic subsampling (2D decimation by 2 referenced 11, 2D decimation by 4 referenced 12). Each of the subsampled streams is then subjected to a temporal decomposition 13 of the MCTF type ("Motion Compensated Temporal Filtering"). A low-resolution version of the video sequence is encoded 14 up to a given bit rate which corresponds to the maximum decodable bit rate for the low spatial resolution (this base level is AVC compatible).

The higher levels are then encoded by subtracting the previous reconstructed and over-sampled level and encoding the residuals as a base level, and optionally one or more enhancement levels obtained by multi-bit coding of bit planes (called FGS for "Fine Grain Scalability", fine grain scalability).

According to this approach, the motion information 17 and the texture information 18 are distinguished. In order to achieve a rate adaptation, the texture information 18 is coded using a progressive scheme: coding of a first level of minimum quality (called "Base Layer" in English, or Base Layer); coding of progressive refinement levels (called "Enhancement Layer" or "Enhancement Layer").

With reference to FIG. 1, the texture information 18 feeds a coding module of the quantization base layer 19. The coded data, at the output of the module 19, are used to feed a block 21 for spatial transformation and entropic coding, which is working on the levels of signal refinement. The output data of the module 21 feeds an interpolation 20 from the base level. This interpolation is used as a prediction in the coding module 19 of the level above. A multiplexing module 22 orders the different sub-streams generated in a global compressed data stream 23. The compressed stream 23, at the output of the coder, is structured in data units called "NALU" (for "Network Abstraction Layer Unit"). The NALUs are organized in blocks of data units called subsequently "AU" (for "Access Units" in English). An AU comprises all the NALUs corresponding to the same time instant, that is to say having the same DTS (for "Decoding Time

Stamp "in English, or" timestamp information for decoding "), and therefore belonging to the same image. Each NALU is associated with an image resulting from spatio-temporal decomposition, a spatial resolution level, and a quantization level. This structuring in units of data makes it possible to adapt to bit rate and / or space-time resolution by eliminating the NALUs of too great spatial resolution, or of too high temporal frequency or even of too much encoding quality.

2.3 Technique described in the French patent application FR2854018

The French patent application published under No. FR2854018 describes a technique for controlling data packet traffic at the input of a network. The traffic comprises N flows and / or substreams each associated with a priority level, N> 2. Each packet is marked with the priority level associated with the stream or sub-stream to which it belongs.

In a particular application, the traffic comprises N substreams each corresponding to one of the N hierarchical levels of a hierarchical flow or an aggregate of hierarchical flows. This is for example an audio / video hierarchical stream comprising the following sub-streams: an audio sub-stream, a basic video sub-stream and a video enhancement sub-stream.

The goal is to treat bursts to control congestions in a network. Indeed, congestions mainly affect bursts, which contain the most important information of an MPEG encoded stream. This control is particularly suitable for ADSL type networks when the nominal flow rate is close to the maximum flow of the link.

The general principle of this known technique is to implement a multi-level chip bucket mechanism (MLTB, for "Multi Layer Token Bucket").

Each MLTB level is used to process one of the N priority levels. Each packets undergoes processing according to a marking corresponding to its priority level: it is accepted or refused according to whether or not it is possible to give it tokens according to its priority level. The accepted packets are placed in a buffer (buffer) of packets to be sent, which forms a means of managing a queue. Rejected packets are discarded or, in an alternative embodiment, placed in the buffer after being marked again with a lower priority level (i.e. after being reclassified to a lower priority level). . In an exemplary embodiment, the aforementioned MLTB is placed at the input of the packet buffer to be transmitted, for burst control, and a single-level bucket is placed at the output, for the smoothing of traffic (TBTS, for "Token Bucket"). Traffic Shaper ").

A disadvantage of this known technique is that the reclassification of a packet is performed independently of the notion of dependency between packets. In other words, it does not take into account that a packet is reclassified for processing subsequent packets that depend on this reclassified packet. Another disadvantage of the known technique is that the reclassification of a packet involves a new marking of this packet (that is to say a modification of this packet).

Another disadvantage of the known technique is that it does not offer an optimal solution for the reclassification of a packet in a lower priority level.

Another disadvantage of the known technique is that a packet is processed only when it is at the input of the packet buffer to be transmitted. It does not propose any processing on the packets already present in the buffer, and therefore does not require means of distinction between packets present in the buffer and having the same level of priority (no weighting of the packets).

Yet another disadvantage of the known technique is that it does not envisage that the packets leaving the packet buffer to be transmitted are directed to different transmission means according to their priority levels. Thus, in the aforementioned embodiment, the single-level TBTS does not handle the simultaneous sending of multiple streams. Moreover, the independence between the MLTB and the TBTS potentially risks to lead to an incompatibility between the filling and the exit. The quality of the output signal would be impaired the time to return to normal operation. 3. STATEMENT OF THE INVENTION

In a particular embodiment of the invention, there is provided a method of adapting a scalable data stream organized into blocks of data units each comprising at least one basic data unit and at least one unit of data. boost data, for defining a plurality of quality and rate levels according to the number and type of data units used, each data unit being initially classified in a selected initial level from said plurality of levels. This method implements a control mechanism such that if a data unit is reclassified to a reclassification level different from the initial level of said data unit, all the data units that depend on the decoding of said reclassified data unit are also reclassified in reclassification levels where all the data units necessary for their decoding are accessible. The general principle of the invention therefore consists in dynamically taking into account the impact of the reclassification of a unit of data on the classification of the other data units which depend on this reclassified data unit. Thus, it is ensured that the stream can be decoded correctly by the user's terminal.

It should be noted that the reclassification of a data unit does not imply a new marking (and therefore no modification) of this data unit. This reclassification is therefore not carried by the reclassified data unit, and is managed entirely and solely by the device implementing the adaptation method of the invention. However, as indicated below, in a particular embodiment of the invention, this reclassification results in a change of transmission means to which the reclassified data unit is directed.

It should also be noted that if the processing of an input data unit of a FIFO buffer causes a reclassification, there is a first embodiment of the invention in which it is this data unit. which is reclassified (see hereinafter the detailed "simplified compensation" algorithm), and a second embodiment of the invention in which there may be one or more reclassified data units among this data unit and all of the data units already present in the buffer (see below the detailed "full compensation" algorithm).

Advantageously, each data unit is directed to an associated transmission means at the level in which said control mechanism has finally classified or reclassified said data unit. M different transmission means are used, with l≤M≤R, where R is the number of quality and throughput levels.

Thus, the distribution of the units of data is modified either in a single transmission means or among several transmission means. In the second case, without being re-tagged (i.e. without being modified), the data units can be redirected from one transmission means to another, and the use of the different means is optimized. of transmission.

In an advantageous exemplary embodiment, each regulation level is associated with a separate transmission means (M = R).

Advantageously, at least one of the reclassification levels is such that the data units reclassified therein are not transmitted. This is for example a fictitious level forming trash, not part of the plurality of levels of quality and flow.

According to an advantageous characteristic, said regulation mechanism comprises a multi-level chip bucket (MLTB) type algorithm. In other words, one can combine the general concept of the present invention (reclassification of data units taking into account dependency between data units classified in different levels) with that of the MLTB, and thus enjoy all the advantages related to it. at the MLTB.

Advantageously, a data unit is reclassified if a rate resource available for the initial level of said data unit is not sufficient.

It should be noted that a unit of data may also be reclassified for other reasons, for example related to the resources of the terminal of the user receiving the stream.

Advantageously, when the flow comprises a reference data unit, it is canceled, for the data units that follow said reference data unit. in the flow, reclassifications resulting from reclassifications already made for data units that precede said reference data unit in the flow.

Advantageously, said regulation mechanism manages any dynamic modification of a flow rate resource assigned to one of said levels. Thus, one can immediately improve (for example in real time) the adaptation of the flow.

Advantageously, said regulation mechanism takes into account the actual transmission time of the data units, to manage the flow resources assigned to the levels. In this way, one can adjust the resources (eg flow resources) assigned to the different levels and indirectly take into account the transport protocol headers that are unknown to the regulation.

Advantageously, said regulation mechanism is such that it manages in a cumulative mode the flow resources assigned to the levels, so that at least one part of an unused resource of one level can benefit at another level. whose available throughput resource is not sufficient. This makes it possible to further improve the adaptation of the flow, without exceeding the overall predicted flow rate.

According to an advantageous variant, said regulation mechanism is such that it manages in an independent mode the flow resources assigned to the levels, so that each resource assigned to one level can not benefit on another level. This variant is easier to manage. In a first advantageous embodiment of the invention, said regulation mechanism performs a simplified regulation comprising the following steps if the available bit rate resource for the selected initial level is not sufficient: a) a new level selected is selected which has a lower priority than the previous level selected; b) if the available rate resource for the newly selected level is sufficient, then said data unit is reclassified to the new selected level; c) otherwise: if the new level is not the lowest of the said plurality of quality and flow levels, then step a) is returned, otherwise the said data unit is reclassified to a fictitious trash level. This first embodiment (simplified regulation) is simple to implement but only acts when the data units are input into the device implementing the method of the invention.

In a second advantageous embodiment of the invention, said regulation mechanism performs a complete regulation comprising the following steps if the available rate resource for the selected initial level is not sufficient: a) said data unit is classified in the level initial selected; b) the selected level is selected as the current level; c) if the flow rate resource for the current level is not exceeded, the current level is considered to be regulated and step d); otherwise: selecting at least one data unit to be reclassified from the data units classified in the current level, according to at least one predetermined reclassification criterion, reclassifying said at least one data unit selected in a lower priority level than the current level, and repeating step c); d) if the current level is not the least priority among the said plurality of quality and flow levels, a new current level with a lower priority than the previous current level is chosen, then it returns to step b), otherwise said regulation complete is complete. This second embodiment (complete regulation) is more complex than the first but offers the advantage of acting on all the data units already input and not yet output device implementing the method of the invention. As detailed later, the complete regulation is more efficient than the simplified regulation because it allows a better reactivity and a smoothing of the quality of the decoded flow.

In the case of the aforementioned second embodiment, advantageously, for each data unit, the method further comprises a step of assigning to said data unit a weighting among a plurality of possible weights each associated with a relative priority rank. within one level, said assignment being made according to at least one assignment rule making it possible to establish a correspondence between on the one hand said at least one marking information carried by data units and on the other hand one of the possible weights. In addition, said step of selecting at least one data unit is performed at least according to the weighting of the data units classified in said current level.

The choice of weights is consistent with the dependencies between data units. Taking weighting into account facilitates the selection of excess data units in one level, following a reclassification.

Advantageously, said step of selecting at least one data unit is performed with the following double criterion: the highest weighting, among weightings assigned to each of the data units; and for the same weighting, the most recent time reference.

The invention is advantageously applied in the case where said scalable data stream is a scalable video stream. In this case, the initial classification of each data unit in an initial level is a function of at least one marking information carried by said data unit and comprising a n-tuple of at least two indicators among the indicators belonging to the group. comprising: a priority indicator (P); a dependency indicator (D), relating to a spatial resolution; a time resolution indicator (T); and- an indicator of quality and / or complexity (Q).

Thus, the existing distinctions between data units (NALUs) are exploited by marking each unit of data with a quadruplet of {P, D, T, Q} indicators from an MPEG4-svc encoder.

Advantageously, the method is implemented by at least one of the following network equipment: stream server and intermediate node of a transmission network.

In another embodiment, the invention relates to a computer program product downloadable from a communication network and / or recorded on a computer readable medium and / or executable by a processor, this computer program product comprising instructions program code for executing the steps of the aforementioned method, when said program is executed on a computer. In another embodiment, the invention relates to a device for adapting a scalable data stream organized in blocks of data units each comprising at least one basic data unit and at least one unit of data of enhancement. , for defining a plurality of quality and rate levels according to the number and type of data units used, each data unit being initially classified in a selected initial level from said plurality of levels. This device comprises regulating means such that, if a reclassification condition is satisfied for a data unit, then said data unit is reclassified to a reclassification level less priority than the initial level of said data unit, and all Data units that depend on the decoding of said reclassified data unit are also reclassified into reclassification levels where all the data units necessary for their decoding are accessible.

More generally, the adaptation device according to the invention comprises means for implementing the adaptation method as described above (in any one of its various embodiments).

In another embodiment, the invention relates to a network equipment comprising an adaptation device as mentioned above.

Advantageously, the network equipment belongs to the group comprising stream servers and intermediate nodes of a transmission network. 4. LIST OF FIGURES

Other features and advantages of embodiments of the invention will appear on reading the following description of a preferred embodiment of the invention, given by way of indicative and nonlimiting example (all the embodiments of the invention). of the invention are not limited to the features and advantages of this preferred embodiment), and the appended drawings, in which: FIG. 1, already described in relation with the prior art, presents a block diagram of a generating encoder; an MPEG4-svc stream (scalable video stream); FIG. 2 presents a block diagram of a scalable data stream adaptation device according to one embodiment of the invention; FIG. 3 shows the three scalability axes of an MPEG4-svc stream; Figure 4 shows an example of dependencies between NALUs of different coding levels; FIG. 5 illustrates the operating principle of the regulation buffer and the reserve buffer appearing in FIG. 2; FIG. 6 shows a more detailed view of the regulation module appearing in FIG. 2; FIG. 7 illustrates a double referenced data structure, between the regulation buffer and the MLTB, according to a particular embodiment of the invention; FIGS. 8 and 9 illustrate two modes of operation, independent and cumulative respectively, of the MLTB implemented in the regulation module appearing in FIG. 2; Figure 10 shows another example of dependencies between different levels of coding; FIG. 11 presents a flowchart of a particular embodiment of a simplified regulation algorithm implemented by the regulation module appearing in FIG. 2; FIGS. 12 and 13 show a flowchart of a particular embodiment of a simplified control algorithm implemented by the regulation module appearing in FIG. 2 (FIG. 13 details the step of adjusting the regulation levels. which appears in Figure 12); and Figure 14 shows an example of distribution of the NALUs in the different levels of regulation. 5. DETAILED DESCRIPTION In all the figures of this document, the elements and identical steps are designated by the same numerical reference. 5.1 Device (Figure 2)

As illustrated in FIG. 2, in a particular embodiment, the scalable data flow adaptation device according to the invention 1 comprises the following elements (which are described in detail below): a reserve buffer 3 (hereinafter referred to as reserve buffer), which receives an MPEG4-svc 2 stream; a control buffer 4 (hereinafter called regulation buffer), the input of which is connected to the output of the reserve buffer 3; a static classification table 5; a dynamic ranking table 6; a resource adjustment module 7; a regulation module 8, to guarantee the different requested rates (arrow referenced 26) for the different levels of regulation; and a switching module 9, for routing each of the NALUs present at the output of the regulation buffer 4 as a function of the regulation level in which the regulation module 8 has classified them. The routing module generates several sub-streams 24 each corresponding to one of the regulation levels.

In an exemplary embodiment, at least some of the aforementioned means 3 to 9 included in the adaptation device are software means, resulting from the execution of a computer program by a processing unit. In this case, the device comprises a non-volatile memory storing a computer program implementing the flux matching method according to the invention, and a processing unit (microprocessor for example) controlled by this computer program. At initialization, the code instructions of the computer program are for example loaded into a volatile memory before being executed by the processor of the processing unit.

5.2 Classification Tables (Figures 3 and 4)

The classification tables 5 and 6 contain the different configurations allowing to distribute the NALUs in the different levels of the regulation. In fact, they contain correspondences between elements present in the additional information of the data and the regulation levels managed in the regulation module 8.

In the case of NALUs used in MPEG4-svc, these elements are indicators present in the header of each NALU, in the form of a quadruplet: {P,

D, T, Q} where: P indicates the priority;

D indicates the dependence (superset of the spatial resolution S); T indicates the temporal resolution; and Q indicates the quality and / or complexity. The ranking is made from the scalability properties of a stream

MPEG4-svc.

As illustrated in FIG. 3, an MPEG4-svc stream has three possible scalability axes: spatial axis S (image resolution), frequency axis T (image frequency) and Q axis, also called SNR axis (quality), translating the level of detail obtained by the truncation of the coefficients resulting from the coding. Each NALU is represented by a frame.

Figure 4 shows an example of dependencies between NALUs of different coding levels. In this example, it is assumed that: the first level of regulation (base level B) is associated with QCIF15 and QCIF30 coding; the second level of regulation (first level of enhancement R1) is associated with the CIF coding 15; and the third regulation level (second enhancement level R2) is associated with the CIF30 encoding. Each coding level includes, for a given spatial / temporal resolution (for example CIF / 15Hz), NALUs of different SNR qualities. It can therefore be said that each coding level comprises a plurality of SNR levels. In other words, a coding level corresponds to a rate range that can be adjusted by taking more or less NALUs with different SNRs. Figure 4 shows the dependencies between NALUs to access a specific level of quality. In this example, the CIF30 coding level is based on the CIF coding level 15, which itself relies on the QCIF or QCIF30 coding level.

In some cases, all SNR levels of a coding level are not routinely used by the higher quality coding levels. For example, in FIG. 4, among the SNR levels of the CIF coding level 15: those used by the coding level CIF30 are referenced 41 and those that are not 42. In the sample tables presented below, the NALUs not needed for the above levels are identified by a high weight (value 255).

When a need for reclassification appears, this high value has the effect, in the case of a complete regulation, to encourage their rejection (reclassification in the fictitious trash level) or their reclassification in a very low priority level.

Furthermore, the patent application No. FR0507690 specifies that several fallback modes (also called filtering paths) are possible. It should be noted that, in the embodiment of the present invention, the static and dynamic classification tables 5 correspond to a given fallback mode. For example, the table examples below give priority to the fluidity of the images (QCIF 15 and QCIF30 at the first level) over the resolution of the display (CIF 15 at the second level). Another fallback mode would prioritize display resolution (QCIF 15 and CIF 15 at the first level and QCIF30 at the second level). In the table examples below, the simultaneous presence of QCIF30 and QCIF in the first level is justified by their need for decoding CIF30.

It is of course possible to change the fallback mode, and therefore the static (reference) classification table, on the arrival of a reference image. In this case, the dynamic ranking table is not updated with the previous static table, but with the new static table. Then, the adaptation of the video stream to the rate (s) modifies this dynamic ranking table as indicated above.

The ranking tables therefore contain a list of correspondences in the form: {P, D, T, Q} -> (regulation level, weighting within the flow). In other words, each type of NALU, defined by a particular quadruplet {P, D, T, Q}, is associated with information "level of regulation" and information "weighting".

The information "level of regulation" corresponds to the level of regulation in which the NALU will be affected. The information "weighting within the flow" makes it possible to be able to order the NALUs within the same level of regulation. This weighting information is independent of ranking in a regulation level. In a particular embodiment, this weighting information is equal to the value P of the NALU header. The static ranking table 5, referred to as a reference, is provided with metadata included in the accompanying data of the audiovisual content. It contains the ranking levels defined by the provider of the content.

The purpose of the dynamic ranking table 6 is to take into account the dependencies between the NALUs. It intervenes in case of modification of the classification by the regulation module. Its goal is to maintain flow consistency. Since rate control can place N-level NALUs in a higher level (or delete them: classification in a fictitious trash level), this dynamic ranking table makes it possible to avoid placing NALUs in decoding relies on NALUs placed at level N + 1 (or deleted). This dynamic ranking table may not contain the weighting elements.

The dynamic ranking table is reset with the values from the static classification table to each reference image (IDR or I image).

The content of the static and dynamic ranking tables 5 is not static. The correspondences evolve as and when regulation is carried out by the regulation module, in order to take into account all evolutions of the constraints. It is up to an external module, decision, to dynamically modify their content. As a general rule, at any time an incoming NALU of value {P, D, T, Q} can be reclassified in a level N + k less priority, with k> l, in order to limit the flow of the level N. day of the dynamic classification table is such that all the NALUs which then enter the regulation buffer and which depend on the decoding of this reclassified NALU will be classified in a level N + m, with m> k and N + m≤Nmax where Nmax is the last possible level (that is, the lowest priority level). Only the fictitious trash level is greater than Nmax. The classification of a NALU from the N level to the N-I level is meaningless since the previous NALUs of the video stream necessary for decoding this NALU

NALU are not all present. In other words, the N-I level is independent of the N level, while the N level is dependent on the N-I level (hierarchical principle).

The following table shows an example of a static table for three-level control. The "Target" column 51 presents, for information purposes, a possibility of flow characteristic (coding level). Target IPIDITIQI Level | Weighting j

The following table shows an example of an example of a dynamic classification table according to the invention, obtained by updating the static table example above after regulating the first regulation level. In this example, we observe that this update concerns two types of NALUs: NALUs having a PDTQ {3,0,1,2} (fifth line) will have to be reclassified in level 4 of regulation (fictitious level forming trash). ; and the NALUs with a PDTQ {2,0,1,1} (twelfth line) will have to be reclassified in level 2 of regulation.

5.3 Buffer and reserve buffer (Figure 5)

FIG. 5 illustrates the operating principle of the regulation buffer and the reserve buffer.

The regulation buffer 4 contains the data coming directly from the source. It is essentially used to keep the order of arrival of the data in order to facilitate their processing during the broadcast. Indeed, all the data having the same DTS must be transmitted at the same time, this information setting the emission. It is a FIFO file whose arrow referenced 70 indicates the direction. The associated constraint is the rate calculation of each adjustment level of the regulation, its size must be fixed throughout the process and it must be constantly filled to be able to perform this calculation. The notion of filling is done on the basis of the temporal information of the data. Thus, for the data contained in the regulation buffer, we can say that: DTSi - DTSo = constant, with DTSi the DTS value of the last AU input in the regulation buffer, and DTSo the DTS value of the first AU to be output of the regulation buffer. In other words, in normal operation, when the regulation buffer is normally full, it contains a constant number of images. In order to synchronize the inputs and outputs, the reserve buffer 3 is added to the input of the regulation buffer 4. The presence of this reserve also allows the grouping of the NALUs to form the AUs. In fact, the data are processed at the AU level in the regulation buffer 4, but at the NALU level in the regulation module 8. This reserve also makes it possible to introduce the problem of supercharging and under-powering of the regulation buffer. Overfeeding is an input of a larger number of data than the regularization buffer can consume. In this case, the additional data will be stored in the reserve, until saturation thereof. Finally, if the reserve is considered to have reached its maximum size, the data will no longer be accepted and rejected from the system. Underrange means faster data consumption than entering data into the control buffer. The buffer regulation therefore does not have its optimal size, and the flow calculation remains possible, but on an incomplete basis because all the expected data are not present. However, the operation is still possible, as long as the control buffer is not completely empty because there is always data to be sent on the output network. Eventually, the program stops if there is no more data, only this case is problematic because it will not be possible to recover the lost flow. 5.4 Control module (Figures 6 to 14) 5.4.1 General principle of control In order to guarantee the different requested bit rates, the regulation module 8 is implemented. It: receives as input the references of the NALUs contained in the regulation buffer

4 according to a pre-established classification; - regulates the NALUs between its levels in order to guarantee the requested flows; updates the dynamic classification table 6. Indeed, during a reclassification of a NALU from a level N to N + k, it is necessary that the data dependencies are also reclassified at least at this level N + k .

A resource adjustment can be established based on the targeted rates and feedback (arrow referenced 27 in Figure 2) from the broadcast. This adjustment is performed by the resource adjustment module 7 which can, for example, refine the rates to take into account the headers of the NALUs transmitted.

Only the NALUs present in the regulation buffer 4 are managed in the regulation module 8. The referral module 9 directs the NALUs at the output of the regulation buffer.

Each NALU is routed to the transmission means corresponding to the regulation level in which it is classified. At each regulation level corresponds for example a different means of transmission, in order to separate the levels. This differentiation can be done at the IP / port address, TOS mark, track within a multiplexing ... In a variant, there are also several transmission means but several levels of regulation are associated with the same means of transmission . In yet another variant, there is a single transmission means with which all the regulation levels are associated. 5.4.2 Regulation by an MLTB algorithm To carry out the regulation, the regulation module 8 implements, for example, a regulation mechanism comprising a multi-level chip bucket type algorithm (hereinafter called MLTB).

In the example of FIG. 6, the MLTB manages three levels of regulation N1, N2 and N3. The arrows referenced 81, 82 and 83 symbolize the flow resources assigned to each of the regulation levels. The arrow referenced 84 the regulation between the levels

N1 and N2. The arrow referenced 85 the regulation between the levels N2 and N3. Optionally, the MLTB allows a dynamic change of the flow rate of any regulation level (without modifying the flow rates of the other levels). Thus, changing the rate of a level instantaneously changes the tokens allocated at this level, thus adapting the regulation to the new rate setpoint. The data within the MLTB are references (pointers) to the regulation buffer. The smallest information is NALU. AU integrity and temporal notions are managed only indirectly via the regulation buffer. Only the notion of token is used in MLTB calculations.

With respect to the content of the regulation buffer, the data present in the reserve are not taken into account in the MLTB.

As detailed below, the MLTB can operate in independent or cumulative mode. 5.4.2.1 Organization of the data between the regulation buffer and the MLTB

FIG. 7 shows a dual referenced data structure between the regulation buffer and the MLTB, according to a particular embodiment of the invention.

Each level of the MLTB contains first pointers to the NALUs stored in the regulation buffer 4. For the record, each NALU contains a header

(including the {P, D, T, Q}} indicators and coding data. In FIG. 7, the first pointers of level N1 are referenced 9I ₁ , 9I ₂ , ..., those of level N2 are referenced 92 _ls 92 ₂ , etc.

In each level of the MLTB, the first pointers of NALUs are ranked in order of weighting (as symbolized by the arrow referenced 93). For the same weighting level, the pointers to the oldest DTS NALUs are placed first. This organization makes it possible to immediately find the lowest priority NALUs (in the case of complete regulation, described below). The addition of new NALUs in the MLTB respects this rule.

The buffer buffer data is ordered by AU, and following the DTS

(as symbolized by the arrow referenced 94). The order of the NALUs within an AU is indifferent for the regulation carried out by the MLTB (the MLTB contains the order of priority). This makes it possible to dispense with the order of arrival of the NALUs, for example in the case of cooperation of networks with different transfer times.

In addition to the actual data, the control buffer 4 contains second pointers to the first pointers within the MLTB. This double referencing structure (first and second pointers), symbolized by the referenced arrows

95, makes it possible to efficiently find the NALUs of an AU which are distributed in the MLTB. It is possible to find the information of the NALUs within the buffer of regulation, starting from the first pointers included in the MLTB, and conversely, to find the position of the NALUs within the MLTB, from the second pointers included in the buffer of regulation.

5.4.2.2 Calculating the chips

In the MLTB, a bucket of chips is assigned for each regulation level, with a number of tokens set according to the requested rate for this regulation level and the size of the regulation buffer. For example, one of two ways to calculate the released tokens is used:

Tokens released = AU size: the MLTB behaves as a simple average on the regulation buffer, and the bit rate is AU level;

Tokens released = flow output * time out AU: the MLTB integrates the IP headers in its calculation. The rate corresponds to the IP level.

These two ways of calculating the flow can be applied indifferently to independent mode in levels, or to the cumulated mode which takes into account the tokens available in the levels of lower indices (and thus of higher priorities).

Figures 8 and 9 illustrate these two modes of operation, independent and cumulative respectively.

C1 and C2 represent the target flow rates of the respective levels 1 and 2. D1 and D2 represent the actual flow rates.

In the independent mode (figure 8), we have the following relations, relative to the constraints of flows: Dl <Cl, and D2 <C2. Flow constraints clearly show an underutilization of the bandwidth. In the cumulative mode (figure 9), level 2 can take advantage of level 1 reserves to improve the quality of its transmission. We therefore have the following relations: Dl <Cl, and Dl + D2 <C1 + C2. The overall throughput constraint is respected, thus optimizing quality through better use of bandwidth. The accumulated mode can be used in another regulation mode than the

MLTB. Indeed, it is sufficient that the regulation can use, for each level, the available resources of the lower level.

In some cases, it is interesting to use an MLTB operating with a mixture of independent and cumulative modes. Thus, in the example illustrated in FIG. 10, an SD base flow (of target bit rate C1) having HD enhancement flux in 72Op mode (of target bit rate C2) and another HD enhancement mode in FIG. 1080i mode (target rate C3), these two enhancement modes are independent. The flow constraints are then: D1 + D2 ≤ C1 + C2 D1 + D3 ≤ C1 + C3

It would also be possible to add a 1080p mode (target rate C4) based on the modes 72Op and 1080i such that: D1 + D2 + D3 + D4 <C1 + C2 + C3 + C4. 5.4.2.3 Organization of the data in the MLTB

In the use of data from SVC feeds, some NALUs have priority. As a result, the lowest priority NALUs will be the first to be reclassified. To define the reclassification order, the weighting information present in the classification tables is used as a priority. 5.4.3 Complete regulation and simplified regulation

Two types of regulation that can be performed using an MLTB type algorithm are presented below: a complete regulation, implementing all the NALUs contained in the regulation buffer for each regulation level; and a simplified regulation acting only at the input of the regulation buffer.

In both cases, the timing is done at the exit of an AU. Thus, to respect the fixed DTS size constraint of the regulation buffer in the normal case of use, a new AU is entered in the regulation buffer at the output of an AU. The regulation can then be performed.

When the output of an AU, the MLTB indicates the distribution of the NALUs of this

AU in each of the control levels (flow levels), to calculate the adjustment of resources by regulation level.

5.4.3.1 Simplified regulation

The simplified regulation only acts when entering the data (AU grouping NALUs) into the regulation buffer. When the quantity of tokens freed by the data output of the regulation buffer is not sufficient to allow the entry of a new datum by taking into account only the classification operation, this data is reclassified in the upper levels of the MLTB. . To account for dependency issues, the dynamic ranking table is updated.

We now present, in connection with the flowchart of Figure 11, a particular embodiment of the simplified control algorithm. In a step 131, an event causes the output of an AU of the regulation buffer. In a step 132, the tokens that were associated with the NALUs of this AU become available again for the MLTB (according to the calculation of chips detailed above).

In a step 133, an AU taken from the reserve buffer is imported into the regulation buffer. In this standard scenario, an AU output is an AU input. The ranking is applied sequentially to each NALU of the AU entry, thanks to the chips available in the MLTB.

Then, in a step 134, it is detected whether all the NALUs composing the AU input have been processed. If so, go to the end step 135. Otherwise, in a step 136, a NALU is to be processed, then in a step 137 it is detected whether the available tokens of the requested N regulation level (according to dynamic classification table) suffice for the NALU.

If the available tokens of level N are sufficient, then in a step 138 the

NALU is ranked in the N level and N level tokens are awarded. Then, we return to step 134. If the available tokens of the level N are not sufficient for the NALU, then in a step 139 the regulation level N + 1 is chosen, of lower priority than the N level. Then, in a step 1310, it is detected whether the available tokens of the N + 1 regulation level suffice for the NALU: if the available tokens of the N + 1 level are sufficient, then one goes on to the 1311 modification step of the dynamic ranking table (to take dependencies into account) before proceeding to step 138 so that the NALU is reclassified in the level N + 1 and the tokens of the level N + 1 are allotted to it; if the available tokens of the N + 1 regulation level are not sufficient for the NALU, one returns to the step 139. The last passage by the step 139 consists in choosing the level of fictitious regulation forming trash, and in this case the The following step 1310 is considered as verified and therefore, step 1311 is taken.

The advantage of this simplified regulation is the low complexity of the algorithm. It acts only on the incoming data in the system and is only based on the consultation and possibly an update of the ranking tables. 5.4.3.2 Complete control Complete control implements the entire control. When entering a new AU, it is put in the MLTB from the classification table. The regulation is then done on all the contents of the MLTB by selecting the most appropriate NALUs for a reclassification. In fact, when there is a need for reclassification, there may be a NALU in the system that is more likely to be reclassified than the new AU.

We now present, in connection with the flowchart of Figure 12, a particular embodiment of the complete control algorithm.

In a step 141, an event causes the output of an AU of the regulation buffer. In a step 142, the tokens that were associated with the NALUs of this AU become available again for the MLTB. In a step 143, an AU taken in the reserve buffer is imported into the regulation buffer.

Then, in a step 144, it is detected whether all the NALUs composing the AU input have been processed. If so, we go to the end step 145. Otherwise, in a step 146 we take a NALU to be processed, then in a step 147 we detect if the available tokens of the N regulation level requested (according to the dynamic classification table) suffice for the NALU. If the available tokens of the N level are sufficient, then in a step 148 the NALU is ranked in the N level and the tokens of the N level are assigned to it.

If the available tokens of the N level are not sufficient for the NALU, then in a step 149 a regulation of the set of NALUs of all the regulation levels of the MLTB is carried out, then in a step 1410 the dynamic classification table is modified. (in order to take dependencies into account) before returning to step 144.

With reference to FIG. 13, the detail of step 149 of adjustment of the regulation levels (regulation of the set of NALUs of all the MLTB regulation levels) is now presented. In a step 151, the input NALU is added to the N level. The number of tokens available in this level of regulation is then likely to become negative if, before the addition, there were not enough tokens available. NALU is nevertheless added because it must be taken into account in future research.

Then, in a step 152, it is detected whether all the regulation levels to be processed (namely the level N and the lower priority levels) have been. If so, we go to the end step 153. If not, in a step 154 we take a new current regulation level to be processed (we start with the level N and then at each iteration we go to the next level N + 1 of lower priority), then in a step 155 it is detected whether the number of available tokens for the current level is correct (that is, greater than or equal to zero).

If the number of chips available for the current level is correct, the level is considered to be regulated and it returns to step 152.

If the number of tokens available for the current level is incorrect, proceed to a step 156 of NALU selection, among those classified in the current regulation level, having the least dependence. A selection criterion is to take the NALU with the highest weighting and the most recent DTS. In a step 157, the selected NALU is reclassified at the N + 1 level, thus releasing the tokens associated with it. Then we return to step 155, to check if the reclassification of this NALU was sufficient. The last pass through step 157 consists of a reclassification into the dummy trash control level, and in this case the test of step 155 which follows is considered as checked, and therefore step 152 is returned.

Figure 14 shows an example of distribution of NALUs (represented by black squares) in three possible control levels N1, N2, N3 and a dummy trash control level N4.

Each AU is represented by a column of NALUs that may have a different number of NALUs. During regulation, the NALUs are selected to be assigned to a regulation level. The number of NALUs classified in one level of regulation may vary from one DTS to another, depending on the flow rate assigned to this level of regulation. Thus, in this example, there is a decrease in the flow rate assigned to the Nl level, between the second and third DTS (starting from DTSo), then an increase in this rate between the fifth and sixth DTS. Complementarily, there is an increase in the bit rate assigned to the N2 level, between the second and third SDRs, and then a decrease in this bit rate between the fifth and sixth SDRs. On the other hand, the flow rate assigned at level N3 remains constant.

When changing the rate, the dependencies are taken into account when selecting the NALUs. Thus, one NALU will not have a dependency on another NALU placed in a higher level (lower priority). The dashed arrow referenced 161 illustrates the order of travel for the reclassification of the NALUs. The NALUs reclassified first are those with the highest weighting and the most recent DTS (close to DTSi).

Each of the arrows referenced 162 and 163 illustrates the dependence between two NALUs. It is recalled that a NALU of a regulation level N can only depend on a NALU of the same N regulation level or a N 'less priority regulation level (N'> N).

This figure 14 also illustrates the (re) classification of two NALUs, according to the example of dynamic classification table above (see the fifth and twelfth lines), following a reduction of the flow rate assigned to level N1: - IaNALU referenced 164 is (re) ranked in level 2; and

IaNALU referenced 165 is (re) classified in level 4 (trash). The complexity of the algorithm of the complete regulation is increased compared to that of the simplified regulation, because it must consult, or even modify, all the NALUs contained in the buffer of regulation, and this for each introduction of a new given. The advantage of the complete regulation is the reactivity. Indeed, all the data present in the system are regulated during the change of constraints. Thus, the data transmitted just after the change will already respect these new constraints.

Another advantage of this complete regulation is to smooth the quality of the decoded stream, by regulating the classification of NALUS over the entire regulation buffer. The order of selection of reclassified NALUs optimizes the quality in an order that respects dependencies.

Claims

A method of adapting an organized scalable data stream into blocks of data units each comprising at least one basic data unit and at least one enhancement data unit, for defining a plurality of quality levels. and flow rate (N1-N3) according to the number and type of data units used, each data unit being initially classified in a selected initial level among said plurality of levels, characterized by implementing a mechanism control system (8) such that, if a data unit is reclassified to a reclassification level less priority than the initial level of said data unit, all data units that depend on the decoding of said reclassified data unit are also reclassified. in reclassification levels where all the data units necessary for their decoding are accessible.

Method according to claim 1, characterized in that each data unit is directed (9) to a data transmission means associated with the level at which said control mechanism has finally classified or reclassified said data unit, and that M different transmission means are used, with l≤M≤R, where R is the number of levels of quality and flow.

3. Method according to any one of claims 1 and 2, characterized in that at least one of the reclassification levels is such that the data units reclassified there are not transmitted.

4. Method according to any one of claims 1 to 3, characterized in that a data unit is reclassified if a flow rate resource available for the initial level of said data unit is not sufficient.

5. Method according to claim 4, characterized in that said regulation mechanism performs a simplified regulation comprising the following steps if the flow rate resource available for the initial level selected is not sufficient: a) a new regulation level is chosen selected less priority than the previous regulation level selected; b) if the available rate resource for the newly selected level is sufficient, then said data unit is reclassified to the new selected level; c) otherwise: if the new level is not the lowest of the said plurality of quality and flow levels, then step a) is returned, otherwise the said data unit is reclassified to a fictitious trash level.

6. Method according to claim 5, characterized in that said step of selecting at least one data unit is performed with the following double criterion: the strongest weight, among weights assigned to each of the data units; and for the same weighting, the most recent time reference.

7. Method according to any one of claims 1 to 6, characterized in that said scalable data stream is a scalable video stream, and in that the initial classification of each unit of data in an initial level is a function of minus marking information carried by said data unit and comprising a tuple of at least two indicators among the indicators belonging to the group comprising: a priority indicator (P); a dependency indicator (D), relating to a spatial resolution; a time resolution indicator (T); and an indicator of quality and / or complexity (Q).

8. Computer program product downloadable from a communication network and / or recorded on a computer-readable medium and / or executable by a processor, characterized in that it comprises program code instructions for the execution of the steps method according to at least one of claims 1 to 7, when said program is run on a computer.

A device for adapting a scalable data stream organized into blocks of data units each comprising at least one basic data unit and at least one enhancement data unit, for defining a plurality of quality levels. and bit rate according to the number and type of data units used, each data unit being initially ranked in a selected initial level among said plurality of levels, characterized by comprising regulating means such that, if a reclassification condition is satisfied for a data unit, then said data unit is reclassified to a reclassification level lower priority than the initial level of said data unit , and all data units that depend on the decoding of said reclassified data unit are also reclassified to reclassification levels where all the data units necessary for their decoding are accessible.

10. Network equipment, characterized in that it comprises an adaptation device according to claim 9.