EP1867100A1

EP1867100A1 - Method for managing decisions, method for constructing a decision tree, central manager, intermediate manager, terminal and corresponding computer programme products

Info

Publication number: EP1867100A1
Application number: EP06725429A
Authority: EP
Inventors: Servane Bonjour; Christophe Dousson; Mai Trang Nguyen Thi
Original assignee: France Telecom SA
Current assignee: Orange SA
Priority date: 2005-04-04
Filing date: 2006-03-30
Publication date: 2007-12-19
Also published as: US20090119392A1; FR2883997A1; WO2006106067A1; US7991869B2

Abstract

The invention concerns a method for managing decisions between a central manager (GM) and at least one terminal (TM) in an architecture of networks distributed and prioritized according to N levels, with N = 2. The central manager is included in a level of rank N. The terminal is included in a level of rank 1. The method includes a step of distributing a decision tree constructed by the central manager and supported by the architecture. The step of distributing the decision tree itself includes the following steps, for a given terminal: based on the decision tree, called rank N tree, the central manager creates a simplified tree of rank N-1; if N = 2, the central manager transmits the simplified tree of rank N-1 to the given terminal, so that the latter takes a decision based on the simplified rank N-1 tree at the given terminal and executes same; if N = 3, the following steps are performed, after initializing k at N-1: a) the central manager transmits the simplified rank k tree to an intermediate manager of rank k (GA1, GA2) included in the level of rank k; b) the intermediate manager of rank k creates a simplified tree of rank k-1 based on the simplified tree of rank k; c) if k = 3, the intermediate manager of rank k transmits the simplified tree of rank k-1 to an intermediate manager of rank k-1 included in the level of rank k-1 and step b) is resumed after decrementing k by one unit; if k = 2, the intermediate manager of rank k transmits the simplified tree of rank k-1 to the given terminal, so that the given terminal takes a decision based on the simplified tree of rank k-1 and executes same.

Description

Decision management method, decision tree construction method, central manager, intermediate manager, terminal and corresponding computer program products.

FIELD OF THE DISCLOSURE The field of the invention is that of decision trees.

More specifically, the invention relates to a decision management method between a central manager and at least one terminal within a distributed network architecture and hierarchical N levels, with N> 2. It is assumed that the central manager is understood in a level of rank N and that the terminals are included in a level of rank 1.

Conventionally, a decision tree is constructed from a set of rules, which define a decision policy, and from a set of variables comprising a subset of explanatory variables (measurable variables) and a subset of target variables (variables to be deduced). The invention applies in particular, but not exclusively, to a network architecture in which a network operator wishes to implement a decision policy, such as for example a mobility decision policy, security or even quality of service.

In the particular case of a mobility decision policy, the object of the invention is to enable a network operator to implement rules based handover policies that the operator defines himself. The invention applies in all types of data network and is independent of the access network technology (GPRS ("General Packet Radio Service", ie "general packet radio service") , Universal Mobile Telecommunications System (UMTS), Wireless Local Area Network (WLAN), Ethernet, etc. ). These rules take into account various parameters related to the operator, the mobile terminal, the access networks and the user.

2. PRIOR ART Disadvantages of the techniques of the prior art are now presented through the particular case of a known technique for implementing a control strategy. mobility decision within a packet mode network architecture.

A mobility manager implementing handover control over the network is presented in Y. Khouaja, K, "Hierarchical Mobility Controlled by the Network" (or "Hierarchical Mobility Controlled by the Network" in French). Guillouard, P. Bertin and JM. Bonnin, published in "Multiaccess, Mobility and Teletraffic for Wireless Communications" published in 2002 by Kluwer Academi Publishers.

This document defines a mobility manager that has the ability to initiate and guide the execution of a handover, based on information transmitted by the operator, the network and the mobile nodes. This mobility manager is located in a cellular network and makes it possible to choose target cells according to the radio data (measurements reported by the mobiles), network (load, type of traffic, quality of service) and operator (subscriber profiles, network parameters, trip thresholds, operating status of access points). Each mobility manager manages a set of radio cells (i.e., an access point set). The operation of the mobility manager is described below. The mobility manager (GM) transmits to the mobile node (mobile terminal), a list of neighboring access points (PA). The mobile node (NM) transmits to the mobility manager, the quality measurements of the radio link. The mobile node requests the mobility manager to change the access point without specifying the new target access point. The mobility manager selects the target access point by consulting its database (BD). This database contains various information that can assist handover decision making. The mobility manager manages the access point change, by transmitting the data packets to the mobile node, simultaneously to the two access points involved in the handover, as soon as the mobility manager knows about the access point. imminent execution of handover. This duplication of data limits the loss of data packets.

The mobility manager described above enables implementation of rule-based mobility management that the network operator defines. This known technique, however, has several disadvantages. First, the centralization of a mobility manager has the disadvantage of reassembling information from one or more hierarchical levels to the central point, which increases the implementation time. In addition, handover decisions are made for all terminals that must move, which makes the implementation time proportional to the number of terminals to be moved. With a centralized manager, the number of terminals is very important and therefore the implementation time too. Finally, the travel time of a decision tree is proportional to the number of information contained in the tree. However, in a centralized manager, the number of information to be taken into account is very important. Moreover, for the distribution of rules, we can consider defining a system of deduction (expert system) at each hierarchical level, but this approach does not allow to take into account decision rules mixing the different categories of parameters (local or global). ). In this case, the rules could only contain local parameters. This solution requires the user to write rules that work only one level and distribute them "by hand" then.

In other words, the hierarchical structure of the management of the policy influences the writing of the rules of policy (in particular, in case of change of hierarchical structure without change of policy, one must rewrite the rules). In addition, no decision tree distribution means is described. 3. Objectives of the invention

The invention particularly aims to overcome these disadvantages of the state of the art.

More precisely, one of the objectives of the present invention is to provide a technique for constructing a decision tree containing a decision policy to be implemented within a distributed and hierarchical architecture, as well as a technique use of this decision tree to execute the decision policy it contains, both of these techniques for optimizing the implementation of the decision policy.

The invention also aims to provide such techniques that are simple to implement and inexpensive. Another objective of the invention, in the particular case of a mobility decision policy within a network architecture, is to provide such techniques making it possible to optimize the management of the mobility of the terminals, in particular by decreasing the deadline for making intercellular transfers. 4. Summary of the invention

These various objectives, as well as others which will appear later, are achieved according to the invention using a decision management method between a central manager and at least one terminal within a network architecture. distributed and hierarchical according to N levels, with N> 2, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1. According to the invention, this method comprises a distribution step a decision tree constructed by the central manager and based on said architecture, the distribution step of the decision tree including itself the following steps, for a given terminal: - from the tree of decision, said tree of rank N, the central manager creates a simplified tree of rank NI; if N = 2, the central handler transmits the simplified tree of rank N-I to the given terminal, so that the given terminal makes a decision from the simplified tree of rank N-I and executes it; if N> 3, the following steps are carried out, after initializing k to N-I: a) the central manager transmits the simplified tree of rank k to an intermediate manager of rank k included in the level of rank k; b) the intermediate manager of rank k creates a simplified tree of rank k-1 from the simplified tree of rank k; (c) if k> 3, the intermediate manager of rank k transmits the simplified tree of rank k-1 to an intermediate manager of rank k-1 included in rank level k-1 and returns to step b) after decrementing k by one unit; if k = 2, the intermediate manager of rank k transmits the simplified tree of rank k-1 to the given terminal, so that the given terminal makes a decision from the simplified tree of rank k-1 and executes it.

It should be noted that a simplified tree may comprise a single leaf. The technique of the invention is more efficient than the known technique based on the use of a centralized manager. Indeed, the technique of the invention does not require the reassembly of information from one or more hierarchical levels to a central manager, since each manager of a given rank level processes information available at this rank. given.

In addition, the overall processing time of the tree (until a leaf is reached) is reduced because the managers involved successively use less and less complex trees (a tree of rank k-1 being obtained by reducing a tree of rank k). Advantageously, if in step b) the simplified tree of rank k-1 is reduced to a leaf node, then step c) is replaced by the following step: c ') the intermediate manager of rank k takes a decision and transmits it to the given terminal, so that the given terminal executes it.

In a preferred embodiment, the transmission of a simplified tree of rank j-1 by a manager of rank j, with je {N ... 2}, is based on a simplified global tree structure comprising a linked list elementary structures each defining a given node of the simplified tree of rank j-1 and each comprising: a flag indicating whether, for the simplified tree of rank j-1, the given node is a root node, an intermediate node or a decision node; an identifier of the elementary structure; if the given node is a root node, an intermediate node, or a decision node that is not a leaf node:

* an explanatory variable; * an operation to be performed on the explanatory variable, constituting a test on the value of the explanatory variable;

* for each possible value of the test, a field "next elementary structure" containing the identifier of the elementary structure defining the next node of the simplified tree of rank j-1; if the given node is a decision node which is a leaf node:

* at least one target variable; for each target variable, an operation relating to the target variable, constituting the supply of a value of the target variable;

* for each target variable, a "next elementary structure" field containing an end indicator. Thus, when he traverses the simplified tree of rank j-1 that he has received, the rank manager j-1 knows that each decision node which is not a leaf of the simplified tree of rank j-1 is a root node of a simplified tree of rank j-2 to be transmitted.

Advantageously, the decision tree comprises, when traversed from a root node to leaf nodes, a sequence of N sets of nodes, a set of nodes of rank i, ie {N ... 1}, supporting on explanatory variables available at rank i level. The creation by a manager of rank j of a simplified tree of rank j-1 from a tree of rank j, with je {N ... 2}, is carried out taking into account at least one information that the Rank manager j has at least one predictor available at rank level j. When he arrives at a node of the set of nodes of rank j-1 by traversing the tree of rank j, the rank manager j passes the hand to a manager of rank j-1 so that the rank manager j- 1 traverses the simplified tree of rank j-1.

According to an advantageous characteristic, the transmission of a simplified tree of rank j-1 to a manager of rank j-1, with je {N ... 2}, is carried out by the manager of rank j during a preliminary step distribution of all trees of rank j-1 possible between all managers of rank j-1.

In this case, all the simplified trees are distributed a priori once and for all, and then we simply browse through changing manager during node level changes. According to an advantageous variant, the decision tree comprises, when it is traversed from a root node to leaf nodes, a sequence of N sets of nodes, a set of nodes of rank i, ie {N ... 1}, based on explanatory variables available at rank i level. In addition, the transmission of a simplified tree of rank j-1 to a manager of rank j-1, with je {N ... 2}, is performed by the manager of rank j only after the rank manager j arrived, by traversing a tree of rank j, on a node of a set of nodes of rank j-1. Contrary to the previous case, the simplified trees are not distributed a priori but only when necessary during processing, that is to say each time a manager of rank j arrives on a node of rank j-1 ( and at this time sends a simplified tree of rank j-1 to one of the managers of rank j-1 Advantageously, the decision tree contains a decision policy belonging to the group comprising: mobility decision policies said at least one terminal within the network architecture; QoS decision policies for said at least one terminal within the network architecture; and security decision policies for said at least one terminal within the network architecture.

In this particular context of a mobility decision, the invention relates to the distribution of a decision tree based on a distributed mobility management architecture. In the particular case where N = 3, the structure of the tree is decided by the central mobility manager (level 3 manager) and it takes into account the information he has to create simplified trees that he distributes to Access Managers (Tier 2 Managers). They are doing the same job of distributing simplified trees to terminals (Tier 1 level) that will apply the information they possess to implement the final mobility decision. All data included in the mobility policy are taken into account locally and do not need to be reported to the central mobility manager. Only the location information of the mobile terminals goes back to the central mobility manager. This makes it possible to distribute only the trees that are useful for the lower level managers (sending the trees only for the terminals located in an access network). This also makes it possible to optimize the global time of the handover because the parameters are evaluated locally and the decision is made only for the terminals of a single access network at a time.

The invention also relates to a method for constructing a decision tree adapted for implementing the aforementioned method for managing decisions between a central manager and at least one terminal, the decision tree being constructed at from a set of rules defining a decision policy to be implemented within a distributed and hierarchical architecture according to N levels, with N> 2, and a set of variables comprising a subset of explanatory variables, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1, the method comprising a step of choosing an explanatory variable so as to create a new node of the tree s relying on said explanatory variable, the selection step being iterated starting from a root node to go to leaf nodes, the choice made at each new iteration being performed among the explanatory variables, called free explanatory variables, not already chosen during a previous iteration. According to the invention, at each iteration of the selection step, an explanatory variable available at the highest rank level is selected from among the ranks of the levels where the free explanatory variables are available, so that the decision tree includes when traversed from the root node to leaf nodes, a sequence of N sets of nodes, a set of nodes of rank i, ie {N ... 1}, based on explanatory variables available at the level of rank i.

The general principle of the invention therefore consists in automatically transforming all the rules of the decision policy into a distributable decision tree, thanks to an appropriate choice of the explanatory variables on which the nodes of the decision tree are based. . More precisely, based on the fact that each of the explanatory variables is available at one of the levels of the architecture (from the most global level to the most local level), the invention proposes to group the nodes from the decision tree into sets of hierarchical nodes according to the hierarchy of the levels of the architecture. Thus, all the nodes of the same set of nodes are based on explanatory variables available at the same level of the architecture. In addition, the nodes closest to the root node rely on explanatory variables available at the most global level. The further away they are from the root node, the more the nodes rely on explanatory variables available at levels themselves further and further away from the more global level. In other words, the invention makes it possible to rework the rules so as to obtain a distributable decision tree whereas, a priori, the rules mix the two types of parameters and therefore are not distributed as is. The fact that the decision tree is distributable makes it possible to distribute the decision tree during the implementation of the aforementioned decision management method (method of executing a decision policy contained in this tree).

Advantageously, the decision policy belongs to the group comprising: mobility decision policies of said at least one terminal within the network architecture; quality of service decision policies for said at least one terminal within the network architecture; and security decision policies for said at least one terminal within the network architecture.

Advantageously, the decision policy is a mobility decision policy of the at least one terminal within the network architecture, and the architecture is hierarchical according to the following three levels: a tier 3 level comprising a core network ; a rank 2 level comprising at least two access networks each forming a part of the level of rank 2; a rank level 1 comprising a plurality of terminals each forming a part of the level of rank 1.

Advantageously, the subset of explanatory variables comprises: at least one variable available at rank 3 and belonging to the group comprising:

* a variable defining the type of subscription associated with a terminal;

a variable defining the access network to which a terminal is connected; at least one variable available at rank 2, for each access network, and belonging to the group comprising:

a variable defining the load of an access point to the access network; at least one variable available at rank 1, for each terminal connected to a given access network, and belonging to the group comprising: for each access network, a variable defining the availability of the access network; * a variable defining the quality of service required by the terminal;

* a variable defining the quality of current service offered to the terminal; and

* a variable defining the quality of service offered by the given access network. Preferably, the method of constructing a decision tree according to the invention is implemented in the central manager.

The invention also relates to a method of partial processing of a decision tree by a central manager, within the framework of decision management between a central manager and at least one terminal within a distributed and hierarchical distributed network architecture. N levels, with N> 2, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1. According to the invention, the central manager performs the following steps, for a terminal given: from the tree of decision, said tree of rank N, it creates a simplified tree of rank NI; if N = 2, it transmits the simplified tree of rank N-I to the given terminal, so that the given terminal makes a decision from the simplified tree of rank N-I and executes it; if N> 3, it transmits the simplified tree of rank NI to an intermediate manager of rank NI included in the level of rank NI, so that the given terminal receives a simplified tree of rank 1 via a cascade of intermediate managers comprising at least the middle manager of rank NI.

Thus, the central manager of the invention performs only part of the complete processing of the decision tree, this complete processing being distributed between the central manager, the terminal and, possibly, one or more intermediate managers. This distribution of the complete processing makes it possible to avoid the reassembly of all the information to the central manager. Furthermore, the fact that the central manager transmits a simplified tree also simplifies the other parts of the complete processing, performed by the terminal and possibly the intermediate manager or managers. Advantageously, the central manager first performs a step of constructing the decision tree from a set of rules defining a decision policy to be implemented within said architecture, and a set of variables comprising a subset of explanatory variables, the constructing step comprising a step of choosing an explanatory variable so as to create a new node of the tree based on said explanatory variable, the selection step being iterated starting with from a root node to go to leaf nodes, the choice made at each new iteration being carried out among the explanatory variables, called free explanatory variables, not already chosen during a previous iteration. At each iteration of the choice stage, the central manager chooses an explanatory variable available at the level occupying the highest rank among the ranks of the levels where the free explanatory variables are available, so that the decision tree understands, when it travels from the root node to the leaf nodes, a sequence of N sets of nodes, a set of nodes of rank i, ie {N ... 1}, based on explanatory variables available at rank i.

The invention also relates to a method of partial processing of a decision tree by an intermediate manager of rank m, with me {Nl ... 2}, within the framework of decision management between a central manager and at least one terminal within a distributed network architecture and hierarchical N levels, with N> 3, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1. According to the invention, the intermediate manager performs the following steps, for a given terminal: it receives a simplified tree of rank m; from the simplified tree of rank m, he creates a simplified tree of rank m-1; if m = 2, it transmits the simplified tree of rank m-1 to the given terminal, so that the given terminal makes a decision from the simplified tree of rank m-1 and executes it; if m> 3, it transmits the simplified tree of rank m-1 to an intermediate manager of rank m-1 included in the level of rank m-1, so that the given terminal receives a simplified tree of rank 1 via a cascade intermediate managers including at least the middle manager of rank m-1. Like the central manager or the terminal, the intermediate manager of the invention performs only part of the complete processing of the decision tree. The distribution of the complete processing between the different entities (intermediate manager (s), central manager and terminal) makes it possible to avoid the reassembly of all the information to the central manager. Furthermore, the fact that the intermediate manager transmits a simplified tree (even more simplified than the simplified tree that he has received) also simplifies the other parts of the complete processing, performed by the terminal and, possibly, the other intermediate manager (s) of lower rank (s). Advantageously, if the simplified tree of rank m-1 is reduced to a leaf node, then the intermediate manager makes a decision and transmits it to the given terminal, so that the given terminal executes it.

The invention also relates to a method for partial processing of a decision tree by a terminal, in the context of managing decisions between a central manager and at least said terminal within a distributed and hierarchical network architecture according to N levels, with N> 2, the central manager being included in a level of rank N, the terminal being included in a level of rank 1. According to the invention, the terminal performs the following steps: it receives a simplified tree of rank 1 and he makes a decision from the simplified tree of rank 1 and executes it. Like the central manager or the intermediate manager, the terminal of the invention performs only part of the complete processing of the decision tree. The distribution of the complete processing between the different entities (intermediate manager (s), central manager and terminal) makes it possible to avoid the reassembly of all the information to the central manager. Moreover, the fact that the terminal receives a simplified tree also simplifies the part of the complete processing performed by the terminal.

The invention also relates to computer program products, downloadable from a communication network and / or recorded on a computer readable medium and / or executable by a processor, for the execution of each of these methods (partial processing method of a decision tree by a central manager, method of partial processing of a decision tree by a manager intermediary of rank m, method of partial processing of a decision tree by a terminal).

The invention also relates to a central manager for partial processing of a decision tree in the context of decision management between a central manager and at least one terminal within a distributed network architecture hierarchical according to N levels, with N> 2, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1. According to the invention, the central manager comprises: means of creation, for a given terminal , a simplified tree of rank N-1 from the decision tree, said tree of rank N; if N = 2, transmission means of the simplified tree of rank N-I to the given terminal, so that the given terminal makes a decision from the simplified tree of rank N-I and executes it; if N> 3, transmission means from the simplified tree of rank N1 to an intermediate manager of rank N1 included in the level of rank N1, so that the given terminal receives a simplified tree of rank 1 via a cascade of intermediate managers comprising at least the intermediate manager of rank NI.

Advantageously, the central manager comprises means for constructing the decision tree from a set of rules defining a decision policy to be implemented within said architecture, and a set of variables comprising a subscript. set of explanatory variables. In addition, the means of construction comprise means for selecting an explanatory variable so as to create a new node of the tree based on said explanatory variable, the means of choice being used iteratively, starting from a root node to go to leaf nodes, the choice made for each new iteration being carried out among the explanatory variables, called free explanatory variables, not already chosen during a previous iteration. Finally, at each iteration, the means of choice choose an explanatory variable available at the level occupying the highest rank among the ranks of the levels where the free explanatory variables are available, so that the decision tree understands, when it is traversed since the root node to the leaf nodes, a sequence of N sets of nodes, a set of nodes of rank i, ie {N ... 1}, based on explanatory variables available at rank i.

The invention also relates to an intermediate manager of rank m, with m e

{N-1 ... 2}, allowing the partial processing of a decision tree in the context of decision management between a central manager and at least one terminal within a distributed and hierarchical distributed network architecture according to N levels, with N> 3, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1. According to the invention, the intermediate manager comprises: reception means, for a given terminal, a simplified tree of rank m; means for creating a simplified tree of rank m-1 from the simplified tree of rank m; if m = 2, transmission means of the simplified tree of rank m-1 to the given terminal, so that the given terminal makes a decision from the simplified tree of rank m-1 and executes it; if m> 3, means of transmission from the simplified tree of rank m-1 to an intermediate manager of rank m-1 included in the level of rank m-1, so that the given terminal receives a simplified tree of rank 1 via a cascade of intermediate managers including at least the middle manager of rank m-1.

Advantageously, the intermediate manager further comprises: decision-making means, activated if the simplified tree of rank m-1 is reduced to a leaf node; and means for transmitting the decision taken at the given terminal, so that the given terminal executes it.

The invention also relates to a terminal for partial processing of a decision tree in the context of decision management between a central manager and at least said terminal within a distributed network architecture hierarchized according to N levels, with N> 2, the central manager being included in a level of rank N, the terminal being included in a level of rank 1. According to the invention, the terminal comprises: means for receiving a simplified tree of rank 1; decision-making means from the simplified tree of rank 1; and means for executing a decision taken.

5. List of figures

Other features and advantages 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, and the appended drawings, in which: FIG. 1 presents a flowchart of a particular embodiment of the method according to the invention of construction of a decision tree; FIG. 2 shows an exemplary decision tree obtained by implementing the construction method of FIG. 1; FIG. 3 is a generic representation of an elementary structure defining a node of a decision tree according to the invention, making it possible to see the decision tree as a linked list of such elementary structures and thus to transport it; FIG. 4 presents a flowchart of a particular embodiment of the method according to the invention of executing a decision policy contained in a decision tree obtained by implementing the construction method of FIG. 1; FIG. 5 shows an example of a 3-level distributed architecture for mobility management, within which the invention can be implemented; Figure 6 shows an example of a rank level decision tree NI; FIG. 7 shows an exemplary N-2 rank level decision tree; FIG. 8 shows the structure of a central manager according to the invention; FIG. 9 shows the structure of an intermediate manager according to the invention; and Figure 10 shows the structure of a terminal according to the invention. 6. Detailed description The invention therefore relates to a method for constructing a decision tree from, on the one hand, a set of rules defining a decision policy to be implemented within a distributed and hierarchical architecture according to N levels. , N> 2, the level of rank N being the most global level of the architecture and, on the other hand, a set of variables comprising a subset of explanatory variables.

The invention also relates to a method for executing a decision policy contained in a decision tree obtained by executing such a construction method. This method is also called a decision management method between a central manager and at least one terminal within a distributed network architecture hierarchized according to N levels, with N> 2, the central manager being included in a level of rank N , the terminal being included in a level of rank 1.

In general, the method of constructing the decision tree is based on the following elements: a set V of variables, each with a finite domain; a set R of rules or constraints on these variables; a rules-based deduction system or a constraint satisfaction system if the rules express themselves as constraints capable of ingesting the rules R; a subset Vi of V, called set of explanatory variables (also called measurable variables or inputs); a subset Vo of V, disjoined from Vi, called set of target variables

(also called variables to be deduced or outputs).

In the embodiment specific to mobility, the variables of Vi are those specifying the subscription of the terminal (measurable in the client database), the network where the terminal is currently located, the quality of service it obtains. .. The variables of Vo are those specifying whether the handover must be performed or not (required handover worth "true" or "false") and, for each available access network, if this network is a possible candidate for the handover or not (network_candidate [x] worth "true" or "iàlse" (false)). As illustrated by 1 flow diagram of Figure 1, in a particular embodiment, the method according to the invention of construction of the tree comprises the steps detailed below.

In a step 11, we add the set of rules in the deduction system, the latter reduces the domains of the possible values of all the variables of

V.

Then, if all the explanatory variables of Vi have not yet been processed (negative answer to the question of step 12), proceed to a step 13 of choosing an explanatory variable v among those of Vi. This selection step is iterated starting from a root node to go to leaf nodes. In a step 14, we create a new node of the tree based on the explanatory variable v. In a step 15, the deduction system determines the possible values of the explanatory variable v. In a step 16, an arc of the decision tree is created by possible value of the explanatory variable v. In a step 17, for each of the possible values 'a' of the explanatory variable v: we create a new instance of the deduction system by adding the rule v = a (which will propagate new constraints on the possible values of the other variables ); and deleting the explanatory variable v from Vi for the next steps, before returning to step 11.

When all the explanatory variables of Vi have been processed (positive answer to the question of step 12), proceed to a step 18 of recovery of the decision tree in its current unfinalized form. Then, we go to a step 19 during which, for each arc of the decision tree, the deduction system calculates the possible values of the target variables of Vo, and stores them in a sheet of the decision tree. . The end step is referenced 191. The elements referenced 14a, 16a, 18a and 19a connected by dotted lines to the steps

14, 16, 18 and 19, respectively, illustrate the state of the construction of the tree as the construction process is carried out.

According to the invention, at each iteration of the choice step 13, an explanatory variable available at the highest rank level is chosen from the ranks of the levels where the free explanatory variables (i.e. already selected during a previous iteration of step 13). So, the decision tree includes, when traversed from the root node to leaf nodes, a sequence of N sets of nodes, a set of nodes of rank i, ie {N, NI ... 1}, based on available explanatory variables at rank level i.

The choice criterion mentioned above (the explanatory variable chosen is that (or one of those available at the level ranking highest among the ranks of the levels where the free explanatory variables are available) is for example implemented. as follows.

Each of the explanatory variables of Vi is associated with a value corresponding to the rank of the level of the architecture where it is available. For example, the global variables are associated with the smallest value - 0 - and the local variables at the terminal are associated with the highest value. This choice being arbitrary, one can also consider a variant in which the global variables are associated with the highest value. Several variables can be associated with the same value (which means that they are available in the same level of the architecture). Then, during the choice step, the choice of an explanatory variable v among those of Vi is made taking into account the values associated with the explanatory variables. In the above example, the explanatory variable chosen is that (or one of those) associated with the smallest value among the set of values associated with the explanatory variables. Thus, the most comprehensive explanatory variables are chosen primarily.

At the output of the process, we obtain a tree whose nodes closest to the root are the nodes corresponding to the explanatory variables (parameters) of high level (global parameters) and the nodes closest to the sheets correspond to the explanatory variables (parameters). local. The sheets contain the values deduced from each of the target variables of Vo.

For example, in the context of the mobility rules and with a three-level architecture (described below in relation to FIG. 5), we have the following set Vi: variable "Subscription" (value 0 (rank level 3 architecture), variable available at the heart of the network); - "Current network" variable (value 0 (level 3 of the architecture), variable available at the heart of the network); variable "AP wlan load" (value 1 (level 2 of the architecture), variable available in the WLAN access network); variable "Availability [umts]" (value 2 (level 1 of the architecture), variable available at the terminal); - variable "Availability [wlan]" (value 2 (level 1 of the architecture), variable available at the terminal); variable "QoS required" (value 2 (level 1 of the architecture), variable available at the terminal); "Current quality" variable (value 2 (level 1 of the architecture), variable available at the terminal); variable "WLAN quality" (value 2 (level 1 of the architecture), variable available at the terminal).

FIG. 2 shows an example of a decision tree obtained by implementing the construction method of FIG. 1, in the particular case of mobility management and with a three-level architecture.

A sequence of 3 sets of nodes can be seen, comprising: a first set 21 of rank 3 nodes, based on explanatory variables available at level 3 of the architecture (global level); a second set 22 of rank 2 nodes, based on explanatory variables available at level 2 of the architecture (intermediate level); a third set 23 of rank 1 nodes, based on explanatory variables available at level 1 of the architecture (local level).

An example of a decision tree distribution protocol obtained with the construction method according to the invention is now presented in connection with FIG.

The BNF grammar (Backus Normal Form) of the message "Handover Decision Tree" (ADH) is as follows: MESSAGE ADH :: = ARBRE_DECISION_LISTE ARBRE_DECISION_LISTE :: = (TREE DECISION) *

TREE DECISION :: = FLAG NODE DATA (NEXT) * FLAG :: = ROOT | ELEMENT | DECISION

NODE :: = #LABEL

DATA :: = #LABEL

NEXT :: = OPERATION (TREE DECISION | EMPTY) OPERATION :: = (OPERATOR T1 DOMAIN) | (OPERATOR T2 DATA)

OPERATOR T1 :: = IN | OUT

OPERATOR T2 :: = LOWER | SUPERIOR | EGAL | DIFFERENT |

INFERIEUR_OU_EGAL | GREATER THAN OR EQUAL TO

DOMAIN :: = DOMAIN_TYPE NUMBER ELEMENTS (ELEMENT) + DOMAIN TYPE:: = BOOLEEN | NUMERIC | SYMBOLIC

Thus, with the BNF grammar above, a decision tree can be seen as a linked list of elementary structures each defining a given node of the tree. The transport of the tree then simply consists in transmitting all the elementary structures of the linked list. Figure 3 is a generic representation of an elementary structure, defining a given node. It comprises a flag 31, an identifier 32 of the elementary structure, at least one variable 33, an operation 34 to be performed on each variable and a field "next elementary structure" 35.

The flag 31 (called "FLAG" in the BNF grammar above) indicates whether, for the given tree, the given node is a root node (the first node of the tree) (FLAG = ROOT), an intermediate node (FLAG = ELEMENT) or a decision node (FLAG = DECISION).

The identifier 32 of the elementary structure (called "NODE" in the BNF grammar above) is for example obtained by taking the content of the flag and, in the case where the flag indicates "DECISION" or "ELEMENT", a number order for the type of flag concerned. Thus, one obtains for example identifiers such as: ROOT, ELEMENT 2, DECISION 5 ... (see annexes E and F discussed below, in which the first indicated field corresponds to the identifier of the elementary structure; for the sake of simplification, the flag field has not been indicated in these appendices because it is included in the identifier field). The variable or variables 32 (called "DATA" in the BNF grammar above) are the parameter (s) that must be considered in this elementary structure. In the case of an explanatory variable, this may be the quality of a radio link, the availability of an interface of a terminal, etc. For an explanatory variable, different values can lead to different directions in the decision tree. Thus, depending on the different values, one can have two types of action: evaluating another explanatory variable or taking a final decision (obtaining the value of at least one target variable). The flag of an element indicates if the manager continues to consider another parameter (FLAG = ELEMENT) OR if he makes a decision (FLAG = DECISION).

The operation 34 to be performed on each variable is for example defined either with the combination of the fields "OPERATOR T1" 34a and "DOMAIN" 34b, or with the combination of the fields "OPERATOR T2" 34a 'and "DATA" 34b' (DATA 'being another variable, with which the variable considered DATA is compared). The fields "DOMAIN", "OPERATOR T1" and "OPERATOR S" can take different values depending on the variable ("DATA" (for example: integer, character string for the DOMAIN and>, <, =, in, out for OPERATORS).

The field "next elementary structure" 35 contains either the identifier of the elementary structure defining the next node of the tree (content 351 called

"TREE DECISION" in BNF grammar above), ie an end indicator (352 content called "EMPTY" in BNF grammar above).

The filling of the elementary structure can be summarized as follows.

If the given node is a root node, an intermediate node or a decision node which is not a leaf node, the elementary structure comprises an explanatory variable (referenced field 33) and an operation to be performed on the explanatory variable, constituting a test on the value of the explanatory variable (fields referenced 34). For each possible value of the test, the referenced field 35 contains the identifier of the elementary structure defining the next node of the simplified tree of rank j-1. This is the case, for example, of the elementary structures "Racine", "Decision 9", "Decision

10 "and Element 1" to "Element 8" in Annex E. If the given node is a decision node which is a leaf node, the elementary structure comprises at least one target variable and for each target variable, an operation relating to the target variable, constituting the supply of a value of the target variable. For each target variable, the referenced field 35 contains the end indicator "Empty". This is the case, for example, of the elementary structures "Decision 1" to

"Decision 8" in Appendix E.

FIG. 4 shows a particular embodiment of the method according to the invention of executing, within an architecture, a decision policy contained in a distributable decision tree (obtained by setting implementation of the construction method of Figure 1).

As an illustrative example, we assume that N = 3, that is to say that the architecture is distributed and prioritized according to 3 levels. The rank 3 level, the most global of the architecture, includes the central manager who builds the complete decision tree, called a rank 3 tree. The rank 2 level includes several parts each comprising an intermediate manager (rank 2). Rank level 1 includes terminals.

For a given terminal, the method comprises the steps described below. In a step 41, the central manager creates a simplified tree of rank 2 from the rank tree of rank 3, then transmits it to an intermediate manager. In a step 42, the intermediate manager creates a simplified tree of rank 1 from the simplified tree of rank 2, then transmits it to the terminal. In a step 43, the terminal makes a decision from the simplified tree of rank 1 and executes it.

It is clear that the invention is not limited to this particular embodiment.

It may be envisaged, while remaining within the scope of the present invention, that the architecture is distributed and prioritized according to two levels. In this case, the central manager directly creates a simplified tree of rank 1 from the complete decision tree, then transmits it directly to the terminal, so that the terminal makes a decision from this simplified tree of rank 1 and the 'executed.

It can also be envisaged that the architecture is distributed and prioritized according to more than three levels (N> 3). In this case, each intermediate manager of a given rank transmits a simplified tree to the manager of the lower rank, until we reach an intermediate manager of rank 2, who transmits a simplified tree to the terminal. This can be summarized by saying that the following steps are performed, after initializing k to NI: a) the central handler transmits the simplified k-rank tree to an intermediate manager of rank k included in the rank level k; b) the intermediate manager of rank k creates a simplified tree of rank k-1 from the simplified tree of rank k; (c) if k> 3, the intermediate manager of rank k transmits the simplified tree of rank k-1 to an intermediate manager of rank k-1 included in rank level k-1 and returns to step b) after decrementing k by one unit; if k = 2, the intermediate manager of rank k transmits the simplified tree of rank k-1 to the given terminal, so that the given terminal makes a decision from the simplified tree of rank k-1 and executes it.

FIGS. 5 to 7 show a particular embodiment of the two methods according to the invention (method for constructing a distributable decision tree and corresponding execution method, with distribution of the tree). . In the present description, called handover, the passage of a mobile terminal of an access network to another access network of the same technology or not. This intercell transfer is decided and ordered by the central mobility manager.

FIG. 5 shows an example of a 3-level distributed architecture (N = 3) for managing mobility, within which the invention can be implemented. This architecture comprises: a tier 3 level comprising a core network 51 itself comprising a central mobility manager GM; a rank level 2 comprising two access networks 52, 53 (each access network forms part of the level of rank 2). Each of the access networks itself comprises an access manager GA1, GA2; and a rank level 1 comprising a plurality of mobile terminals. For the sake of simplification, a single mobile terminal TM is shown in FIG.

5. The GM central mobility manager includes a PD-GM decision point that updates an R-GM database and builds distributed decision trees based on the information stored in that database. The R-GM database of the central mobility manager GM contains all the information necessary for the implementation of the mobility management policy as well as the global or / and static data as defined for example in Appendix A hereof. - after.

The access managers GA1, GA2 provide the connection to the access networks 52, 53 via the core network 51. Each access manager GA1, GA2 comprises a decision point PD-GA1, PD-GA2 which implements the simplified decision trees received from the central mobility manager GM taking into account local information stored in a database R-GA1, R-GA2. The database R-GA1, R-GA2 of an access manager GA1, GA2 contains the local information of the access network 52, 53 concerned, as defined for example in Appendix B.

Access points PA1, PA2 secure the attachment of the mobile terminals TM to an access manager GA1, GA2 and are connected to a router of an access network.

Each mobile terminal TM contains a PD-TM decision point and an R-TM database. The PD-TM decision point of a mobile terminal implements the simplified decision tree received from one of the access managers GA1, GA2 taking into account local information stored in the database R-TM. The terminal's R-TM database contains the local terminal information as defined, for example, in Appendix C.

In this architecture, the central mobility manager GM does not completely make the handover decision for each terminal. The GM Central Mobility Manager database does not contain all terminal and access point information belonging to specific access networks. On the other hand, the central mobility manager GM has all the knowledge to elaborate the decision trees for the management of handoffs (the algorithm, the rules, the policies for a handover). These trees have been administratively accepted by all member access networks. The base of data of the central mobility manager GM contains static information (subscription, preferences ...) and global information of the access networks.

The main task of the central mobility manager GM is to consult his database to generate decision trees and distribute them to the access managers GA1, GA2 according to the location of the terminals (an access manager receives only the simplified trees corresponding to the terminals that are located in its access network).

Each access manager has a decision point PD-GA1, PD-GA2 and a database R-GA1, R-GA2. The decision point contains the simplified decision trees provided by the central mobility manager GM. The database R-GA1, R-GA2 contains the local information of the access network and some information relating to the terminals present in the access network. The decision point consults its database and follows the simplified decision trees. The result of this journey is transmitted to the mobile terminals. This result can be either a simplified decision tree or a handover decision.

Each TM terminal has a PD-TM decision point and an R-TM database. The decision point contains the decision tree or decision provided by one of the access managers GA1, GA2. The R-TM database contains the local information of the terminal. The PD-TM decision point consults its R-TM database and follows the decision tree and executes the handover decision.

A mobility management policy can be represented by a set of rules chosen for example from those listed in Appendix D. The choice and prioritization of some of these rules (some are exclusive) make it possible to define a global mobility policy. A particular implementation implements for example a set of five rules (see rules 1 to 5 below) which are transformed into decision tree by the construction method defined by the invention (and implemented in a software called COP (Compiler and Propositional Optimizer) which integrates its own deduction engine). The rule language supported by the COP software has been found to be sufficient for carrying out the invention within the mobility described herein. The final decision must whether a handover is required and if so to which access network. The access networks are for example a WLAN 52 and a UMTS 53 network. Rule 1: an access network is a candidate if it is available

Rule 2: A WLAN access network is not a candidate if the quality perceived by the terminal to the access point is less than 2 on a scale of 1 to 5.

Rule 3: Handover is required if the terminal is on a WLAN and the perceived quality on the WLAN is less than 2 on a scale of 1 to 5. Rule 4: Handover is required if the terminal is on a network. cellular network and if the quality of service required is a WLAN QoS. Rule 5: A handover is necessary if the terminal is on an access network

WLAN and it has a non-priority subscription (hereinafter referred to as "Bronze subscription" as opposed to a "Gold subscription" for a priority subscription) and if the access point charge is equal to 2 on a scale of 1 to 5.

The complete tree (also called N-rank tree) corresponding to these rules, generated by the COP software, has 42 sheets and 124 nodes. It has a maximum depth of

7 knots.

The formation of trees of rank NI is now presented. Using the principle described above, it is possible to reduce the tree of rank N by taking into account two parameters (explanatory variables) at the level of the central mobility manager GM. This is the subscription which is a static parameter known to the operator and the access network to which the terminal is connected. This parameter is sent back to the central mobility manager GM by the terminals TM during the power-up and each change of access network 52, 53. Thus, the central mobility manager can calculate simplified decision trees specific to the terminals. . For example, four simplified trees (also called reduced trees) of rank N1 are obtained: if we have a terminal whose subscription is bronze and whose current access network is the WLAN, we obtain a first reduced tree of rank NI of 18 leaves and 5 deep; - for a bronze subscription and a UMTS current access network, a second reduced tree of rank N1 of 12 sheets and a depth of 4 is obtained; for a gold subscription and a WLAN access network, a third reduced tree of rank N1 of 12 sheets and a depth of 4 is obtained; for a gold subscription and a UMTS current access network, we obtain a fourth reduced-rank tree with a rank of 12 sheets and a depth of 4. It is these reduced trees of rank NI which are transmitted to the access managers GA1, GA2 .

By way of example, we detail in Annex E the messages which describe, based on the notion of elementary structure described above in relation to FIG. 3, the first reduced tree of rank N1 (bronze subscription and network of WLAN access.

Figure 6 illustrates this first reduced tree of rank N-I. In this FIG. 6, the first reduced tree of rank NI, which comprises 35 nodes (of which 1 root node and 18 leaf nodes) is described with only the following 19 elementary structures: an elementary structure (whose identifier is "root" in Annex E) defining a root node referenced R; eight elementary structures (whose identifiers are "Element 1" to "Element 8" in Appendix E) defining intermediate nodes referenced E1 to E8; and ten elementary structures (whose identifiers are "Decision 1" to "Decision

10 "in Appendix E) defining decision nodes referenced D1 to D10 (among which only those referenced D1 to D8 are leaf nodes).

It is possible to use a limited number of elementary structures (less than the number of nodes to be defined) because several nodes are defined with the same elementary structure. For example, three nodes E3 are defined with the same elementary structure whose identifier is "Element 3". The formation of N-2 trees is now presented. By using again the principle described above, it is possible to reduce a tree of rank NI by taking into account a parameter (explanatory variable) at one of the access managers GA1, GA2. For the access manager GA1, it is the load parameter ("AP load wlan"). Depending on this parameter, the access manager makes an access network change decision or sends the terminal a reduced tree of rank N-2. By way of example, Appendix F describes the messages which describe, based on the notion of elementary structure described above in relation to FIG. 3, the reduced tree of rank N-2 corresponding to the case where the The variable "AP wlan load" is set to 2. FIG. 7 illustrates this reduced rank tree N-2, which corresponds to a portion of the rank tree NI of FIG. 6. This part is referenced 61 in FIG. In this FIG. 7, the reduced tree of rank N-2, which comprises 11 nodes (of which 1 root node and 6 leaf nodes) is described with only the following 8 elementary structures: an elementary structure (whose identifier is " Root "in Appendix F) defining a root node referenced R '; three elementary structures (whose identifiers are "Element 3", "Element 5" and "Element 6" in Appendix E) defining intermediate nodes referenced E3, E5 and E6; and four elementary structures (whose identifiers are "Decision 1", "Decision 2", "Decision 5" and "Decision 6" in Appendix E) defining decision nodes referenced D1, D2, D5 and D6 (which are all leaf nodes). When it receives a reduced tree of rank N-2, the terminal evaluates the local parameters (explanatory variables) ("quality wlan", "availability umts" and "availability wlan") and makes the final decision. FIG. 8 shows the structure of a central manager according to the invention, which comprises a memory M 81, and a processing unit 80 equipped with a microprocessor μP, which is controlled by a computer program (or application) Pg 82. The processing unit 80 receives all the information 83 necessary for the implementation of a decision policy (for example a mobility management policy), which the microprocessor μP processes, according to the instructions of the program Pg 82, to generate simplified trees 84, which are transmitted to intermediate managers (case N> 2) or terminals (case N = 2).

FIG. 9 shows the structure of an intermediate manager according to the invention, which comprises a memory M 91, and a processing unit 90 equipped with a microprocessor μP, which is controlled by a computer program (or application) Pg

92. Processing unit 90 receives simplified trees from the manager of higher rank (central manager or other intermediate manager), the microprocessor μP processes, according to the instructions of the program Pg 92, to generate simplified trees 94, which are transmitted to lower intermediate managers or terminals, depending on the position the intermediate manager considered within the hierarchical architecture.

FIG. 10 shows the structure of a terminal according to the invention, which comprises a memory M 101, and a processing unit 100 equipped with a microprocessor μP, which is controlled by a computer program (or application) Pg 102 The processing unit 100 receives simplified trees from the higher rank manager (central manager or intermediate manager), that the microprocessor μP processes, according to the instructions of the program Pg 102, to make a decision and execute it. The result of this execution is symbolized by the arrow referenced 104.

Appendix A: Global Data Located in the GM Base

Appendix B: Local Data to Access Networks in GA Bases

Appendix C: Terminal Local Data located in the Terminal Database.

Appendix D: Sample Rules Implemented in a Mobility Management Policy

Prohibit intercellular transfers to prohibited access networks

When an access point becomes unavailable, hand over all terminals on that access point - When the quality on an access network becomes insufficient, make a handover to another available access network

When the load on an access point becomes too large, make a handover of terminals with a lower subscription

When the load on an access point is too large, make a handover of the lowest priority applications

Handover of high speed terminals to access networks with large cell size

Prohibit handover of high-speed terminals to access networks with small cell size - Match the level of security required by applications with the level of security offered by access networks

When the quality of service requested by an application is greater than the quality of service offered by the access network and the quality of service available on another access network is higher then make the handover of the application to this access network

Handover applications when a higher-performance access network becomes available

Handover applications to the access network with the highest quality of service - Handover applications to the least busy access network

Handover all applications from a client to their preferred access network

Handover an application to the least expensive access network Appendix E: Messages (Basic Structures) that describe the first reduced-rank tree (bronze subscription and WLAN access network).

Appendix F: messages (elementary structures) describing the reduced tree of rank N-2 corresponding to the case where the variable "load AP wlan" takes the value 2.

Claims

A decision management method between a central manager (GM) and at least one terminal (TM) within a distributed network architecture hierarchized according to N levels, with N> 2, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1, characterized in that it comprises a step of distribution of a decision tree constructed by the central manager and based on said architecture, the step of distributing the decision tree itself comprising the following steps, for a given terminal: from the decision tree, said tree of rank N, the central manager creates (41) a simplified tree of rank NI ; if N = 2, the central handler transmits the simplified tree of rank N-I to the given terminal, so that the given terminal makes a decision from the simplified tree of rank N-I and executes it; if N> 3, the following steps are performed, after initializing k to NI: a) the central manager transmits (41) the simplified tree of rank k to an intermediate manager of rank k (GA1, GA2) included in the level of rank k; b) the intermediate manager of rank k creates (42) a simplified tree of rank k-1 from the simplified tree of rank k; (c) if k> 3, the intermediate manager of rank k transmits the simplified tree of rank k-1 to an intermediate manager of rank k-1 included in rank level k-1 and returns to step b) after decrementing k by one unit; if k = 2, the intermediate manager of rank k transmits (42) the simplified tree of rank k-1 to the given terminal, so that the given terminal makes a decision from the simplified tree of rank k-1 and the executes (43).

2. Method according to claim 1, characterized in that, in step b) the simplified tree of rank k-1 is reduced to a leaf node, then step c) is replaced by the following step: c ') the intermediate manager of rank k makes a decision and transmits it to the given terminal, so that the given terminal executes it.

3. Method according to any one of claims 1 and 2, characterized in that the transmission of a simplified tree of rank j-1 by a manager of rank j, with je {N ... 2}, is based on a simplified global tree structure comprising a linked list of elementary structures each defining a given node of the simplified tree of rank j-1 and each comprising: a flag (31) indicating whether, for the simplified tree of rank j- 1, the given node is a root node, an intermediate node or a decision node; an identifier (32) of the elementary structure; if the given node is a root node, an intermediate node, or a decision node that is not a leaf node:

* an explanatory variable (33);

an operation (34) to be performed on the explanatory variable, constituting a test on the value of the explanatory variable;

for each possible value of the test, a "next elementary structure" field (35) containing the identifier of the elementary structure defining the next node of the simplified tree of rank j-1; if the given node is a decision node that is a leaf node:

at least one target variable (33);

for each target variable, an operation (34) relating to the target variable, constituting the supply of a value of the target variable;

* for each target variable, a "next elementary structure" field (35) containing an end indicator.

4. Method according to any one of claims 1 to 3, characterized in that the decision tree comprises, when traversed from a root node to leaf nodes, a sequence of N sets of nodes (21, 22, 23), a set of nodes of rank i, ie {N ... 1}, based on explanatory variables available at rank i, in that the creation by a rank manager j of a simplified tree of rank j-1 from a tree of rank j, with je {N ... 2}, is performed taking into account at least one information that the rank manager j has on at least one explanatory variable available at level of rank j, and in that when it arrives on a node of the set of nodes of rank j-1 by traversing the tree of rank j, the rank manager j passes the hand to a manager of rank j-1 so that the rank manager j-1 walks the simplified tree of rank j-1.

5. Method according to any one of claims 1 to 4, characterized in that the transmission of a simplified tree of rank j-1 to a manager of rank j-1, with j e

{N ... 2}, is performed by the rank manager j during a preliminary step of distribution of all trees of rank j-1 possible between all the managers of rank j-1.

6. Method according to any one of claims 1 to 4, characterized in that the decision tree comprises, when traversing from a root node to leaf nodes, a sequence of N sets of nodes (21, 22, 23), a set of nodes of rank i, ie {N ... 1}, based on explanatory variables available at rank i, and in that the transmission of a simplified tree of rank j-1 to a manager of rank j-1, with je {N ... 2}, is carried out by the manager of rank j only after the manager of rank j arrived, by traversing a tree of rank j, on a node d a set of nodes of rank j-1.

7. Method according to any one of claims 1 to 6, characterized in that the decision tree contains a decision policy belonging to the group comprising: mobility decision policies of said at least one terminal within the architecture networks; quality of service decision policies for said at least one terminal within the network architecture; and - security decision policies for said at least one terminal within the network architecture.

8. A method of constructing a decision tree adapted for implementing the decision management method between a central manager (GM) and at least one terminal (TM) according to any one of claims 1 to 7, l decision tree being constructed from a set of rules defining a decision policy to be implemented within a distributed and hierarchical architecture according to N levels, with N> 2, and a set of variables comprising a subset of explanatory variables, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1, the method comprising a step ( 13) of choosing an explanatory variable so as to create a new node of the tree based on said explanatory variable, the selection step being iterated starting from a root node to go to leaf nodes, the choice made with each new iteration being carried out among the explanatory variables, called free explanatory variables, not already chosen during a previous iteration, characterized in that, at each iteration of the selection step (13), a variable is chosen explanatory information available at the highest rank among the ranks of the levels where the free explanatory variables are available, so that the decision tree understands, when it is traversed since the root node to leaf nodes, a sequence of N sets of nodes, a set of nodes of rank i, ie {N ... 1}, based on explanatory variables available at rank i.

9. Method according to claim 8, characterized in that the decision policy belongs to the group comprising: mobility decision policies of said at least one terminal within the network architecture; quality of service decision policies for said at least one terminal within the network architecture; and security decision policies for said at least one terminal within the network architecture.

10. Method according to claim 8, characterized in that the decision policy is a mobility decision policy of said at least one terminal within the network architecture, and in that the architecture is hierarchical according to the three levels. following: a tier 3 level comprising a core network (51); a rank level 2 comprising at least two access networks (52, 53) each forming a part of the rank level 2; a rank level 1 comprising a plurality of terminals (TM) each forming part of the level of rank 1.

11. Method according to claim 10, characterized in that the subset of explanatory variables comprises: at least one variable available at rank 3 and belonging to the group comprising: a variable defining the type of subscription associated with a terminal ;

a variable defining the load of an access point to the access network; at least one variable available at rank 1, for each terminal connected to a given access network, and belonging to the group comprising:

* for each access network, a variable defining the availability of the access network; * a variable defining the quality of service required by the terminal;

* a variable defining the quality of service offered by the given access network.

12. Method according to any one of claims 8 to 11, characterized in that it is implemented in the central manager.

13. A method of partially processing a decision tree by a central manager (GM), as part of the decision management between a central manager and at least one terminal (TM) within a distributed network architecture and hierarchical N levels, with N> 2, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1, characterized in that the central manager performs the following steps, for a given terminal: from the decision tree, said tree of rank N, it creates (41) a simplified tree of rank NI; if N = 2, it transmits the simplified tree of rank N1 to the given terminal, so that the given terminal makes a decision from the simplified tree of rank N1 and executes it; if N> 3, it transmits (41) the simplified tree of rank NI to an intermediate manager of rank NI included in the level of rank NI, so that the given terminal receives a simplified tree of rank 1 via a cascade of intermediate managers comprising at least the intermediate manager of rank NI.

14. The method as claimed in claim 13, characterized in that the central manager first performs a step of constructing the decision tree from a set of rules defining a decision policy to be implemented within said architecture. , and a set of variables comprising a subset of explanatory variables, the constructing step comprising a step (13) of choosing an explanatory variable so as to create a new node of the tree based on said explanatory variable, the selection step being iterated starting from a root node to go to leaf nodes, the choice made at each new iteration being carried out among the explanatory variables, called free explanatory variables, not already chosen by a previous iteration, and that, at each iteration of the choice step, the central manager chooses an explanatory variable available at the highest occupying level rank among the ranks of the levels where the free explanatory variables are available, so that the decision tree understands, when it is traversed from the root node to the leaf nodes, a sequence of N sets of nodes, a set of nodes of rank i, ie {N ... 1}, based on explanatory variables available at rank i.

15. Method of partial processing of a decision tree by an intermediate manager (GA1, GA2) of rank m, with me {N-1 ... 2}, within the framework of decision management between a central manager ( GM) and at least one terminal (TM) within a distributed network architecture hierarchized according to N levels, with N> 3, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1, characterized in that the intermediate manager performs the following steps, for a given terminal: he receives a simplified tree of rank m; from the simplified tree of rank m, he creates (42) a simplified tree of rank m-1

if m = 2, it transmits (42) the simplified tree of rank m-1 to the given terminal, so that the given terminal makes a decision from the simplified tree of rank n

1 and executes it; if m> 3, it transmits the simplified tree of rank m-1 to an intermediate manager of rank m-1 included in the level of rank m-1, so that the given terminal receives a simplified tree of rank 1 via a cascade intermediate managers including at least the middle manager of rank m-1.

16. The method according to claim 15, characterized in that, if the simplified tree of rank m-1 is reduced to a leaf node, then the intermediate manager makes a decision and transmits it to the given terminal, so that the given terminal l 'executed.

17. A method for partial processing of a decision tree by a terminal (TM), as part of the decision management between a central manager (GM) and at least said terminal within a distributed and hierarchical network architecture. according to N levels, with N> 2, the central manager being included in a level of rank N, the terminal being included in a level of rank 1, characterized in that the terminal performs the following steps: - it receives (43) a simplified tree of rank 1; it makes (43) a decision from the simplified tree of rank 1 and executes it.

18. 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 any one of claims 13 and

14 partial processing of a decision tree by a central manager, when said program is run on a computer.

19. 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 executing the steps method according to any of claims 15 and 16 partial processing of a decision tree by an intermediate manager of rank m, when said program is run on a computer.

20. 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 executing the steps method according to claim 17 partially processing a decision tree by a terminal, when said program is run on a computer.

21. Central manager (GM) for partial processing of a decision tree in the context of decision management between a central manager and at least one terminal within a distributed network architecture hierarchical according to N levels, with N> 2, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1, characterized in that the central manager comprises: creation means, for a given terminal, a simplified tree of rank N-

1 from the decision tree, said tree of rank N; if N = 2, transmission means of the simplified tree of rank N-I to the given terminal, so that the given terminal makes a decision from the simplified tree of rank N-I and executes it; if N> 3, transmission means from the simplified tree of rank N1 to an intermediate manager of rank N1 included in the level of rank N1, so that the given terminal receives a simplified tree of rank 1 via a cascade of managers intermediates comprising at least the intermediate manager of rank NI.

22. Central manager according to claim 21, characterized in that it comprises means for constructing the decision tree from a set of rules defining a decision policy to be implemented within said architecture, and a set of variables comprising a subset of explanatory variables, the means of construction comprising means for selecting an explanatory variable so as to create a new node of the tree based on said explanatory variable, the means choice being used iteratively, starting from a root node to go to leaf nodes, the choice made for each new iteration being carried out among the explanatory variables, called free explanatory variables, not already chosen during a previous iteration, and that, at each iteration, the means of choice choose a variable explanatory information available at the highest rank among the ranks of the levels where the free explanatory variables are available, so that the decision tree understands, as it is traversed from the root node to the leaf nodes, a sequence of N sets nodes, a set of nodes of rank i, i G {N ... 1}, based on explanatory variables available at rank i.

23. Intermediate manager (GA1, GA2) of rank m, with m G {N-1 ... 2}, allowing the partial processing of a decision tree as part of the management of decisions between a central manager and the manager. least one terminal within a distributed network architecture hierarchized according to N levels, with N> 3, the central manager being included in a level of rank N, said at least one terminal being included in a level of rank 1, characterized in that the intermediate manager comprises: means for receiving, for a given terminal, a simplified tree of rank m; means for creating a simplified tree of rank m-1 from the simplified tree of rank m; if m = 2, transmission means of the simplified tree of rank m-1 to the given terminal, so that the given terminal makes a decision from the simplified tree of rank m-1 and executes it; if m> 3, means of transmission from the simplified tree of rank m-1 to an intermediate manager of rank m-1 included in the level of rank m-1, so that the given terminal receives a simplified tree of rank 1 via a cascade of intermediate managers including at least the middle manager of rank m-1.

Intermediate manager according to claim 23, characterized in that it further comprises: decision-making means, activated if the simplified tree of rank m-1 is reduced to a leaf node; and means for transmitting the decision taken at the given terminal, so that the given terminal executes it.

25. Terminal (TM) allowing the partial processing of a decision tree as part of the management of decisions between a central manager and at least said terminal within a distributed network architecture hierarchized according to N levels, with N > 2, the central manager being included in a level of rank N, the terminal being included in a level of rank 1, characterized in that the terminal comprises: means for receiving a simplified tree of rank 1; decision-making means from the simplified tree of rank 1; and means for executing a decision taken.