Device and method for determining a distribution of resources of a physical network

A device for determining a distribution of resources of a given physical network among logical links by subdividing physical link capacities into said logical links using an algorithm, comprising:

• first neural network means, in which one part of the algorithm is implemented;
• and
• second neural network means, in which a second part of the algorithm is implemented, said first and second neural network means interworking to compute logical link capacities,
• wherein said first neural network means comprises interconnected neuron means that are arranged in two layers, referred to as an upper layer and a lower layer, wherein said first neural network means is a discrete-time linear programming neural network, operating on said one part of the algorithm.

A device for determining a distribution of resources of a given physical network among logical links by subdividing physical link capacities into said logical links using an algorithm, comprising:

• first neural network means, in which one part of the algorithm is implemented;
• and
• second neural network means, in which a second part of the algorithm is implemented, said first and second neural network means interworking to compute logical link capacities,
• wherein said second neural network means comprises interconnected neuron means that are arranged in two layers, referred to as an upper layer and a lower layer,
• wherein said second neural network means is a discrete-time neural network, operating on said second part of the algorithm, and
• wherein said discrete-time neural network is a link blocking computing neural network (LBCN).

A method for determining a distribution of resources of a given physical network among logical links by subdividing physical link capacities into said logical links using an algorithm, wherein said method further comprises the steps of:

• using first neural network means, in which one part of said algorithm is implemented, and second neural network means, in which a second part of said algorithm is implemented;
• and
• bringing said first and second neural network means to interwork in order to compute logical link capacities whereby the operation of said physical network, given an objective function, is generally optimized according to said objective function,
• wherein said algorithm as input has parameters describing the physical network, the topology of said logical links and traffic demands, and
• wherein said inter-working between said first neural network means, being a continuous-time linear programming neural network, and said second neural network means, being a discrete-time neural network, is an iterative process which comprises the following steps:
• (1) said first neural network means and said second neural network means are initialized,
• (2) said first neural network means completely solves a linear programming task, thereby computing logical link capacities,
• (3) said logical link capacities, computed in step (2), are sent, after suitable signal transformations, to said second neural network means,
• (4) said second neural network means solves a non-linear recursive system of equations, thereby computing link blocking probabilities,
• (5) a convergence test is carried out, and if the convergence conditions are not satisfied, said computed link blocking probabilities are transformed according to a function of said algorithm, and then sent, after suitable signal transformations, to said first neural network means and the steps (2)-(5) are repeated, while if the convergence conditions are satisfied, said logical link capacities computed by said first neural network means in step (2) are the logical link capacities that essentially optimize the operation of the physical network, according to said objective function.

A method for determining a distribution of resources of a given physical network among logical links by subdividing physical link capacities into said logical links using an algorithm, wherein said method further comprises the steps of:

• using first neural network means, in which one part of said algorithm is implemented, and second neural network means, in which a second part of said algorithm is implemented;
• and
• bringing said first and second neural network means to interwork in order to compute logical link capacities whereby the operation of said physical network, given an objective function, is generally optimized according to said objective function,
• wherein said algorithm as input has parameters describing the physical network, the topology of said logical links and traffic demands, and
• wherein said inter-working between said first neural network means, being a continuous-time linear programming neural network, and said second neural network means, being a discrete-time neural network, is an iterative process which comprises the following steps:
• (1) said first neural network means and said second neural network means are initialized,
• (2) said first neural network means completely solves a linear programming task, thereby computing logical link capacities,
• (3) said logical link capacities, computed in step (2), are sent, after suitable signal transformations, to said second neural network means,
• (4) said second neural network means executes one iteration in a non-linear recursive programming task, thereby computing link blocking probabilities,
• (5) a convergence test is carried out, and if the convergence conditions are not satisfied, said computed link blocking probabilities are transformed according to a function of said algorithm, and then sent, after suitable signal transformations, to said first neural network means and the steps (2)-(5) are repeated, while if the convergence conditions are satisfied, said logical link capacities computed by said first neural network means in step (2) are the logical link capacities that essentially optimize the operation of the physical network, according to said objective function.

A method for determining a distribution of resources of a given physical network among logical links by subdividing physical link capacities into said logical links using an algorithm, wherein said method further comprises the steps of:

• using first neural network means, in which one part of said algorithm is implemented, and second neural network means, in which a second part of said algorithm is implemented;
• and
• bringing said first and second neural network means to interwork in order to compute logical link capacities whereby the operation of said physical network, given an objective function, is generally optimized according to said objective function,
• wherein said algorithm as input has parameters describing the physical network, the topology of said logical links and traffic demands, and
• wherein said inter-working between said first neural network means, being a continuous-time linear programming neural network, and said second neural network means, being a discrete-time neural network, is an iterative process which comprises the following steps:
• (1) said first neural network means and said second neural network means are initialized,
• (2) said first neural network means executed one iteration in a linear programming task, thereby computing logical link capacities,
• (3) said logical link capacities, computed in step (2), are sent to said second neural network means,
• (4) said second neural network means executes one iteration in a non-linear recursive programming task, thereby computing link blocking probabilities,
• (5) a convergence test is carried out, and if the convergence conditions are not satisfied, said computed link blocking probabilities are transformed according to a function of said algorithm, and then sent to said first neural network means and the steps (2)-(5) are repeated, while if the convergence conditions are satisfied, said logical link capacities computed by said first neural network means in step (2) are the logical link capacities that essentially optimize the operation of the physical network, according to said objective function.

A device for determining a distribution of resources of a given physical network among logical links by subdividing physical link capacities into said logical links, comprising a neural network configuration that comprises:

• a first individual neural network, in which a first part of a predetermined logical link capacity optimization algorithm is implemented for generating a first result;
• a second individual neural network, in which a second part of said algorithm is implemented for generating a second result;
• means for mutually interconnecting said first neural network and said second neural network, wherein a representation of said first result is transferred by said interconnecting means to said second neural network, and a representation of said second result is transferred by said interconnecting means to said first neural network such that said first neural network and said second neural network are mutually interworking in an iterative manner to compute capacity values of said logical links;
• and
• a control unit for stopping the iterative and mutual interworking of said first neural network and said second neural network when a predetermined condition is satisfied.

A method for determining a distribution of resources of a given physical network among logical links by subdividing physical link capacities into said logical links, comprising the step of:

• using a neural network configuration that includes:
• a first continuous-time linear programming neural network, in which a first part of a predetermined logical link capacity optimization algorithm is implemented,
• a second discrete-time neural feedback network, in which a second part of said algorithm is implemented, and
• means for mutually interconnecting said first neural network and said second neural network,
• wherein said first linear programming neural network and said second neural feedback network are mutually interworking in an iterative manner to compute logical link capacities, and said algorithm uses a given objective function, and said interworking comprises the steps of:
• (1) initializing said first neural network and said second neural network;
• (2) said first neural network completely solving a linear programming task, thereby computing logical link capacities,
• (3) sending said logical link capacities, computed in step (2), after suitable signal transformations, to said second neural network,
• (4) said second neural network solving a non-linear recursive system of equations, thereby computing link blocking probabilities, and
• (5) performing a convergence test, and if convergence conditions are not satisfied, transforming said computed link blocking probabilities according to a function of said algorithm, and sending said transformed probabilities, after suitable signal transformations, to said first neural network and repeating steps (2)-(5), while if said convergence conditions are satisfied, said logical link capacities computed by said first neural network in step (2) are the logical link capacities that essentially optimize the operation of the physical network, according to said objective function.

A method for determining a distribution of resources of a given physical network among logical links by subdividing physical link capacities into said logical links, comprising the step of:

• using a neural network configuration that includes:
• a first continuous-time linear programming neural network, in which a first part of a predetermined logical link capacity optimization algorithm is implemented,
• a second discrete-time neural feedback network, in which a second part of said algorithm is implemented, and
• means for mutually interconnecting said first neural network and said second neural network,
• wherein said first linear programming neural network and said second neural feedback network are mutually interworking in an iterative manner to compute logical link capacities, and said algorithm uses a given objective function, and said interworking comprises the steps of:
• (1) initializing said first neural network and said second neural network;
• (2) said first neural network completely solving a linear programming task, thereby computing logical link capacities,
• (3) sending said logical link capacities, computed in step (2), after suitable signal transformations, to said second neural network,
• (4) said second neural network executing only a few iterations in a non-linear recursive programming task, thereby computing link blocking probabilities, and
• (5) performing a convergence test, and if convergence conditions are not satisfied, transforming said computed link blocking probabilities according to a function of said algorithm, and sending said transformed probabilities, after suitable signal transformations, to said first neural network and repeating steps (2)-(5), while if said convergence conditions are satisfied, said logical link capacities computed by said first neural network in step (2) are the logical link capacities that essentially optimize the operation of the physical network, according to said objective function.

A method for determining a distribution of resources of a given physical network among logical links by subdividing physical link capacities into said logical links, comprising the step of:

• using a neural network configuration that includes:
• a first discrete-time linear programming neural feedback network, in which a first part of a predetermined logical link capacity optimization algorithm is implemented,
• a second discrete-time neural feedback network, in which a second part of said algorithm is implemented, and
• means for mutually interconnecting said first neural network and said second neural network,
• wherein said first neural feedback network and said second neural feedback network are mutually interworking in an iterative manner to compute logical link capacities, and said algorithm uses a given objective function, and said interworking comprises the steps of:
• (1) initializing said first neural network and said second neural network;
• (2) said first neural network executing only a few iterations in a linear programming task, thereby computing logical link capacities,
• (3) sending said logical link capacities, computed in step (2) to said second neural network,
• (4) said second neural network executing only a few iterations in a non-linear recursive programming task, thereby computing link blocking probabilities, and
• (5) performing a convergence test, and if convergence conditions are not satisfied, transforming said computed link blocking probabilities according to a function of said algorithm, and sending said transformed probabilities to said first neural network and repeating steps (2)-(5), while if said convergence conditions are satisfied, said logical link capacities computed by said first neural network in step (2) are the logical link capacities that essentially optimize the operation of the physical network, according to said objective function.