JPH0731704B2 - Routing method using neural network - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an adaptive packet routing system in a packet network, and more specifically, each node constituting a multimedia integrated network holds a neural network and uses it to transmit packets from its own node. It relates to a routing method using a neural network for obtaining the optimum output direction of.

[0002]

2. Description of the Related Art In the future, communication networks will continue to grow in speed and capacity, and will tend to develop into a multimedia integrated network capable of accommodating various media in a mixed manner. In such an integrated network, the required performance such as communication speed, reliability, and real-time performance varies greatly in time and space.Therefore, flexible communication path setting according to changes in the network status and when a failure occurs There is a need for a routing technology that can instantly set up a bypass path.

Application of a packet switching network is considered as a technique for realizing such communication. As a switching method in a packet switching network, there are a datagram method that does not set a logical link between terminals, that is, a method that does not take the concept of a call, and a virtual call method that sets a logical channel between terminals. is there.

The datagram-based routing currently performed in such a packet-switched network is based on a system in which the routing information up to the destination node is periodically rewritten according to the change in the state of the network. This is based on the premise that traffic fluctuations are relatively gentle and the optimum path does not change frequently.

In addition, in the routing of the virtual call system, for example, the amount of traffic accommodated in the network is predicted in advance at the time of designing the network, and it is optimal that the predicted traffic is not concentrated on a specific node between the originating and terminating nodes. The route is obtained, the path information is held in each node, and the call setup of the virtual call is performed based on the path information.

FIG. 52 shows a conventional example of a path setting method in a packet switching network. In the figure, it is assumed that the packet is transmitted from the source node (A) 1 to the destination node (B) 2.
It is assumed that a path ~ n0 exists between the source node 1 and the destination node 2. On the other hand, each node is provided with a routing table as optimum path information for each requested quality of information, for example, for each requested band of media, and path setting is performed based on the contents of these tables.

In addition, in case of failure in a node or link in a virtual call, a plurality of detour routes are prepared between the originating and destination nodes in order to avoid the failure point, and when a failure occurs, the path is re-established between the originating and destination nodes. Then, the virtual call communication is resumed.

FIG. 53 shows a conventional example of setting a fault bypass route in a packet switching network. In the figure, a route passing through the nodes 3, 4 and 5 is set as an optimal route between the terminal A and the terminal B, and a route passing through the nodes 3, 6, 4 and 5 and a node 3 are set as detour routes. 6,
A route through 7 and 5 is set. Also terminal C
Between D and D, a route passing through the nodes 4, 6 and 8 is set as an optimum route, and a route passing through the nodes 4, 5, 7 and 8 is set as a bypass route.

[0009]

As described above, in the conventional datagram type routing, as shown in FIG. 54, it is necessary to periodically update the routing table up to the destination node according to the change in the state of the network. However, in a situation where the scale of the network is large and traffic fluctuations are large, the calculation time for calculating the optimum path explosively increases, and there is a problem that it is not possible to realize the routing adapted to the momentary state change.

Further, even if an optimum path is obtained at the time of inputting a packet by the virtual call method, the range of temporal and quantitative changes of the accommodated media becomes so large that it cannot be compared with the conventional packet switching network. It will be impossible to predict how to use the media and find the optimum path for each media type and network state in advance. In addition, there is a problem that the calculation time for calculating the optimum path explosively increases as the network becomes larger.

If a detour is newly set between the originating node and the destination node in the event of a link or node failure, the communication of the virtual call is interrupted from the occurrence of the failure to the setting of the detour. There is also the problem that it is unprofitable.

In the past, a single medium was often handled in a network, and routing was not necessarily performed so as to satisfy the different quality requirements of various media. However, in the integrated multimedia network, it is essential to realize the routing that satisfies the required quality for each medium.

To solve such a problem, we have already proposed a method for realizing adaptive routing with respect to the state of the network by a control network in which threshold elements are mutually connected (Japanese Patent Application No. 1-507916). Heisei 1
(Filed on July 6, 2012) {US Ser. No. 455,323 (July 6, 19
89)}.

In the above-mentioned application, a threshold element corresponding to a neuron is arranged at each node, and the entire real network constitutes one neural network. In this case, the time until the neural network obtains a stable optimal path depends on the distance between the nodes, so that there is a problem that adaptability to a state change cannot be obtained in a network with a long distance between the nodes.

An object of the present invention is to adaptively realize a routing that satisfies the conditions required by various media and the reliability of communication even in a network with a long distance between nodes.

That is, in the present invention, each node holds a neural network of the same form and closes each node to perform the calculation of the neural network, and this configuration does not depend on the distance between the nodes. , Realizes the routing adapted to the change of the network state.

[0017]

1 to 5 are block diagrams showing the principle of the present invention. FIGS. 1 to 5 are principle block diagrams of a routing method using a neural network in a packet switching network including, for example, an integrated multimedia network 10 which accommodates various media such as voice, images and data in a packet format.

1 to 5, a multimedia integrated network 10 includes a plurality of nodes 11 and a link 1 connecting the nodes.
4 and. Multimedia integrated network 1
The neural network 12 provided inside each node 11 forming 0 determines the output direction of the packet from its own node, and the external stimulus input means 13 is the current state of the integrated multimedia network 10 and the integrated state. External stimuli are input to the neurons in the neural network 12 according to the requirements of the media accommodated in the network 10. In the neural networks 12 of FIGS. 1 to 5, black circles represent nodes in which packets to be routed exist (hereinafter referred to as routing nodes), or source and destination nodes of packets. .

FIG. 1 is a principle block diagram of the first and sixth inventions. The first invention is applied to a multimedia integrated network using a datagram method, and as the current state of the integrated network 10, for example, a packet discard rate and a packet delay time in each link 14 are used. Also, as the media requirement, for example, an allowable packet discard rate, which is specified by a terminal or the like that inputs the packet and indicates how much the packet may be discarded, is used, and external stimulus input is performed according to these values. Using the external stimulus provided by the means 13, the neural network 12 determines the optimum output direction of the packet from the node.

The mutual connection type neural network 12 in FIG. 1 is composed of, for example, two types of neurons, one for each node 11 and one for each link 14, and external stimulus input means 13. Inputs an external stimulus to a neuron corresponding to one-to-one to the link 14, each neuron constituting the neural network 12 adds an input from another neuron and the external stimulus, and thresholds the addition result. The operation of outputting 0 or 1 is repeated, and the node corresponding to the node corresponding neuron whose output is stable to 1 is the node in the packet output direction, and the link corresponding to the link corresponding neuron whose output is 1 is stable is the link of the packet output direction. Is determined as

For a communication medium using the datagram method, every time a packet moves in the integrated multimedia network 10, the optimum output direction of the packet is obtained at the node to which the packet is input. By repeatedly performing, the packet arrives at the destination node via the optimum path.

FIG. 2 is a block diagram of the principle of the second invention. The second invention is also applied to the multimedia integrated network using the datagram method. When this figure is compared with FIG. 1 showing the principle of the first invention, it is different in that a packet passing node number detecting and adding means 15 and a neuron output fixing means 16 are added.

The packet passing node number detecting and adding means 15 detects the number of the node through which the packet has passed from the packet input to the own node, adds the own node number to the packet, and adds the packet to the packet. Send to adjacent node. Further, the neuron output fixing means 16 fixes the output of the neuron, which has a one-to-one correspondence with the node through which the input packet has passed in the neural network 12, at the time of routing the packet from its own node.

As a result, in the second invention, the path passing through the node through which the input packet has passed can be excluded and the input packet can be routed, and the calculation time of the neural network 12 can be reduced. At the same time, it is possible to avoid a path that causes a packet ping-pong phenomenon or a loop phenomenon.

FIG. 3 is a block diagram showing the principle of the third invention. In this figure, state information notifying means 17 is provided as compared with FIG.
And another node state information value reduction means 18 is added. The status information notifying means 17 is the multimedia integrated network 1
Information indicating the state related to the own node among the current states of 0 is notified to all the nodes configuring the integrated network 10. In the third aspect of the invention, the notified state information is the packet discard rate of each link as the state of each output link of the node and the packet delay time. Further, the other node state information value decreasing means 18 is the other node state information notifying means 17
The value of the state information received from the node is reduced according to the distance from the own node to other nodes, and the reduced result is output to the external stimulus input means 13 and used for calculation of the external stimulus.

As a result, in the third invention, when the output direction of the packet is obtained in the routing node, the optimum path can be obtained without being affected by the state of the node away from the routing node. That is, by decreasing the received state information value as the distance from the routing node increases, the influence of the state of the node at a distance can be reduced.

FIG. 4 is a block diagram showing the principle of the fourth invention. Comparing this with FIG. 3 showing the principle of the third invention,
The only difference is that the other node state information value reducing means 18 is provided in place of the other node state information and notification range value reducing means 20.

However, the status information notifying means 19 is
Unlike the status information notifying means 17 in the third invention,
In the present state of the multimedia integrated network 10, not only the state related to the own node, for example, the state of each output link described above, but the range in which the state information is notified inside the integrated network 10 is set to all other nodes. To notify.

According to this, the other node state information and notification range value reduction means 20 reduces the value of the state information from the other node and the value of the notification range according to the distance to the other node, and decreases the notification range. When the result is not 0, the decrease results are notified to the adjacent node and are output to the external stimulus input means 13 in the own node, and when the decrease result in the notification range is 0, the decrease results are output to the adjacent node. It is output to the external stimulus input means 13 without being notified.

As a result, in the fourth invention, the third
In the same manner as in the invention described above, not only is the value of the state information reduced according to the distance, but the notification range is also limited, so the amount of communication for state notification can be reduced.

FIG. 5 is a block diagram of the principle of the fifth invention. Comparing this figure with FIG. 1 showing the principle of the first invention,
The difference is that the state information average / notification unit 21 and the other node state information reception processing unit 22 are added.

The state information averaging / notifying means 21 averages the values representing the states of the output links of its own node among the states of the integrated network 10 in a plurality of types of cycles, and is variable according to the cycles. The value indicating the notification range in the integrated network that conveys the average result and the average result are notified to each node that configures the integrated network in the plurality of types of cycles.

Another node status information reception processing means 22
Reduces the state information average of other nodes and the value indicating the notification range received from the notification means 21 by 1, and when the reduction result is not 0, outputs the received state information average value to the external stimulus input means 13. The notification range decrease result and the state information average value are output to all the adjacent nodes, and when the notification range value decrease result is 0, the state information average value is output to the external node without external node output. Output to.

As a result, in the fifth invention, a plurality of notification frequencies and notification ranges of the link status information from each node to other nodes are provided, and in the routing node, the status of the near node is far from the instantaneous value that fluctuates in real time. For the node state, the optimum route is calculated using the average value of a longer time. Therefore, it is possible to prevent the traffic increase due to the status notification and realize the routing adapted to the network status.

The principle of the sixth invention is the same as that of FIG. 1 as described above. However, the state of the network used for the external stimulus input by the external stimulus input means 13 and the media requirement are different from those of the first invention. The sixth invention is directed to a multimedia integrated network using a virtual call system.
The available bandwidth of each link, the bandwidth in use, the allowable utilization rate of the link, and the packet delay time are used as the status of, and the required bandwidth of the information input from the terminal is used as the media requirement condition. .

In the virtual call system, at the time of call setting at the packet originator node, the mutual connection type neural network 12 held in the own node is used to make a node corresponding neuron whose output is stable to 1 in the same manner as described above. The corresponding node is determined as the packet output direction node, and the link corresponding to the link corresponding neuron whose output is stable at 1 is determined as the packet output direction link. Then, according to this result, the entire path to the destination node is set. Subsequent data packets are sent to the destination node via the set path.

[0037]

EXAMPLE FIG. 6 shows the connection relationship of two neurons in the neural network 12. In the figure, the neuron UV and the neuron XY are integrated by the integration coefficient T _{UV, XY} , and the product of the output V _{UV of the} neuron UV and this coupling coefficient is input to the neuron XY. The output of another neuron adjacent to the neural network 12 and the output of its own neuron are weighted by the coupling coefficient and input to the neuron XY. The external input I _{XY is} also input. The neuron XY takes the sum of these, thresholds the result, and outputs the output V _XY . This output value is either 0 or 1.

In the present invention, each node constituting the integrated multimedia network holds the neural network of the same configuration, and the output of the neuron is exchanged in the closed state in the node to satisfy the condition required by the medium. In addition, the optimum output direction of the packet is obtained by associating the objective function that minimizes the packet transmission delay with the energy function of the neural network in a one-to-one correspondence and minimizing the energy function.

Generally, in a neural network, the operation of each neuron is defined by the following state equation (1), and its output state is stable in that the energy function of equation (3) defined by the entire neural network is minimized. It is known that there is a property to do.

[0040]

[Equation 1]

[0041]

[Formula 2] V _XY = f (U _XY ) (2)

[0042]

[Equation 3]

Where U _XY is the internal state of the neuron XY,
V _XY neuron XY output state, T _{XY, X 'Y'} is the coupling coefficient between neurons XY and neuron X'Y ', I _XY
Is an external stimulus (external input) to the neuron XY, f (X)
Is a threshold function.

In order to find the optimum output direction of the packet from the node in the multimedia integrated network by using the neural network as described above, the outputs of the neurons corresponding to the nodes i and the links ij are V _i and V respectively.
_ij and if the nodes and links are on the path of the optimal path,
These outputs are defined as variables that are 1.

Then, in the neural network in which the outputs of the neurons corresponding to the source node and the destination node are fixed to 1, a path that minimizes the packet transmission delay between the source node and the destination node is calculated using the above-mentioned energy minimization principle. I will ask for it. The objective function for minimizing the transmission delay between the source and destination nodes is given by the following equation.

[0046]

[Equation 4]

Here, s is a source node, d is a subscript representing a destination node, d _ij is a packet delay time when the link ij is passed, and A _ij is 1 when the link ij is available, and is non-existent. It is a variable that becomes 0 when possible.

In the equation (4), the first term to the third term are constraint conditions for the existence of only one optimum path, and the fourth term represents the transmission delay between the originating node and the destination node. The objective function and the energy function of the above-mentioned neural network, that is, the equation (3) are made to correspond one-to-one, and the coupling function of each neuron and the external stimulus to the neuron are obtained.

FIG. 7 shows an example of the connection relation of neurons in the neural network, and this example corresponds to the case where the real network, that is, the multimedia integrated network is composed of three nodes i, j, and k. . Inside each of the three nodes i, j, and k, a neural network having the same structure as shown in FIG. 7 is provided. The coupling coefficient of the neuron in this neural network is equal to the coefficients C ₁ and C ₂ in the objective function in order to make the objective function correspond to the energy function.

Next, the external stimulus to the link-corresponding neuron ij is given by the following equation by referring to the coefficient of the variable V _{ij in} the equation (4).

[0051]

[Equation 5] C ₃ A _ij −C ₄ d _ij (5)

Here, the external stimulus indicates a value input from the external stimulus input means 13 of FIGS. 1 to 5, and in particular, the variable A _ij indicating whether or not the link corresponding to the link corresponding neuron ij can be used is as follows. It becomes possible to model with another neuron according to the equation of state of.

[0053]

[ _Equation 6] A _ij = f [L-PL _ij ] (6)

Here, f [X] is a threshold function, where L is a permissible packet discard rate designated by the input terminal of the packet as a media requirement, and PL _ij
Is the packet discard rate on link ij. The allowable packet discard rate (L) is the packet discard rate (PL) of the link.
_ij ), the packet can be accepted on the link, and thus the variable A _ij indicating whether or not the link is available becomes 1.

FIG. 8 is a configuration block diagram of an embodiment of a node in the first invention. In the figure, a node is a neural network 31 for determining the output direction of a packet from the node, a network state and a request information holding unit 32 for holding a state of a multimedia integrated network and a requirement condition of a medium, a network state. And an external stimulus input processing unit 33 that outputs an external stimulus to the neural network 31 using the storage result in the request information holding unit 32, an optimum path display unit 34 that holds the output result of the neural network 31 and displays the optimum path, A packet output direction determining unit 35 that determines the output direction of a packet from the own node, a reception packet processing unit 36 that receives an input packet from an adjacent node or a terminal connected to the self node, and a packet that the reception packet processing unit 36 receives Is a status notification packet, process it State notification packet processing unit 37 that outputs the packet delay time and the packet discard rate of each link of the other node to the network state and request information holding unit 32, and the network state information regarding the own node, State information generation / output to the request information holding unit 32 and generation of a state notification packet to another node
It is composed of the notification packet generator 38.

FIG. 9 is a configuration block diagram of an embodiment of the external stimulus input processing section in the first invention. In the figure,
In the external stimulus input processing unit, a neuron 41 is provided for each link corresponding to i = 1 to n, j = 1 to n, and an allowable packet discard rate L for each neuron,
And the packet loss rate PL _ij of each link are input,
The value of A _ij is obtained by the equation (6), and the coefficient units 42 and 43 and the adder 44 are calculated from this value and the packet delay time d _ij.
Using, the external stimulus to the neuron corresponding to the link in the neural network 31 of FIG. 8 is output in accordance with the equation (5).

FIG. 10 is a block diagram of an embodiment of the neural network in the first invention. In the figure, the neural network includes a neuron 45 that corresponds to a node one-to-one and a neuron 4 that corresponds to a link one-to-one.
6, and between each neuron, coefficient units 47 and 48 representing coupling coefficients C ₁ and C ₂ are provided between the neurons as described with reference to FIG.
External stimuli are input from the external stimulus input processing unit 33 to the neurons corresponding to each link.

FIG. 11 shows a data packet format embodiment in the first invention. In the figure, the data packet includes an identifier indicating that the packet is a data packet, the numbers of the source node and the destination node of the data packet, the numbers of the destination node, and the requirements of the media accommodated in the packet. The allowable packet discard rate of is stored.

FIG. 12 shows a format embodiment of a status notification packet for status notification from each node to another node. In the figure, the status notification packet includes an identifier indicating that the packet is a status notification packet, the node number that sent the packet, and the sequence number at the sending node for the status notification packet, followed by the node Corresponding to each output link, the link number, the packet delay time in the link, and the packet discard rate are stored.

FIG. 13 shows an embodiment of the contents of the network status holding unit that holds the packet delay time and the packet discard rate in the network status and request information holding unit 32 in FIG. In the figure, for example, the own node 1
The number of output links from is 4 and the link number is 1
1 to 14, and the output links of other nodes, for example, node 2, are numbered 21 to 24, and the packet delay time and the packet discard rate are stored for each link.
In the network status and request information holding unit 32,
Since the allowable packet discard rate has only one value for the packet to be routed, as described with reference to FIG. 11, the value is held in a buffer (not shown).

FIG. 14 is a configuration block diagram of an embodiment of the state information generation / notification packet generation unit 38 in FIG. In the figure, the state information generation / notification packet generation unit includes an output link packet waiting capacity holding unit 50 that holds a waiting capacity of a packet corresponding to each output link from the own node, and a capacity of each output link from the own node. An output link capacity holding unit 51 for holding a packet, a delay time calculation unit 52 for calculating a packet delay time for each link using the contents of the two holding units 50, 51, and a packet discarded for each output link Input packet capacity holding unit 5 in the output link that holds the capacity of the input packet and the capacity of the input packet
3, output of the discard rate calculation unit 54 that calculates the packet discard rate using the contents of the storage unit 53, the sequence number storage unit 55 that stores the sequence number for the status notification packet, the delay time calculation unit 52, and the discard rate calculation unit 54 Is used to add a sequence number output from the sequence number holding unit 55 to assemble a state notification packet to another node and output the packet to an adjacent node.

The packet delay time is obtained by dividing the packet waiting capacity on the output link by the capacity of the corresponding output link, and the result is input to the status notification packet assembling unit 56 and the network status and request information of FIG. It is also output to the holding unit 32 as a delay time for the output link of its own node. Similarly, the packet discard rate is obtained by dividing the packet discard capacity in each link by the input packet capacity, and the result is output to the network status and request information holding section together with the status notification packet assembling section 56.

FIG. 15 shows the optimum path display section 34 in FIG.
It is a configuration block diagram of an embodiment of. In the figure, the optimum path display unit is a first register 58 (Reg 1) for storing the output of the neuron corresponding to the node in the neural network 31 and the output of the neuron corresponding to the link in the neural network 31. Of the second register 59 (Reg 2) for storing In the figure, dotted arrows indicate the correspondence between a node and an output link of that node.

FIG. 16 is a flow chart showing an embodiment of the packet output direction decision processing by the packet output direction decision unit 35 in FIG. In the figure, in step S61, a pointer is set in the second register 59 of FIG. 15 in the row corresponding to the node number i of the self node, and in step S62, the neuron corresponding to the output link in the row is one-to-one. Of the output V _ij is searched for the column number j of “1”, and the packet is output to the output link having the adjacent node number j in S63.

Next, an embodiment of the second invention will be described. The second invention is applied to the multimedia integrated network using the datagram method as described above. However, the ping-pong phenomenon in which a packet is repeatedly exchanged between two nodes, and the same route is repeated many times. In order to avoid the looping phenomenon, the number of the node through which the packet passed is added to the header of the packet, and the output of the neuron, which has a one-to-one correspondence with the node number at the routing node, is fixed at '0'. By placing the packet, the calculation time of the neural network is reduced except for the path passing through the node through which the packet has passed. As an example, it is possible to limit the number of nodes through which the packet has passed to five, to limit the number of nodes to which the packet has been output to the adjacent node, or to limit only the source node of the packet. it can.

The embodiment of the second invention is almost the same as the embodiment of the first invention including the configuration of the node shown in FIG. 8, and only different parts will be described. FIG. 17 is an example of the header information in the data packet in the second invention.
In (a) of the figure, the number of passing nodes of the packet (maximum value 5) and the number of the passing node are stored at the beginning of the header (next to the identifier), and after that, as other header information, for example, The originating and terminating node numbers in 10 and others are stored.

FIG. 17B shows that a node that outputs a packet holds only its own node number in the packet, and this number corresponds to the adjacent node number to the node that received the packet. To do. Hereinafter, this number is called an adjacent node number, and other header information is stored following this adjacent node number. Further, in (c) of the figure, only the source node number of the packet is held at the beginning of the header, and other header information is stored after that.

FIG. 18 is a block diagram showing the configuration of the first embodiment of the node number detection and neuron output fixing unit in the second invention. In the figure, the node number detecting and neuron output fixing unit is provided between the reception packet processing unit 36 and the packet output direction determining unit 35 in the node of FIG.

In FIG. 18, the node number detecting and neuron output fixing unit has a node number holding unit 65 that holds the number of the passing node held in the received packet, and the number of passing nodes held in the packet is "1". And the passing node number stepping unit 66, the maximum number of nodes held in the packet, here, the maximum node number holding unit 67 holding “5”, the passing node number stepping unit 66 and the maximum node number holding unit 67. Of the output of the comparator 68, the own node number holding unit 70 that outputs the own node number to the determination circuit 69, and the determination circuit 69.
And the determination circuit 69, which updates the number of transit nodes stored in the packet based on the output of
Output from the node number adding unit 72 that adds the node number to the packet, and the output of the node number holding unit 65 to the neuron corresponding to the passing node number, an external stimulus that sets the output to "0" It is composed of an external stimulus holding unit 73 for inputting.

Here, the external stimulus output from the external stimulus holding unit 73 to the neuron corresponding to each node on a one-to-one basis is "0" for the node through which the packet has not passed.
Also, for a node through which the packet has passed, it is a negative number'I 'having a large absolute value for making the output of the neuron corresponding to that node' 0 '.

In FIG. 18, the difference between the addition result of the number of passing nodes and "1" and the maximum number of nodes, here "5", is calculated by the comparator 68, and the difference value is positive (not including 0).
Is 1 when it is, and 0 when it is negative (including 0).
9 is output.

When the output of the comparator 68 is 1, the judgment circuit 69 has the maximum number of passing nodes held in the packet, which is 5 in this case, and it is not necessary to update the number of passing nodes. The own node number from the own node number holding unit 70 is the node number addition unit without outputting the addition result of the number of passing nodes input from the node number incrementing unit 66 and “1” to the passing node number updating unit 71. It is output to 72 and the own node number is added.

In this case, the passing node number updating unit 71 shifts the five node numbers shown in FIG. 17A to the left one by one, and creates a free area on the rightmost side. Then, the node number adding unit 72 sets the own node number in the empty area, and the packet is input to the packet output direction determining unit 35.

On the other hand, when the output of the comparator 68 is 0, there is a vacancy in the storage area of the node number, and the value of the pass node number + 1 is updated from the judging circuit 69 to update the pass node number. It is output to the updating unit 71 and the number of transit nodes in the packet is updated. The stored node number is not shifted, the packet is output to the node number adding unit 72, and the number of the own node is added to the position of the updated number of passing nodes.

FIG. 19 is a block diagram of the second embodiment of the node number detection and neuron output fixing unit. In the figure, the second embodiment is similar to the first embodiment of FIG. 18 from the external stimulus holding unit 73 and the received packet processing unit 36 that output external stimuli to the neurons corresponding to the nodes one to one. Of the input packet, the adjacent node number detecting unit 76 detects the adjacent node number, and the node number updating unit 77 updates the passing node number stored in the packet.

Then, the adjacent node number detecting unit 76 detects the adjacent node number that has output the packet to its own node, and outputs a negative value I having a large absolute value to the neuron corresponding to the adjacent node according to the result. ,
The output of that neuron is fixed at 0.

The node number updating unit 7 is added to the input packet.
After the number of its own node is stored as the adjacent node number by 7, the packet is output to the adjacent node via the packet output direction determining unit 35.

FIG. 20 is a block diagram of the third embodiment of the node number detection and neuron output fixing unit. This figure shows an embodiment in which only the source node number is held in the packet, as shown in FIG. 17 (c). In the figure, the source node number detection unit 78 detects the source node number from the input packet output from the received packet processing unit, and from the external stimulus holding unit 73 to the neuron corresponding to the detected node on a one-to-one basis. An external stimulus I that fixes the output of the neuron to 0 is output.

Next, the third embodiment of the invention, the principle of which is shown in FIG. 3, will be described. In the third invention, when determining the output direction of a packet in the routing node, in order to obtain an optimum path without being affected by the state of the node having a long distance, the value of the state information notified from another node is A method is used that decreases according to the distance to the node.

Since the third invention is similar to the first embodiment described with reference to FIGS. 8 to 16 in many respects including the node configuration, the parts different from the first embodiment will be described. Only explained.

FIG. 21 is a block diagram showing the configuration of the status notification packet processing unit of the third invention, that is, the block 37 of FIG. In the figure, the status notification packet processing unit 1
Reference numeral 10 is a sequence number check unit 1 for checking the sequence number added to the received packet output from the received packet processing unit 36 of FIG.
11 and the state information value reduction processing unit 11 that reduces the value of the state information of another node stored in the received state notification packet according to the distance from the own node to the other node.
2 and.

FIG. 22 shows the sequence number check unit 111.
It is a configuration block diagram of. In the figure, the sequence number checking unit holds the latest sequence number holding unit 113 holding the number of the packet received most recently, and the sequence number of the packet input from the received packet processing unit in the latest sequence number holding unit 113. When the sequence number of the received packet is less than or equal to the latest sequence number held by the comparison of the sequence number comparison unit 114 and the sequence number comparison unit 114, the received packet is discarded and the received packet is received from the latest sequence number. When the sequence number is larger, the relay discard processing unit 115 that relays the received packet to the state information value reduction processing unit 112 is provided.

FIG. 23 is a block diagram showing the configuration of the state information value reduction processing unit in the third invention. In the figure, the state information value reduction processing unit 112 has a notification data latch unit 117 that latches notification data from the state notification packet sent from the sequence number check unit 111, and the number of hops from its own node to another node in the network. Hop count table 118 holding the information, notification data latch unit 117
The notification data reducing unit 119 is configured to reduce the notification data latched in.

In FIG. 23, when a status notification packet is input from the received packet processing unit via the sequence number check unit 111, the packet is relayed to an adjacent node, and the node number from which the packet is transmitted is The number of hops from the own node to the node is retrieved from the hop number table and the state information stored in the packet, that is, the notification data is latched in the notification data latch unit 117.

The notification data latched in the notification data latch unit 117, here, the packet delay time d in each output link of the packet sending node, and the packet discard rate PL are determined by the hop count H _i stored in the hop count table 118. The notification data reduction unit 119 performs division, and the division result is output to the network state holding unit in the node, block 32 in FIG.

Next, an embodiment of the fourth invention, the principle of which is shown in FIG. 4, will be described. In the fourth invention, as in the third invention, the value of the status information from another node is reduced according to the distance to the node, but at the same time, the status notification packet is a notification that should notify the packet. The range is held, the value of the notification range is also reduced according to the distance, and the relay of the status notification packet is stopped when the notification range becomes '0'.

Similar to the third invention, only the points different from the first and third inventions will be mainly described.
FIG. 24 shows an example format of a status notification packet in the fourth invention. This is shown in FIG. 12 for the first invention.
Compared with, the only difference is that the notification range hop count H is stored next to the sequence number.

FIG. 25 shows a state information generation / notification packet generation unit for notifying other nodes of the state information, and in FIG. 8, a state notification packet assembling unit in the block 38, that is, a configuration block of the embodiment of the block 56 of FIG. It is a figure.

In this figure, the state notification packet assembling section 121 is connected to the sequence number holding section 55 and the notification range hop count holding section 122 as in FIG. Creates a status notification packet that stores the value H of the notification range held by the notification range hop count holding unit, using the delay time of each input output link and the packet discard rate,
The packet is output to the adjacent node.

FIG. 26 is a detailed block diagram of the state information reduction processing unit in the fourth invention. This state information reduction processing unit is provided in the state notification packet processing unit as in FIG.

In FIG. 26, the state information value reduction processing unit has a notification data latch unit 125 for latching notification data from another node from a received packet input from the received packet processing unit via the sequence number check unit. hop count checking unit 127 for checking the status notification the notification range hop number H ₁ stored in the packet '1' only notified decreasing range hops countdown processing unit 126, the output of the notification range hop count countdown processing unit 126, As a result of the check by the hop count check unit 127, the countdown value H _{1 of the} notification range
If the value of -1 is not 0, the hop count check unit 12
7, a hop count updating unit 128 that updates the hop count in the status notification packet and outputs the status notification packet to an adjacent node, for example, a status in the network that is fixed for a certain period depending on the network status. The notification range holding register 129 that holds the value H _R of the notification range of the notification packet and the notification data reduction unit 130 that decreases the value of the state information latched by the notification data latch unit 125 are configured.

In the notification data reduction unit 130, the status information of the other nodes latched by the notification data latch unit 125, that is, the packet delay time d and the packet discard rate PL.
And are divided by H _R -H ₁ +1 and the division result is output to the network state and requirement holding unit. Here, H _R is the number of hops of the notification range held in the notification range holding register 129, and H ₁ is the number of notification range hops stored in the received status notification packet. For example, when the number of nodes that notify the status notification packet is limited to adjacent nodes in the network, the notification range holding register 12
The value of H _R held in 9 becomes 1.

Further, in the status notification packet sending node, since the notification range hop number H stored in the packet is set to 1 and the packet is output, the hop number H ₁ stored in the received packet becomes 1 and the result is notified. The data reduction unit 130 outputs the value of the status information received from the adjacent node as it is to the network status and requirement holding unit.

In FIG. 23 for the third invention, the notification data reduction unit 119 divides the notification data from the adjacent node by the hop number 1 to the adjacent node and outputs the divided data.
That is, it is the same as outputting the value of the state information from the adjacent node as it is. For example, H in the notification range holding register
_When the value of _R is 5 and the sending node of the status notification packet outputs the packet with the notification range hop number H set to 5, the value of H ₁ detected from the received packet at the second node from the sending node is The value of the status information received as a result is divided by 2 and used for packet routing.

Next, a fifth embodiment of the invention, the principle of which is shown in FIG. 5, will be described. In the present invention, the notification range of the status information is designated and transmitted in the status information notification packet as in the fourth invention, and the node receiving the packet reduces the value of the notification range and relays it to the adjacent node. However, the value of the status information is not reduced.

However, a plurality of notification ranges are provided, and the notification frequency of packets is changed corresponding to the plurality of notification ranges. As a result, a node that is distant from its own node is notified of an average state value for a long time, and a nearby node is notified of an instantaneous value that fluctuates in real time as a state value. And packet routing is performed according to network conditions.

FIG. 27 is an explanatory view of the concept of the fifth invention. In the fifth invention, the value of the information indicating the state of a certain link, for example, l is averaged for each of a plurality of time intervals, and the average value with a short time interval is notified to a node close to the own node in the time interval. , The average value with a long time interval is notified to the node distant from the own node.

The time interval of this status notification and its notification range are classified by class in this embodiment, and the average of the link status is taken for each of class 1, class 2 ,. The notification time interval and the notification range are divided into a plurality of stages.

That is, in FIG. 27, classes 1, 2, ...
Time average 132 against the state 131 of link l is provided for each of ..., notification interval 133 _1, 133
₂ , the notification interval control 134 is performed according to the following, and when the notification time interval has passed, a time average 132
State value as an output of, hop count 135 _1, 1
The range specified by 35 ₂ , ... Is notified by the packet transmission 136. Here, classes 1, 2
The hop counts h ₁ , h ₂ , ... For .. indicate the relay range of the packet, and the state information stored in the packet is not always used in all the relayed nodes. This will be described with reference to FIG. 28 below.

FIG. 28 is an explanatory view of the relay range of the status notification packet in the fifth invention and the routing range thereof. As will be described later, the status notification packet contains
For each of the classes 1, ₂ , ... In addition to the relay ranges h ₁ , h ₂ , ... Of the packet, the number of hops from the node that uses the state information stored in the packet at the time of routing h _1s , h _2s , ... Are stored.

[0102] In FIG. 28, the first packet of the state notifying packet is stored is 1 as 3, h _1s as h _1, the state notifying packet by 1 the value of h ₁ upon reaching the adjacent node It is relayed while being reduced, and is used for routing in the adjacent node of the packet generation node, the second node, and the third node.

On the other hand, in the second packet, h ₂ is 5 and h _2s is 4, and the number of hops from the packet generation node is 5 while the value of h ₂ is decreased by 1 in the packet. Although the packet is relayed to the node, since h _2s is 4, the packet is only relayed from the node and is not used for routing, but is used for routing with the node.

FIG. 29 is a configuration block diagram of an embodiment of the state calculation circuit of the link l in the class i which takes the time average of the state of the link L in FIG. In the figure, the state calculation circuit includes a latch circuit 141 for latching the state 140 of the link 1, an adder 142 for adding the output of the latch circuit 141 every time t, a down counter 143 for instructing the end of the addition period, and a latch. A clock circuit 144 that supplies a clock at a time interval t to the circuit 141 and the down counter 143, and a latch circuit 14 that latches the output of the adder 142 when the down counter 143 counts out.
5, the output of the latch circuit 145 is notified the number of interval clocks _ni
The division circuit 146 divides by.

The number of clocks n _i as an interval for notifying the time average and the state value is preset in the down counter 143 and the dividing circuit 146, and the notifying time interval is n _i × t. This state calculation circuit is provided for each class by (number of output links from its own node) × (type of network state value). That is, assuming that the network state value is two types, that is, the packet delay time and the packet discard rate of each output link of the own node, and the number of output links from the own node is three, this state calculation circuit 6 are provided for each.

FIG. 30 is a detailed block diagram of the state information generating / notifying packet generating unit in the node, that is, the block 38 in FIG. Comparing this figure with FIG. 14, the sequence number holding units 55 ₁ and 55 ₁ for each class are provided.
₂ , 55 ₃ and the status notification packet assembling section 56 ₁ ,
56 _2, 56 ₃ are provided, and differs in that state calculating circuit for further calculating the state information of each link have been added.

In this figure, the number of output links from the own node is 3, and the class is from 1 to 3. The state calculation circuit shows the packet delay time d of each link for each class. State calculation circuit 148 for calculating the average value
Are provided, and three state calculation circuits 149 for calculating the average value of the packet discard rate PL are provided.

FIG. 31 is a detailed configuration block diagram of the status notification packet processing unit in the fifth invention, that is, the block 37 in FIG. In the figure, the sequence number check unit 111 first checks the sequence number for the status notification packet from the received packet processing unit, as in FIG. 21 for the third invention.
The packet is processed.

In the processing of the packet, first, the hop number judgment processing section 152 performs the judgment processing described later,
When the state information stored in the received packet is used for routing in the own node, the information is stored in the notification data latch unit 153, and when the notification range hop count is not 0, the state notification is sent. The packet is relayed.

As a result of the judgment by the hop number judgment processing unit 152, when the status information is not used for the routing in the own node, only the status notification packet is relayed if necessary. The processing after the output of the notification data latch unit 153 will be described later.

FIG. 32 shows an embodiment format of the status notification packet in the fifth invention. Comparing this figure with FIG. 24 for the fourth invention, it is different in that the number of hops h _s from the node that uses the state information at the time of routing is added. This means that the state information is not used at the time of routing from the packet generating node to the node having the hop number h _s -1.

FIG. 33 is a detailed block diagram of the hop number determination processing unit 152 in FIG. In the figure,
The hop number determination processing unit holds a hop number table 155 that holds the number of hops from its own node to each node in the network, and a hop number table 155 that stores the hop number h _s from the state information using node stored in the received packet. A hop count comparison unit 156 that compares the search hop count H _i retrieved from the packet, the relay range hop count stored in the packet,
For example, the relay range hop number countdown processing unit 157 that subtracts ₁ from h ₁ ′, the hop number count checking unit 158 and the checking unit 15 that determine whether to relay the status notification packet received by the output of the processing unit 157.
When it is determined that 8 relays the status notification packet,
The hop count updating unit 159 updates the relay range hop count in the received data packet input via the check unit 158 and outputs the status notification packet to the adjacent node.

In FIG. 33, the hop count h _s from the node that uses the status information stored in the packet for routing is detected from the received status notification packet, and this is detected by the hop count comparison unit 156. The search hop count H _i as the hop count from the own node to the packet generating node stored in 155 is compared.

Then, h stored in the packet_sThe value of the
Is the search hop count H from the hop count table 155_iLess than
Status information stored in the packet when
31 is used for routing, the notification data link in FIG.
The notification unit 153 outputs the notification data. to this
To h_sIs the search hop count H from the hop count table _i
If larger, the notification data is used for routing
Hold in the notification data latch unit 153 without being locked
It remains as it was.

In the fifth invention, for example, when the distance from the own node is large and the notification interval of the status information is long,
The state information is predicted in order to predict the change in the state information value during the notification interval.

FIG. 34 is an explanatory diagram of the concept of state information prediction. In the figure, the status 161 of the link 1 is stored in the first status register 163 or the second status register 164 via the switch 162. It is assumed that the first state register 163 stores the state value at the time t−1, and the second state register 164 stores the state value at the time t, and the prediction function unit 1 uses these two state values.
The state value at time t + 1 is predicted by 65, and the state value is given to the external stimulus input means 13 to be used for packet routing.

Here, as the prediction function used by the prediction function unit 165, for example, the following equation is used for performing linear prediction.

[0117]

[Equation 7]

In FIG. 31, the notification data latch unit 15
Of the state information output from 3 for use in routing of the own node, the packet delay time d is the prediction function unit 1
66 and the delay time latch unit 167, and the packet discard rate PL is calculated by the prediction function unit 171 and the discard rate latch unit 1.
72 and 72.

At the same time, from the notification data latch unit 153, the corresponding link number is stored in the link number storage units 168 and 1.
Given to 73. The data latched by the delay time latch unit 167 and the discard rate latch unit 172 is stored in the delay time holding table 169 and the discard rate holding table 174 with a delay of one time.

That is, assuming that the state information input to the prediction function units 166 and 171 is at time t, the state information output from the tables 169 and 174 is time t-.
It becomes the thing in 1.

Using the values of the state information at these two times, the prediction functions 166 and 171 output the predicted values of the packet delay time and the packet discard rate at time t + 1, respectively, and these predicted values are stored in the network state holding unit. Is output to.

FIG. 35 shows the prediction function units 166 and 17 of FIG.
It is an example of the prediction function used in 1. (A) in the figure shows the packet delay time, and (b) in the figure shows the prediction function used to predict the packet loss rate. Linear prediction is performed in these prediction functions.

FIG. 36 shows an embodiment of a neural network as a prediction function section. (A) of the figure shows the configuration at the time of learning. For example, the value of the packet delay time d at time t-1 is given to the input layer unit of the three-layer neural network 181, and the predicted value at time t from the output layer unit. Is output.

An error between the predicted value and the value of the packet delay time at the actual time t is taken by the subtractor 182, and the back propagation 183 is performed based on the error.
The weighting coefficient in the three-layer neural network 181 is changed by.

(B) of the figure shows after learning, and the packet delay time d at the time t is the three-layer neural network 1
81, and the value of the packet delay time at time t + 1 is output from the neural network. For the packet discard rate, a predictive function unit using a three-layer neural network can be constructed in the same manner.

In FIG. 36, the packet delay time and the packet discard rate are input to different three-layer neural networks, but it is naturally possible to input them all to the same three-layer neural network.

FIG. 37 shows an embodiment of the prediction function unit for inputting the packet delay time and the packet discard rate into the same three-layer neural network. In the figure, the three-layer neural network 186 has a number of units that is twice the number of output links from its own node in both its input layer and output layer. Each unit has a packet delay time and a packet delay time in each link. Values such as the discard rate are input, and predicted values of those values are output from the output layer.

Then, at the time of learning shown in FIG.
These errors are taken by 7, and the weighting coefficient is changed by the back propagation 188 using the errors, and the state value at the current time t is output to the external stimulus input means 13 and used for packet routing. To be

After learning (b) in the figure, the current state value at time t is input to the three-layer neural network 186, and the neural network outputs time t.
The state value at +1 is output and the external stimulus input means 1
Given to 3.

Next, an embodiment of the sixth invention will be described. The block diagram of the principle of the sixth invention is the same as that of FIG. 1 showing the principle of the first invention, but the sixth invention is applied to a virtual call type multimedia integrated network, and external stimulus is applied. As the state of the integrated network given to the input means 13, the capacity of each link in the integrated network, the currently used bandwidth of each link, the utilization rate of each link, and the packet delay time at each link are used, and the requirements of the media are used. The required bandwidth of the media input from the terminal is used as.

FIG. 38 is a block diagram showing the structure of an embodiment of the node in the sixth invention. Comparing this figure with FIG. 8 in the first invention, the packet output direction determining unit 35 and the state information generating / notifying packet generating unit 38 do not exist, and instead, the call setting packet generating unit 85 and the delay time calculating unit 8
8. A status notification packet generation unit 89 is provided, and the capacity of each link, the current bandwidth used, the link utilization rate, the packet delay time, and the media required bandwidth are held in the network status and media request information holding unit 82. Is different.

FIG. 39 is a block diagram showing the configuration of an embodiment of the external stimulus input processing section in the sixth invention. In the virtual call system, the variable A _ij indicating whether or not the link ij can be used is obtained according to the following equation.

[0133]

## _EQU8 ## A _ij = f [α (C _ij −F _ij ) −T] (8)

Here, T is the required bandwidth of the medium, C _ij is the capacity of the link ij, F _ij is the bandwidth in use, and α is the maximum allowable link utilization rate. In FIG. 39, a neuron 9 for inputting an external stimulus to a neuron corresponding to a link one-to-one in a neural network 81 in a node.
1, the capacity C _{ij of} each link, the used bandwidth F _ij , the link utilization rate α, and the required bandwidth T of the media are input,
The value of the variable A _ij is calculated according to the equation (8), and the coefficient unit 92,
The external stimulus of equation (5) is obtained by 93 and the adder 94, and is output as the external stimulus input of the neuron corresponding to the link.

FIG. 40 shows an embodiment of the stored contents of the network status holding unit in the sixth invention. In the figure,
Similar to FIG. 13 in the first invention, the link capacity, the used band, and the packet delay time are stored for each of the output links 11 to 14 of the own node and the output links 21 and subsequent ones of other nodes.

FIG. 41 shows a format embodiment of the status notification packet in the sixth invention. The packet format of the figure is the same as that of FIG. 12 in the first invention,
The difference is that the network status stored for each output link is the capacity of the link, the bandwidth used, and the packet delay time.

FIG. 42 is an explanatory diagram of the processing by the delay time calculation section. In the virtual call system, the packet delay time d _ij of each link is calculated by the following equation for each link, and each node notifies all other nodes of this value using a status notification packet.

[0138]

[Equation 9]

However, C _ij is the link capacity, and F _ij is the used bandwidth. In FIG. 42, the delay time calculation unit 88 is the network state and media request information holding unit 8 in the node of FIG.
The packet delay time d _ij is calculated by the equation (9) using the link capacity C _ij and the used bandwidth F _ij held in 2,
The result is output to the network status and media request information holding unit 82.

FIG. 43 shows an embodiment format of a call connection request and call setup packet. In the virtual call system, a call as a virtual call is connected by first relaying a call setup packet from a source node to a destination node as a destination node in the multimedia integrated network.

As shown in (a) of FIG. 43, first, a call connection request packet is sent from the source terminal of the packet to the source node accommodating the terminal, and the packet becomes a call setup packet at the source node. It is corrected and transmitted to the destination node via the relay node.

In (b) of the figure, first, an identifier indicating that the packet is a call connection request packet as a packet identifier at the source terminal of the packet, the number S of the source node, and the number d of the destination node Requested bandwidth T
A call connection request packet in which is set is created and sent to the calling node. The number of the relay node is added to the subsequent area at the source node, and the packet identifier is converted into an identifier indicating that the packet is a call setup packet, and a call setup packet is created.

FIG. 44 is an explanatory diagram of processing by the reception packet processing unit 86 in the node of FIG. In the figure, the packet identifier is determined at 95 for the packet received from the adjacent node or the terminal of the own node.
When the packet identifier indicates a call connection request, the call connection request packet is sent to the call setup packet generator 85, converted into a call setup packet, and output to the adjacent node.

When the packet identifier indicates that it is a status notification packet from another node, the status notification packet is output to the status notification packet processing unit 87. If the packet identifier indicates that it is a call setup packet, the left and right values of the origination / destination node number and the relay node number field stored in the call setup packet are set in 96 in the routing table described later, and in 97. The required bandwidth T of the medium stored in the call setup packet is added to the used bandwidth F corresponding to the output link number of the packet in the network state holding unit described in FIG.
In the call setup packet, the relay node number stored in the relay node number field is shifted to the left by one, and the leftmost relay node number is taken out. Since this number represents the adjacent node number, the call setup packet is output to the output link of this adjacent node number.

FIG. 45 shows an embodiment of the stored contents of the routing table held in the received packet processing unit. (A) of the figure is an example of the routing table held in the node 4 in the topology network as shown in (b) of the figure. For example, for a virtual call having a source node number of 1 and a destination node number of 6, 6 is held as an output adjacent node number and 46 is held as an output link number.

By storing such contents in the routing table, after the call setup packet is transferred, the data packet is routed according to the contents of this routing table.

In FIG. 44, when the packet identifier judgment 95 by the received packet processor determines that the received packet is a data packet, the originating and destination node number is detected for the packet, and the result is used in 99 for routing. The table is searched, and the packet is output to the output link corresponding to the adjacent node number stored in the routing table.

FIG. 46 is a flowchart showing an embodiment of call setup packet generation by the call setup packet generator 85 in the sixth invention. The display content by the optimum path display unit 84 in the sixth invention is exactly the same as that in FIG. 15 in the first invention.

In FIG. 46, first, in step S101, the own node number, that is, the originating node number S is set in the pointer on the second register 59 of FIG. Then, in S102, the column number j corresponding to the pointer value in the row of the second register 59 where V _ij is 1 is searched, and the value of j is set in the pointer in S103. The value is left-justified in the relay node number field in the call setup packet, that is, set to the leftmost.

Next, in S104, it is determined whether or not the value of the pointer is equal to the value of the destination node number. If they are not equal, the processing from S102 is repeated to add the relay node number. However, for example, in the second processing of S102, the value of the pointer is the number of the column searched in the first S102, and among the neurons corresponding to the output links of the adjacent nodes corresponding to this column number, The column number indicating the neuron whose output is 1 is searched,
In S103, the value of the search result is set in the second area from the left in the relay node number field.

When the pointer value becomes equal to the destination node number value in S104, the leftmost node number in the relay node number field in the call setup packet in S105,
That is, the adjacent node number of the calling node is held in the routing table, and the call setup packet is output to the output link corresponding to the adjacent node number in S106. Prior to the output of the call setup packet, the packet identifier is changed from the call connection request to the call setup as described above.

FIG. 47 is an explanatory view of a concrete example of the call setting packet generating process of FIG. In the network shown in (a) of the figure, if a call is set up from the source node to the destination node via the relay node, first, S1 in FIG.
A link-corresponding neuron whose own node number is set to 01, here is set to a pointer, and whose output is 1 in the row of the second register corresponding to the node in S102 as shown in FIG. 47 (b) Is retrieved and 2 is obtained as its column number.

The column number is set in the pointer in S103 and stored in the left end of the relay node number field as shown in FIG. Since the value 2 of the pointer is not equal to the value 4 of the destination node number in S104, S1 is again set.
Returning to 02, the row corresponding to the value 2 of the pointer, that is, the column number of the link-corresponding neuron whose output is 1 in the second row of the second register, which is 4 here, is searched.

The column number is set in the pointer in S103 and is stored in the second in the relay node number field. Since the pointer value 4 is equal to the destination node number value 4 in S104, the processes of S105 and S106 are executed. After that, the call setup packet generation process ends.

FIG. 48 shows an example of call setting in a virtual call. In the figure, in the originating node A, an optimum path satisfying the required quality is obtained by using a neural network according to the required quality of the medium and the current state of the network, and that path is set as the path for the virtual call.

FIG. 49 shows an embodiment of a simulation model for optimum path setting in a real network consisting of 6 nodes. In the figure, the path of the virtual call that minimizes the packet delay time from the source node to the destination node was obtained by simulation.

FIG. 50 shows the result of the simulation.
This figure shows the change of the energy function with respect to the number of state updates of the neuron until the optimum path is obtained in the network of FIG. 49. After 19 times of the state update of the neuron, the output of the neuron becomes stable and the corresponding neuron changes. The nodes where the output V _{i of} is stable at 1 are and
The path that uses these nodes as relay nodes was found as the optimal path with the minimum delay time.

Generally, for example, the link capacity C _ij
Is not equal to the capacity C _ij of the reverse link, but in FIG. 49, assuming that the path is set from the source node to the destination node, the same path is set in the reverse direction on the assumption of bidirectional communication. Was done.

FIG. 51 shows an example of setting a bypass path when a link failure occurs. In the figure, when a failure occurs in the link used for communication during virtual call communication, the failure detection node that detects the failure uses the neural network in its own node to set the detour path, and the destination node Communication of the virtual call to B continues uninterrupted.

Also in the third invention, although the status information is notified, the status change is affected only by the call setup and cancellation, the time interval of the status change is longer than that of the datagram, and the notified status is It always reflects the state at that time. Therefore, if the state information is changed as in the invention of the third to fifth state notification information, the correct state information will not be transmitted, which is not suitable for the datagram method.

[0161]

As described in detail above, according to the present invention, in a multimedia integrated network accommodating various media, it is possible to adapt to the ever-changing network condition and to meet the media requirements. It is possible to determine the optimum packet output direction in various datagrams and set up a virtual call.

Further, by limiting the range of status information notification in the network and by reducing the status information value from other nodes according to the distance between the nodes, it is possible to prevent the traffic increase due to the status notification and to separate from the other nodes. Adaptive routing that is not affected by the state of the node can be realized, and it contributes to effective use of the network and improvement of communication reliability.

Furthermore, the present invention can be utilized for packet routing in an integrated multimedia network which accommodates various media including voice, image and data in a packet format. In addition to the multimedia integrated network,
It is naturally possible to apply it to a network in which data is transmitted in bursts, such as a computer network.

[Brief description of drawings]

FIG. 1 is a principle block diagram of first and sixth inventions.

FIG. 2 is a principle block diagram of a second invention.

FIG. 3 is a principle block diagram of a third invention.

FIG. 4 is a principle block diagram of a fourth invention.

FIG. 5 is a principle block diagram of a fifth invention.

FIG. 6 is a diagram showing a connection relationship between two neurons.

FIG. 7 is a diagram showing a configuration of an embodiment of a neural network in each node.

FIG. 8 is a block diagram showing a configuration of an embodiment of a node in the first invention.

FIG. 9 is a block diagram showing a configuration of an embodiment of an external stimulus input processing unit in the first invention.

FIG. 10 is a diagram showing a configuration of an embodiment of a neural network in the first invention.

FIG. 11 is a diagram showing an embodiment format of a data packet in the first invention.

FIG. 12 is a diagram showing an example format of a status notification packet in the first invention.

FIG. 13 is a diagram showing an example of contents of a network state holding unit in the first invention.

FIG. 14 is a block diagram showing a configuration of an embodiment of a state information generation / notification packet generation unit.

FIG. 15 is a block diagram showing a configuration of an example of an optimum path display unit.

FIG. 16 is a flowchart illustrating an embodiment of packet output direction determination processing.

FIG. 17 is a diagram showing an example of header information in a data packet in the second invention.

FIG. 18 is a block diagram showing a configuration of a first embodiment of a node number detection and neuron output fixing unit in the second invention.

FIG. 19 is a block diagram showing a configuration of a second embodiment of a node number detection and neuron output fixing unit.

FIG. 20 is a block diagram showing a configuration of a third embodiment of a node number detection and neuron output fixing unit.

FIG. 21 is a block diagram showing a configuration of a status notification packet processing unit in the third invention.

FIG. 22 is a block diagram showing a configuration of a sequence number check unit.

FIG. 23 is a block diagram showing a configuration of a state information value reduction processing unit.

FIG. 24 is a diagram showing an example format of a status notification packet in the fourth invention.

FIG. 25 is a block diagram showing a configuration of an embodiment of a status notification packet assembling unit in the fourth invention.

FIG. 26 is a block diagram showing a detailed configuration of a state information reduction processing unit.

FIG. 27 is a diagram illustrating the concept of the fifth invention.

FIG. 28 is an explanatory diagram of a relay range of a status notification packet and its routing use range.

FIG. 29 is a block diagram showing a configuration of an embodiment of a state calculation circuit.

FIG. 30 is a block diagram showing a detailed configuration of a state information generation / notification packet generation unit in the fifth invention.

FIG. 31 is a block diagram showing a detailed configuration of a status notification packet processing unit.

FIG. 32 is a diagram showing an example format of a status notification packet in the fifth invention.

FIG. 33 is a block diagram showing a detailed configuration of a hop count determination processing unit.

FIG. 34 is an explanatory diagram of a concept of state information prediction.

FIG. 35 is a diagram showing an example of a prediction function.

FIG. 36 is a diagram showing an example of a neural network as a prediction function unit.

FIG. 37 is a diagram showing an embodiment of the prediction function unit when the packet delay time and the packet discard rate are input to the same three-layer neural network.

FIG. 38 is a block diagram showing the configuration of an embodiment of a node in the sixth invention.

FIG. 39 is a block diagram showing the configuration of an embodiment of an external stimulus input processing unit in the sixth invention.

FIG. 40 is a diagram showing an example of stored contents of a network state holding unit in the sixth invention.

FIG. 41 is a diagram showing an example format of a status notification packet in the sixth invention.

FIG. 42 is an explanatory diagram of processing by a delay time calculation unit.

FIG. 43 is a diagram showing an example format of a call connection request and call setup packet.

FIG. 44 is an explanatory diagram of processing by a received packet processing unit.

FIG. 45 is a diagram showing an example of stored contents of a routing table.

FIG. 46 is a flowchart illustrating an example of call setup packet generation.

FIG. 47 is an explanatory diagram of a specific example of call setting packet generation processing.

FIG. 48 is a diagram showing an example of call setting in a virtual call.

FIG. 49 is a diagram showing an example of a simulation model of optimum path setting.

FIG. 50 is a diagram showing simulation results.

FIG. 51 is a diagram illustrating an example of setting a bypass path when a link failure occurs.

FIG. 52 is a diagram showing a conventional example of a path setting system in a packet switching network.

[Fig. 53] Fig. 53 is a diagram illustrating a conventional example of fault bypass route setting in a packet switching network.

FIG. 54 is a diagram illustrating a problem of a conventional routing method.

[Explanation of symbols]

 10 multimedia integrated network 11 node 12, 31 neural network 13 external stimulus input means 14 link 15 packet passing node number detecting and adding means 16 neuron output fixing means 17, 19 state information notifying means 18 other node state information value reducing means 20 other Node status information and notification range value reduction means 21 Status information averaging and notification means 22 Other node status information reception processing means 32 Network status and request information holding section 33 External stimulus input processing section 34 Optimal path display section 35 Packet output direction determination section 36 Received packet processor

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI technical display location H04Q 11/04

Claims (27)

[Claims] 1. A communication node constituting a packet switching network (10) for accommodating media in a packet format for communication.
In (11), the mutual connection type neural network (12) for determining the output direction of the packet from the own node , the present state of the packet switching network (10) and the switching network (10) are accommodated. An external stimulus input means (13) for outputting an external stimulus to neurons constituting the neural network (12) is provided in accordance with the requirements of the media, and each node (11) has the packet switching network. (10) A routing method using a neural network in a communication node , characterized in that the optimum direction for the current state and the requirement of the medium is the output direction of a packet from the own node . 2. The neural network (12)
However , there is a one-to-one correspondence with each link between each node (11) constituting the packet switching network (10),
Link to which external stimulus is input from hard input means (13)
Corresponding neuron, said neural network
All neurons in (12) have multiple entries in each.
Forces are weighted and added, and the addition result is subjected to thresholding to 0 or
Output 1 as a result of repeating the operation of outputting 1.
The link corresponding to the stable link-corresponding neuron is
To be decided as the link in the packet output direction
Link compatible neuron and the packet switching network (1
0) corresponding to each node (11) that has a one-to-one correspondence.
Node corresponding to the neural network
All the neurons in (12)
The inputs are weighted and added, and the addition result is thresholded to 0.
Or as a result of repeating the operation of outputting 1,
A node corresponding to the node-corresponding neuron that is stable to 1
Is determined as the node in the packet output direction
Two types of neurons, which are the node-corresponding neurons
Communication No of claim 1, wherein the al constituted
A routing method using neural networks in the network. 3. The external stimulus input means (13) is the
Packets move within the packet switching network (10 )
Each time a packet is input to the network, the neural network
Outputs external stimuli to the neurons that make up the peak (12)
The neural network (12)
Is an external stimulus output from the external stimulus input means (13).
Each time a packet is input to its own node,
From the destination node to the destination node
To determine the output direction of the packet
A network, claims, characterized in that to reach the packet to the destination node via the optimum path adapted to real time, and satisfies the requirements of the media to a state change of the packet switching network (10) 1 or 2
A routing method using neural networks in communication nodes . 4. The external stimulus inputting means (13) , as a state of the switching network (10), shows a packet discard rate and a packet delay time for each link constituting the switching network (10), and also the medium. As a required condition, the acceptable packet discard rate indicated by the medium is received, and the switching network (1
0) depending on the current status and media requirements
4. A routing method using a neural network in a communication node according to claim 1, wherein the routing method is means for outputting an external stimulus . Wherein said external stimulus input means (13), a neuron corresponding to neurons and one-to-one corresponding to each link (14) constituting the switching network (10), before <br/> serial when the packet from the media to the input of a packet discard rate and media specify allowed packet loss rate of each link (14) can accept the respective link 1, the neuron outputs 0 when unable Prepare,
And the output of the neuron the packet delay time by adding weighting, communication according to claim 4, wherein the addition result, characterized in that a means for outputting an external stimulus to the neuron corresponding to respective links (14) A routing method using a neural network in a node . Wherein said each node (11) detects from the input packet to the further number of the node that the packet has been passed, the packet passes through node number to be sent to the neighboring node by adding the self-node number to the packet and detecting and adding means (15), wherein when the routing of the input packet by neural network (12), said thus to perform routing of the input packet by excluding a path through nodes input packet has passed neural <br/> To reduce the calculation time of the network (12) ,
There is a one-to-one correspondence with the node through which the input packet has passed.
Fixed output of a neuron that fixes the output of a neuron to 0
3. A routing system using a neural network in a communication node according to claim 2, further comprising means (16) . 7. The packet passing node number detecting and adding means (15) is means for detecting only a node adjacent to the own node among the packet passing nodes, and the neuron output fixing means (16) is the adjacent node. 1 to 1
The routing method using a neural network in a communication node according to claim 6, characterized in that it is a means for fixing the output of the neuron corresponding to 0 to 0. 8. The packet passing node number detecting and adding means (15) is means for detecting only the number of the source node of the packet among the packet passing nodes, and the neuron output fixing means (16). A hand that fixes the output of the neuron, which has a one-to-one correspondence with the source node, to 0
7. The communication node according to claim 6, wherein the communication node is a stage.
Routing scheme using definitive neural network. 9. Communication by accommodating media in a packet format
Each communication node constituting the packet switching network (10) for performing
In (11), the mutual connection type neural network (12) for determining the output direction of the packet from the own node , the present state of the packet switching network (10) and the switching network (10) are accommodated. Depending on the requirements of the media, the external stimulus input means (13) for outputting external stimulus to each neuron constituting the neural network (12) and the state of each output link of its own node are set to the switching network ( 10) state information notifying means (17) for notifying all nodes constituting the node, and the value of the state information notified from the state information notifying means (17) of another node from the own node to the other node. until a other node state information value decreasing means for decreasing (18) depending on the distance, under conditions to reduce the effect of the state of the other node with increasing distance to said other node, Use a neural network in a communication node each node (11), characterized in that the output direction of the packets of the current state and the direction of the ideal and requirements of the media of the packet switching network (10) from its own node The routing method that was used. 10. In each node (11) , the status information notified by the status information notifying means (17) is a packet delay time and a packet discard rate for each output link of the own node, and the other node status. The state information notification packet processing unit that constitutes the information value reducing means (18) respectively indicates the packet delay time and the packet discard rate for each output link of the other node as the state information notified from the other node, respectively. 10. The neural network in a communication node according to claim 9, wherein the result is divided by the number of hops as the number of relay links from the node to the other node, and the result of the division is output to the external stimulus input means (13). The routing method used. 11. Media are stored in a packet format for communication.
Each communication node constituting the packet switching network for performing signal (10)
In the packet (11), the mutual connection type neural network (12) for determining the output direction of the packet from the own node , the current state of the packet switching network (10) and the accommodation state in the switching network (10). The external stimulus input means (13) for outputting an external stimulus to each neuron constituting the neural network (12) and the state of each output link of its own node according to the required conditions of the media. State information notifying means (19) for notifying all the nodes constituting the switching network (10) about the range of notifying the state in (10), and the state information notifying means (1) of other nodes.
9) The value of the status information and the value of the notification range notified from
Decrease according to the distance from the own node to the other node,
When the decrease result of the notification range is not 0, the decrease result of both values is notified to the adjacent node and is output to the external stimulus input means (13) in the own node, and the decrease result of the notification range is 0 , The other node state information and the notification range value reducing means (2) which are output to the external stimulus input means (13) in the own node without notifying the adjacent node of the decrease result of the both values.
0) and each node under the condition that the influence of the state of the other node is reduced as the distance to the other node increases.
The node (11) determines, from its own node , the optimum direction for the current state of the packet switching network (10) and the requirements of the media.
No communication characterized by setting the output direction of the packet
A routing method using neural networks in the network. 12. A medium for storing media in a packet format for communication.
Each communication node constituting the packet switching network for performing signal (10)
In the packet (11), the mutual connection type neural network (12) for determining the output direction of the packet from the own node , the current state of the packet switching network (10) and the accommodation state in the switching network (10). The external stimulus input means (13) for outputting external stimulus to each neuron constituting the neural network (12) and the state of each output link of its own node in accordance with the required condition of the media according to a plurality of types of cycles. A value indicating a relay range in the network that relays the average result, which is changed according to the cycle, a notification range as a value that specifies a range of nodes that use the average result for routing, and State information average and notifying means (21) for notifying the average result to each node constituting the switching network in the plurality of types of cycles, and state information average and notification for other nodes The value indicating the relay range received from the stage (21) is decreased by 1, and when the decrease result is not 0, the relay range decrease result, the notification range, and the state information average value are output to all the adjacent nodes, and , And another node status information reception processing means (22) for outputting the status information average value to the external stimulus input means (13) when the own node is in the notification range, while preventing traffic increase due to status notification. , The nodes (11) are connected to the switching network (10)
The optimal direction for the current state of the media and the requirements of the media
A routing method using a neural network in a communication node, which is characterized in that the output direction of packets from its own node is used. 13. The other node status information reception processing means (22), when the reception interval for receiving the status information from the other node is long, of the status information previously received during one cycle of the reception interval. The state value of another node at any time is predicted from the value, and the predicted value is input to the external stimulus input means (13).
13. The routing method using a neural network in a communication node according to claim 12, further comprising a prediction function unit for outputting to. 14. The routing method using a neural network in a communication node according to claim 13, wherein a neural network having a learning function is used as the prediction function unit. 15. A packet switching network (1) for accommodating media in a packet format and performing communication by a virtual call system.
0) , each of the communication nodes (11) has a mutual connection type neural network (12) for determining the output direction of the packet from its own node , and the present of the packet switching network (10). And an external stimulus input means (13) for outputting an external stimulus to the neurons constituting the neural network (12) according to the state of the above and the requirement of the medium accommodated in the exchange network (10). , the
Each node (11) determines its own optimum direction according to the current state of the packet switching network (10) and the requirements of the media.
Communication characterized by the output direction of packets from the
A routing method using neural networks in nodes . 16. The neural network (12)
However , there is a one-to-one correspondence with each link between each node (11) constituting the packet switching network (10),
Link to which external stimulus is input from hard input means (13)
Corresponding neuron, said neural network
All neurons in (12) have multiple entries in each.
Forces are weighted and added, and the addition result is subjected to thresholding to 0 or
Output 1 as a result of repeating the operation of outputting 1.
Link corresponding to stable該Ri link corresponding neurons in the
To be decided as the link in the packet output direction
Link compatible neuron and the packet switching network (1
0) corresponding to each node (11) that has a one-to-one correspondence.
Node corresponding to the neural network
All the neurons in (12)
The inputs are weighted and added, and the addition result is thresholded to 0.
Or as a result of repeating the operation of outputting 1,
A node corresponding to the node-corresponding neuron that is stable to 1
Is determined as the node in the packet output direction
Two types of neurons, which are the node-corresponding neurons
Communication of claim 15 wherein the al constituted Bruno
Routing method using neural network in code. 17. A source node of the virtual call, the external stimulus input means (13), outputs an external stimulus in response to the state of the requirements of the media of the virtual call and the current switching network (10) Is a means to
The neural network (12) is
Neural network that finds the entire path to the destination node
16. The method according to claim 15 or 1, which is a network work.
A routing method using the neural network according to 6. 18. The link (1) between the nodes (11) constituting the switching network (10) as the current state of the packet switching network (10) by the external stimulus inputting means (13).
Receive and capacity 4), and the band in use, allowable utilization, and a path <br/> packet delay time and the bandwidth the media requires a requirement for media of the virtual call
The current status of the switching network (10) and media requirements
It is a means to output the external stimulus according to the situation
And the neural network (12) sends the packet
Birch adapted in real time to the state of the switching network (10)
The packet output direction is determined according to the call path.
The routing system using a neural network according to claim 15 or 16, wherein the routing system is a defined neural network. 19. The external stimulus input means (13), before
Corresponding to each link (14) that forms the exchange network (10)
Which have a one-to- one correspondence with
Capacity of each link (14), bandwidth in use, allowable usage
Rate and requested bandwidth of the media for said virtual call
The input bandwidth of the media to each link
Outputs 1 when it can be accepted, 0 when it cannot
To comprise a neuron, characterized in that the external stimulus in response to said output of said neuron packet delay time of each link is means for outputting an external stimulus to the neuron corresponding to the respective link (14) A routing method using the neural network according to claim 18. 20. The originating node of the virtual call
The external stimulus input means (13) is
Outputs an external stimulus that satisfies the required bandwidth of the call media
And the neural network
(12) becomes a delay time of a packet from source node of the virtual call to the destination node is minimal, and media
A neural network that finds all paths that satisfy the required bandwidth
16. A network according to claim 15 , wherein the network is a network.
A routing method using the neural network according to 6. 21. The originating node of the virtual call
In the neural network (12), a node corresponding to a node corresponding neuron whose output is stable at 1 is a relay node which constitutes a path of the virtual call,
A call setup packet holding the number of the relay node is generated, and the call setup packet is sent to an adjacent relay node in order to reach the destination node of the virtual call.
A call setup packet generator that sends out the call setup packet
The routing system using a neural network according to claim 16 , further comprising a stage . 22. When a failure of the link or node in communication by the virtual call is generated, a node that detected the failure, the pseudo-node which is the virtual call originating node, the neural networks
(12) is a link-compatible or node-compatible node.
Of the virtual loop with the output of the neuron corresponding to the faulty link or faulty node fixed to 0
A net that asks for the entire failure bypass path to the call destination node.
Network and further the network (12)
Corresponds to a node-corresponding neuron whose output is stable in 1
As a relay node configuring the failure bypass path,
Generate a call setup packet holding the number of the relay node
The call setup packet to the destination number of the virtual call.
To the adjacent relay node to reach
A call setup packet generator that sends out the call setup packet
The routing system using a neural network according to claim 16 , further comprising a stage .
that's all 23. In a packet switching network (10) for communicating media in a packet format, each node (11) constituting the packet switching network (10) determines an output direction of a packet from the own node. Mutual connection type neural network (12) and the packet switching network (1
External stimulus input means (13) for outputting external stimulus to neurons constituting the neural network (12) according to the current state of (0) and the requirements of the media accommodated in the exchange network (10). A routing method using a neural network, characterized in that the packet output direction is the optimum direction for the current state of the packet switching network (10) and the requirements of the media. 24. In a packet switching network (10) for accommodating media in a packet format, said packet switching network (1)
0) each node (11) determines an output direction of a packet from its own node, and a mutual connection type neural network (12) and the packet switching network (10).
Of the neural network (1) according to the current state of the network and the requirements of the media accommodated in the switching network (10).
In order to notify the external stimulus input means (13) for outputting an external stimulus to each neuron constituting 2) and the state of each output link of its own node to all the nodes constituting the exchange network (10). Other node status information value that decreases the value of the status information notified from the status information notification means (17) of the other node and the status information notification means (17) of another node according to the distance from the own node to the other node. Reducing means (18) for reducing the influence of the state of the other node as the distance to the other node increases, and an optimum direction for the current state of the packet switching network (10) and the requirements of the medium. A routing method using a neural network, which is characterized in that the packet output direction. 25. In a packet switching network (10) for accommodating media in a packet format, the packet switching network (1)
0) each node (11) determines an output direction of a packet from its own node, and a mutual connection type neural network (12) and the packet switching network (10).
Of the neural network (1) according to the current state of the network and the requirements of the media accommodated in the switching network (10).
The external stimulus input means (13) for outputting an external stimulus to each neuron constituting 2) and the state of each output link of the own node, and the range for notifying the state in the exchange network (10), The status information notifying means (19) for notifying all the nodes constituting the switching network (10), and the value of the status information and the value of the notification range notified from the status information notifying means (19) of the other node. , The external stimulus input means (13) in the own node while notifying the adjacent node of the decrease result of the both values when the decrease result of the notification range is not 0, according to the distance from the own node to the other node. ), And when the decrease result of the notification range is 0, the other node state information is output to the external stimulus input means (13) in the own node without notifying the adjacent node of the decrease result of both values. And notification range value reducing means (20) , The influence of the state of the other node is reduced as the distance to the other node increases, and the optimum direction for the current state of the packet switching network (10) and the requirements of the medium is the packet output direction. A routing method using a neural network. 26. A packet switching network (10) for accommodating media in a packet format, said packet switching network (1)
0) each node (11) determines an output direction of a packet from its own node, and a mutual connection type neural network (12) and the packet switching network (10).
Of the neural network (1) according to the current state of the network and the requirements of the media accommodated in the switching network (10).
The external stimulus input means (13) for outputting an external stimulus to each neuron constituting 2) and the state of each output link of the own node are averaged over a plurality of types of cycles, and are changed according to the cycle. A value indicating the relay range in the network that relays the average result, a notification range as a value that specifies the range of nodes that use the average result for routing, and the average result in the plurality of types of cycles in the switching network. The state information average and notifying means (21) notifying each of the constituent nodes and the value indicating the relay range received from the state information average and notifying means (21) of another node are decreased by 1, and the decrease result is 0. When it is not, the relay range reduction result, the notification range, and the state information average value are output to all the adjacent nodes, and when the own node is in the notification range, the state information average value is input to the external stimulus. The other node status information reception processing means (22) for outputting to the means (13) is provided to prevent traffic increase due to status notification, and is optimal for the current status of the switching network (10) and the requirements of the media. A routing method using a neural network, characterized in that the packet output direction is the direction of. 27. A packet switching network (1 ) for accommodating media in a packet format and performing communication by a virtual call system.
In 0), each node (11) constituting the packet switching network (10) and the mutual coupling type neural network (12) for determining the output direction of the packet from the own node, and the packet switching network (10) ), And external stimulus input means (13) for outputting an external stimulus to neurons constituting the neural network (12) according to the current state of (1) and the requirements of the medium accommodated in the exchange network (10). And a packet switching network (10)
A routing method using a neural network, characterized in that the optimum direction for the current state of the media and the requirements of the media is the packet output direction.