Stochastic generative model of cost-effective OADM using a three-dimensional neural network in a WDM access network

For more than a decade twelve years, all-optical networks have greatly matured. The WDM access network is a high performance network with various channels for providing offering different services, including data, voice, video and others. Rings permit slots to be synchronized: check meaning even at extremely high data rates; hence, they support efficient and flexible use of the available bandwidth for packet communication. The WDM access network can provide multiple channels and increase bandwidth to improve its performance. The access node (such as optical add/drop multiplexer (OADM)) transfers data between the feeder network and the distribution network. The OADM has four main parts-multiplexing (MUX), demultiplexing (DEMUX), the 2 × 2 optical switch and the electronic IP router. Reducing the delay of the OADM can increase the performance of the WDM access network. This work uses a three-dimensional neural network algorithm with increased performance to develop new stochastic generative model that aggregates data on cost-effective OADM. It is based on Kohonen self-organizing maps because they are unsupervised. The self-organizing map neural network has three input parameters-source node, destination node and the number of channels are considered to evaluate the performance of the network; they are translation time, switching times and conversion times. When a packet is transferred to the OADM on the channel and then transferred to another OADM, the same channel must be free; otherwise the 3-D neural network algorithm is used to select another free channel to transfer it. The random packet generation processor generates packets randomly from a distribution network. In the simulation, when the number of the access node increased, the load becomes heavier and collisions occur frequently. When the load is low (the maximum number of randomly generated packets is 50), the performance of a access network that uses a three-dimensional neural network is similar to that of one that uses a random generating network. However, when the load is heavy (maximum number of randomly generated packets is 500), an access network that uses the three-dimensional neural network outperforms one using a randomly generating algorithm or the two-dimen- sional neural network.