Spike neural network as a controller in SDN network

T he paper proposes a methodology for predicting packet flow at the data plane in smart SDN based on the intelligent controller of spike neural networks(SNN). This methodology is applied to predict the subsequent step of the packet flow, consequently reducing the overcrowding that might happen. The centralized controller acts as a reactive controller for managing the clustering head process in the Software Defined Network data layer in the proposed model. The simulation results show the capability of Spike Neural Network controller in SDN control layer to improve the (QoS) in the whole network in terms of minimizing the packet loss ratio and increased the buffer utilization ratio.


INTRODUCTION
In recent years, it has become been widely expected to use software-defined networking (SDN) in wireless sensor networks (WSNs), so the WSNs' have great dynamic performance. SDN is a novel type of network structural design, which can overcome the drawbacks of existing network structures (Gelberger, et al., 2013). It acts on separating the data plane (DP) and control plane (CP) of a network, meaning that Smart Controller conFig.s forwarding elements (FEs) using precise forwarding procedures for data packets of dissimilar streams (F., Richard Yu, et al., 2020). The controller gets enough information to perform the mission. That's why the control protocols distributed are not required anymore between (FEs). Besides optimizing the network, the controller might interact with applications (Buurman, Ben, et al., 2020). WSN is composed of sensor nodes having capabilities to communicate, compute and sense. Most sensing nodes contain batteries, eventually leading to shortening the lifetime of nodes. It is essential to simplify sensing nodes' operation for saving power and managing wireless sensor networks with a robust controller having full vision over the whole network instead of distributed control protocols (Hayes, et al., 2016). This paper contains other sections that can be listed as the following: Section two presents the SDN controller and artificial intelligence overview. Section three presents the method and algorithm of the proposed intelligent control. Section four presents the results of proposed intelligent SDN and compare with other technique. Section 5 contains the conclusions and suggestions for future work

A. MOTIVATION
Data flowing volume in the data plane can be a highly significant matter in traffic management in software-defined networking. Since the sensor number connecting to switches goes up in the data plane, that will be faced with an increase in the traffic load in queuing buffer. Furthermore, when the switch number in a software-defined networking Gateway goes up, the centralized controller's performance in its controlling plane would not succeed in processing the entire requests that come from the switches. More importantly, in recent years, using machine learning and artificial intelligence networks with software-defined networking has gained increased attention.

B. CONTRIBUTIONS
The paper has introduced an SNN as a congestion controller in the model proposed. A spiking neural network can be regarded as a Spiking neural network. Furthermore, with a different controller, which is based upon Artificial Neural Network, it is introduced for managing sensors in the spike Smart SDN network. However, the paper's essential contributions can be listed as the following: We proposed a spike intelligent Smart SDN model with an intelligent controller in a software-defined network control plane. Based on (SNN), that intelligent controller functions for estimating the volume of packets flowing in the network. A model of smart queuing for estimating buffer size capacity in the spike intelligent software defined network was proposed.

RELATED WORK AND BACKGROUND THEORETIC 2.1 RELATED WORK
In the last few years, many researches of WSN to increase the performance of the network are done. ( proposed an interface protocol, which can also be referred to as Wireless Sensor and Actuator Network Flow, in charge of the whole communications among SDN controller and SDN oriented end devices. In this approach, the SDN controller possesses the intelligence of the network. The results have shown that the current Wireless Sensor and Actuator Network system's total performance is optimized. The processes of control and management are made simple through the model proposed. (Mohammed J.F. Alenazi, et al., 2019) have presented a new nodes metric, node-disjoint path (NDP) that is utilized for measuring the significance of the node regarding its varied connection to other nodes. According to a node-disjoint path, they propose two types of algorithms (NDP-global) and (NDPcluster) to locate k-controllers to increase the network's robustness in the face of targeted attacks. In addition, the results indicated that the algorithm of the NDP-cluster has a similar delay performance to the k-median algorithm and can provide higher resilience to the SDN network. show that this algorithm performs better than SA and QA about network performance by reducing the average distance among controllers sensors and up to 9% and 13%.

SOFTWARE-DEFINED-WIRELESS SENSOR NETWORK (SD-WSN)
Here, an overview SD-WSN will be given. First, a broad description of SD-WSN architecture and explain its differences regarding non-SD-WSN. After that, a comparison with SD-WSN and wireline software-defined wireless was made. Lastly, an overview of software tools of SD-WSNs is given (

THE PROPOSED MODEL: ARCHITECTURE AND SPECIFICATIONS
In this system, the intelligent controllers are proposed in the SDN platform. These controllers are able to compute the data flow of the physical area. Moreover, one of the functions in a proactive way is an SNN for estimating the sensing area's packet flowing. The other controller functions are reactive and are assigned to choose the head of cluster and cluster members. The controller of SNN is efficiently able to choose the head of a cluster and members of the cluster in the sensor zone. The proposed model can compute the rate of data flowing, contributing to coordinating the available buffer capacity by having a set of active sensing nodes in the network to prevent buffer overrun. In this paper, an SDN smart controller was proposed based on SNN Controller. Spiking Neural Networks possess a structure similar to conventional neural networks. However, dissimilarity existing in structural design can be outlined numbers of the presynaptic ending (synaptic terminals) within each layer of nerve cells. The architecture of neural networks contains a diverse layer called J, I, and H. These initials stand for the three layers: (1) input layer, (2) hidden layer (3) output layer separately, as presented in Fig. 2 (A). The structure also consists of a feed forward fully-connected SNN with several delayed the presynaptic ending. Synapses are Journal of Engineering Volume 27 September 2021 Number 9 68 signal connectivity consisting of a constant number of the presynaptic end; each one acts as sub-connectivity related with a dissimilar weight and delay as Fig. 2 (B).  A typical architecture of layer feed-forward neural network FF NN contains neuronal layers communicating only with neurons in successive layers. That is to say, organized in layers, and each unit in a layer is connected with the units in the whole preceding layer. Basically, there are three layers of this type of network, which can be illustrated as follows: the input layer behaves as network input; in fact, it contains no neurons, since no actual processing is involved, the socalled input-neurons must possess a particular output. In spiking neurons, the input-neurons work on firing spike-train defined in advance. Output-layer, the neurons' spike-trains create the network output. In between the input and output-layer there could be any number of hidden layers. Each sub-connection possesses its weight variable that is remaining fixed, but is updated by SpikeProp, which is among the first basic machine learning rules and applies the backpropagation principle of feed-forward SNN. The variables ℎ (act) , (act) and (act) are representing certain neurons' actual spike times in the particular layers. The spike response function for a single subconnection from a neuron to neuron membrane in the following layer is ( ) delayed through its spike time (act) and delay , and 0 for ≤ (act) + a neuron MP is (t), equivalent to the weighted sum of incoming delayed Spike Response Functions(SRF) from the preceding layer: (2) Spike-Prop is involved only with the first spike. Therefore it is possible to ignore successive membrane dynamics. Error-function represents half the smallest mean squares, where represents the training target spike time of output neuron : Eq. (3) Spike-Prop is a method of gradient descent, acts on adjusting each weight in the direction reducing . (54) To train the network, weights must be updated each between the output layer and hidden layers for minimizing the error: Eq. (7) The firing process of a neuron, which can refer to action potential (AP), is conducted either by entire or complete operation or effect.

RESULTS AND DISCUSSION
In this paper, MATLAB® R2018b was used in the proposed simulation. This paper proposed a smart SDN controller by using multi spike neural network, in this work, 100 sensors randomly placed in a sensing area of square shape. The implementation for the area is shown in Fig. 5. At the start of every scheduling period, the sensor generates traffic; in other words, sensor(s) conducts low-flow to high-flow. After that, the traffic is directed to Forwarding Cluster Head. The controller of SNN reduces the level of congestion. However, Forwarding Cluster Heads can be categorized as congestion when this ratio goes beyond the threshold level. Table 1 shows the Simulation parameters of the proposed system. In this work, the random distribution achieved by generation random select 100 random point map (100*100) and distrusted around the base station at (50,50) as it is illustrated in Fig. 4 . After distributing node, the K-mean cluster was applied to select the best location of FCH, where that achieve to get all the nodes (100) and enter to k mean cluster and cluster it to 3 groups, where each group belong the subset of nodes. The Center point will represent the location of a head cluster, as shown in Fig. 4. Table 2 shows the position of FCH based on the k-mean cluster, where the position of FCH represent the three center point and through Table 4 below note each node connected to any point of FCH. After distributing the node and determining the FCH, then connecting the FCH with the node, the model will be the LEACH protocol as shown in Fig. 4 Fig. 9. It can be seen that SNN can reach the error goal. Table 3 shows a comparsion of this work compare work with other works. Improved the sink's position sort and network controllers.
Improved on a controller, By replacing the controller with one that uses artificial intelligence, specifically SNN technology 6 Wei Ren et al., 2018 The results demonstrate that it roughly minimizes the control delay The results demonstrate that it roughly minimizes the overflow, energy, loss of training, and data loss 7 Ali Burhan Al-Shaikhli, et al., 2018 Proposed an interface protocol, the results have shown that of the current Wireless Sensor and Actuator Network system's total performance is optimized The proposed intelligent controller use SNN technology; the results have shown that the Network system's total performance is optimized 8 Mohammed J.F. Alenazi, et al., 2019 Have presented a new nodes metric, node-disjoint path (NDP). The results indicated that the algorithm of the NDPcluster has a similar delay performance to the k-median algorithm and can provide higher resilience to the network. The simulation results demonstrated that this solution could reduce memory consumption compared with various algorithms concerning placing controllers in dissimilar topologies of the SDN network.
Developed an optimization algorithm, the results of Simulation have shown that this algorithm performs better than traditional controller. The algorithm improves overflow, energy, memory, density, loss of training, and data loss.

CONCLUSIONS
The paper proposes smart SDN architecture for utilization in wireless sensor network applications. The proposed model presents smart controllers in the Software-defined networking control plane, and these controllers are able to estimate the sensing area's packet flow. Moreover, one of the functions is in a proactive way in an SNN for estimating the sensing area's packet flow. Simulation results show that the quality of service is optimized in the smart SDN network. The SNN controller can efficiently choose the head of a cluster and its members in the sensor area, as displayed in the quality of service results. More importantly, the proposed model estimates the rate of the packet flow. It contributes to coordinating the available buffer Capacity with a set of active sensing nodes in the network for preventing a buffer overrun. The network control by this model is more accurate than SNN (single spike) due to the proposed training algorithm's spiking power.