Switching of Interfaces and Self-optimization of weights using Backpropagation ANN in WHN Enviornment

doi:10.21203/rs.3.rs-2626450/v1

Download PDF

Research Article

Switching of Interfaces and Self-optimization of weights using Backpropagation ANN in WHN Enviornment

https://doi.org/10.21203/rs.3.rs-2626450/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

In today’s scenario, mobile communication is facing a healthy competition due to different networks, interfaces, channels, and many more available in wireless heterogeneous environment. The problem arises when customers/users get the availability of many interfaces at the same time. At that time users need an intelligent or smart mechanism to connect them to the best services according to their requirements/preferences. Interface management manages available interfaces and connects the user with the best. In this paper, Interface management with Artificial Neural Network (ANN) allows the smart use of different radio accesses/interfaces. The selection is made with different parameters of different networks. This paper proposed a backpropagation neural network that is used for the switching in between different networks-3G, WLAN, 4G and 5G. The different parameters of a network are used as the selection parameters with assigning proper weights. Weights are initialized with fuzzy AHP and optimized with Back Propagation Neural Network (BPNN). The target value and the actual value is compared and their difference used as the adjusting value for the weights to get the optimum value. The backpropagation is used to train the network. The comparison among the projected algorithm and the existing algorithm shows the valuablity of the new method and the best connectivity of the network.

Artificial Intelligence and Machine Learning

Wireless Heterogeneous Network

Artificial neural network

Back Propagation

Interface Management

Today, the world has become a competitive world due to no. of options/choices available in the market. Mobile technologies are also facing such competition due to different networks, interfaces, channels and many more in wireless heterogeneous environments. The problem arises when customers/users get the availability of many interfaces at the same time. The process of managing the switching between networks efficiently and perfectly is considered the foremost requirement and strong challenge. Different interfaces are managed with different techniques –traditional or modern. As the complexity of the communication system increases, new algorithms/ techniques are developed. Artifical intelligence is the new era of software where one can have numerous choices.

1.1 Artifical Neural Network (ANN)

ANN has the capability to activate any wireless device and intelligently observe its surroundings by learning and predicting the progress of the various environmental features. A neural network is a group of connected input/output units with each connection having an associated weight. It helps to make prophetic miniatures from big databases. The mini-brain is a small imitation of the human nervous system. It helps direct image understanding, mortal literacy (reading and writing), computer address—and other information-processing tasks.

Neural networks are sets of algorithms aimed to recognize patterns and interpret data through clustering or labeling. A training algorithm is a system used to execute the neural network's literacy process. An artificial neural network (ANN) is the element of artificial intelligence that's meant to pretend the execution of a natural brain. Processing units are the building blocks that form ANNs, and each processing unit corresponds to one input or manual worker. The inputs are what the ANN

learn from to produce the asked affair. NNs are considered as an important machine learning fashion applied to take an intelligent and selfdependent decision for automatic selection of the stylish available seeker wireless technology.

1.2 Backpropagation Neural Network ( BPNN)

Backpropagation is a unremarkable approach of training an artificial neural networks. It consists of a set of learning rules designed to reduce training errors in a neural network using some gradient-based method. The weights are fine-tuned by adjusting the connections between neurons based on errors made during a prior iteration (or epoch). To make a model reliable, it must be tuned properly so that the error rate is minimized and generalization is maximized. Backpropagation is a fast, simple and easy-to-program way to train neural networks. It has no adjustable parameters aside from the number of inputs. It is a flexible method as it does not require prior knowledge about the network. It does not need any special mention of the features of the function to be learned.

A BP network is a feedforward, backpropagation, multi-layer network which starts with random weights. The structure of a BP network with five inputs is shown in Figure 1. The network has three layers- input, hidden and output. It is possible to have more than one hidden layer. For simplicity only one hidden layer is shown in this structure. Inputs X, are modeled using randomly selected weights. Every neuron's output is calculated from the input layer to the hidden layers, and finally to the output layer.The difference in the outputs called Error _BP (Actual Output – Desired Output) is calculated. To minimize the error adjust the weights in the hidden layers and repeat the process until the desired output is achieved. A differentiable activation function is required for the backpropagation algorithm, and tan-sigmoid and log-sigmoid are the most commonly used ones. During the learning process, the weightings are adjusted in order to reduce the error E. This is done in order to make the outputs of the system more closely resemble the actual outputs. The network learning process is completed when the error rate falls below the allowable level. In this manner, the network is ensured to accurately reflect the real world [1].

1.3 Fuzzy Analytical Hierarchy Process (FAHP)

Weights related to each input, communicate the importance of that input in estimating the output value. Weights show the relationship between a particular feature in the dataset and the target value. Determining the most suitable weights for different criteria is also a difficult task. Input parameters with small weight values have lesser significance as compared with high weight values which have more significance. Weights play an important role in changing the orientation or slope of the line that separates two or more classes of data points. There are some general weight initialization techniques used in a system: Zero and random initialization techniques. Mobile users can benefit from MADM algorithms due to their nature as a solution for determining the best network in terms of quality of service (QoS). There are many methods available in MADM, for example, analytical hierarchy processes (AHP), fuzzy analytic hierarchy processes (FAHP), analytic networks (ANP), fuzzy analytic networks (FANP), and a technique for order preference based on similarity to ideal solution (TOPSIS. In this paper, we use the FAHP method to assign weights for input parameters. The fuzzy AHP method is the most appropriate method for quality and multicriteria decision-making problems. FAHP method provides more efficient, flexible, and realistic decisions based upon the available criteria and alternatives than other methods. Fuzzy Analytic Hierarchy Process is a combination of Analytic Hierarchy Process (AHP) and fuzzy logic theory. Both have the same method to evaluate the weights of the network except the Fuzzy AHP method sets the AHP range into the fuzzy triangle range i.e low, medium and high.

This paper is framed as heading I describes the introduction of artificial neural networks,backpropgation, and FAHP. Heading II is related to the literature survey. Heading III described the system model and hierarchy of the work. Heading IV described the interface management with the algorithm. Heading V described the implementation and simulation result of the technique. Heading VI discussed the conclusion of the work and the future scope of this research.

In a wireless heterogeneous environment number of network selection algorithms and weights, techniques have been extensively studied in the literature in different contexts.

Several construction choices and algorithms for intelligent access network selection in a multi-access environment were discussed in [2]. The major challenges in uninterrupted connectivity and service skill of future heterogeneous 5G networks were also discussed. RSSI-based and load-aware techniques were compared to show the best network selection. In [3] authors proposed to initialize weights randomly but at the distinct interval for neural network and a Resilient Backpropagation algorithm was used to train the network. The authors showed the better efficiency of the system than the general random method as the networks could achieve a minimum of the error practically during training. A novel network selection algorithm was proposed with signal strength, available bit rate, signal to noise ratio, achievable throughput, bit error rate and outage probability metrics as criteria for network selection [4]. The selection metrics were combined with PSO for relative dynamic weight optimization. The proposed algorithm was implemented in a typical heterogeneous environment of EDGE (2.5G) and UMTS (3G). The switching rate of the user between available networks had been used as the performance metric and a utility function was used to maintain desired QoS. A neural network was used [5] to select the network for handover, based on different aspects like network related, terminal related, user related, application related and service related. The performance of the three networks was evaluated based on voice over the internet and compared with the traditional method. Afuzzy logic-based approach for an automated network selection based on real-network implementations and measurements was proposed [6]. The proposed network selection model was based on fuzzy inference rules considering features that affect the selection decision. The authors [7] lightened the highlights of new QoS, QoE and SON feature associated with the 5G evolution. They also discussed the limitations of earlier networks & advantages of 5G for different scenarios like M2M, D2D, IOT, and many more. An author proposed, implemented and evaluated a strategy[8] that allows an entire 802.11 Basic Service Set (BSS) to dynamically hop between the available channels always selecting the "best" one. The selection not only guarantees the hopping BSS performances well but also minimizes the interference toward other BSSs so that the overall performance of the system maximizes resource utilization. Particle Swarm Optimization (PSO) method was used to optimized weights in [9]. The second criterion consists of a fuzzy logic system fed with similar metrics as inputs and targeted towards the same output. The final decision of network selection was taken by the blend of these two criteria. Simulation results indicated that the proposed scheme based on the Cost function, PSO and, Fuzzy system (C-P-F) provided better performance in terms of minimizing the unnecessary handoffs (network selection rate), utility degree and load balancing. The proposed algorithm (C-P-F) significantly reduces the network selection rate by 50%. A FAHP technique [10] to calculate the weights and MADM algorithms to score the different networks. A network selection algorithm was designed [11] with the combination of a utility function, FAHP and MADM technique: SAW, MEW & TOPSIS. The system worked based on service uniqueness and consumer choices. Utility functions were used for the utility value of each parameter. The objective weight and subjective weight was calculated with the Fuzzy-analytic hierarchy process (FAHP) and entropy method. The different MADM techniques were used to rank the networks and the network with the highest rank was selected as the best network. An energy consumption as a criterion in the Flow/Interface Association (FIA) was considered [12]. A novel method called Smart Tabu Search (STS) was proposed that took into account network conditions, the monetary cost of the network, QoS requirements of the applications, user preferences, and energy consumption of the mobile device to select the optimal FIA that achieved the best trade-off among all the considered criteria. The research paper [13] focused on the potential application of machine learning, one of the applications and part of artificial neural networks (ANNs), in wireless communications. The different types of neural networks such as feedforward, recurrent, and deep learning neural networks their basic architecture, training procedure, challenges and opportunities related to them were overviewed. The neural network approach was used to estimate the bandwidth [14]. To minimize the cross-layer interference of multiple base station cells, a cluster-based cooperative interference management scheme for a multi-interface environment was proposed in [15]. Several conditions for applying ML to wireless communication were discussed in [16]. This paper answered to beginners related to usage, different kind of ML techniques, advantages and how it is better than traditional approaches. A novel method was proposed [17] for calculating weights in a neural network for linear projections. A user-oriented intelligent access selection algorithm in HWNs [18] with five modules (input, user preference calculation, candidate network score calculation, output, and learning. A utility function to calculate the utility value of the judgment parameter; the user preference calculation module calculates the weight of the judgment parameter using the fuzzy analytic hierarchy process (FAHP) approach; the candidate network score calculation module calculates the network score through a fuzzy neural network; the output module calculates the error between the actual output value and the expected output value. A neural network-based scheme was proposed [19] for minimizing the redundant handover (unnecessary handovers) numbers. It enhanced network efficiency and increased the handover speed by scanning the quality of all signal between the mobile user and all neighboring stations. It predicted the next target station for heterogeneous network handover. Authors [20] showed switching and selection of best networks among different networks like 3G, 4G and 5G based on input parameters. The backpropagation neural network algorithm is used for implementation. Assignment of weights and applications were not taken specifically in this paper. An access selection algorithm was proposed [21]. In order to maximize the transmission rate of a wireless network, an optimal value of bandwidth should be allocated to users. For this purpose, we developed a model of the link transmission rate and used it in conjunction with dynamic programming theory to calculate an optimal value of bandwidth for HWNs. The prediction-based energy-efficient techniques and their limits were discussed in [22]. The authors also recommended context-specific energy-efficient research efforts in the design of wireless communication networks. A technique called an Artificial Neural Network (ANN) was proposed [23] to predict Signal-to-Interference-and-Noise Ratio (SINR) in mobile networks. Radio resource scheduling was generally achieved based on estimated channel conditions, i.e., SINR with the help of Sounding Reference Signals (SRS). The Non-Linear Auto-Regressive External/Exogenous (NARX)-based ANN aims to minimize the rate of sending SRS and achieves an improved accuracy. A distributed framework for dynamic network selection was proposed considering dynamic parameters such as user mobility and battery life [24]. A deep Multi-Agent Reinforcement Learning (DMARL) algorithm, was proposed to optimize cost and energy-efficient manner. A distributed machine learning approach for a multiuser mobile edge computing (MEC) network was investigated [25] to optimize the system performances - latency and energy consumption.

In the multi-interface environment, the different available networks with their parameters are chosen according to their downloading speed as the total usage of the internet in mobile phones is 90% which essentially requires the high speed for data audio/voice and video downloading. The highest download speed means the faster the network, the more selection is made. The selection of a network has to go through the following steps:

a) Normalization of Input Parameters

In ANN approaches, data normalization is necessary before starting the training process to ensure that the influence of the input variable in the course of model building is not influenced by the magnitude of its original values, or its range of variation. The normalization technique usually consists of a linear transformation of the input/output variables to the range (0, 1). There are two most commonly used techniques to normalize the values. The first method bifurcates the inputs into beneficial and non-beneficial then divides beneficial with maximum value and non-beneficial with minimum value ( minimum value should be at numerator). In other methods the normalized value X_nis calculated with the formula as shown in equation (1).

In this paper we followed the very simple steps to normalize the input parameters.

b) Weight Initialization Method

Weights are used with inputs to show the preferences and weightage given by users. The more value of weights shows the higher preference of the user and vice-versa. Weight initialization aims to prevent layer activation outputs from exploding or vanishing during a forward pass through a deep neural network. Now, weights can be initialized with defined values with many MCDM techniques such as TOPSIS, AHP. The Fuzzy analytic hierarchy process (FAHP) is used in this paper to initialize the weights as it has more advantages than other methods.

c) Initialize Network. The network is initialized with its parameters and their respective weights. Each neuron has a set of weights that are maintained by the neural system. One weight for each input connection and an additional bias weight for the neuron's activity.

d) Forward Propagate: Forward propagation is a network-learning process that takes an input signal and propagates it through each layer, until the last layer outputs a predicted value. The first step is to calculate neuron activation as a weighted sum of the inputs. Then different activation functions are used to check the actual output.

e) Back Propagate Error: The backpropagate error, which is simply the difference between the outputs from the target outputs. This total error then propagated backward through the network from the output layer to the hidden layer, is used to update the weights using a backward pass.

f) Train Network and Predict:. Creating a neural network requires multiple iterations of exposing a training dataset to the network and for each row of data forward propagating the inputs, backpropagating the error, and updating the network weights. This process is broken down into two sections: update weights and train network. The function that returns the class prediction can be more useful than simply outputting an integer value. We can turn these values into a crisp prediction by selecting the class value with a larger probability value. We call this argmax() after the function in Python.

The proposed system is made for the different available networks and their parameters like download speed, latency, bandwidth, packet loss rate and jitter. The efficiency of any network depends mainly on these parameters, no doubt others also play a significant role. These parameters have different values for different networks from minimum to maximum. In Table 1. the parameter range of the different wireless networks is shown.

Table 1

Parameters of different networks [20]
Networks	Network download speed (Mbps)	Latency (ms)	Bandwidth	PLR (%)	Jitter (ms)	Target output (maximum download speed) Mbps
3G	0.1 Mbps-3Mbps	30–200	144kbps-2Mbps	2–10	10–30	3-7.2
WLAN	2Mbps – 10 Mbps	80–300	1Mbps-8Mbps	4–15	30–80	450–600
4G	15Mbps − 90Mbps	20–150	200Mbps-1Gbps	4–20	15–40	100–300
5G	150Mbps – 10Gbps	1–10	1Gbps-10Gbps	0–10	10–20	1000-10,000

In this paper, different values of input parameters were used to test the performance of the networks. Different network configurations were tested, and the configuration with five inputs and one output showed the highest download rate.

The users have different preferences; some users want good video quality or some may require better audio quality. Searchers need good speed for web browsing and some require fast & uninterrupted data transfer. So in this paper, we calculate the weights for each application separately shown in Table 2. The different inputs with their weights for particular applications are given to a fully connected neural network. The backpropagation network calculates the error (target value – actual output) and is system is trained to get the minimum value of error by changing the weights accordingly.

Table 2

Normalized Weights for different Applications
Application	DS	L	BW	PLR	Jitter	Total
Video	0.3101	0.1661	0.2912	0.0519	0.1808	1
Audio	0.4427	0.2843	0.1127	0.0639	0.0964	1
Data transfer	0.5232	0.1924	0.1331	0.1152	0.0361	1
Web Browsing	0.3686	0.3743	0.1242	0.0868	0.0461	1

Five algorithm for each networks are run. The best network is selected by taking the final outputs of five networks and comparing them with each other. This process is repeated for all applications separately. The graphs of the output of networks are shown and discussed in the next section.

We use pseudocode to determine the which network provides best output .

// Program to optimize weights for network selection using back propagation neural network //

// GET random values of input parameters.

// OBTAIN normalize values of inputs.

// INIT weights using FAHP.

// INIT the bias value ( -0.5 to 0.5)

// SET learning rate (0 to 1)

// SET momentum factor

// SET target value T (randomly)

DETERMINE Zin = X_n*W_ij

COMPUTE output node value (Y).

COMPUTE Error = T – Y

// Backpropagate the error value to get new updated value of weights and bias.

IF error = = 0 THEN

END

DISPLAY output(Y)

ELSE

REPEAT the process

// Compare the output of all networks

MAX (OUT1, OUT2……..OUTn)

DISPLAY network with maximum value.

// REPEAT the process for all applications.//

MATLAB is used to simulate the proposed prediction schemes and determine their effectiveness. The results are obtained for different networks for different applications. Here we have four networks: 3G, WLAN, 4G, and 5G five parameters and four applications: audio, video, web browsing and data transfer. The target value is set differently according to applications. We compare the performance of the algorithms in the maximum download rate, which is one measure of the strength of a network.

The different weights are assigned to different applications for example in video applications weights 0.3101, 0.1661, 0.2912, 0.0519, and 0.1808 are assigned to download speed, latency, bandwidth, packet loss rate, and jitter respectively with FAHP. Same for other applications audio, web browsing and data transfer, weights are calculated with the FAHP method. The basic BPNN algorithm produces output with a minimum error value.

Figure 3 represents the result of video applications where network 3G was selected two times ( 0.62, 0.93), 4G was selected three times( 0.96, 0.63, 0.83), 5G was selected four times( 0.71, 0.97, 0.99, 0.60) and WLAN was selected one time(0.76 ). The weights were calculated with FAHP method as: 0.4427, 0.2843, 0.1127, 0.0639, 0.0964 for audio. In Fig. 4 result of audio applications was shown in which 3G was selected two times (0.63,0.92), 4G was selected five times, (0.92, 0.64, 0.88, 0.83, 0.91) 5G was selected two times(0.79, 0.67) and WLAN was selected one time(0.82).

In Fig. 5 result of web browsing applications was shown in which 3G was selected two times (0.60, 0.94), 4G was selected five times (1.00, 0.63, 0.82, 0.90,0.86) 5G was selected two times (0.62, 0.67) and WLAN was selected one time (0.77). The weights were calculated with FAHP method as: In Fig. 6 result of data applications was shown. The weights were calculated with FAHP method as: 0.5232, 0.1924, 0.1331, 0.1152, 0.0361. For this application 3G was selected two times (0.58, 0.92), 4G was selected four times(1.00, 0.81, 0.86, 0.64) 5G was selected three times (0.73, 0.89, 0.73) and WLAN was selected one time(0.83).

For different applications: video, audio, web browsing, and data transfer the switching of networks depends upon these outputs, higher the output corresponding network is selected.We have selected random inputs for all networks and target was set as variable. As the changes happened in output the corresponding network with highest value at particular point was selected.

Interface management systems, which play an important role in wireless heterogeneous networks and IoT devices, informatics and security systems, are crucial to all of these fields. Recently, the various multimedia applications like audio, video on demand, data transfer and web surfing have drawn significantly more interest with advancements in technology. These applications require maximum use of internet services. Interface management provides the availability of an interface that will be the best for that particular application. Backpropagation neural network is used as a prominent technique of artificial intelligence(AI). Weight initialization with FAHP provides the more realistic weights for the implementation. As the algorithm run and simulations are done, weights are updated to achieve the minimum error. The results showed that this technique provides the switching of networks more efficiently than existing techniques in different multimedia applications. In future, more networks and more input parameters can be taken to get more efficient and precise results.

COMPETING INTEREST: NO

https:// www.guru99.com/ backpropagation neural-network-html
Andreev, S., Gerasimenko, M., Galinina, O., Koucheryavy, Y., Himayat, N., Yeh, S.P. and Talwar, S., 2014. Intelligent access network selection in converged multi-radio heterogeneous networks. IEEE wireless communications, 21(6), pp.86-96.
Sodhi, S.S. and Chandra, P., 2014. Interval based weight initialization method for sigmoidal feedforward artificial neural networks. AASRI Procedia, 6, pp.19-25.
Ahuja, K., Singh, B. and Khanna, R., 2014. Particle swarm optimization based network selection in heterogeneous wireless environment. Optik, 125(1), pp.214-219.
Alotaibi, N.M. and Alwakeel, S.S., 2015, December. A neural network based handover management strategy for heterogeneous networks. In 2015 IEEE 14th international conference on machine learning and applications (ICMLA) (pp. 1210-1214). IEEE.
Abbas, N. and Saade, J.J., 2015, January. A fuzzy logic based approach for network selection in WLAN/3G heterogeneous network. In 2015 12th Annual IEEE Consumer Communications and Networking Conference (CCN C) (pp. 631-636). IEEE.
Agiwal, M., Roy, A. and Saxena, N., 2016. Next generation 5G wireless networks: A comprehensive survey. IEEE Communications Surveys & Tutorials, 18(3), pp.1617-1655.
Kunarak, S., 2016, April. A Dynamic Channel Allocation Algorithm Based on Back-Propagation Neural Network for Vertical Handover in HetNets. In 2016 UKSim-AMSS 18th International Conference on Computer Modelling and Simulation (UKSim) (pp. 287- 292). IEEE
Ahuja, K., Singh, B. and Khanna, R., 2018. Network selection in wireless heterogeneous environment by CPF hybrid algorithm. Wireless Personal Communications, 98(3), pp.2733-2751.
Goyal, R. K., Kaushal, S., & Sangaiah, A. K. (2018). The utility based non-linear fuzzy AHP optimization model for network selection in heterogeneous wireless networks. Applied Soft Computing, 67, 800-811.
Liang, G. and Yu, H., 2018. Network selection algorithm for heterogeneous wireless networks based on service characteristics and user preferences. EURASIP Journal on Wireless Communications and Networking, 2018(1), pp.1-16.
Senouci, M.A., Senouci, H., Senouci, M.R., Ferdosian, N. and Mellouk, A., 2019. Flow/Interface Association for multi-connectivity in heterogeneous wireless networks: e-Health case. Ad Hoc Networks, 94, p.101942.
Chen, M., Challita, U., Saad, W., Yin, C. and Debbah, M., 2019. Artificial neural networks-based machine learning for wireless networks: A tutorial. IEEE Communications Surveys & Tutorials, 21(4), pp.3039-3071.
Khangura, S.K., Fidler, M. and Rosenhahn, B., 2019. Machine learning for measurement-based bandwidth estimation. Computer Communications, 144, pp.18-30.
Meng, Y. and Liu, X., 2019. Resource allocation and interference management for multi-layer wireless networks in heterogeneous cognitive networks. EURASIP Journal on Wireless Communications and Networking, 2019(1), pp.1-12.
Sun, Y., Peng, M., Zhou, Y., Huang, Y. and Mao, S., 2019. Application of machine learning in wireless networks: Key techniques and open issues. IEEE Communications Surveys & Tutorials, 21(4), pp.3072-3108.
Gao, Z., Chen, Y. and Yi, Z., 2020. A novel method to compute the weights of neural networks. Neurocomputing, 407, pp.409-427.
Liang, G., Sun, G., Fang, J., Guo, X. and Yu, H., 2020. An Access Selection Algorithm for Heterogeneous Wireless Networks Based on Optimal Resource Allocation. Wireless Communications and Mobile Computing, 2020.
Hosny, K.M., Khashaba, M.M., Khedr, W.I. and Amer, F.A., 2020. An Efficient Neural Network-Based Prediction Scheme for Heterogeneous Networks. International Journal of Sociotechnology and Knowledge Development (IJSKD), 12(2), pp.63-76.
Tan, X., Chen, G. and Sun, H., 2020. Vertical handover algorithm based on multi-attribute and neural network in heterogeneous integrated network. EURASIP Journal on Wireless Communications and Networking, 2020(1), pp.1-21.
Liang, G., Guo, X., Sun, G. and Fang, J., 2020. A User-Oriented Intelligent Access Selection Algorithm in Heterogeneous Wireless Networks. Computational Intelligence and Neuroscience, 2020.
Ogbebor, J.O., Imoize, A.L. and Atayero, A.A.A., 2020. Energy efficient design techniques in next-generation wireless communication networks: Emerging trends and future directions. Wireless Communications and Mobile Computing, 2020.
Ullah, R., Marwat, S.N.K., Ahmad, A.M., Ahmed, S., Hafeez, A., Kamal, T. and Tufail, M., 2020. A Machine Learning Approach for 5G SINR prediction. Electronics, 9(10),p.1660
Allahham, M. S., Abdellatif, A. A., Mhaisen, N., Mohamed, A., Erbad, A., & Guizani, M. (2022). Multi-Agent Reinforcement Learning for Network Selection and Resource Allocation in Heterogeneous multi-RAT Networks. IEEE Transactions on Cognitive Communications and Networking.
Guo, Y., Zhao, R., Lai, S., Fan, L., Lei, X., & Karagiannidis, G. K. (2022). Distributed machine learning for multiuser mobile edge computing systems. IEEE Journal of Selected Topics in Signal Processing.

Download PDF

Version 1

posted

You are reading this latest preprint version

Switching of Interfaces and Self-optimization of weights using Backpropagation ANN in WHN Enviornment

Status:

Version 1

Abstract

Figures

1. Introduction

2. Background And Related Work

3. System Model

4. Managing Interfaces With Proposed Algorithm

5. Simulation And Results

6. Conclusion And Future Scope

Declarations

References

Status:

Version 1