Traffic and mobility aware resource prediction using cognitive agent in mobile ad hoc networks

https://doi.org/10.1016/j.jnca.2016.06.010Get rights and content

Abstract

Mobile Ad hoc NETwork (MANET) characteristics such as limited resources, shared channel, unpredictable mobility, improper load balancing, and variation in signal strength affect the routing of real-time multimedia data that requires Quality of Service (QoS) provisioning. Accurate prediction of the resource availability assists efficient resource allocation before the routing of such data. Most of the published work on resource prediction in MANET focuses on either bandwidth or energy without considering mobility effects. Adoption of intelligent software agent such as Cognitive Agent (CA) for the accurate resource prediction has a significant potential to solve the challenges of resource prediction in MANET. The intelligence provided in CA is similar to the logical thinking like a human for decision-making. The predominant CA architecture is the Belief-Desire-Intention (BDI) model, which performs the various tasks on behalf of the human user as an assistant.

In this paper, we propose a CA-based Resource Prediction mechanism considering Mobility (CA-RPM) that predicts the resources using agents through the resource prediction agency consisting of one static agent, one cognitive agent and two mobile agents. Agents predict the traffic, mobility, buffer space, energy, and bandwidth effectively that is necessary for efficient resource allocation to support real-time and multimedia communications. The mobile agents collect and distribute network traffic statistics over MANET whereas a static agent collects the local statistics. CA creates static/mobile agent during the process of resource prediction. Initially, the designed time-series Wavelet Neural Networks (WNNs) predict traffic and mobility. Buffer space, energy, and bandwidth prediction use the predicted mobility and traffic. Simulation results show that the predicted resources closely match with the real values at the cost of little overheads due to the usage of agents. Simulation analysis of predicted traffic and mobility also shows the improvement compared to recurrent WNN in terms of mean square error, covariance, memory overhead, agent overhead and computation overhead. We plan to use these predicted resources for its efficient utilization in QoS routing is our future work.

Introduction

Mobile Ad hoc NETwork (MANET) is a peer-to-peer infrastructure-less network consisting of nodes within the range of each other that dynamically configure themselves for operating as the end host as well as routers. Unique characteristics of MANET (such as dynamic configuration due to mobility, shared wireless medium, distributed multi-hop communication) yield dynamic, unpredictable, and unreliable environment for real-time multimedia applications. Dynamic nature of MANET affects routing path rapidly and unexpectedly as nodes share a common broadcast radio channel that operates at 2.4/5 GHz. The limited radio spectrum results into limited available bandwidth for communications in MANET. The IEEE 802.11 Medium Access Control (MAC) protocol tries to address this issue by managing and maintaining communications between nodes by coordinating access to the shared channel through Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) (Siva Ram Murthy and Manoj, 2004). The nodes ready to transmit the data in the scenario contend for the channel simultaneously and the winning node gain access to the channel. Gaining the access to the channel is not fair among the nodes ready to transmit during heavy traffic conditions. The throughput degrades exponentially with the increase in the number of nodes during heavy traffic conditions, which cause the increase in packet delay of the node that has a frame to transmit. The unfair access opportunities, throughput degradation, and increased packet delay affect QoS support in MANET. Future resource availability assists in achieving the efficient and fair resource access, which contributes to enhance the performance. An approach discussed in this paper predicts the future resource availability considering future traffic and mobility of nodes.

Nodes in MANET admit traffic without considering residual resources that result in QoS deterioration. Resources in MANET include buffer space, energy, bandwidth, and CPU. The efficient and fair access is possible if the available resources are known in advance, but MANET has inconsistent availability of resources. The balance between traffic injected and required available resources help to release the overloaded resource contention. Excess usage of resources due to injected traffic than available resources results in the increase of end-to-end delay, and packet drop ratio, which leads to decrease in network throughput. Residual resource computation is important for network functionality evaluation. Residual resources must meet the application demand.

Residual resources predictions for future transmission contribute towards an efficient and fair resource access with QoS provisioning. The existing available resource prediction approaches focus on the estimation of available bandwidth and energy, neglecting other resources. Adopting some intelligent prediction strategies such as software agents for available resource prediction are suitable where the environment is dynamic, unpredictable and unreliable (Padgham and Winikoff, 2004).

Software agents execute tasks created from intelligent programs automatically without disturbing the normal activities (Eberhart and Shi, 2011). Software agents use knowledge base maintained at the node to meet the specified goals. Static and mobile agents are the classes of software agent technology. Static agents are always available at a node to carry out the complex work in real-time, where mobile agents are dynamically created and immediately migrated from the source node to other nodes to collect the network statistics in real-time. The decision-making at the software agent is part of intelligence. The use of cognitive concepts such as beliefs, knowledge, desires, and intentions for decision-making are similar to human logical thinking and speaking. Belief-Desire-Intention (BDI) model includes modular data structures for defining beliefs, intentions, and desires as a part of model procedure and manipulated by applying a suitable theorem prover (Rao and Georgeff, 1995). The cognitive concept usage involves the design of various tasks based on the theorem prover, procedural approach, or finite-state machine. The cognitive concepts represented as the finite-state machine operate in the restricted environment.

A software agent with the cognitive property, called as Cognitive Agent (CA), resolves the complex domain of resource prediction in MANET. CA is a static, multi-agents with high-level mental attitudes, such as beliefs, goals, and plans (Rao and Georgeff, 1995). Such CA is a software entity with at least one of the following characteristics (Uhrmacher and Weyns, 2009). (1) CA executes continuous functions autonomously in a specified environment. (2) Activities of CA use a flexible and intelligent way. (3) CA responds to the changes in the specified environment. (4) Reinforcement learning is used for executing the activities of the specified environment. (5) They communicate and co-operate with other agents in the environment to meet the goals. (6) Even though they are proactive in nature, they move from one place to another place to meet the goal. (7) They have high-level mental attitudes, such as beliefs, goals, and plans.

Architectures of CA emulate human's cognitive concepts like decision-making, representing information, and learning from experiences. A set of beliefs specifies knowledge in the environment of BDI model. Desire in the BDI model indicates the state of the environment the agent prefers at a particular time. Intentions yield the state of the environment the agent is trying to meet. Benefits of BDI model architecture include (1) adapting and learning quickly in MANET environment, (2) updating beliefs regularly in a dynamic environment like MANET, (3) independent decision-making on available options related to an event, (4) generating commitments and execute them according to intentions, and (5) offering a precise model of teamwork, which is critically required for MANET.

The belief, desire and intention states relate the agent perceptions and actions as shown in Fig. 1. Knowledge representation in BDI model encapsulates BDI modules, allowing only query/update interface. Interaction rules specify the interactions between BDI modules. Application of an interaction rule changes the agents’ states in an atomic way. Interpreter selects and executes an arbitrarily applicable interaction rule. Designing such a BDI model requires following four stages. (1) The application scenario includes relevant classes for each role. The hierarchy of these classes is also developed. (2) Each role involves the functionality in terms of the responsibilities associated, the services required and provided. Each service determines the associated goals. (3) The belief structure of the system determines the information required for each role to meet the goal. (4) The planned sequence of action meets each goal along with the appropriate environment.

The autonomous intelligent action could be computed through one of the computational intelligence techniques such as genetic algorithm, swarm intelligence, fuzzy logic, machine learning, and neural networks (Eberhart and Shi, 2011). Linear or nonlinear prediction uses computational intelligence techniques. Linear prediction model requires the linear stationary resources, which makes its usage unsuitable for MANET. The non-linear prediction models such as Wavelet model and Neural Network (NN) model are suitable for MANET resource prediction. Wavelet model is more accurate compared to NN model. Wavelet model does not predict the real-time resource availability and recursive characteristics. The NN model evaluates and predicts the behavior of non-linear and non-stationary systems, which relies on the observed data rather than on an analytical data. NN can estimate any function in an efficient and stable way, when the underlying data relationships are unknown. The characteristics of NNs include nonlinear mapping, generalization ability, robustness, fault tolerance, adaptability, parallel processing ability, etc. NN converges slowly and gets only local sub-optimal solution. Combination of Wavelet and NN model for resource prediction shows better performance in training and adaptation efficiency. Wavelet function usage at the hidden layer of NN results in Wavelet Neural Network (WNN) that yields more accurate real-time prediction (Alexandridis and Zapranis, 2014). WNN model has good prediction properties in MANET environment.

The proposed CA-based Resources Prediction mechanism considering Mobility (CA-RPM) at MAC layer uses CA and WNN to support real-time and multimedia communication in multi-hop MANET with QoS provisioning. To meet real-time constraints, the mechanism uses priority-based differentiated MAC scheduling. The use of this developed mechanism in routing protocol offers an efficient network management.

This section discusses some of the related works. Determining the future status of the resources in the wireless network is not straightforward since many external factors play an important role in the determination of variations in the traffic generated by the users as well as interferences. A passive residual resource estimation method based on medium utilization, collision and back-off is described as part of MANET admission control in Wu et al. (2013). The length of the HELLO message is modified by adding the Network Allocation Vector (NAV) duration field, which does not generate any serious impact on the MANET performance. Even though the estimated results are accurate, it estimates the network bandwidth without considering energy and buffer space. Determination of the available resource is the heart of admission control, which admits new traffic whose QoS requirements are satisfied without violating the existing traffic (Hanzo and Tafazolli, 2009). Resource requirements for the requesting traffic are not considered in the resource estimation based admission control method. There is a need to predict a collision, congestion, route failure, or inaccuracies for the requesting traffic. Resource availability is required before reserving it to provide QoS for high-priority traffic at all intermediate nodes along the route from source to destination (Yu et al., 2013). Every resource reservation scheme needs to find available resources along a route.

Prediction of traffic is important to improve network resource allocation, QoS provisioning, network traffic management, congestion control, bandwidth efficiency, the performance and availability of the MANET. Network traffic in MANET is unpredictable and chaotic. Computational intelligence is used in network traffic prediction model (Yadav and Balakrishnan, 2014). It uses an Adaptive Neuro-Fuzzy Inference System (ANFIS) and Auto-Regressive Integrated Moving Average (ARIMA) for the wireless network traffic modeling. The comparison in terms of computational complexity indicates the ANFIS model suitability for wireless traffic prediction. Hidden Markov model based traffic measurement at the packet-level, and application layer is modeled for fixed as well as variable packet size using real test traces in Maheshwari et al. (2013).

Combining NN with the bio-inspired techniques such as genetic algorithm, ant-colony algorithm, and particle swarm optimization for traffic prediction improves the convergence speed and learning ability of the NN, which is described in Lan (2013). It does not explain any overhead due to bio-inspired technique. Recurrent WNN has built in characteristics like fast learning, good generalization and information storing ability, which is used for network traffic prediction (Zhang et al., 2012). The authors claim that the model operates well with the much higher prediction precision compare to traditional Back-Propagation NN (BPNN) model. In Liu et al. (2013), fusion model by combining Stationary Wavelet Transform (SWT), Quantum Genetic Algorithm (QGA), and BPNN are described for wireless traffic predictions. SWT is used to transfer non-stationary data to stationary components while BPNN is optimized with QGA. Wireless ad hoc network traffic is modeled according to a Poisson process (Afshar et al., 2011) that is used for prediction of future traffic with NN under the assumption that the environment is the collision-free, and no packet loss can occur.

The performance of MANET is highly sensitive to the mobility of nodes. Mobility prediction with linear model is used if the nodes keep on moving in the same direction with constant speed. Most of the location prediction techniques assume that nodes move according to the Random Waypoint Mobility (RWM) model. Mobility models are based on different parameter settings related to node movements such as starting location of a node, movement direction, velocity range, and speed changes over time. As a result, another mobility model usage for mobility prediction can lose its accuracy and efficiency. In addition, Global Positioning System (GPS) device is used for determining the current speed and direction on each node in MANET, which is used in many mobility prediction algorithms for calculating link expiration time.

Stable link selection during route discovery for adaptive routing in route maintenance is described in Dung et al. (2013) which modifies RREQ and RREP to add mobility information. Duration of link stability between two nodes is calculated without using mobility information for classification of neighbors into eight zones, one for each direction (Al-Akaidi and Alchaita, 2007). Hello message rate is adjusted by predicting the mobility using an auto-regressive model for predicting the location, speed and direction using previous location data in Hello messages (Li et al., 2011). The Hello message transmitted only when that node's predicted position may lead to topology changes. Linear model may not give the most accurate results in real-time MANET scenario. The mobility prediction technique developed in Kaaniche and Kamoun (2010) is independent of the mobility model used by nodes. They assume that each node in the MANET is aware of its location using an on-board GPS receiver. So, it can periodically record its geographical location. The particle swarm algorithm can be used to manage the connectivity between the node (Dengiz et al., 2011). The authors of Ghoutia et al. (2013) explain the usage of extreme learning machines in neural network to predict the mobility (position, speed, and movement direction angle) of arbitrary nodes in MANET. The authors claim that an excellent prediction performance is achieved without any quantitative measure. The mobility prediction given in Torkestani (2012) relies on the history of the mobility using Gauss–Markov random process. The accuracy of prediction is high, but long time is taken to learn the mobility prediction model, which may cause the delay.

The buffer is one of the important resources in MANET, which is maintained as the queue for processing packets. The maximum queue size is equal to the total buffer space at MANET node. If a node has no buffer space remaining, it drops the arriving packets for which it is not a destination. This queue management scheme is called as Drop Tail. In popular IP4 Active Queue Management (AQM) scheme, the sending node is notified to stop sending data or lower the data rate due to queue space unavailability and later the sending node can be allowed to send more packets when the enough space is again available in the queue. AQM for MANET nodes lacks efficient queue management due to packet broadcasting, which makes it difficult to locate the sender of the packet easily by resource constrained MANET nodes. Considerable work has been done on the packet queue management in wired and wireless networks as explained in Richelle (2013).

Lightweight and efficient AQM scheme discussed in Kulkarni et al. (2006) and Kumar et al. (2010) for MANET is PAQMAN, which uses Recursive Least Squares (RLS) algorithm to predict the average queue length in the next prediction interval. The average queue length is used as the congestion indicator. The performance of PAQMAN has been compared only with Drop Tail not with other well-known AQM algorithms. Ad Hoc Hazard RED (AHRED) has been proposed by authors of Abbasov (2004), which is compared with RED, REM and SRED. The packet dropping probability characteristic follows a Weibull model of the hazard rate function, with queue length as the congestion indicator. However, packet dropped itself changes with queue length. MANET still uses Drop Tail policies for packet queue management. Data sent from low data rate sources does not get a fair share in queues. In Aamir and Zaidi (2013), buffer management scheme for fixed and mobile nodes is explained, which assigns buffer space dynamically to all neighboring nodes in proportion to the number of packets received from them. Initially, the equal buffer space share is allocated to each neighbor. Later, according to the designed algorithm, the allocation is dynamically adjusted according to the instantaneous share of neighbors in the buffer and the gap between the occupied and allocated buffer space. Maximum and minimum buffer space limits on a node also help to allocate a fair share. This scheme does not consider packet priority. High-priority data from high data rate nodes may not get buffer space. Packet drop is the main issue of queue management in MANET environment, which need a highly responsive solution.

MANET nodes are powered by small-sized batteries that are difficult to replace. Sources of energy consumption include radio communication (involves energy consumption due to transmission and reception), data processing (involves the usage of CPU, memory, and hard drive), retransmission (due to collision, packet loss), node overhearing (node picks up packets that are destined for other nodes) and protocol overhead (network control packets and header bits of data packets). Power consumption for data processing is relatively very small compared to radio communication. Energy consumption due to mobile nodes is an issue in MANET. Existing literature (Singh and Kumar, 2010) focus on various energy-efficient protocols at physical layer, MAC layer, network layer and transport layer of MANET (Mohsin et al., 2012). Energy conservation-related problems are solved by these protocols but still research is ongoing to maximize the energy conservation at various layers. In any case, a precise estimation of energy consumption helps in efficient energy management. There is not much work published in the literature on prediction-based evaluation of energy consumption in MANET.

The entire capacity of a channel cannot be used during the transmission of the packet because some amount of bandwidth is needed for communication-related overheads (such as initiating communication, neighbor node interferences, etc.), which reduces the node's Available Bandwidth (AB) (Alzate et al., 2008). AB in MANET is affected by several factors such as the channel utilization, idle period synchronization at sender and receiver, RTS/CTS overhead, acknowledgment delay, collision probability, frame retransmission count and back-off duration. These factors can be computed by collecting network statistics at the router level before estimating AB. There is a trade-off between accuracy and overhead with respect to AB estimation (Zhao et al., 2013). The challenges for accurate AB estimation in MANET include identification of the nodes in the carrier sense range, intra-flow contention, synchronization of an idle period at the sender and receiver, and estimation of the collision probability. The effort has been made to improve the accuracy of estimated AB for wireless networks.

The existing AB estimation techniques are categorized as passive, active and model-based techniques (Chaudhari and Biradar, 2015a). In passive techniques, AB is estimated by developing models using the network statistics such as the synchronized idle period at the sender–receiver pair before data transfer, collision probability, and random waiting time. Passive techniques do not disturb the network traffic but waste limited computation and storage resources causing the delay. In active techniques, probe packets are sent to recover link-level information (such as the number of collisions, end-to-end capacity, etc.) in a path from source to destination, which may disturb and delay the normal network traffic. The accuracy of these techniques is still not satisfactory. In model-based techniques, the mathematical model is developed, which requires complex mathematical structure. Analytical/Mathematical models help in providing the quantitative analysis of the network protocol performance and also helps to predict the result if the network parameters are changed. This is not possible either with active or passive techniques. Building an accurate analysis model for a multi-hop wireless network is not an easy job. Analytical models are proposed using the operation of the Distributed Co-ordinated Function (DCF) in ad hoc networks each with their own set of assumptions. Mathematical model mechanism sends packet trains at a rate lower than the AB.

Passive techniques are more suitable in MANET where the node activities and network behavior is monitored to estimate AB. The QoS-AODV (de Renesse et al., 2004) is one of the early passive estimation techniques, where each node calculates the bandwidth efficiency ratio between the number of transmitted and received packets that is broadcasted among the one-hop neighbors through Hello messages. The QoS-AODV performs one-hop AB estimation ignoring interferences due to carrier sense range. BRuIT (Chaudet and Lassous, 2001) reserves the bandwidth for ad hoc networks that takes into account the existence of the interferences. Each node derives AB in its two-hop neighborhood, but lacks synchronization between sender and receiver for their availability. In CACP (Yang and Kravets, 2005), each node computes the local idle time fraction by monitoring the radio medium. In addition to interference, it notices intra-flow contention, but does not consider wastage of AB due to the random period such as RTC/CTS. Extensions to CACP are AAC (de Renesse et al., 2007), ABE (Sarr et al., 2008) and IAB (Zhao et al., 2009) also consider such interfering problem.

In AAC, each node considers the set of potential senders as a single node and the channel idle period between them is totally overlapped, which results in the overestimation of AB. AAC also takes into account the intra-flow contention problem. AAC does not consider the idle period synchronization at sender and receiver, which is considered in ABE. ABE calculates the overlap probability of idle time at the sender and receiver assuming that each node's medium occupancy is uniform, which ignores the dependency of interfering around the sender and receiver. This may result in inaccuracy of AB. The cPEAB (Tursunova et al., 2010) considers the additional overheads caused by ACK frames at random waiting time, which are not considered in the AAC and ABE. APBE (Park and Roh, 2010) additionally considers the overhead caused by the RTS and CTS. The IAB is similar to ABE except that it considers synchronization between sender and receiver by differentiating the channel busy caused by transmitting or receiving from that caused by carrier sensing. It improves the accuracy of estimation using the overlap probability of two adjacent node's idle time. Lagrange polynomial is calculated for any scenario, which is used by all nodes in any scenario in the IAB and ABE. The collision probability of each node/scenario may not be same practically, so there is a need to compute the Lagrange polynomial at each node in each scenario separately before actual transmission of data, which we have been proposed in DLI-ABE (Chaudhari and Biradar, 2014b). DABE (Peng and Yan, 2012) estimates AB in a distributed way by the method of channel monitoring, collision estimating and back-off duration predicting. For the random appearance of the frame collision and back-off procedure, the monitoring result cannot reflect the future status of the link. Because of this, a collision between frames and back-off-related information should be evaluated when the link bandwidth is estimated.

The proposed resource prediction mechanism in MANET, CA-RPM, is an extension of our previous work given in Chaudhari and Biradar (2014a), where we presented the fundamental concepts of various resource predictions in MANET using WNN-based traffic and mobility models. Initially, traffic and mobility prediction models use the corresponding WNN model. The predicted traffic and mobility are used to predict the buffer space, energy, and bandwidth. Limitations of the previous work (Chaudhari and Biradar, 2014a) are as follows. (1) It lacks usage of the intelligent technique, which helps to improve the performance of an application in MANET environment. (2) Formulation of the components and the problem require detailed description. (3) The accuracy of all predicted resources needs a clearer description. (4) It lacks the computation of various overheads for each resource prediction in MANET environment. The use of the intelligent software agents helps to solve these shortcomings in CA-RPM. The CA-RPM includes clear formulation of the modules for predicting various resources along with the various overhead computations. The use of a BDI model in CA-RPM performs the various resource prediction tasks on behalf of the human user as an assistant.

Our contributions in this paper are as follows. (1) Formulation of traffic and mobility models in MANET. (2) Designing buffer, energy, and bandwidth prediction models. (3) Designing CA-RPM software agency for predicting resources in MANET using resource prediction models. (4) Employing BDI based approach for detection of various events such as decision-making, prediction, and collection or distribution of node traffic. (5) Engaging CA for predicting the future availability of resources. (6) Simulation analysis of the proposed work in terms of accuracy of the predicted resource, memory overhead and computational overhead over the network. (7) Comparing the simulation results of traffic prediction, mobility prediction and AB prediction with the methods given in Zhang et al. (2012), Torkestani (2012), and ABE (Sarr et al., 2008) respectively.

The rest of the paper is organized as follows. Traffic and mobility-aware resource prediction models are explained in Section 2. Agent-based resource prediction mechanism is explained in Section 3. Evaluation of CA-RPM with the non-agent based approach is described through simulations in Section 4. Finally, Section 5 presents the conclusions of this research.

Section snippets

Traffic and mobility aware resource prediction model

A node willing to transmit the information successfully is required to do so by finding the adequate resource availability. It becomes important to predict the future availability of resources since the resources are limited in MANET. Resource prediction assists efficient resource utilization yielding the longer lifetime of the MANET. The challenge is to predict future availability of resources in dynamic and unpredictable MANET environment. The fundamental requirement to predict future

Cognitive agent based resource prediction mechanism

The proposed CA-RPM activates agent platform using the resource prediction agency to predict various resources for future usage at a node, which consists of four agents; one static agent, one static CA and two mobile agents. The static agent is known as Local Statistics Collection Static Agent (LSCSA), which collects statistics of a node like available energy, available buffer space, current location, speed of mobility, available bandwidth, and current traffic. Resource Prediction Cognitive

Simulation

Agent-based resource prediction mechanism is simulated using event-driven simulation in various wireless network scenarios to assess the performance and effectiveness of the proposed approach. Event-driven simulation executes various functions at discrete events in a chronological order. This section explains the simulation model used to test our proposed solution with the performance of it. In this section, we discuss the simulation model used to test the proposed scheme, simulation procedure,

Conclusion

MANET traffic is nonlinear and time varying, which results in the lack of QoS provisioning. The guaranteed QoS for MANET application expects the resource prediction-based routing. This paper proposed cognitive agent-based resource prediction mechanism in MANET through the resource prediction agency. The agency effectively performs various functionality related to the prediction of traffic, mobility, buffer space, energy, and bandwidth, which is necessary for efficient resource allocation to

Acknowledgments

The authors wish to thank Visvesvaraya Technological University (VTU), Karnataka, India, for funding the part of the project under VTU Research Scheme (Grant No. VTU/Aca./2011-12/A-9/753, Dated: 5 May 2012.

References (44)

  • Chaudhari, Shilpa Shashikant, Biradar, Rajashekhar C., 2014. Resource prediction using wavelet neural network in mobile...
  • Chaudhari, Shilpa Shashikant, Biradar, Rajashekhar C., 2014. Collision probability based available bandwidth estimation...
  • Shilpa Shashikant Chaudhari et al.

    Survey of bandwidth estimation techniques in communication networks

    Wirel. Pers. Commun.

    (2015)
  • Shilpa Shashikant Chaudhari et al.

    Available bandwidth estimation using collision probability, idle period synchronization and random waiting time in MANETscognitive agent based approach

    Wirel. Pers. Commun.

    (2015)
  • de Renesse, R., Ghassemian, M., Friderikos, V., Aghvami, A.H., 2004. QoS enabled routing in mobile ad-hoc networks. In:...
  • Ronan de Renesse et al.

    Cross-layer cooperation for accurate admission control decisions in mobile ad-hoc networks

    IET Commun.

    (2007)
  • Dung, Le The, Ha, Sue Hyung, An, Beongku, 2013. A practical adaptive scheme for enhancing network stability in mobile...
  • Russell C. Eberhart et al.

    Computational Intelligence: Concepts to Implementations

    (2011)
  • Ghoutia, Lahouari, Sheltamia, Tarek R., Alutaibib, Khaled S., 2013. Mobility prediction in mobile ad-hoc networks using...
  • I.I. Hanzo et al.

    Admission control schemes for 802.11-based multi-hop mobile ad-hoc networksa survey

    IEEE Commun. Surv. Tutor.

    (2009)
  • Heni Kaaniche et al.

    Mobility prediction in wireless ad-hoc networks using neural networks

    J. Telecommun.

    (2010)
  • Kulkarni, P., Nazeeruddin, M., McClean, S., Parr, G., Black, M., Scotney, B., Dini, P., 2006. Deploying lightweight...
  • Cited by (20)

    • A deep stochastical and predictive analysis of users mobility based on Auto-Regressive processes and pairing functions

      2020, Journal of Network and Computer Applications
      Citation Excerpt :

      So, we will consider a generic mobile network, in which all the nodes are free to move among a given geographical region (e.g. Vehicular Ad-hoc NETworks - VANETs, Wireless Sensor Networks - WSNs, Mobile Ad-hoc NETworks - MANETs, etc.). As introduced in (Wu, 2018; Chaudhari and Biradar, 2016; Wang et al., 2019), each time a mobile network needs to know the future positions of mobile users, mobility patterns are stored in a centralized/distributed database. Then, a predictive model is built-up by analyzing the structure, and features of the stored patterns and as well as the model are exploited to obtain the so-called “next-places”.

    • Enhanced resource allocation in mobile edge computing using reinforcement learning based MOACO algorithm for IIOT

      2020, Computer Communications
      Citation Excerpt :

      Mohammad Aazam et al. proposed that the technologies which are directly or indirectly connected to the IoT devices are assumed to be of the multimedia based IoT services in the establishment of the growing tendencies in the IoT with the social awareness in the SIoT paradigm [32]. The QoE Quality of Experience is measured in the dynamically allocated region to follow the incorporating QoE and the net promoter score is established within the mathematical model to frame the QoE parameter estimation in the CloudSim simulator in the regions using java and its impact with the proposed model [8]. Md SalikParwez et al. presented that the Wireless paradigm has the connectivity using the Internet of Things (IoT) with the next-generation wireless network in the predicted resources are being utilized with the virtualization to enhance the resource utilization with the support of sharing the content and the approach is elaborated with the use of Virtual networks and henceforth the mobile edge computing is assumed to be of the fast computing resource utilization platform in the cache resources [33].

    • Bandwidth estimation for aggregate traffic under delay QoS constraint based on supermartingale theory

      2018, Computer Communications
      Citation Excerpt :

      In [9] and [10], the bandwidth requirement was predicted by the historical information of bandwidth consumption. In [11], the authors proposed the cognitive agent-based resource prediction mechanism in mobile ad hoc networks, and the time-series Wavelet Neural Networks (WNNs) were deigned to predict traffic performance including bandwidth. In [12], quaternion-based Kalman filter was adopted to predict the required bandwidth, in which the prediction equation was formulated via analyzing the prior and posterior network states.

    • Assessing carbon emissions from road transport through traffic flow estimators

      2018, Transportation Research Part C: Emerging Technologies
      Citation Excerpt :

      Data clustering combined with genetic programming (Elhenawy et al., 2014) is a computationally efficient method that significantly reduces the prediction error in the instantaneous and historical average. Chaudhari and Biradar (2016) compare General Regression Neural Network (GRNN) to vector autoregression and Historical Average models to forecast traffic flow. GRNN is found to be the most stable method in missing data conditions.

    • A novel efficient MAKA protocol with desynchronization for anonymous roaming service in Global Mobility Networks

      2018, Journal of Network and Computer Applications
      Citation Excerpt :

      The rapid development of mobile networks has greatly facilitated people's lives. When a user moves from one place to another, Global Mobility Network (GLOMONET) ensures that a travelling wireless device is kept connected to network without breaking the connection (Chaudhari and Biradar, 2016; Du et al., 2017). With the help of the home agent, Global Mobility Network (GLOMONET) can identify legitimate users and allows legitimate mobile users to use roaming service anytime anywhere.

    View all citing articles on Scopus
    View full text