Next project was one of my first networking Science articles. It is a short look at latest achievements from some of the biggest networking scientists today with some short comments from my mentor and me. The work was presented at 19. International scientific conference of International Federation of Communication Associations. International scientific conference “DIT 2012” accepts and publishes scientific and professional papers and the results of interdisciplinary scientific research, whose area of interest is the development of society, education, science and technology.

Okaj, let’s go on…


dr. sc. Božidar Kovačić & Valter Popeškić (me)
University of Rijeka – Department of Informatics
Theme: 3. New ICT technology, media and e-education;

Abstract or simply, intro…

Today’s computer networks will not be able to resolve the tangled problems that emerge from increasingly throughput-demanding services after their need for resources exceed the capabilities of today’s networking technologies. Cognitive networks have the means to resolve this issue incorporating intelligence to the network functions. Introduction to Cognitive network as a concept brings the view into the future of communication, information and learning using modern technology.

A cognitive network is a network consisting of elements that reason and have the ability to learn. In this way they self-adjust according to different unpredictable network conditions in order to optimize data transmission performance. In a cognitive network, judgments are made to meet the requirements of the network as an entire system, rather than the individual network components. The main reason of the emergence of cognitive networks is to achieve the goal of building intelligent self-adjustable networks and in the same time improve the performance. Intelligent self-adjustable networks will be able to use intelligence to determine ideal network operating state for many tunable parameters.

Cognitive Networks

The world is a living witness of the beginning and development of wireless technology. From communication to networking, wireless technology has been part of the personal and professional lives of people. Today’s world is obviously operated by technology with the presence of wireless gadgets ranging from big to small sizes with varying costs. Since problems arise due to the increasing demands in services concerning networking technologies, cognitive networks can be used as a resolution. The present technology used in data networking is limited and not open to possible changes leading to low quality performance and results. The limitations of the current data networking technology include the limitations in response, state and scope mechanism, which lead to the incapability of network elements such as nodes, protocol layers, policies and behaviors to perform intelligent adaptations. With this, the concept of cognitive networks is to be presented to the people for their comprehension on its benefits on modern ways of communication, information and learning.

Thomas[1] defined cognitive networks (CN) as a network with a cognitive process that can perceive current network conditions, plan, decide, act on those conditions, learn from the consequences of its actions, all while following end-to-end goals.  The authors also added that the loop also referred as the cognition loop can has the ability to sense the environment, plan action as based on the inputs from sensors and network policies, decide the best scenario for its end-to-end purpose with its reasoning engine and act on the selected scenario. Cognitive networks are said to learn from the former situations, plans, decisions and actions and can use its knowledge to improve future decisions. Cognitive networks are formed when a collection of communication nodes organize to achieve network-level goals with the aid of some form of cognitive processing. The concept is an outgrowth of research in cognitive radios, which couples intelligent, or cognitive, processing with a platform that provides the ability to both sense the radio environment and react by changing its configuration (Friend, 2009).

According to Mathur and Subbalakshmi[2], cognitive radio is characteristically related with the opportunistic use also termed as secondary users who utilize the unused parts of primary users’ spectrum. The secondary users are usually with license but have the ability to sense white spaces in the spectrum of licensed primary users. These secondary users must not cause harmful interference to primary users. Priority access to the spectrum is given to primary users this requires the secondary users to vacate the bands when primary users ask to.

The term cognitive is related to the ability of a network to be aware of its operational status and adjust its operational parameters to fulfill specific tasks, such as detecting changes in the environment and user requirements (Kliazovich & Granelli, 2010). In a cognitive network, autonomous and adaptive radios select their operating parameters to achieve individual and network-wide goals (Komali et al. 2010). Research in cognitive networks seeks to provide a solution to the difficulty of electively using the expanding capabilities of wireless networks by embedding greater degrees of intelligence within the network itself (Friend, 2009). Cognitive networking is rich with opportunities for novel research. The areas of learning and reasoning are particularly ripe and central to the success of cognitive network implementations (Friend et al. 2007).

Figure 1: Spectrum pooling idea (Mahamuni et al. 2010)
Spectrum pooling ideaFigure 2: Cognitive radio system
Cognitive radio system


Mahmoud[3] mentioned the different definitions of Cognitive Networks from various authors. Sifalakis et al. (2004) defined cognitive network as a networks with a capability to adapt itself in response to conditions and events based on reasoning and the prior knowledge that it has acquired. Whereas Boscovic’s (2005) definition for cognitive network was a network that can perform dynamic alteration on its topology and/or operational parameters to answer the necessities of a specific user while enforcing operating and regulatory policies and optimizing overall network performance. In general, cognitive network is a type of network, wireless or not, that has a capability to think learn and remember. To add, cognitive networks are unique because of its capabilities such as self optimization,self monitoring, self repair, self protection, self adaptation and self healing. In order to optimize network operation, reconfiguration, management, and improving performance,  a proposal to introduce self-awareness, self-management, and self-healing properties by bringing “intelligence” into the network, creating a new paradigm in networking, referred to as cognitive networking (Kiliazovich & Granelli 2010). One of the key requirements of a cognitive network is to learn the relationships among network protocol parameters spanning the entire stack in relation with the operating network environment (Quer, 2011). Certain cognitive networks are structurally similar to one another. The network and cognitive element objectives align in a manner that, when particular strategies are used by the cognitive elements, the network can be assured of converging to desirable network and cognitive element performance. When it is possible to identify this similar underlying structure, we say that the cognitive networks that share this similarity are part of the same class (Thomas, 2007).

Boutaba and Xiao (2007), discussed the capability of cognitive network to self-manage.  The self management vision cited IBM’s research on “autonomic computing”, a term defined as a distributed system where a set of software/network components can regulate and manage themselves in areas of configuration, fault, performance and security in order to achieve some common user-defined objectives. According to the self management vision, the four self properties of autonomic computing are: self configuration which enables the entities to automate system configuration following high level specifications while self organizing desirable structures and/or patterns; self optimization allows entities to constantly look for improvement on their performance and efficiency while adapting to environmental changes without inputs from humans; self healing enables automatic detection, diagnosis and recovery from faults resulting from internal errors or external inconsistencies; lastly is the self protection capability which allows the entities to have automatic defence against malicious attacks while isolating attacks for the prevention of system-wide failures.

            Self managing networks face challenges. These networks must undergo laborious and careful engineering to assure the system’s validity, consistency and correctness from the time of its development until operation and removal. Monitoring should also be performed as it has a supreme importance in self managing networks. This may answer questions like when and where to monitor. Since cognitive networks are known for learning and reasoning, knowledge is considered to be critical to autonomic reasoning, analysis and decision making. Questions concerning the determination of over-all system behaviour from interactions of individual components, coordination resulting to achievement of system-wide objectives and the likes are answered by interactions among autonomic components or autonomic systems.

System stability, robustness, resilience and correctness are also challenges faced by self managing networks. Although not easily understood, Boutoba and Xiao (2007) explained these in the form of questions like how does localized reasoning based on limited information lead to stable global behavior? How does one avoid global inconsistencies and undesirable system behaviors in autonomic systems? As mentioned by other studies, cognitive networks must adapt to environmental changes. This characteristic will provide consistency and maintenance of optimality of the system. Successful network design should make assumptions explicit, handle common failures, understand the interactions between different control loops, make goals explicit, make systems more user friendly and understandable and have built in validation according to Boutaba and Xiao (2007).

Self optimization is a characteristic that enables the system to provide data service requests the optimum resource distribution amongst a group of data servers. Self optimization may be related to QoS performance management, resource scheduling, queuing modelling and management of multi-tier network applications. Efficient self-healing and protection mechanisms require the autonomic system to be predictive about the status of its environment and itself, and be able to reason about the changing environment for the purpose of fault localization and recovery (Boutoba & Xiao, 2007). Self organization is necessary for the achievement of comparable constructs including clustering and election in networks utilizing stigmergy concepts. As the primary function of a network is to transport traffic between source and destination nodes with efficiency, it is important to examine the concept of stability in the context of network paths (Boutoba and Xiao, 2007). The idea of self-organization is implicit in society design; furthermore software agents are designed to operate in large, distributed environments (Thomas, 2007).

Cognitive radio[4] (CR) today includes a relatively wide range of technologies for making wireless systems computationally intelligent. This has resulted from an interdisciplinary integration of complementary but somewhat isolated technologies: perception, planning and machine learning technologies from artificial intelligence on the one hand, and on the other hand software radio technologies that had come to include self-description in the extensible markup language, XML (Mitola, 2007).

Killiazovich and Granelli (2010) cited the similarities of cognitive networks to cognitive radio. According to the authors, cognitive radio provides efficient and dynamic spectrum access by adaptively changing transmission and reception parameters to avoid interference with other communication systems. In contrast, cognitive radios operate locally at the radio link level whereas cognitive networks perform with an end-to-end optimization. Thomas (2007) cited Mitola (2007) views of cognitive radio which according to him was built upon the foundation of system defined radio (SDR) and defined as “a radio that employs model-based reasoning to achieve a specified level of competence in radio-related domains”. Another citation used by Thomas (2007) was Haykin’s (2005) who defined cognitive radio as an “intelligent wireless communication system that is aware of its surrounding environment (i.e., outside world), and uses the methodology of understanding by building to learn from the environment and adapt its internal states to statistical variations in the incoming radio frequency stimuli by making corresponding changes in certain developing parameters (e.g., transmit power, carrier frequency, and modulation strategy) in real time, with two primary objectives in mind: highly reliable communications whenever and wherever needed and efficient utilization of the radio spectrum”.  Cognitive radios offer the requisite agility in employing dynamic spectrum selection by rapidly adapting itself to changes in spectral usage with interoperability among multiple interface standards (Pitchaimani et al. 2007). Emerging cognitive radio technology has been identified as a high impact disruptive technology innovation, that could provide solutions to the “radio traffic jam” problem and provide a path to scaling wireless systems for the next 25 years (Steenkiste et al. 2009).

Thomas (2007) also discussed that cognitive radio’s knowledge are comprised within a modelling language like Radio Knowledge Representation Language (RKRL). RKRL is defined as an instantiation of Knowledge Query Markup Language (KQML) which is a language of interaction created for communication between software agents. KQML is said to be based on Standard Generalized Markup Languange (SGML) which enables expression and exchange of information between intelligent agents and other entities. KQML is also intended to make and react to information requirements and identification of qualified agents.

According to Mitola (2007), RKRL is composed of components including the mappings between the real world and the various models formed by the cognitive process, a syntax defining the statements of the language, models of time, spaces, entities and communications among entities such as people, places  and things, an initial set of knowledge including initial representation sets, definitions, conceptual models, and radio domain models, and lastly, mechanisms for modifying and extending RKRL. Cognitive networking is different from cognitive radios or cognitive radio networking in that the latter two typically apply cognition only at the PHY layer to dynamically detect and use spectrum holes, and focus strictly on dynamic spectrum access (Quer, 2011).

The main motivation behind cognitive radios has been to increase spectrum utilization by allowing the unlicensed (secondary) users to opportunistically access the frequency band actually owned by the licensed (primary) user. In contrast to other network security architectures, in cognitive radios networks, the users are categorized into two distinct classes: primary users and secondary users.

Figure 3: Cognitive wireless network with multiple network-layer overlays (Pitchaimani et al. 2007)
Cognitive wireless network with multiple network-layer overlays (Pitchaimani et al. 2007)
Mitola added that awareness stops short of perception and it is required for adaptation. However, awareness does not give guarantee that adaptation would be performed. One example the author has prepared is “embedding a GPS receiver into a cell phone makes the phone more location-aware, but unless the value of the current location is actually used by the phone to do something that is location-dependent, the phone is not location-adaptive, only location aware. These functions are a subset of the CRA that enable adaptation”.  Perception functions of radio enables the identification of track knows, unknowns and backgrounds in an available sensor domain. The author defined backgrounds as subsets of a sensory domain that share common features that entail no particular relevance to the functions of the radio. The sensory functions of radio is described as hardware and/or software capabilities that allows a radio to measure sensory domain features. These sensory domains are anything that can be sensed including audio, video, vibration, temperature, time, power, fuel level, ambient light level, sun angle, barometric pressure, smell etc. According to Mitola (2007), sensory domains for fixed infrastructure could include weather features like ultraviolet sunlight, wind direction and speed, humidity, traffic flow rate or rain rate. All of these roles are under the CRA which allows perception.

The cognitive cycle discussed by Mitola (2007) stated with a stimuli entering the cognitive radio. This interrupts the sensory and results to removal of the cognition cycle creating a response. The cognitive radio’s flow of control may possibly move in the cycle from observing the action in a single processor inference system. Whereas, in a multi-processor system, there are parallel and complex temporal structures of sensing, pre-processing, reasoning and acting. CNs represent the evolution of the CR concept that has swept the radio communications field by storm. Although the concept of CNs may seem an extension of CRs, it should be noted that most CR research has focused on how changes to the physical layer affect the radio (Thomas, 2007). Cognitive radios offer the promise of being a disruptive technology innovation that will enable the future wireless world. Cognitive radios are fully programmable wireless devices that can sense their environment and dynamically adapt their transmission waveform, channel access method, spectrum use, and networking protocols as needed for good network and application performance (Steenkiste et al. 2009).

The field of machine learning attempts to characterize how such changes can occur by designing, implementing, running, and analyzing algorithms that can be run on computers (Dietterich & Langley, 2003). The discipline adapted by cognitive networks are based on ideals from various fields such as cognitive psychology, information theory, logic, complexity theory, statistics and operation research. All of these disciplines are intended to achieve the cognitive networks’ goal to understand and to acquire learning for a better and more concrete reasoning in the future.

Mitola et al (2007) stated that the cognitive cycle consisted of six processes including observation, orientation, planning, learning, decision making, and action. Likewise, Kiliazovich and Granelli (2010) discussed that the fundamental techniques that enable cognitive properites of networking algorithms are summarized into functions such as observation, analysis, decision making and action. Functions including analysis and decision making receive feedback depending on the observation to give action commands to the action elements. Cognitive networks should use observations of network performance as input to a decision-making process and then provide output in the form of a set of actions that can be implemented in the modifiable elements of the networks. Ideally, a cognitive network should be forward-looking, rather than reactive, and attempt to adjust to problems before they occur (Friend, 2009).

Cognitive networks are intended to create dynamic wireless networks with no topology set in advance and to launch dependable communications using large scale wireless networks. A typical cognitive network has the following four major phases of operation: Observe, Learn, Plan and Decide, and Act. The Observe phase focuses on sampling the network protocol parameters as a function of time (Quer, 2011). Learning is considered as the core of the cognitive system. It gives the functional elements the capability to execute various actions depending on past experiences. Learning is a function of perception, observations, decisions and actions. Initial learning is mediated by the Observe-phase perception hierarchy in which all sensory perceptions are continuously matched against all prior stimuli to continually count occurrences and to remember time since last occurrence of the stimuli from primitives to aggregates (Mitola, 2007).  The learning process includes an important portion which is the feedback adaptation loop. This loop enables the cognitive system to examine the reactions towards an action performed by doing observations, analysis and decision making. Upon the performance of these actions, a conclusion would be yielded which will be stored in the system for future references.

The reasoning and learning mechanism distinguishes the cognitive process from the adaptive process. In the cognitive routing algorithm, the reasoning and learning entity evaluates the execution results of routing policy and then amends the policy selection algorithm and routing policy itself (Hongyan et al. 2010).

Thomas (2007) used the definition of Thathachar (1999) of machine learning which is any algorithm that improves its performance through experience gained over a period of time without complete information about the environment in which it operates. This definitions gives the cognitive network a comprehensive possibility of probable mechanisms used in learning. Aside from learning, reasoning is one of the characteristic of a cognitive network. Thomas (2007) used the definition of reasoning provided by Papadimitriou (2001) which is the immediate decision process, using historical and current knowledge, which chooses the set of actions for the network. Whereas learning is considered as a long-term process of gaining knowledge from former actions resulting to improvement of future reasoning. According to Thomas (2007), the actual reasoning process may have various forms like pure or hybrid versions but any reasoning method should have the capacity to give solutions faster than the changes on network state.

Figure 4: Relationship between learning, performance, knowledge, and the environment (Dietterich & Langley, 2003)

Dietterich and Langley (2003), discussed the three broad formulations of machine learning. According to the authors, the most common formulation focuses on learning knowledge for the performance task of classification or regression. Classification comprises the test case assignment to one of a finite set of classes while regression involves the prediction the value of cases on some continuous variable or attribute. This formulation enables a description of specific situation in terms of node connections and whether numeric attributes at one node are higher than those at an adjacent node. Another under this formulation are unsupervised learning which anticipates training cases without related classes or values for the attribute to be predicted given to the learner whereas supervised learning is described otherwise. A formulation called semi-supervised learning according to Blum and Mitchell (1998), this formulation falls between the two approaches which are the supervised and unsupervised learning. According to the authors, some of the training instances come with associated classes or value for predicted attributes while typically or majority of the classes do not possess information in this framework.

The second broad formulation according to Dietterich and Langley (2003) addresses learning of knowledge for selecting actions or plans for an agent to carry out in the world. Approaches under this formulation can be used form problem solving, planning and scheduling. These approaches are also utilized in making cognitive choices for future actions instead of instant actions in the environment. The last broad formulation is concentrated on learning knowledge that enables a user to interpret and comprehend situations or events. To sum up, these formulations recommend various approaches to the diagnostic task during the process of learning to support the task. These are also used as criteria for evaluation of the success of the learning component.

According to Thomas, complexity, wireless networking and Quality of Service (QoS) are the problems from which the requirements and motivations for cognitive networks were born. Thomas (2007) defined complexity as a term used to describe many large, disordered systems of interactions; wireless networking as a rapidly growing area of networking that exhibits many of the features of complexity and; QoS[5] as the original motivation for end to end network control. Network complexity involves the functions of the number of nodes, communication medium, protocols and alternate routes working in particular network. The relation of the wireless nodes to the network complexity is its capacity to change the notion of connectivity. . In cognitive networks, all nodes track the spatial, temporal, and spectral dynamics of their own behavior, as well as of the environment. The information gathered is used to learn, plan and act in a way that meets network or application Quality of Service (QoS) requirements (Quer, 2011). QoS has seen a dilution in meaning, with many previously unrelated metrics being including in the definition. This dilution may be related to the fact that while networks operate as collections of connections, for the user (regardless of what application they are using) the only connection that matters is his end-to-end connection. This desire for an end-to-end viewpoint can be seen as desirable for most network (as opposed to network node) operations (Thomas, 2007). Cognitive networks are motivated by complexity. Particularly in wireless networks, there has been a trend towards increasingly complex, heterogeneous, and dynamic environments. While wired networks can also take on any of these characteristics and are not excluded from potential cognitive network applications, wireless networks are a natural focus of research in complex networks because of the significant additional complexity introduced by the wireless medium (Friend, 2009).

Dietterich and Langley (2003) also discussed the three capabilities that the vision of the Knowledge Plane (Clark, 2002; Partridge, 2003) assumes in terms of cognitive functionalities. These three capabilities are anomaly detection and fault diagnosis, responding to intruder worms, and rapid configuration of networks. Anomaly and fault diagnosis answers to the detection of unusual and undesirable behaviors, diagnosis of faults, isolation of sources, and repair of problem. The first activity called anomaly detection is the capability to recognize the unusual and undesirable occurrence in a particular network. An anomaly is defined as a low probability state of network (Dietterich and Langley, 2003). Fault isolation entails the identification of locus of an anomaly or fault within a network. Whereas, diagnosis is the process of making some conclusions about the cause of the anomalous behaviour. This activity may constitute the detection of old problems that had existed before or a new recognized problem. It is important to know that once diagnosis is performed, repair should follow. Responding to intruders and worms may be performed by prevention tasks which enable the manager to minimize the possibility of future intrusions by performing security audits. These security audits involve testing of the computers’ weaknesses, vulnerabilities and exposures to intrusions. Next method of responding to intruders and worms is the detection task, wherein the detection of intrusion has already been established. This activity aims to reduce the window of permeability. Lastly, the response and recovery tasks is done when a diagnostic intrusion is available. This activity aims to reduce the effect of the intrusions on the operations by trying to narrow the window of compromisibility. As cognitive networks rely on knowledge acquisition and exchange between network nodes to ensure proper function, it is likely that the knowledge management processes will be the target of attacks (Debar, 2007).

The adaptation of criteria for effective Cognitive Specification Language (CSL) was performed by Thomas (2007). The author stated that the criteria that were adapted are just measured of successful CSL and not requirements. The criteria included expressiveness, cognitive process independence, interface independence and extensibility. Expressiveness is a characteristic of a CSL which enables it to specify a wide variety of end-to-end goals. These expressions include constraints, goals, priorities and behaviors towards a cognitive element in a specific process. In addition, the author said that expressiveness of a CSL should enable it to express new goals without the need of a new language. Cognitive process independence makes the CSL independent from the cognitive process architecture and functionality’s dictations. This characteristic allows the system to be reused with slight alterations. Interface independence enables the user to be presented as an abstract interface regardless of the distribution or centralization in operation of the cognitive process. Just like the cognitive process independence, this characteristic results to reusability of the system. Lastly is the extensibility which enables CSL to adapt to new network elements, applications and goals. The overall goal of any technology is that it meet some need in the best way possible for the least cost. With the rest half of this goal in mind, a cognitive network should provide, over an extended period of time, better end-to-end performance than a non-cognitive network (Friend, 2009).

Kiliazovich and Granelli (2010) stated that there are two main driving forces for cognitive networks. These are technological and business oriented. The authors mentioned that from the technological perspective, cognitive networking is seen as a logical evolution towards the definition of a united QoS aware environment while covering multiple technologies currently present in the wireless network domain. The perspective of the business side on cognitive networks is that it will increase profits for wireless service providers through reduction of cost and development of new revenue streams by having the heterogenous wireless access solutions. According to the authors, the benefits of cognitive networks on businesses are the possibility to rely on common hardware and software platforms while simultaneously supporting the evolution of radio technologies, development of new services, minimization of infrastructure upgrades, accelerated innovation and maximization of return-on-investment through the reuse of currently available network equipment. Efficient service preparations and implementations can also be achieved with the use of cognitive networks by continuously analysing the configuration and performance of a network segment.

The perspective of business on cognitive networks relates to Strassner’s (2007) statements on the role of autonomic networking in cognitive networks. The author stated that the network equipment vendors, service providers and network operators responded to the desire of consumers to have more for less by creating and manufacturing smarter devices packed with features.

Strassner (2007) included at least five sources of complexity concerning the autonomic networking in cognitive networks.  First complexity is the separation of specific information relating to both business and technology followed by inherently different network management data which cannot be harmonized. Third complexity is the lacking of common design philosophy used by all components which results to the inability to cope with new functionality and new technologies. Fourth complexity is the isolation of common management data into separate repositories, and lastly, the inability of the user to respond to environmental changes over the course of the system lifecycle.

According to Strassner (2007), autonomics is the first and foremost a way to manage complexity. It is said that the networking analogy of autonomics are the manual which can be compared to the functions of the body, the time consuming tasks executed by the network administrator, planning and optimization of the network which are comparable to the complex cognitive functions of the brains. In general, these autonomics are created to perform tasks that are time consuming to make humans more resourceful. However, this does not mean that autonomics can replace humans. Complexity, particularly in wireless networks, is a problem that cannot be solved or understood easily using local and reactive networking protocols. The layered approach to the network stack at times prevents the network from achieving end-to-end goals. Although other networking concepts have addressed some of these shortcomings, none have addressed all of them (Thomas, 2007).

While it is tempting to convert the most human-like cognitive architectures to cognitive network architectures, we must not forget that the purpose of the cognitive network is to exchange data between users and applications within the network. Therefore, it may be appropriate to simplify a human-like cognitive architecture to remove the elements that are superfluous to the achievement of that purpose (Thomas, 2007). Kliasovich and Granelli (2010) mentioned the different researches on cognitive networks. First project that the authors cited was the E2R (End-to-end Reconfigurability) by Bourse et al. (2003) which benefit from the Next Generation Network (NGN) and utilizes a wide range of network technologies including cellular, fixed and WLAN. Next is the m@ANGEL platform by Demestichas et al. (2006) which brings in a special method of solution for mobility problems in heterogenous network environments with the assistance of cognition. m@ANGEL platform is said to have two planes which are the infrastructure plane comprised of reconfigurable elements like hardware transceivers, base stations and network core and the management plane is composed of the m@ANGEL entities. The functions of these entities are for monitoring, resource brokerage, goals management and reconfigurable element control. CogNet (Cognitive Complete Knowledge Network) is a research that proposes new cognitive network architecture for the maintenance of layered abstraction of TCP/IP protocols stack. CogNet’s exceptional propose architecture is that the cognitive functions executed in intra-layer cognitive elements are disseminated between various protocol layers. Whereas the proposal of Thomas et al. (2006) comprised of three horizontal layers wherein the top layer represents the elements responsible for specification and interpretation of user/application requirements into goals that can be understood by cognitive process. Gelenbe et al. (1999) suggested a cognitive packet networks which moves routing and flow control capabilities from network nodes into packets.

Communication networks have traditionally been engineered following the principle of protocol layering. This means designing specific network functionalities (such as flow control, routing, medium access) in isolation from each other, and putting together the complete system through limited interfaces between the layers performing these specific tasks (Srivastava & Motani, 2007). The CN must support trade-offs between multiple goals, whereas cross-layer designs typically perform single objective optimizations. Cross-layer designs perform independent optimizations that may not account for the end-to-end performance goals. Trying to achieve each goal independently is likely to be sub-optimal, and as the number of cross-layer designs within a node grows, conflicts between the independent adaptations may lead to adaptation loops (Thomas, 2007). Cross layer design proposal’s principles are concentrated on rapid prototyping, transferability and efficient application of cross-layer entities while maintaining TCP/IP modularity. The following are several cross layer signalling architectures have been proposed under this framework as discussed by Kliazovich and Granelli (2010): Interlayer signalling pipe which is one of the first approaches utilized for the implementation of cross-layer signalling. This architecture enables the propagation of signalling messages layer-to-layer along with the packet data flow. However, the disadvantage of interlayer signalling pipe is the limited propagation of the signalling information to the course of the packet flow which results to inappropriate cross-layer schemes which require instant signalling. Next architecture is the Direct interlayer communication which is another signalling method proposed by (Wang & Abu-Rgheff, 2003) which aims to improve the interlayer signalling pipe method by introducing signalling shortcuts which are performed beyond control. In this way, the proposed Cross-Layer Signaling Shortcuts (CLASS) approach allows non-neighboring layers of the protocol stack to exchange messages, skipping processing at every adjacent layer (Kliazovich & Granelli, 2010).  Central cross layer plane is said to be the most widely proposed cross layer signaling architecture as it is executed in parallel to the protocol stack. In the study of Quer (2011), it was stated that adapted cognitive networking architecture does not adopt a fully cross-layer approach, but keeps the layered structure of the network intact. At the same time, however, it uses some cross-layer principles, as the core of the cognitive architecture is the BN that represents the relationships among network parameters belonging to different layers.

Figure 5: Architectures and Cross-Layer Design for Cognitive Networks (Thomas, 2007)
Architectures and Cross-Layer Design for Cognitive Networks (Thomas, 2007)
Bayesian networks (BNs) are methodologies comprised of reasoning under uncertainty. The uncertainty may be a consequence of limited observations, noisy observations, unobservable conditions or unsure states or uncertain relationships concerning inputs, states and outputs found in a system. All of these causes for uncertainty are common in communications networks. In particular, the ability of cognitive networks to potentially control parameters at different layers in the protocol stack gives rise to concern over interactions between different protocol layers, interactions that are currently not well understood.




This work is written to be an early insight of the new emerging technology called Cognitive Network and whole new technology field Cognitive Networking. Today in the world we have some examples of the implementation of cognitive networking but only in the form of Wireless sensor networking. In these technologies each network node has the ability to modify his state based on the intern calculations that are based on the information from other sensors in surroundings. The word cognitive is simply taken to represent this ability to make decision about state change by itself.

The Cognitive Networks are more than that in every way but Wireless sensor networks are the simplest cognitive network system and the only one in use today. In the future we will have much more complex network nodes that will be able to make our network run faster, safer and in the same time have the ability to adapt and change the complete way of designing network infrastructure. To be completely honest, maybe there will not be nodes at all, in the way we are used to se the network nodes today.

This is what is this document all about. Its about giving a summary of all the best Cognitive Network ideas in the worlds biggest networking minds with a bit of my imagination on the top of them.

Throughput-demanding services in the future will eat the whole internet network alive. The technology use today to make the communication across the network possible is simply not enough to make computers from the future connect and communicate in right way. Computers in the future and most likely beginning to be simple consoles with big bright screens and all the computational part of the applications are on the server side. Welcome to the CLOUD. The buzzword in the IT world that will make todays network unusable in couple of years. All the applications will need more and more throughput and todays networking technology is not able to expand the speed on the network much more without serious change in the networking technology and design.

In this situation we are pointing our view to Cognitive Networking which will make our network go faster, be smarter and change we do networking.



  1. Qusay H. Mahmoud (2007.), Cognitive networks: towards self-aware networks ISBN 978-0-470-06196-1
  2. Beasley, Jeffrey S. (2009.), Networking, Second Edition ISBN-13: 978-0-13-135838-6 (hardcover w/cd)
  3. Ryan W. Thomas (2007.), Cognitive Networks
  4. Rajiv Ramaswami, Kumar N. Sivarajan, Galen H. Sasaki (2007.), Optical Networks: A Practical PerspectiveISBN: 978-0-12-374092-2
  5. Erol Gelenbe, Ricardo Lent, Alfonso Montuori , and Zhiguang Xu School of Electrical Engineering and Computer Science University of Central Florida
Orlando, Cognitive Packet Networks: QoS and Performance
  6. A. Manzalini, P.H. Deussen, S. Nechifor, M. Mamei, R. Minerva, C. Moiso, A. Salden, T. Wauters and F. Zambonelli, Self-optimized Cognitive Network of Networks

List of Images

  1. Figure 1: Spectrum pooling idea (Mahamuni et al. 2010)
  2. Figure 2: Cognitive radio system
  3. Figure 3: Cognitive wireless network with multiple network-layer overlays (Pitchaimani et al. 2007)
  4. Figure 4: Relationship between learning, performance, knowledge, and the environment (Dietterich & Langley, 2003)
  5. Figure 5: Architectures and Cross-Layer Design for Cognitive Networks (Thomas, 2007).

[1] Qusay H. Mahmoud (2007.), Cognitive networks: towards self-aware networks

[2] Qusay H. Mahmoud (2007.), Cognitive networks: towards self-aware networks

[3] Qusay H. Mahmoud (2007.), Cognitive networks: towards self-aware networks

[4] Ryan W. Thomas (2007.), Cognitive Networks

[5] Erol Gelenbe, Ricardo Lent, Alfonso Montuori , and Zhiguang Xu School of Electrical Engineering and Computer Science University of Central Florida
Orlando, Cognitive Packet Networks: QoS and Performance

UPDATE on 14 June 2016:
Few smaller changes to make the work more legible were made. Article was added to new Scientific & Academic category where all my science work will be published.

One Response

  1. Keerthiraj December 13, 2014

Leave a Reply