NEURAL: A Self-organizing Routing Algorithm for Ad Hoc networks Vicente E. Mujica V., Dorgham Sisalem and Radu Popescu-Zeletin Fraunhofer FOKUS Institute. Berlin, Germany. {mujica,sisalem,zeletin}@fokus.fraunhofer.de Abstract This paper evaluates a self-organizing routing protocol for Ad Hoc network, called the NEUron Routing ALgorithm (NEURAL). NEURAL has been designed taking into ac- count the learning and self-organizing abilities of the brain. More precisely, it was inspired by the synapses process be- tween neurons, when a signal is propagated. Basically, the most significant characteristic of NEURAL is the uniform distribution of the information around the node’s location based on the current changes in its neighborhood. Using a 2-hop acknowledgment mechanism, local information is monitored in order to be used for route selection method, classification procedures and learning algorithms. 1 Introduction Inspired by the biological nervous system, Artificial Neural System (ANS) and neural networks are being ap- plied to study a wide variety of problems in the areas of engineering and business [7, 22, 18]. A neuron is the indi- vidual computational element that makes up most artificial neural system models [11]. A neuron presents three major parts called dendrites, the cell body and a single axon. Den- drites are nerve fibers that are connected to the cell body or soma, where the nucleus is located. Extending from the cell body is a single long fiber called the axon. At the ends of the neuron, terminal branches, synaptic junc- tions or synapses connect the axon with the dendrites of the following neurons [16]. In a ANS system, the information is propagated between neurons using electrical stimulation along dendrites. High stimulation signal produces an out- put to the other neighbor neurons and so the information takes the right way to the destination, where a reaction will occur. Otherwise, low stimulation signal will be blocked by the neurons and the information will be not transported. The synapse is defined as the communication process be- tween neurons, when a signal is propagated. This means that the information is forwarded using electrical stimula- tions along dendrites. The modification of synaptic weights provides the traditional method for the design of neural networks. Such an approach is the closest to linear adap- tive filter theory, which is already well established and suc- cessfully applied in such diverse fields as communications, control radar, sonar, seismology and biomedical engineer- ing [14]. The characteristics described above are desirable in the context of Ad Hoc networks. The first association is the synapse process between neurons as the capacity of a pro- cessing element to communicate with others (Routing). Thus, the amount of neighbors around a node can be repre- sented as a probabilistic weight value or “synaptic weight”. Nodes are free to move randomly and organize themselves arbitrarily. This means that the ad hoc network’s topol- ogy changes rapidly and unpredictably as well as synaptic weights are changed in the local environment. The advan- tage of Artificial Intelligent algorithms for control problem in complex systems is that the weights can be found by ex- amining the performance of a network as controller rather than by providing correct control signals for various input data [24]. The goal behind this paper is described a novel routing Algorithm called NEuron Routing ALgorithm (NEURAL), which it has been inspired by the synapses process. The design of NEURAL is based on three main phases, which apply some algorithms used in the area of neural networks. The Pre-processing phase involves a classification rule for Pattern Recognition called the K-Nearest Neighbor Rule. Afterwards, the Route Discovery phase considers an self- organizing algorithm based on the Kohonen model. And finally, the Learning phase employs a Trust and Reputation mechanism, which it has been integrated to the extension of the Kohonen model. The rest of the paper is organized as follows: Sec- tion 2 introduces a background about Classification, Self- organizing and Artificial Intelligent System theory. The NEURAL architecture was developed in section 3 using the performance of three main modules. The conjuction of these modules provide tools to accomplish the implementa- tion for a network simulator in section 4, and finally, con- clusions in section 5.