• No results found

anfis based data rate prediction for cognitive radio - ethesis


Academic year: 2023

Share "anfis based data rate prediction for cognitive radio - ethesis"


Loading.... (view fulltext now)

Full text

This is to confirm that the thesis titled "ANFIS BASED DATA RATE PREDICTION FOR COGNITIVE RADIO" submitted by Shrishailayya M Hiremath, Roll No. Here, ANFIS and neural network methods can help a cognitive radio system select the best radio configuration to operate in.



The idea of ​​cognitive radio goes beyond making productive use of unused parts of the spectrum and being able to make human-like decisions to transmit without obstruction. This definition also allows a spectrum band to be used for operation in different RATs in accordance with the flexible spectrum management concept [4]. Cognitive radio involves different stages which are explained in later chapters.


With the aim of achieving the above capability, cognitive radio behaves reactively or proactively based on external environmental information, as well as on their goals, principles, skills, experience and knowledge. In this work, various neural network-based data rate predictions are tested based on previous work and an ANFIS model-based technique is proposed for cognitive radio-assisted data rate prediction.

Thesis Layout

3 . achievable data rate), taking into account recent observed information, as well as past experience and knowledge?" in terms of helping to predict the data rate a specific radio configuration could achieve if selected for use.


Brief History of CR

He described how a cognitive radio can increase the flexibility of personal wireless services through a new language called It resulted from many technologies coming together to result in Cognitive Radio technologies, due to the authenticity that exists among its applications.

Definitions of CR

The regulatory bodies which focus on the operation of transmitter like FCC defines cognitive radio as: “A radio that can change its transmitter parameters based on interaction with the environment in which it operates.” [9] [10]. SDR forum which is one of the highly associated with CR and SDR , that works on CR [10]application defines CR as “An adaptive, multi-dimensionally aware, autonomous radio (system) that learns from its experiences to reason, plan, and decide future actions to meet user needs.”.

CR Tasks

  • Radio-scene analysis
  • Channel-State Estimation
  • Predictive Modeling
  • Distributed Transmit-Power Control
  • Dynamic Spectrum Management

Throughout the spectrum sharing process, the transmission power controller keeps track of the bit load over the spectrum slots being used. To meet this requirement, the dynamic spectrum management algorithm must include a traffic model of the primary user occupying a black space.

Figure 2. 1 Basic cognitive cycle.
Figure 2. 1 Basic cognitive cycle.

Cooperation in Cognitive Radio

  • Virtual Capacity
  • Power Consumption
  • Cost
  • Reliability

In contrast to the conventional single-link communication, the use of cooperative multi-link communication offers higher reliability. Moreover, the probability of losing all the links in a multilink communication is extremely low compared to the probability of losing a single link.

Emergent Behaviors of Cognitive Radio

This is caused by the fact that the effect of the loss of a link from the multilink group is much smaller than the total loss in the case of the loss of a link in the case of single-link communication.

Emerging Cognitive Radio Standards and Deployments

  • IEEE 802.22
  • IEEE 802.11h

Orientation – Based on these observations, the WLAN must determine whether it is operating in the presence of a radar installation, on a bad channel, in the satellite band, or in the presence of other WLANs. Decision – Based on the situation the WLAN is facing, the WLAN must decide to change the operating frequency (Dynamic Frequency Selection), adjust the transmit power (Transmit Power Control), or both.

Cognitive Radio Applications and Drawbacks

Reviewing most of the definitions from earlier, learning or “recalling and relating past actions, environments, and performance” alone is not required as part of the standard. However, if we shift the requirements of the standard to the expected implementations, it seems reasonable that many vendors will include and make use of some memory of past observations (useful for detecting outliers), meaning that both definitions of cognitive radio will be met.

Important Institution and forums working on CR

This work presents a learning mechanism introduced in the channel estimation phase and a prediction module to predict the data rate of a given radio configuration, which was previously implemented in [4]. In this chapter, the first section discusses the need for a learning mechanism for CR, the second section gives an overview of different neural networks used for prediction, the fourth section discusses the motivation for data rate prediction using NN, the fifth section gives the algorithm used for data rate prediction rates in the basic scheme and the simulation results for the same, the sixth section does not use the data rate prediction algorithm for the extended case and the simulation results, and the last section gives the conclusion.

Need For Learning Mechanism

Channel evaluation and predictive modeling: during which the capabilities of configurations are discovered (discovery process) and evaluated accordingly based on the measurements of the previous phase; in addition, past experience and knowledge can be used at this stage. Problem statement: Given a candidate radio configuration, what are its expected capabilities (e.g. in terms of achievable data rate), taking into account recent sensed information and past experience and knowledge?” The main objective is to examine two learning schemes, “basic” and "extended", which are based on neural networks and are designed to improve the learning capabilities of the cognitive terminal in terms of helping to predict the data rate that a particular radio configuration could achieve if chosen for operation, and finally providing a neural network benchmark. These neural network schemes are used to compare the final thesis contribution, which is ANFIS-based learning.

Figure 3. 1 Simplified representation of cognitive radio cycle.
Figure 3. 1 Simplified representation of cognitive radio cycle.

Overview of Different Neural Nets Used for Data Rate Prediction

  • Definition of Neural Network
  • Architecture of neuron model
  • Neural Networks Architecture
    • Feed forward networks architecture and operation
    • Recurrent Neural Networks (RNN)

If P1 is the number of neurons in the first hidden layer, then each element of the output vector of the first hidden layer can be computed as , . Elman networks are two-layer back-propagation networks, with the addition of a feedback connection from the output of the hidden layer to its input. The output of Elman's network is determined by a set of output weights, V, and is calculated as, .

Figure 3. 3 Neuron model.
Figure 3. 3 Neuron model.

Motivation for Data rate prediction using Neural Network

What is gained is that by associating each configuration with a predictable, feasible data rate, NN-based learning schemes can facilitate the cognitive terminal to make its decision regarding the configuration in which it should operate, and select the best among a set of candidates. choose. Consequently, two neural network-based learning schemes were set up and tested: the "basic" and the "extended" one. In both cases, several types of NNs with a considerable number of adjustable parameters were investigated.

Basic NN-based Data Rate Prediction

  • Preparation procedure
  • NN Pattern selection for Basic Scheme
  • Simulation Results and Discussion

Weights βmj actually represent the number of occurrences of each of the reference values ​​mj in M ​​within the time window. A "Validation Set" (invisible data) which is used to measure the performance of the network by holding its parameters constant. Moreover, the number of hidden layers and/or neurons plays a crucial role in the learning process and has a great influence on the performance of the network.

Figure 3. 7 Neural network for the basic learning scheme.
Figure 3. 7 Neural network for the basic learning scheme.

Extended NN -based Data Rate Prediction

  • Preparation procedure
  • Results and discussion

As shown in table .3, for the first time zone, the third time zone and the fourth zone, respectively, the best results are found with 10 hidden layers, while for the second zone, 15 hidden neurons are required. The Levenberg-Marquardt optimization function was used for training. Also, the same 1000 sample data points as in the previous test set were used for training. Due to the complexity of the problem (multiple time zones), a network with two hidden layers works better.

Figure 3. 12 Weight values per time slot for the extended scheme.
Figure 3. 12 Weight values per time slot for the extended scheme.



Basics of Fuzzy Modeling

Fuzzy if-then rules or fuzzy conditional statements [20] are expressions of the form IF A THEN B, where A and B are labels of fuzzy sets characterized by appropriate membership functions. Both types of fuzzy if-then rules have been used extensively in both modeling and control. Fuzzy if-then rules form a core part of the fuzzy inference system, which is discussed in the next section.

Figure 4. 1 Fuzzy Inference System.
Figure 4. 1 Fuzzy Inference System.


The firepowers w1 and w2 are usually obtained as the product of the membership degrees in the premise part, and the output f is the weighted average of each line's output. The output of each node in this layer is simply the product of the normalized firepower and a first-order polynomial (for a first-order Sugeno model). From the ANFIS architecture in Figure 4.3 it is observed that given the values ​​of starting parameters, the overall output f can be expressed as linear combinations of the resulting parameters:.

Figure 4. 2 First order sugeno model
Figure 4. 2 First order sugeno model

ANFIS Learning Method

The basic learning rule of ANFIS is the backpropagation gradient descent [23], which computes error signals (the derivative of the squared error with respect to each node's output) recursively from the output layer backward to the input nodes. Specifically, the form of membership functions in the "If" part of the rules is determined by a limited number of parameters. These parameters are called prerequisite parameters, whereas the parameters in the "THEN" part of the rules are referred to as consequent parameters.

Membership Function and Rules Selection for ANFIS

The initial values ​​of premise parameters are set so that the centers of the MFs are equidistant from each other along the range of each input variable. The grid partitioning approach of fuzzy systems has the serious drawback that the highly regular partitioning of the input space may not be able to produce an acceptable size rule set capable of properly handling a given data set. This process is repeated until all data is within beam distance of a cluster center.

Figure 4. 4 Grid partitioning of input space for two input sugeno fuzzy model with nine rules
Figure 4. 4 Grid partitioning of input space for two input sugeno fuzzy model with nine rules

ANFIS based data rate prediction: Basic Scheme

  • Preparation procedure
  • Simulation results and discussion

The conventional ANFIS membership function before training and after training is presented in Figures 4.6 and 4.7. RMSE plot for training and validation case is shown in Figure 4.8, while Figure 4.9 and 4.10 show prediction accuracy in the case of training and validation case. Figure.4.9 and Figure 4.10 depict that prediction accuracy of conventional ANFIS is 91% during training and 89% in validation. The results for optimum cluster size are presented with the best prediction accuracy and RMSE in the case of training and validation.

Figure 4. 6 Memberships plot for each input before training.
Figure 4. 6 Memberships plot for each input before training.

ANFIS based data rate prediction: Extended Scheme

  • Preparation procedure
  • Simulation results and discussion

And Figures 4.22 to Figure 4.26 show detailed graphs of MF before and after training, error curves, and prediction accuracy. It can be seen that the prediction accuracy is very high compared to the neural network method. The table shows that ANFIS has better performance in terms of prediction accuracy and RMSE error.

Figure 4. 21 ANFIS model for the extended learning scheme
Figure 4. 21 ANFIS model for the extended learning scheme





Scope for Future work

10] James O'Daniell Neel, "Analysis and Design of Cognitive Radio Networks and Distributed Radio Resource Management Algorithms," Virginia Polytechnic Institute og State University, doktorafhandling 6. september 2006. 11] Mohamed Gafar Ahmed Elnourani, "Cognitive Radio And Game Theory: Overview And Simulation," Blekinge Institute of Technology, Master of Science-afhandling 2008. Shellhammer, Winston Caldwell, "IEEE 802.22: The First Cognitive Radio Wireless Regional Area Network Standard," IEEE Standards In Communications, pp.

MSE curve for –extended scheme


Figure 3. 1 Simplified representation of cognitive radio cycle.
Figure 3. 2. Cognitive radio engine.
Figure 3. 3 Neuron model.
Figure 3. 4 MLP Architecture.


Related documents

TELANGANA STATE ELECTRICITY REGULATORY COMMISSION #11-4-660, 5th Floor, Singareni Bhavan, Red Hills, Khairatabad Hyderabad 500 004 PUBLIC NOTICE In the matter of passing