Machine learning (ML) has been suggested as a promising tool in the design of telecommunication systems to overcome the challenges of traditional methods and meet the requirements regarding future cellular networks. As an enabler of 5G communications, Non-orthogonal multiple access (NOMA) has emerged to adaptively manage the existing resources and received huge attention in terms of performance analysis to confirm its efficiency in resource management. This work addresses the applications of nonlinear symbolic regression (NLSR) techniques to evolve mapping models for bit error rate (BER) prediction in the downlink NOMA systems with perfect and imperfect successive interference cancellation (SIC) receivers. In this regard, a multi-stage structure is proposed to evaluate the BER using the related inputs as well as the output of the previous stages for imperfect SIC signal decoding. Each step consists of an NLSR block with different inputs and an output, which inputs include the effective parameters and the output of the previous stages. Experimental results confirm that the NLSR techniques are able to establish modeling expressions with good prediction accuracy compared to other benchmark methods. In particular, these ML-based techniques overcome the limitation of the classical approaches in modeling a closed-form BER formula, especially for the NOMA systems with more than two users. In addition, this BER prediction tool can be utilized by the NOMA base station as new side information for more optimal resource management.

Machine Learning (ML) has been proposed as a powerful tool for designing future cellular networks to meet their requirements. In the realm of 6G communications, IRSNOMA has emerged as an adaptive resource management solution for its performance in resource management. This study explores the use of ML methods to develop prediction models for bit error rate (BER) in IRS-NOMA systems, particularly in impulsive noise environments. The performance of the derived models has been investigated and compared by performing various simulation tests, revealing the effectiveness of ML models in predicting BER for both near and far users in IRS-NOMA systems.

The user grouping has a significant impact on the performance of non-orthogonal multiple access (NOMA) systems. The present work is focused on downlink NOMA user grouping leveraging the type-2 fuzzy set (T2FS). The main drawback of conventional user grouping methods is low efficiency for middle users which degrades the overall system performance. To overcome this problem, a 2-step user grouping process is proposed that includes identifying group candidates and selecting the most qualified one. The first step is handled by introducing a group membership competency criterion based on T2FS modeling due to its capability to handle extra uncertainty in real-world phenomena. The second step involves adding the most qualified candidate to the desired group, given the additional power imposed on the group. The key contribution of the paper is twofold: (i) it affords a multi-stage structure to evaluate the competency of users to join a particular group relying on T2FS modeling. It allows groups of different sizes to be formed, depending on the network channel status and quality of experience (QoE) requirements, and (ii) the additional power imposed on the group is exploited as a measure to select the final group. In this regard, the analysis of interference that each user brings to others in the same group is taken into account. The performance of the proposed scheme is evaluated by simulations and compared with other methods. The obtained results indicate the efficiency of the proposed approach in terms of total power consumption and symbol error rate.

In this article, the problem of radio resource management is addressed by adjusting the tradeoff between two key quality of experience (QoE) influencing factors, namely transmission rate, and service price. To this end, a piecewise utility function is employed to assess the user QoE in three different quality classes. It comprises the combination of two utility functions corresponding to the transmission rate and the service price. A low-complexity fuzzy-based approach is then introduced to reallocate additional resources to dissatisfied users to increase the number of satisfied users. The simulation results indicate the efficiency and applicability of the novel approach in terms of increasing the service provider revenue while providing approximately the same overall QoE and power consumption. The results also show proposed approach achieves an improvement of at least 15% in total service provider revenue on average compared to other methods.

This article addresses linear precoder design for the peak‐to‐average power ratio (PAPR) reduction of the generalized frequency division multiplexing (GFDM) scheme. This design relies on minimizing various statistical parameters of the instantaneous power of the GFDM signal—including variance, second and third moments—using gradient‐based optimization methods. In this regard, a general framework of four different scenarios, which employs the steepest descent and conjugate‐gradient methods and utilizes two different approaches of fixed and dynamic step‐sizing, is proposed. In the case of fixed step‐sizing, the results demonstrate that the algorithm diverges in some scenarios and provides low‐speed convergence for the others due to its inability in meeting the terminating threshold regarding minimizing objective function to a desirable level. However, dynamic step‐sizing using the Wolf line search rule circumvents these drawbacks, outperforms the existing research, and converges in all scenarios to a precoder providing advantages in terms of design speed and achievable PAPR, while keeping the symbol error rate and out‐of‐band emissions at the ranges of the un‐precoded GFDM systems. Moreover, the results confirm the comprehensiveness of the proposed framework in adapting to various GFDM statistical parameters and gradient methods.

As a dramatic advancement in mobile communications, 5G has put together several state of the art technologies including Cognitive Radio (CR) and Quality-of-Experience (QoE). While CR is to overcome frequency scarcity, how to maintain QoE for all the connected users is one of the paramount issues in such integration, especially for multimedia communications. As a key technology for 5G, Non-Orthogonal-Multiple Access (NOMA) is combined with Orthogonal Frequency Division Multiplexing (OFDM) to improve the spectral efficiency. In this paper, the QoE requirements for typical applications in a CR platform are characterized, based on which a method for user grouping and power management in an OFDM–NOMA system is proposed. The performance of the proposed method is analyzed and evaluated by simulating a typical network. The evaluations show noticeable improvement in transmit power reduction while the requested users’ perception levels are maintained.

This paper addresses linear precoder design for Peak-to-Average Power Ratio (PARP) reduction of Generalized Frequency Division Multiplexing (GFDM). A general framework, which is composed of four different scenarios and utilizes Gradient-based iterative methods to reduce PAPR through minimizing statistical parameters of the instantaneous power of GFDM signal including variance, power, and third moment, is suggested. Numerical results confirm when the step-size of the Gradient method is dynamically computed using the Wolf line search rule, the suggested algorithm circumvents drawbacks of existing studies and converges to a precoder providing advantages in design speed, obtained PAPR, symbol error rate, and out-of-band emission.

Technology incubators, where new early-stage ventures accommodate in a supportive environment, are younger than 15 years of age in Iran. Nevertheless, it is necessary to localize the technology incubator models based on such parameters as culture, human resources, level of technology, and education system so as to meet an appropriate effectiveness. To achieve this goal, the present paper firstly introduces a three-stage incubation model considering special characteristics of the studied country. In this proposed model, the pre-incubation stage is the same as other currently used models but the incubation stage breaks down into two new stages namely technology incubation and technology development. The new model enhances market concentration and encourages incubator clients to finalize their products/services. This model has been successfully implemented in Kerman Technology Incubator and our experimental studies and evidences show the effectiveness of the proposed approach in improving the performance of the incubator. At the second phase, a machine learning evaluation model is developed with an aim to measure the incubator’s client performance. This model utilizes the advantages of classification algorithms for mapping the business success factors into quality of client level. Hence, different classification methods are applied and their performances have been compared together. Results show the efficiency of the developed model in terms of accuracy.

This paper introduces a multimodal emotion recognition system based on two different modalities, i.e., affective speech and facial expression. For affective speech, the common low-level descriptors including prosodic and spectral audio features (i.e., energy, zero crossing rate, MFCC, LPC, PLP and temporal derivatives) are extracted, whereas a novel visual feature extraction method is proposed in the case of facial expression. This method exploits the displacement of specific landmarks across consecutive frames of an utterance for feature extraction. To this end, the time series of temporal variations for each landmark is analyzed individually for extracting primary visual features, and then, the extracted features of all landmarks are concatenated for constructing the final feature vector. The analysis of displacement signal of landmarks is performed by the discrete wavelet transform which is a widely used mathematical transform in signal processing applications. In order to reduce the complexity of derived models and improve the efficiency, a variety of dimensionality-reduction schemes are applied. Furthermore, to exploit the advantages of multimodal emotion recognition systems, the feature-level fusion of the audio and the proposed visual features is examined. Results of experiments conducted on three SAVEE, RML and eNTERFACE05 databases show the efficiency of proposed visual feature extraction method in terms of performance criteria.

This paper presents an automatic method for finding optimal channels in Brain Computer Interfaces (BCIs). Detecting the effective channels in BCI systems is an important problem in reducing the complexity of these systems. In this research, Improved Binary Gravitation Search Algorithm (IBGSA) is used to automatically detect the effective electroencephalography (EEG) channels in left or right hand classification. To do this, at first, data is filtered with a bandpass filter in order to reduce the amount of different types of merged noise. Then, the electrooculography (ECG) and electromyography (EMG) artifacts are corrected based on Blind Source Separation (BSS) algorithm. Data is epoched according to the left or right hand motor imageries and central beta frequency band is isolated for Event Related Synchronization (ERS) analysis. Feature extraction process is carried out by analyzing EEG signals in time and wavelet domains. The logarithmic power of each channel is computed in time domain and the features of mean, mode, median, variance, and standard deviation are calculated in wavelet domain. IBGSA is employed to detect the optimal channels to achieve better classification results. Support Vector Machine (SVM) is used as the classifier. The maximum accuracy of 80% and average accuracy of 76.24% were obtained for eight subjects in BCI competition IV dataset. The results of this research confirm that automatically detecting effective channels can enhance the practical implementation of BCI based systems and reduce the complexity. (C) 2016 Elsevier Ltd. All rights reserved.

The main aim of this study is to develop a low-complexity non-intrusive quality prediction model in Voice over Internet Protocol (VoIP) systems. In order to gain this goal, a 2-level structure for predicting the quality of speech is proposed. Furthermore, the capabilities of multi-gene genetic programming are investigated through developing a number of parallel models and different feature vectors. These models are utilized in two hierarchical levels to construct the final model. To consider the transmission media and speech signal characteristics in quality measurement process, both network impairments and per-frame features are employed simultaneously for developing models. Several experiments are performed based on the proposed structure while different combinations of speech feature types in the cases of noise free and noisy speech signals are examined. The obtained results indicate that using parallel models in a 2-level structure enhances the accuracy of derived models as compared with 1-level structure and common single-gene GP models.

This paper presents a novel method for modeling the one-way quality prediction of VoIP, non-intrusively. Intrusive measures of voice quality suffer from common deficiency that is the need of reference signal for evaluating the quality of voice. Owing to this lack, a great deal of effort has been recently devoted for modeling voice quality prediction non-intrusively according to quality degradation parameters, while among the past proposed methods, intelligent techniques have been remarkably successful due to their abilities for modeling the non-linear processes. The present study introduces a procedure for developing fuzzy models, employing Genetic Algorithm (GA) and Adaptive Neuro-Fuzzy Inference System (ANFIS). The proposed method is able to generate optimized fuzzy models in terms of accuracy and complexity. The efficiency of this procedure is compared with and contrasted against 13 regression methods implemented in KEEL as one machine learning tool. Moreover, several experimental results are performed over voice data from 10 different languages. In order to complete the experiment, a comprehensive statistical comparison is also drawn between our proposed method and other previous ones. The results apparently show the efficiency and applicability of this novel method in terms of generating accurate and simple fuzzy models for estimating the VoIP quality.

In this study a novel process for measuring quality of speech is introduced which employs multiple-level decomposition of signal method. In this way, the Discrete Wavelet Transform (DWT) is used to decompose original speech signal into different frequency sub-band signals. Then the feature vectors are obtained by extracting MFCC features from each sub-band. In order to investigate the capabilities of ensemble learning methods, various ensemble regression models are studied and results are compared with individual models. Also, to prepare training and test dataset, a simulation environment is set up which distort speech signal by different speech impairments. At last, different experiments are performed to illustrate the efficiency of ensemble methods. Results demonstrate that using a group of base learners (ensemble model) improve the performance of models in comparison with single learner.

In real-time multi-media services, that uses internet infrastructure for transferring data traffics, the quality of service and consequently the level of user satisfaction are significant parameters. Our objective in this paper is to investigate the capability of Bayesian classifiers for estimating the quality of perceived voice in VoIP (Voice over IP) system. In this study, some quality parameters have been utilized to estimate the level of user satisfaction. The employed classifiers operate non-intrusively that means there is no need for original signal to estimate the quality of the perceived voice. For this purpose, a data set has been provided by simulation environment based on PESQ (that is an intrusive method that compares original and degraded signal for evaluating the quality of voice). Finally, we compare the performance of Bayesian classifiers with some other classification approaches in terms of estimating accuracy. For this purpose, the WEKA tool is used that contains implementation of many algorithms for classification problems. The results obtained, show the efficiency of Bayesian classifiers comparing to the other methods in terms of accuracy and computational time.

This paper presents a comparative study for modeling the quality of VOIP based on one of the most commonly used intrusive method for assessing voice quality called PESQ. Intrusive measures of voice quality against the non-intrusive methods need original speech signal to do a comparison with degraded signal and measure perceived voice quality. The need to have original signal will limit this method for real-time traffic monitoring. Owing to this weakness, some efforts have been recently performed for modeling the voice quality according to speech and IP network parameters like packet loss, codec type, gender and language of talker. Among the past proposed methods, intelligent techniques such as neural networks have been very successful models. In this study our main intention is twofold: Firstly developing a nonintrusive Neuro-fuzzy model based on an intrusive method (PESQ) and secondly comparing the performance of Neuro-fuzzy model to other well-known intelligent modeling approaches. Several experimental results were done and reported for more illustration.