Sparse network coding was introduced to reduce the computational complexity of the Random linear network coding. In this method, most of the decoding matrix coefficients are zero. Partial decoding means the possibility of decoding a part of the raw packets is one of the capabilities of the sparse network coding method. We introduce three different models of sparse coding method as an approach to reduce decoding latency in real-time communication. More precisely, we first evaluate a sparse network coding for a no feedback configuration in terms of the performance of the total number of transmissions required, and the average packet decoding delay for a generation of raw packets, by introducing a Markov chain-based model. Then we evaluate the accuracy of the proposed model using extensive simulation and show that the proposed model can accurately estimate the number of required transmissions and decoding delay for a generation of packets. The results also evaluate the accuracy of the model in the erasure channel. In the following, we introduce the feedback-based model and we show that this model can create a better balance between the functions of the number of transmissions and the average decoding delay per packet. Finally, by focusing on the problem of finding the Random spanning tree, we present a graph-based model for analyzing sparse network coding and show that although the proposed model is valid only for grade 2 sparsity, it also has the capacity to develop for lower sparsity.

Accurate travel time prediction is one of the important issues in the field of traffic and transportation that can significantly affect the daily life of people and organizations. In this research, four different machine learning methods including linear regression, multivariate regression, Random forest and deep artificial neural network were trained to predict travel time. The purpose of this research is to predict travel time for use in intelligent traffic systems and to use and compare several new methods, including deep neural network and Random forest regression, as well as considering new parameters in the computations such as weather conditions, traffic flow, travel time, and accidents and the traffic locking points compared to other studies are the innovation and comprehensiveness of this study compared to other studies. In the design and implementation of this research, real traffic data taken from Google map was used and analyzed. This data includes information such as traffic conditions, season, time of day, weather conditions, and route characteristics. The results of this research show that the deep neural network (DNN) model with R2 equal to 0.833 has a very good performance among the investigated models. This model explains 0.833% of the variance of the data and the distribution of the residuals in it is relatively central with a mean of zero and a distribution close to normal. The linear regression model with R2 equal to 0.615 has a poorer performance than DNN and explains 0.615% of the data variance. But the Random regression model with R2 equal to 0.955 has one of the best performances in competition with DNN and explains 0.955% of the data variance. MSE and RMSE parameters were also used to evaluate the performance of the models, and as a result, a multidimensional comparison was made between the models, and the Random forest model resulted in the lowest error values. Since in the collected traffic data, traffic accidents and consequently traffic locking points are also used in the models, and considering that the Random forest model is more effectively adapted to the data despite the presence of noise and anomaly, the R2 value of this model is higher than R2 of Deep neural networks, due to the overfitting nature of Deep Learning methods.

We study the capacity of point-to-point erasure networks under a restricted form of network coding to which we refer as spatial network coding. In this form of coding, the nodes can not perform coding on successive packets which are received from one incoming link. The coding at a node is restricted to the packets received at the same time slot from different incoming links to the node. In other words, the temporal aspect of coding is absent. We prove that the capacity of a unicast session under spatial network coding is the statistical average of the minimum cut of the Random graph corresponding to the erasure network. Then, we consider a network with a complete, directed and acyclic graph in which the nodes are erased independently. We prove that the capacity of the network under spatial network coding is the same as its capacity under general network coding. This shows that temporal coding has no improving effect for this network. Finally, we compare the capacities under different coding schemes for the complete graph with the edges that are erased independently.

In recent years there has been a growing interest to employ non-linear predictive techniques and models in speech coding to further reduce bit-rate and therefore channel bandwidth. Usually neural nets are used for this purpose that result in an additional up to 3dB reduction in the excitation signal energy. Non-linear prediction can also be performed based on Volterra series expansion wherein the expansion is usually limited to first and second terms, for simplicity (quadratic prediction). Early studies have shown that employing Volterra filters results in a much higher reduction in excitation signal energy (6 to 10 dB), as compared with neural nets. But, because of instability, this reduction can not be materialized in terms of bit-rate reduction or signal to noise improvement. This instability in the decoder is triggered by computational errors (i.e. due to quantization of the excitation signal) and high sensitivity of algorithms to these errors.In the original work, presented here, the instability in the codec is studied in both forward and backward prediction schemes using LS and LMS algorithms respectively. It is shown that stability can be obtained at the cost of losing most of saving in excitation signal energy where final reduction level is as much as for neural nets. With forward prediction, after stabilizing, in spite of a small increasing in the operational complexity for 20 to 45% of frames including the quadratic term will be beneficial. So a scheme is developed to perform non-linear prediction only on these frames. This algorithm results in an improvement of up to 4 dB in final signal to noise ratio. Sequential backward quadrant prediction, although much more interesting from implementation point of view, does not lead to an appreciable better performance over linear prediction.

The purpose of this study is to determine the areas with groundwater potential using artificial neural network (ANN), Random forest (RF), support vector machine (SVM) and linear regression (GLM) models. In the present study, 14 parameters affecting groundwater potential including altitude, slope, slope direction, curvature, distance to stream and fault, stream and fault density, lithology, average rainfall, land use, topographic position index (TPI), relative slope position (RSP) and topographic wetness index (TWI) were used. From a total of 10, 624 springs, Randomly 70% as test data and 30% as validation data were classified. The RF model was also used to determine the most important parameters. Alignment test between parameters was performed using SPSS software. The Receiver operating characteristic was used to Predictive power of models and the Seed Cell Area Indexes (SCAI) was used to accurately distinguish between classes. The results showed that there is no alignment between the parameters. The results of RF model showed that the parameters of height, land use, slope, and distance from fault, TWI and lithology are the most important factors affecting groundwater potential, respectively. Also, based on the ROC curve in both training and validation, the ANN model had the highest accuracy and the RF, SVM and GLM models were in the next categories. Also, the results of the seed cell area index showed that all four models have separated the classes with appropriate accuracy. According to the ANN model, 31. 4% of the basin has high and very high groundwater potential.