To achieve an eective communication relay mission, the position of the relays in a complex environment has a critical impact on the effectiveness of the wireless communication quality, because wireless communication varies significantly with a small amount of movement of mobile robots in such an environment. The indoor experiments are conducted to show that the proposed approach using mobile relays significantly improves the communication performance of the complex network with the accurate channel prediction.
Background and Motivation
Achieving suitable positions for relay nodes is still challenging due to the complex nature of the wireless signal in indoor/urban environments. Therefore, the exciting subject has challenged and motivated me, so that I ended up working with the subject throughout all the years of my studies.
Related Work
The artificial potential field generates a possible next position of the relays and then the LOS in the relay chain is checked. The relay movement towards a neighboring node results in the increase of the received signal quality from the node to the direction of movement and the decrease of the received signal quality from the other node.
Contribution of the Thesis
They used the distance-based model as well as the learning-based Wi-Fi propagation model using Gaussian Process (GP). The shortest path using the RRT (rapidly exploring random tree) method from the base station to the client was generated, and the relays were controlled along the generated path by moving towards one of the neighboring nodes whose transmission signal quality is weaker than the others' . .
Outline of the Thesis
The overview includes the scenarios we consider and the outline of the relay positioning algorithm.
Scenarios
Algorithm Overview
Network Topologies
An important advantage of the tree topology is the simplicity of adding additional nodes to the network. In other words, a small number of failed links is unlikely to lead to a network outage.
Wireless Radio Communication
5 found in the literature [1] shows the strength of the signals obtained with the three main dynamics. The constant represents how fast the signal strength decays over the distance between two nodes, and is usually within the range [1.5,5.5]. This remote decay model shows one of the key dynamics of wireless signal behavior, path loss.
The amount of signals attenuated due to shadowing depends on the size of the obstacles, and many experiments in the literature have shown that it follows a log-normal distribution model well, meaning that the shadow shows the normal distribution on a logarithmic scale. However, the calculation of the method is prohibitive and accurate maps of the environments of interest with the characteristics of the materials are required. Of course, the device is threatening to fade many roads due to the scattering of surrounding structures.
In addition, it also varies with the angle of the incoming signals, frequency and polarization.
Channel Prediction From a Fixed Node
In the case of channel prediction for a fixed node, communication maps created by the Gaussian process (GP) are used. Channel prediction between mobile nodes uses an empirical communication model, because it is difficult and impractical to collect enough communication signals between arbitrary positions of mobile nodes to implement the GP approach. 7; in a network, the GP-based channel prediction is performed between the base and relay nodes, while the model-based communication quality prediction is performed between the relay and client nodes.
The GP can learn from observations by setting the mean and covariance functions relevant to the observations. The GP's hyperparameters, the parameters used in the mean and covariance functions, during the training phase. In general, the GP method suffers from growing computational burden as the number of observations increases; the computational complexity of general GP methods is O(Nt3) where Nt denotes the number of measurements (i.e. the size of training data).
Since the non-parametric nature of GP can alleviate this problem, we use the concept of GP to predict channel quality from a fixed node.
Channel Prediction Between Mobile Nodes
A constant mean function m(x0,x) = c, where c is a hyperparameter to be optimized, and the quadratic exponential covariance function with the spatially varying length scale parameters [39] are used in this study. Although another mean function, such as a typical communication model, could perform slightly better than the constant mean function, it is preferred due to simplicity of implementation, since zero mean functions, a specific type of constant mean functions, are commonly used in various problems [ 40]. RS =−20 logd+Ct,r+αdobs (21), where α is the obstruction penalty coefficient and dobs indicates the sum of the blocked line segments within the internodal line (Fig. 9).
The ITU-R P recommendation takes into account the obstacle distance in a logarithmic scale and the ground loss breakthrough factor. The research agent was initially close to the transmitting node and the RSSI value was around -40 dBm. In addition, according to the wavelength (λ), antenna (GrandGt) and transmission power (Pt) of the communication modules, Ct,r was calculated as 27.85 dBm.
Although this empirically derived model-based approach can handle any type of nodes and is computationally light, the GP-based approach is still attractive and useful for channel prediction as it does not require geometric maps of the environment.
Discussion: Mean Functions of Gaussian Process for Channel Prediction 18
12 shows the road, the positions of the three base stations and the map of interest. The top 6000 samples are taken for the first set, and the rest of the samples from each base node are for the test set (the second set). While the reconnaissance agent remains in the area where a large NLOS exists, the errors of the.
However, the accuracy of the GP methods was not as accurate as the case in the validation. In this case, the result in the test area is interesting, and the rest of the results show consistency with the previous cases. This aberrational performance gap can be attributed to the relative distance between the test area and the position of the base stations.
This area can be difficult to estimate using GP and depends more on the hyperparameters of the covariance function and the values of the mean function.
Communication Performance Metrics
In this section, the relay positioning problem is formulated in terms of optimization, and the appropriate optimization method is introduced. In addition, the proposed algorithm is compared with a recent indoor relay positioning approach and its variant in simulations, which confirm the argument that the global search on the map outperforms local searches in a limited area, such as on an RRT path. It is worth noting that WCC concentrates on improving the weakest link, while GMC tries to improve the quality of all links.
This study mainly uses the WCC metric for simulations and experiments, but other performance metrics could be used without problems.
Relay Position Optimization
Optimizer for the Relay Positioning Problem
For the sake of a clear understanding of this paper, the concept of SL-PSO is introduced in this thesis. Each particle in SL-PSO learns from a randomly chosen particle from its demonstrators whose tness is higher than itself and the average behavior of the swarm. In this study, SL-PSO is adopted as the optimization tool due to the following three reasons.
First, there is less burden of parameter settings since SL-PSO provides a dimension-dependent parameter control method. Finally, SL-PSO occupies less memory than other PSO variants since SL-PSO does not store the historical data of all particles in the swarm. The particle in SL-PSO learns the behavior of one of its demonstrators and the swarm.
In fact, the author of the paper [49] has shown that SL-PSO is scalable in different dimensions due to the dimension-dependent parameter, the learning probability.
Comparison of Algorithm Performance
In a particle, each odd dimension indicates the x-position of the corresponding relay and each even dimension indicates their position. For the comparison of the RRT-based approach with our algorithm, two environments are considered and the algorithms are implemented with one to three relays. In all cases, the averages of the optimized WCC (OWCC) cost show that the network performance is significantly improved compared to the case without relays.
Note that although the mean OWCC values for the RRT-based approach and the proposed approach are similar, the standard deviations of the RRT-based algorithm are much larger than those of the proposed algorithm. Interestingly, regardless of the use of relays, the average cost of the RRT-based OWCC algorithm was worse than the performance of the WCC network even without relays. Placing an obstacle causes a dramatic change in the RRT path, but with minor changes in the quality of communication between nodes.
On the other hand, the proposed approach gives much better performance and quite consistent results regardless of the number of relays.
Experiment Setup
The XBee modules are handled by using the XbeeMav package [51] which handles congurations and communication of XBee modules in ROS environments. The map was obtained before the mission using the simultaneous localization and mapping (SLAM) algorithm found in [52], which uses Rao-Blackwellized particle filters and adaptively reduces the number of particles. The robot system toolbox in MATLAB is used as a bridge between ROS and MATLAB.
Experiment Results
The SL-PSO cluster optimization took 76.2 seconds for 89 iterations as shown in Fig. The channel quality prediction performance of the algorithm is consistent with the case of the first mission as shown in Table 7. The normalized error in the Client-Relay link is relatively larger than that of the other.
This case also shows a consistent result of the signal quality prediction compared to the previous cases as shown in Table 8. The cost of the communication performance is maximized with respect to the relay positions by the heuristic optimizer. Morris, Capacity of ad hoc wireless networks, in Proceedings of the 7th Annual International Conference on Mobile Computing and Networking.
Zill, Routing in multi-radio, multi-hop wireless mesh networks, in Proceedings of the 10th Annual International Conference on Mobile Computing and Networking. Kruskal, On the shortest spanning subtree of a graph and the traveling salesman problem, Proceedings of the American Mathematical Society, vol Mendes, Population structure and particle swarm performance, in Evolutionary Computation, Proceedings of the 2002 Congress on, vol.