Recent Advances in RF Propagation Modeling for 5G SystemsView this Special Issue
Outage Analysis of Multihop Wireless Backhaul Using Millimeter Wave under Blockage Effects
We consider multihop millimeter-wave (mm-Wave) wireless backhaul communications, by which small cell base station (SBS) clusters can connect to a macrocell base station (MBS). Assuming the mm-Wave wireless backhaul links suffer from outage caused by obstacles that block the line-of-sight (LoS) paths, we derive the statistics of a perhop distance based on the blockage model using stochastic geometry and random shape theory and analyze the multihop outage probability using the statistics of a perhop distance. We also provide an optimal number of hops to minimize the end-to-end outage performance between the MBS and the destination SBS cluster when the end-to-end distance is given.
Dense deployment of small cells over traditional macrocell s is considered as a key enabling technique for the emerging fifth-generation (5G) cellular networks [1–4]. For the small cell network deployment, millimeter-wave (mm-Wave) wireless backhaul is a cost-effective and scalable solution with large contiguous bandwidths. Moreover, the line-of-sight (LoS) nature of mm-Wave communication with directional antenna arrays can help to control the signal interference .
However, mm-Wave signals are more sensitive to blockage effects than signals in lower frequency bands, as indicated by the measurement data in [6–8]. Thus, the performances of the mm-Wave cellular systems are vastly affected by blockage effects . There are two widely used approaches to incorporate the impact of blockages into signal propagation: ray tracing and stochastic modeling. In the ray tracing approach, blockages are characterized explicitly by their sizes, locations, and shapes. Therefore, this method is appropriate for environment-specific simulations based on electromagnetic simulation tools, which perform complex numerical calculations for ray tracing [10, 11]. On the other hand, in the stochastic models, the statistics of blockages are characterized with smaller number of parameters compared to ray tracing. Therefore, the stochastic models are used to analyze general networks with acceptable accuracy.
In , the authors propose mm-Wave channel model incorporating the blockage effects based on stochastic geometry and random shape theory. Stochastic geometry has been a powerful technique to evaluate system performance in the conventional cellular networks , which reveals the impacts of multiple system parameters such as base station density, transmit power, and path-loss exponent. The key idea in  is to model random obstacles (e.g., buildings) as rectangles with random sizes and orientations whose centers form a Poisson point process (PPP) on 2-dimensional space. The model proposed in  can capture distance-dependent characteristics of the blockage effects, which is more realistic compared to the conventional shadowing model using lognormal distribution.
Using this mm-Wave propagation model in , the authors in  present a framework to derive signal-to-interference-plus-noise ratio (SINR) distributions, which can be used to analyze coverage and rate performances. Also, the outage probability of a macrodiversity system with multiple base stations (BSs) that are connected by wire to each other is analyzed in , where an outage occurs when there is no LoS path from all the base stations to the user. The analysis in  presents that this macrodiversity coming from the unblocked BS selection can be exploited to mitigate blockage in mm-Wave cellular systems.
In , the outage performance of mm-Wave wireless backhaul links between a macrocell base station (MBS) and small cell base stations (SBSs) is also studied using the channel model in . Specifically, in , the multiple SBSs in a cluster have wired connections to each other, and the wireless backhaul link between the MBS and the SBS cluster is assumed to be reliable as long as there exists one or more SBSs that have blockage-free LoS paths from the MBS. In other words, the MBS can selectively choose an unblocked SBS to construct a wireless backhaul link by beam steering, which is equivalent to the macrodiversity in .
However, while  only considers the one-hop communication between the MBS and a single SBS cluster, the 5G cellular networks may have multiple SBS clusters, which require multihop transmissions to improve the cell coverage [1, 2]. Considering the distance dependence of the blockage effects (i.e., the likelihood of a blockage event increases as the distance increases) at mm-Wave, multihop communication can be an effective solution to build mm-Wave wireless backhaul systems. In this context, motivated by the limitation in , we extend the single-hop wireless backhaul system in  to a multihop scenario with multiple SBS clusters. Moreover, it is noteworthy that the wireless backhaul links between two SBS clusters studied in this paper are distinct from the system model in  because we consider multiple-points-to-multiple-points (SBSs-to-SBSs) links, while  is focused on single-point-to-multiple-points (MBS-to-SBSs) links. Therefore, with different distance statistics from , the intercluster SBS-to-SBS communication can benefit from higher order of spatial diversity compared with the MBS-to-SBS communication in .
The contributions of this paper are fourfold. First, we derive the probability distribution, mean, and variance of a perhop distance (i.e., the distance between two randomly located SBSs in two adjacent SBS clusters), considering the spatial diversity. Second, we analyze the perhop outage performance of mm-Wave wireless backhaul links between two adjacent SBS clusters. Third, the outage analysis is extended from the perhop link to multihop systems. Lastly, we present an optimal and suboptimal hop count to minimize the end-to-end outage performance between the MBS and the destination SBS cluster for a given end-to-end distance, where the suboptimal hop count is derived based on only the perhop outage performance.
This paper is organized as follows. We introduce the system model in Section 2. We derive the probability distribution, mean, and variance of the distance between two SBS clusters in Section 3. The outage performances of perhop and multihop cases are analyzed in Section 4. In Section 5, we show optimal hop distance and hop count to minimize the outage performance and propose suboptimal hop distance and hop count that give close enough outage probabilities to the optimal ones. Finally, conclusions are provided in Section 6.
2. System Model
We consider a 5G network with a single MBS and multiple SBS clusters as an example illustration in Figure 1. In the figure, the macrocell is indicated by the gray hexagon, and the SBS clusters are represented by the sky blue circles. Each SBS cluster has multiple SBSs, which are connected to each other through wires to enable spatial diversity in SBS cluster [15, 16]. The MBS is connected to the core network, and SBSs can access the core network via the MBS using wireless backhaul, which is indicated by the dotted blue lines in the figure. Moreover, some SBS clusters, which are far from the MBS, trigger multihop transmissions for the backhaul links to overcome path loss and blockage effects. For example, in Figure 1, SBS Cluster 2 communicates with the MBS in two hops over Cluster 1.
As shown In Figure 2 focusing on the multihop wireless backhaul links, we assume that SBSs in each cluster are distributed over a circle. In the figure, the areas of the multihop clusters are denoted by Areas , , …, and with radii of , respectively. Also, the hop distances are denoted by , where . As in , we assume that in -th cluster multiple SBSs are uniformly distributed with intensity according to a homogeneous point process (PPP) , which is a widely used model for various types of cellular networks . For the end-to-end wireless backhaul link with hops, the outage probability can be expressed aswhere is the outage probability of -th hop . Assuming the outage events in different hops are mutually independent, the outage analysis over multiple hops can be decoupled into multiple intercluster links, while the first-hop outage can be obtained based on the analysis in . When we consider the multihop performance , we assume uniform SBS clusters and equidistant hops for the mathematical tractability which we will explain in detail in Section 4.2. In general, however, the densities of the SBS clusters are dissimilar, which means variable and hop distances for different . Even in this case, the perhop analysis in this paper still can be used to optimize the end-to-end outage , which is a function of the hop distances and the corresponding end-to-end hop count .
To analyze the outage performances of the following hops, we consider the intercluster outage for as shown in Figure 3. Clusters 1 and have and SBSs, which are uniformly distributed over and with intensities and , respectively. These intensities correspond to the average numbers of SBSs per unit area (i.e, SBSs/m2). In other words, the numbers of the SBSs in the two clusters are assumed to be and , which are independent random variables following Poisson probability distributions as where denotes the probability that and is a nonnegative integer. Therefore, the average numbers of the SBSs in the two clusters are given by and , where implies the expectation operator. We assume the centers of the two clusters are separated by with fixed ratios to the two radii as , where and (i.e., there is no intersection between and ) (we assume the dense deployment of SBSs with wired connections in each cluster for coverage extension, especially to solve the hotspots problem, and the geographically disjoint SBS clusters. In addition, considering the construction cost, we assume that the SBSs in different clusters communicate with each other through the multihop wireless backhaul.).
Let be the distance between the mm-Wave SBSs and in and , respectively, as shown in the figure. Also, as in [12, 14–16], the probability of no blockage in an individual SBS-to-SBS (S2S) path is , where is the parameter that captures density and size of obstacles. The greater means obstacles with higher density and larger sizes, which results in lower . In this paper, we assume can be interpreted as the probability that the communication link between the SBSs and in the two clusters is reliable.
As in , we assume that the multiple SBSs in the same cluster are connected to each other through wires and thus the wireless backhaul link between the two clusters is reliable if one (or more) of the individual S2S links between the two clusters is reliable. An outage event of the intercluster mm-Wave wireless backhaul link is caused by blockage of LoS path. Before data transmission, with beam steering, each SBS in Cluster can selectively connect to one of the SBSs in Cluster whose LoS path is blockage-free. Also, once one SBS finds an unobstructed LoS path, the SBS notifies the other SBSs in the same cluster via the wired intracluster connections.
3. Probability Distribution of S2S Distance between Two SBS Clusters
In order to investigate the performance of the mm-Wave wireless backhaul system, we need to consider probability distribution of the S2S distance between the two SBS clusters, since is a function of , which is a random variable. As shown in Figure 4(a), suppose the distance between the center of the left SBS cluster and SBS in the right SBS cluster is . To derive the probability distribution of , we first find the probability distribution function (PDF) of using an example illustration in Figure 4 (because is the distance between a random location in Area and the fixed point , the PDF of is identical to the PDF of the distance of the single-hop wireless backhaul in .), and then conditioned on we will obtain the PDF of . In Figure 4(b), the shaded area (Area ) is the feasible region of in the circle (Area ). Consequently, the cumulative density function (CDF) of is expressed as the ratio between Areas and asWhere , , and . As a result, the PDF of iswhere . As shown in , when , we can derive an approximate PDF by treating the area enclosed by the path as the region bounded by the path in Figure 4(b):where . With this approximated PDF, we can obtain and .
Then, going back to Figure 4(a), conditioned on , the PDF of can be derived as the ratio of the area within a closed path to the entire circle , which is given bywhere , , and . Hence, the corresponding conditional PDF of is given byAs the PDF of , for a large enough relative to and , this conditional PDF has an approximate formwhere . Hence, the PDF of is given bywhere . This approximate PDF gives the mean and variance as follows:These two statistics will be used to derive the perhop outage probability in the next section.
4. Outage Analysis
4.1. Intercluster Outage Analysis
In this section, we derive the probability that there is no reliable S2S LoS link between the two SBS clusters and . We define an outage, if all the LoS S2S paths are blocked by obstacles. Let be a Bernoulli random variable aswhere if , it means that the mm-Wave link between SBS in and SBS in is reliable. Assuming that blockage is impenetrable, as in [12, 14–16], an outage event of the mm-Wave SBS occurs, if all ’s are zeros. In other words, if there is at least one LoS mm-Wave link with for any pair and , the mm-Wave communication between the two SBS clusters is reliable. Therefore, the outage in th hop can be defined aswhere and .
Following [12, 14–16], for analytical tractability, we assume the outage events indicated by ’s in (12) are mutually independent for all and . Strictly speaking, the blockage events on different links are not always independent. Therefore, as in [15, 16], the outage probability in this paper may be a lower bound as a reference of system design and analysis. However, we note that the numerical results in  show that the error caused by the independent link assumption is minor and acceptable in accuracy. With the assumption of the independent blockage events, we havewhere the distance between SBSs and from and is a random variable with the approximate PDF in (9). Letting , it can be approximated by Taylor expansion aswhere . Thus, the outage probability in (12) is given byThus, we can interpret in (14) as the expected , when the number of SBSs in the two clusters is given ( and ). Unfortunately, it is difficult to obtain a closed form expression of (15), but based on (14) and (15), we can observe some important properties of the intercluster outage probability as follows.
Property 1. For given positive , , , , and , as (because and ) (as described in Section 2, we assume the two SBS clusters are disjoint by with and . Thus, when , the sizes of the two clusters also become zero, which triggers an outage since the clusters do not have any SBSs.). Moreover, as , . This property can be proved by , as or .
Property 2. For fixed , , , , and , as increases (i.e., and increase), increases, because increases.
Property 3. When the other parameters are fixed, the greater gives the higher .
Property 4. As and increase, decreases because the expected numbers of the SBSs and increase.
Figure 5 displays simulation results of the intercluster outage, where the horizontal axis indicates the hop distance , while the vertical axis represents the intercluster outage probability . Assuming , , and , the solid line is the baseline, while the other three graphs correspond to changes in , , and , respectively. We can observe all the four properties in Figure 5. To be specific, all the four graphs in the figure are convex functions of , which corresponds to Property 1. Also, as described in Property 2, compared to the solid line (i.e., the baseline case), the dash-dot graph, which indicates the higher , shows the higher outage rates. Property 3 can be observed by the comparison between the solid and the dotted curves. Lastly, Property 4 is found by the dashed line in the figure.
4.2. Multihop Outage Analysis with a Constant Hop Distance
In this section, we extend the outage analysis to the multihop wireless backhaul assuming the uniform SBS clusters (i.e., , , and for all ). With this assumption, the end-to-end outage probability in (1) between the MBS and the -th SBS cluster is simplified aswhere is the outage probability for the first-hop link between the MBS and the first SBS cluster presented in  and is given in (15). Because the height of the MBS is typically much greater than that of SBS, based on , we need to use a smaller blockage parameter for the first hop, where . Thus, . When the end-to-end distance between the MBS to the center of the -th SBS cluster is , then the hop distance is .
Figure 6 shows multihop simulation results with , where the - and -axes are the number of hops and the outage probability, respectively. The three different curves with the different markers (circle, -marker, and triangle) represent the outage probability results for , 3, and 5 km, respectively. Also, the solid and dotted lines indicate the outage probabilities for the multihop (MH) and the perhop (PH) cases in (16) and (15), respectively. First, we can observe there exists an optimal hop count, denoted as , minimizes the outage probability for a given . By the exhaustive search, we obtain , 12, and 19, for , 3, and 5 km, which means that the corresponding optimal hop distances are almost the same for the three values of . Also, the outage curves for both the MH and PH have the same because of (i) almost negligible first-hop error (because ) and (ii) the same convex curves of for as in Figure 6. Based on these observations, in the next section we propose the suboptimal hop count by mathematically deriving the suboptimal hop distance , instead of the exhaustive search of .
5. Suboptimal Hop Count
As shown in Figure 5, there is an optimal hop distance between two SBS clusters to minimize . Assuming and are fixed, we derive the suboptimal by treating the intercluster S2S path as two-hop link between the two SBS clusters. For example, in Figure 3, suppose that we place an imaginary relay at on the straight line connecting and , assuming that the two clusters communicate with each other only through the relay. Suppose that the outage rates “from the left cluster to the relay” and “from the relay to the right cluster” are and with the distances and , respectively. Then, as finding the center of mass, the best location of the relay corresponds to . Using the outage probability between SBS clusters to the MBS in , and are expressed asTherefore, when , and are, respectively, obtained aswhere and . If plugging and back into (17), the overall outage rate of the two-hop link with the imaginary relay is . The proposed suboptimal hop distance is that minimizes . In other words, corresponds to the solution of , while . This suboptimal can be readily obtained by numerical calculation for given , , and . If we assume the same SBS cluster sizes (i.e., ), we can obtain the suboptimal in (20). Therefore, if the hop distances are the same with the uniform PPP intensities (i.e., for all ), the suboptimal hop count to reach the distance is , where means the nearest integer to .
Figures 7(a) and 7(b) show the simulation results of and , respectively. In Figure 7(a), the horizontal axis is the blockage factor , while the vertical axis is the optimal and suboptimal hop distances and . As shown in the figure, both and , which are indicated by the solid and dotted lines, decrease, as increases, because of the greater blockage effects. Also, comparing the two graphs, with the small gap in the entire range of , which validates the proposed suboptimal hop distance. In Figure 7(b), the -axis represents the distance from the MBS to the -th SBS cluster (i.e., the destination SBS cluster), while the -axis indicates the optimal and suboptimal hop counts and , which are denoted by the solid and dotted curves. In the figure, we can observe that and are almost linearly increasing, as increases. Also, with the small difference, we observe that because . Thus, based on with a small , the suboptimal can be used to reduce the exhaustive search range of .
Figure 8 shows the end-to-end outage probabilities versus , where the solid and dotted lines correspond to the outage probabilities with and , respectively. The markers represent different and . As shown in the figure, the gap between the outage probabilities with and becomes smaller as and decrease. Especially, when km, the outage performance with becomes extremely close to that with as increases, which can be expected from the simulation results in Figure 7(b).
In this paper, we study the outage performance of the multihop wireless backhaul systems at mm-Wave in the presence of the blockage effects. Assuming that multiple SBSs are distributed in a cluster with the wired connection to each other and considering the spatial diversity in a cluster, the outage probability between two SBS clusters is derived. Through analysis and simulation, it is shown that there exists an optimal hop distance that minimizes the intercluster (perhop) outage probability. Then, assuming uniform SBS clusters we extend the perhop outage analysis to the multihop scenario, where we find that there exist optimal hop distance and hop count that minimize the end-to-end outage probability between the MBS and the destination SBS cluster when the end-to-end distance is given. Using the same trend of the perhop and multihop outage performances with respect to the perhop distance, we propose the suboptimal hop distance and corresponding suboptimal hop count. The suboptimal hop count shows a small gap to the optimal hop count, which means that the exhaustive search range of the optimal hop count can be significantly reduced. Potential extensions of this paper include addressing a wider scenario with dissimilar sizes and densities of the SBS clusters.
The authors declare that they have no competing interests.
This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (Grant no.: NRF-2016R1D1A1B03930060).
T. S. Rappaport, F. Gutierrez, E. Ben-Dor, J. N. Murdock, Y. Qiao, and J. I. Tamir, “Broadband millimeter-wave propagation measurements and models using adaptive-beam antennas for outdoor Urban cellular communications,” IEEE Transactions on Antennas and Propagation, vol. 61, no. 4, pp. 1850–1859, 2013.View at: Publisher Site | Google Scholar
P. Billingsley, Probability and Measure, John Wiley & Sons, 2012.