Table of Contents Author Guidelines Submit a Manuscript
Wireless Communications and Mobile Computing
Volume 2019, Article ID 6918637, 19 pages
https://doi.org/10.1155/2019/6918637
Research Article

Toward a Unified Framework for Analysis of Multi-RAT Heterogeneous Wireless Networks

1Computer and Information Systems Engineering Department, NED University of Engineering and Technology, Pakistan
2Telecommunications Research Laboratory, Toshiba Research Europe Ltd., Bristol BS1 4ND, UK

Correspondence should be addressed to Adnan Aijaz; gro.eeei@zajia.a.nanda

Received 15 August 2018; Revised 29 November 2018; Accepted 16 December 2018; Published 3 January 2019

Academic Editor: Antonio De Domenico

Copyright © 2019 Murk Marvi et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

The increased penetration of different radio access technologies (RATs) and the growing trend towards their convergence necessitates the investigation of wireless heterogeneous networks (HetNets) from coverage and capacity perspective. This paper develops a unified framework for signal-to-interference-plus-noise ratio and rate coverage analysis of multi-RAT HetNets, with each RAT employing either a contention-free or a contention-based channel access strategy. The proposed framework adopts tools from stochastic geometry, with the location of APs and mobile users modeled through independent Poisson point processes (PPPs). We specifically focus on a two-RAT scenario (i.e., cellular and Wi-Fi), where for multi-tier Wi-Fi RAT, with contention-based channel access like CSMA/CA, the location dependent distribution of interfering APs has been approximated through a homogeneous PPP. Moreover, by using some simple yet realistic set of assumptions, the distance to nearest active AP has been defined which results in simplified expressions. The medium access probability for a random and a tagged AP under a multi-tier Wi-Fi RAT has also been derived and discussed. By keeping in view the tremendous effect of temporal domain on overall network performance, the stable queue probability has been derived by assuming a non-saturated traffic model. The results have been validated through extensive simulations and compared with existing approaches. Some useful insights have also been presented that shed light on design and analysis of multi-RAT HetNets and provide motivation for further research in this direction.

1. Introduction

According to Cisco’s forecasts [1], billion hand-held devices and billion machine-to-machine (M2M) devices will be connected by 2021. The number of connected devices would clearly exceed the expected global population of billion by that time. The monthly global mobile data traffic is expected to reach 49 exabytes and the annual traffic will exceed half a zettabyte by 2021. Extreme densification and offloading in wireless heterogeneous networks (HetNets) is the key technique for meeting the ever increasing capacity requirements which can potentially bring the 1000× improvement in aggregate data rate as envisioned for 5G networks [2]. With networks becoming denser, more traffic can be offloaded to small cells, especially because a significant proportion of traffic is generated by users in indoor locations (home, offices, etc.). Offloading to Wi-Fi networks is a natural and most effective technique for reducing congestion on cellular networks [3]. The integration of cellular and Wi-Fi networks has been a topic of interest, in both industrial and academic communities, since early 2000s. Wi-Fi is experiencing a trend towards ubiquity with operator-deployed Wi-Fi networks becoming a norm. According to recent projections [1], of mobile data traffic would be offloaded from 4G networks to Wi-Fi by 2021.

Witnessing these technological trends, it can be easily inferred that understanding the fundamental limits of HetNets comprising different radio access technologies (RATs) and employing different channel access techniques becomes particularly important. Analytical tools for HetNets can provide valuable insights to system designers in meeting the future coverage and capacity demands through integration of different RATs, with multiple tiers, and by exploiting licensed and unlicensed resources in an effective manner.

1.1. Related Work and Motivation

Over the last few years, tools from stochastic geometry have been used for analysis of single-tier and multi-tier cellular networks [4, 5], where mostly Poisson point process (PPP) is used for modeling the spatial location of access points (APs). The PPP assumption provides a lower bound, whereas the traditional grid model provides an upper bound on coverage. Other point processes like Gibbs [6], Strauss [7], Ginibre [8], and Detriminantal [9] have also been used for coverage analysis of cellular networks. Such processes provide better accuracy as compared to PPP but at the cost of limited tractability. Therefore, PPP has been a popular choice for analysis of cellular networks. On the other hand, when it comes to Wi-Fi RAT, PPP cannot be used for spatial modeling of active APs due to contention-based nature of carrier sense multiple access with collision avoidance (CSMA/CA) scheme. However, approximated solution for SINR coverage has been presented in [10] by exploiting Modified Matern Hard Core Process (MMHCP) for estimating the set of active APs, where the APs are originally distributed using a PPP. The work is further extended from spatial averages to spatial distributions in [11], and throughput analysis has also been conducted against various parameters of interest. Due to approximation of the set of interfering APs through a non-homogeneous PPP, the resulting expressions in [10, 11] are extremely complicated. The asymptotic expression, for outage probability of general ad-hoc networks, has been obtained in [12]; however, the results are limited to high signal-to-interference ratio (SIR) cases.

Although closed-form expressions for analysis of cellular RAT, under special cases, are available in literature [4, 13], no such results are reported for Wi-Fi RAT due to difficulty in characterizing the interference effect of active APs [10, 11]. This limits the in-depth analysis of multi-RAT HetNets which could be an important tool for coverage and capacity planning of future wireless networks. A general model for multi-RAT HetNets has been presented in [14] by assuming independent PPPs for distribution of APs and users in a given region. However, the effect of channel access schemes, associated with different RATs, has not been taken into account which directly affects the interference and, hence, the coverage analysis. Thus, such a framework cannot provide accurate insights into the different characteristics exhibited by multi-RAT HetNets.

Recently, the coexistence problem of multi-RAT networks has received considerable attention. In [15, 16] the coverage and capacity analysis has been presented by assuming the coexistence of cellular and Wi-Fi RAT in unlicensed spectrum. After investigating various transmission mechanism, the authors in [15] reported that the LTE can coexist with Wi-Fi but under certain conditions. On the other hand, in [16] a similar issue has been addressed by assuming the operation of LTE users in both licensed and unlicensed band. Two different user association schemes, i.e., crossing and non-crossing-RAT, have been considered, where the users of licensed RAT can access unlicensed band by exploiting opportunistic CSMA/CA scheme. The crossing-RAT user association is shown to provide better performance as compared to non-crossing-RAT user association. However, the coexistence of LTE and Wi-Fi users in unlicensed band cannot bring significant gain in performance because increase in the capacity of LTE RAT is achieved at the cost of decrease in the capacity of Wi-Fi RAT [16]. Therefore, in this research, the coexistence of different RATs, in unlicensed band, has not been covered.

Some recent studies have investigated the fundamental limits on densification of cellular network by assuming different path loss models and fading distributions [1720]. According to [17], the SINR coverage decreases after a certain threshold, due to increased interference with smaller path loss exponents, which is in contrast to widely used results as reported in [4]. A detailed investigation has been presented in [18] by exploring four different performance regimes while transitioning from sparse to dense networks. Results similar to [17] are reported, where, after a certain threshold, the SIR coverage no longer remains constant; however, the area spectral efficiency increases linearly. According to [19], the SIR coverage can be increased by exploiting the idle mode capability of APs, under dense scenarios. Further, in [20] energy efficiency analysis has been presented by defining the optimal transmission power, for APs, as a function of RAT density. However, in all of the mentioned investigations, a single-tier dense cellular network, with small cells, has been considered. Moreover, in few of the existing studies [21, 22], multi-RAT networks have been investigated by assuming operation of APs in different bands. In [21] the sub-6GHz macrocells are overlaid with mm-wave small cells, and in [22] the small cells can operate on both bands. Various cell association schemes, considering both uplink and downlink channels, have been investigated in [21]. Following a similar thought, in [22] a biasing based strategy has been proposed for load balancing across a multi-RAT network. In contrast to traditional approaches, in [22] two biasing thresholds are exploited; one for offloading users from macrocells to small cells and other for offloading users from sub-GHz band to mm-wave band. In a nutshell, the integration of mm-wave communication into existing infrastructure is one of the potential contributors for increasing capacity of future wireless networks. However, the communication in mm-wave exhibits the characteristics which are different from sub-GHz band, and, hence, its realization requires a lot of enhancements both at component and architecture level [23].

Dense multi-RAT HetNets would be a key aspect of future wireless networks [2]. According to some studies [17, 18], the degradation in coverage, provided by dense small cell networks, after a certain limit is expected. Therefore, in order to meet the demands in coming future, the focus must be shifted from single-RAT to multi-RAT HetNets. Although in few of the recent studies [21, 22] the multi-RAT networks have been analyzed by exploiting sub-GHz and mm-wave band, this work is focused around the analysis of multi-RAT HetNets, where each RAT operates on a different pool of resources and can use either contention-based or contention-free channel access schemes. As cellular, contention-free, and Wi-Fi, contention-based, RATs are already deployed at a wide scale, and, hence, their integration can be considered as a potential contributor for improving the capacity of future wireless networks, without many modifications into existing infrastructure. Thus, the key motivating factor behind this work is the growing convergence of the two RATs, and, as opposed to standalone RATs, their integration can lead to better network performance.

1.2. Contributions and Outline

With the aforementioned background and motivation, the key contributions of this work can be summarized as follows.(i)A unified framework for multi-RAT HetNets: using tools from stochastic geometry, we develop a unified framework for SINR and rate coverage analysis of multi-RAT HetNets, where RATs can operate on either contention-free or contention-based (CSMA/CA) channel access schemes. It differs from existing framework [14] due to incorporation of contention-based channel access scheme. More specifically, we focus on a two-RAT HetNet scenario which includes a cellular and a Wi-Fi RAT. The Laplace transform of interference for cellular RAT can be derived easily and it is available in existing literature [4, 13]. The main difficulty arises while modeling the cumulative interference effect under Wi-Fi RAT [10, 15], which operates on a contention-based channel access scheme. Thus, by exploiting a few approximations, we derive the Laplace transform of interference for Wi-Fi RAT which provides accuracy comparable to existing studies [15].(ii)Analysis of heterogeneous Wi-Fi RAT: we present a tractable solution for SINR and rate coverage analysis of a multi-tier Wi-Fi RAT, by exploiting a few approximations. To the best of the authors knowledge, the notion of multi-tier Wi-Fi RAT has not been studied in existing literature. However, as new techniques like dynamic carrier sensing and extreme densification are emerging, it is important to analyze the effect of heterogeneity in Wi-Fi RAT. We have derived the medium access probability (MAP) for a random and a tagged AP under multi-tier Wi-Fi RAT, and results show that, under dense network conditions, the MAP for a typical AP approaches that of a tagged AP.(iii)Stable queue probability: by assuming a non-saturated traffic model, we derive the stable queue probability for a user under an AP of a RAT. In order to avoid the problem of interacting queues, similar to [24, 25], we assume a dominant and a modified system, where results for each case have been reported and analyzed. It has been found that, for low packet arrival rate, the stable queue probability of a user under Wi-Fi RAT is slightly higher compared to cellular RAT. However, for higher packet arrival rate, the stable queue probability of a user under cellular RAT is better.(iv)Various insights: we provide various insights by analyzing different HetNet scenarios with the aid of proposed framework. It has been shown that the integration of femto-tier with Wi-Fi tier provides reasonable SIR coverage as compared to multi-tier cellular or Wi-Fi RAT; however, the rate coverage starts declining as the user association with Wi-Fi RAT exceeds the cellular RAT. Further, the SIR coverage increases and gradually approaches unity, as a function of Wi-Fi RAT density; this insight is in contrast to existing results reported in [14]. Although the SIR coverage provided by Wi-Fi RAT is better than cellular RAT, the stable queue probability of a user under cellular RAT is overall better than Wi-Fi RAT. We also explore the trade-off between user and AP density, and the results show that the rate coverage decreases by increasing the AP density of Wi-Fi RAT while maintaining a constant average load per AP. Under such circumstances, it has been suggested to increase the number of non-overlapping channels for Wi-Fi RAT, as it can greatly improve the rate coverage.

The rest of the paper has been organized as follows. Section 2 introduces the underlying system model in detail along with the considered channel access schemes and performance metrics. The MAP metric has been covered under Section 3. The main results of the paper have been covered in Section 4, where a unified framework has been presented for SINR and rate coverage analysis of multi-RAT HetNets, and stable queue probability for a user under a RAT has been derived. Various results have been reported and discussed in Section 5. Finally, Section 6 concludes the paper.

The notation used in the paper and associated details have been provided in Table 1. The general parameters considered for generating various results, under Sections 3, 4, and 5, have been provided in Table 2.

Table 1: Notation summary.
Table 2: General parameters and settings.

2. System Model

We consider a -RAT, -tier HetNet scenario wherein RATs can employ either contention-free (OFDMA, TDMA, CDMA, etc.) or contention-based (CSMA/CA) channel access scheme. We specifically consider a two-RAT scenario which includes a cellular and a Wi-Fi RAT each with number of tiers. More RATs can be considered provided that each RAT operates on a different pool of resources and the user equipment supports connection to all considered RATs. We adopt a homogeneous PPP , with density , for drawing the locations of APs belonging to the tier of the RAT whereas and . Another independent PPP , with density , has been considered for the distribution of users in a given region. We assume that and denote the set of all APs under cellular and Wi-Fi RAT, respectively. Moreover, all APs provide open access; i.e., there is no closed subscriber group, and denotes the set of all APs deployed in the given region.

We consider a downlink channel wherein single resource block (i.e., time, frequency, and code) is utilized in every cell of cellular network. For Wi-Fi RAT, we assume single downlink channel. A saturated traffic model has been considered, where APs transmit continuously even without any packet in queue for transmission. Further, APs of one RAT cannot interfere with those of the other RATs as they operate in different pools of wireless resources. However, APs of different tiers under the same RAT interfere with each other due to shared resources. All APs of transmit at the same power over the bandwidth . We consider both large-scale path loss and small-scale fading. Free space path loss (FSPL) model with reference distance of 1 meter, as given by , has been assumed for all links; here, and denote the operating wavelength and path loss exponent, respectively. The fading channels are Rayleigh distributed with average power of unity i.e., . The noise is assumed additive with power corresponding to the RAT. We assume that user association is based on the maximum average received signal strength. However, it can be easily extended to a generic user association scheme, as given in [14], by just introducing a weight or bias variable. For simplification, normalized parameters for a pair with respect to serving pair have been defined as , , and . Similar to [13, 14], the probability density function (PDF) of the distance between a typical user and the tagged AP is given bywhere is the probability that a typical user associates with an AP of pair , and it can be given asDue to assumption of FSPL model, the association of a user to an AP of pair is dependent on the operating frequencies of RATs, as clear from (1) and (2). The normalized component for standalone RATs becomes unity as we have assumed that all tiers under a single-RAT share the same resources; hence, for such cases, we get simplified expressions for user association, which are similar to those in [4, 13].

2.1. Channel Access

Contention-free channel access schemes are employed by cellular RAT, where some of the operators deploy frequency reuse factor of unity and others go for fractional frequency reuse. Under contention-based channel access schemes, like CSMA/CA used by Wi-Fi RAT, only the APs with different contention domains are allowed to transmit simultaneously, and, therefore, the set of active APs can be less than the deployed one. Under such a scheme for channel contention, each AP maintains a random back-off timer and waits for its expiry, when the channel is sensed as free. Meanwhile the transmission starts if no other AP accesses the channel. Otherwise, it freezes the timer and repeats the procedure. Due to various reasons, a collision may occur when two APs, in the same contention domain, transmit simultaneously. However, there are defined procedures in Wi-Fi for handling such situations.

Under cellular RAT, with contention-free channel access, all deployed APs are active; therefore, the original PPP can be used for capturing the cumulative interference effect. However, under Wi-Fi RAT, with CSMA/CA channel access, APs sharing the same contention domains are not allowed to transmit simultaneously. Therefore, the original homogeneous PPP , used for drawing the location of Wi-Fi APs across a given region, cannot be used for interference modeling. In literature, Modified Matern Hard Core Process (MMHCP), also known as MHCP-2, is widely used for estimating the set of active APs [10, 15]. MMHCP is basically obtained by mark dependent thinning of original PPP , where represents the back-off timer of an AP located at . Thus, any point of the original PPP is retained, only if it has a mark smaller than all marks associated with the APs in its contention domain; i.e., . MMHCP does not take into account the effect of variable back-off timer window size or collisions. However, in [10, 11] it has been proved that the model provides a reasonable conservative representation of active APs, by comparing it against an actual CSMA/CA networks.

2.2. Performance Metrics

We consider four performance metrics described as follows.

2.2.1. Medium Access Probability

For cellular RAT, the MAP, denoted by , is unity as all APs are allowed to transmit simultaneously. On the other hand, due to contention-based channel access, the MAP for Wi-Fi RAT, denoted by , can be less than unity. According to MMHCP defined in [10, 15], a random AP under Wi-Fi RAT can access medium only if it has the smallest mark among all the APs in its contention domain (3). Hence, the medium access indicator for an AP is given byFor further details, please refer to Section 3.

2.2.2. SINR Coverage

A typical user is said to be under coverage if the received SINR from a tagged AP of pair , located at , is greater than some defined threshold , and it is given bywhere denotes the channel gain from a tagged AP located at distance from the user, and is the cumulative interference from all APs of serving RAT-tier pair , outside the disk of radius with center at origin. By using total probability theorem, the overall SINR coverage provided to a randomly located user can be given asFurther details are covered under Section 4.1.

2.2.3. Rate Coverage

The probability that a user, which is associated with an AP of pair , receives a rate greater than a certain threshold is given bywhererepresents the rate of a user, denotes the number of users served by an AP of pair , and represents the MAP for a tagged AP. By exploiting total probability theorem, the overall rate coverage provided to a randomly located user can be given asFurther details are covered under Section 4.2.

2.2.4. Stable Queue Probability

The stable queue probability has been defined as the probability that a user queue under an AP of a RAT is stable. A queue is stable only if the provided service rate is greater than the arrival rate of packets during a time-slot.However, the service rate provided by the network is dependent on the queues status, and vice versa is also true. This creates the problem of interacting queues and it becomes difficult to analyze the combined effect of spatial and temporal domain on overall performance of the network. Thus, in order to avoid this issue, the concept of dominant and modified systems has been exploited in existing literature [24, 25]. Where the dominant system provides a lower bound on performance, by assuming full buffer model for interfering APs, and modified system provides an upper bound by assuming that the active probability of APs is equal to the packet arrival rate of users, hence, the packets not transferred successfully are dropped. Further details are included under Section 4.3.

3. Medium Access Probability

According to the given definition (3) for MAP, a Wi-Fi AP cannot transmit whenever any of its contender AP has a smaller back-off timer, which is similar to one in [10, 15]. As we have assumed a multi-tier Wi-Fi RAT, the APs operate at different power levels based on the tier to which they belong to; hence, it is possible that the APs operating at higher power levels do not sense the presence of low power APs in their vicinity. This effect needs to be captured carefully in order to derive the MAP for a multi-tier Wi-Fi RAT. For better illustration, a two-tier Wi-Fi RAT scenario has been shown in Figure 1, where “” and “” represent the sensing radius for APs operating at high and low power levels, respectively. The sensing radius has been obtained by using (14) which does not include small-scale fading; however, this is just an illustration of possible effects on contention domains while considering multi-tier Wi-Fi RAT. The contention domain of each AP, for scenario shown in Figure 1, is -0:, AP-1:, AP-2:, AP-3:, AP-4:. It must be clear that AP-2 is not part of the contention domain of AP-0, as the received signal strength at AP-0 is below the required threshold . On the other hand, AP-0 is in the contention domain of AP-2. AP-1 is sufficiently close to AP-0, and the required threshold is maintained; hence, it belongs to the contention domain of AP-0. AP-3 and AP-4 are at a far distance from AP-0 such that the received signal strength is less than the required threshold. If AP-0 get a chance to access medium, AP-1 and AP-2 remain in silent mode. On the other hand, if AP-2 access the medium, then AP-0 can also transmit given it has a smaller back-off timer than AP-1, as it cannot detect the presence of AP-2. Thus, the MAP, under multi-tier case, can easily be obtained by exploiting the given definition (3).

Figure 1: Contention domains of APs under a two-tier Wi-Fi RAT.

Lemma 1. Given a Wi-Fi RAT with -tiers, each with transmission power and sensing threshold , then the MAP for a typical AP is given bywhere

Proof. See Appendix A.

Remark 2. If either or , . Furthermore, decays at a faster rate with respect to as compared to .

Remark 3. The MAP for any random AP is the same irrespective of the tier to which it belongs. As clear from Figure 1, the contention domain of an AP operating at either high or low power level includes both low and high power APs within the sensing range “” and “”, respectively.

The obtained expression (11) can be approximated by following expression which provides a lower boun on MAP, where andis the sensing radius of APs belonging to the tier. Based on the parameters listed in Table 2, the MAP for a single-tier and a two-tier Wi-Fi RAT has been plotted in Figure 2, against density parameter. The numerical and analytical results are obtained by using (11) and (13), respectively, whereas the simulation results are generated by using given definition (3). It must be noted that the simulation results are closely following the numerical ones. The results of analytical expression (13) are fairly close and providing a lower bound. As tier-3 operates at a lower power as compared to tier-2, under single-tier scenario, the MAP for tier-3 is higher as compared to tier-2. In accordance to Remark 2, it must be clear from the reported results that, with gradual increase in or , the MAP approaches .

Figure 2: Comparison of numerical, analytical, and simulation results for the MAP of a random AP against Wi-Fi RAT density.

Remark 4. The approximated expression (13) provides a lower bound on ; therefore it is reasonable to say that ; here takes into account only large-scale path loss, whereas also considers the effect of small-scale fading. This implies that the expected sensing area, or equivalently sensing radius, for an AP is small when fading effects are taken into account; hence, the expected number of contenders are less which results in improved MAP; i.e., .

According to orollary 1 of [15], the MAP of a tagged AP is the biased version of the MAP for a typical AP. However, we argue that, as the density or power of tier increases, the MAP for a tagged AP approaches the MAP for a typical AP. For better illustration please refer to Figure 3, where three different cases are considered i.e., low, moderate, and high density, by assuming single-tier scenario. Part shows moderate density case because the distance between a user and its tagged AP is . As the user associates with the nearest AP, the shaded region does not include any AP other than the tagged one. That is why, in [15] it has been suggested that the MAP for a tagged AP is the biased version of MAP for a random AP. Now let us consider the sparse case in part of Figure 3 where . Although the MAP is high in this case, the link between user and its tagged AP is of no use because the received signal strength is less than the required threshold , assuming that the received signal strength required for user is the same as that for the tagged AP. Thus, under sparse condition the MAP for a tagged AP and even for a random AP approaches unity but at the cost of decrease in received signal strength. Finally moving to dense case, part of Figure 3 where , it must be clear that as density of the RAT increases, decreases; hence, the shaded region starts shrinking and the MAP for a tagged AP approaches that of a typical AP.

Figure 3: The relationship between approximated sensing range of a tagged AP and its distance to user.

Lemma 5. The MAP for a tagged AP belonging to the tier of Wi-Fi RAT, with transmission power and sensing threshold , is given by where is defined in (12), and is given by (1).

Proof. ee Appendix B.

Remark 6. By using total probability theorem, the overall MAP for a tagged AP in can be given as .

As we have assumed a multi-tier Wi-Fi RAT scenario, Lemma 5 provides the MAP for a tagged AP which belongs to the tier of Wi-Fi RAT. It is an extension of emma 2 from [26] in which the retention probability for an associated AP has been defined, when LTE APs coexist with single-tier Wi-Fi RAT in unlicensed band. In Figure 4, the numerical results have been plotted for a tagged and a random AP, under single-tier and multi-tier scenarios, against density parameter. It must be clear that, under low density with smaller power of transmission , the MAP for a tagged AP is slightly higher than random AP. However, as the density or power of transmission increases , the MAP for a tagged AP approaches that of a random AP. Further, in Figure 5 the void probability, given in [4], for no AP within a region of radius has been plotted, and the approximated sensing radii for tier-2 and tier-3 are also denoted with markers. It must be clear that, under sparse case when = 100 AP/km2, the probability that the distance between a user and the tagged AP is greater than the corresponding sensing radius is around for tier-3 and for tier-2. As already mentioned while discussing Figure 3, such an event does not provide a successful connection to a user because of low received signal strength. As density increases to 1500 AP/km2, the probability of such an event approaches zero, and the MAP for a tagged AP approaches that of a random AP which is evident from Figure 4.

Figure 4: Comparison of MAP for a typical and a tagged AP against Wi-Fi RAT density for single-tier and multi-tier scenarios.
Figure 5: Probability that the distance from a typical user to a tagged AP is greater than approximated sensing radius of an AP.

Remark 7. Under dense network scenario, it is reasonable to approximate by , whereas by dense here we mean that the probability of no AP within the approximated sensing region approaches zero; hence, the required received signal strength for a successful connection is fulfilled across the region. This can be achieved by either increasing the transmission power of APs or density of the RAT.

4. Coverage

Under this section we cover the rest of the three performance metrics, namely, SINR coverage, rate coverage, and stable queue probability. The key factor which plays an important role for derivation of each of the mentioned metric is the Laplace transform of cumulative interference. We have assumed a multi-RAT HetNet scenario, where APs can access channel by using either contention-free or contention-based schemes; therefore the interference distribution vary under each RAT, and hence, the corresponding Laplace transform. Moreover, it is also important to consider if the user equipment can support multi-RAT connection. Thus, in this work we specifically focus on a two-RAT scenario, by assuming a cellular and a Wi-Fi RAT, each with -tiers such that the APs of tier-1 have maximum and tier- have minimum power of transmission. Please note that the framework is generalized and can be extended to more RATs.

4.1. SINR Coverage

Cellular RAT is deeply investigated in existing literature by using tools from stochastic geometry; therefore, we refer to [4, 13] for the Laplace transform of cumulative interference under cellular RAT. Due to contention-based nature of channel access in Wi-Fi RAT, it is hard to characterize the cumulative interference effect. As the distribution of interfering APs is non-independent thinning of , the Laplace transform of interference is not known in closed-form [10, 15]. Therefore, in [15] the set of interfering APs, under Wi-Fi RAT, is approximated by non-homogeneous PPP with certain density which has been defined by exploiting the conditional MAP and Bayes’ rule. On the other hand, in [26] the set of interfering APs has been approximated by a homogeneous PPP with density , and it has been assumed that the repulsion among APs is captured by , which is reasonable as per discussions in [10, 27]. Two main factors for capturing the cumulative interference effect are (1) the density of active APs and (2) the distance to those APs. In this work, similar to [26], we approximate the conditional MAP for an interfering AP by the conditional MAP of a tagged AP . As per an alternative definition, given in [10], the MAP represents the probability of successful simultaneous transmissions. This implies that if a tagged AP transmits, then on average the number of simultaneous transmissions, and hence, the number of active APs in a given region remain constant. Thus, we can approximate the set of interfering APs by a PPP with density . The other important factor, in modeling the interference effect, is the distance to nearest active AP. As heavy portion in interference is mainly contributed by the closest active APs, the distance to nearest interfering AP has been approximated by using some simple yet effective set of assumptions. The following lemma provides an approximated Laplace transform of cumulative interference for Wi-Fi RAT. Although our framework is based on a few approximations, it provides reasonable accuracy when compared with simulated and existing results.

Lemma 8. The Laplace transform of cumulative interference for Wi-Fi RAT with -tiers is approximated by where represents the mean sensing radius for a tier with lowest power of transmission and and are defined in (C.7) and (C.8), respectively.

Proof. Se Appendix C.

Following Lemma 8 and existing studies [4, 13], for Laplace transform of cumulative interference under cellular RAT, the SINR coverage for a typical user has been defined in the following theorem.

Theorem 9. The SINR coverage of a randomly located user under a multi-RAT HetNet, as defined in Section 2, is approximated by where , is the SINR threshold for the tier of the RAT, and

Proof. By following given definition (4) for SINR coverage, we getwhere is the result of deconditioning with respect to and assumption that , follows from an approximation for , and an assumption that for , follows from independent random variable , and is the Laplace transform of interference. We refer to existing results from [4, 13] for . By using Lemma 8, we get an approximated for Wi-Fi RAT, and the final expression (18) is obtained by using total probability theorem (6) which completes the proof.

Corollary 10. By assuming an interference-limited scenario, i.e., , with and , the SIR coverage of a randomly located user under a single-tier Wi-Fi RAT is given by where

Proof. Substituting given parameters in (18), performing some mathematical operations, and re-arranging variables proof the given corollary.

In Figure 6, the numerical results obtained through (18) are compared against the simulated ones for two single-tier (, ) and two multi-tier cases under Wi-Fi RAT. The simulation environment was created by randomly deploying APs of given density in a region of size 1 km × 1 km. The results were averaged over number of iterations, and under each iteration the SIR was evaluated for 2000 randomly chosen points. It must be clear that the approximated expression (18) is closely following the simulated results and provides a lower bound on coverage, which is according to discussions under Lemma 8 and Theorem 9. Although the interfering APs are very close to the tagged one, under high density regime, the distance between a user and tagged AP is also very less as compared to the sensing radius of APs; that is why in Figure 6, the numerical results provide an upper bound on SIR coverage for tier-2 as density of APs increases. Further in Figure 7, the numerical results are plotted for various network configurations including both standalone and multi-RAT HetNets. Standalone cellular and Wi-Fi RAT, each with two tiers, have been considered, where is providing a lower bound, and it is according to reported results [4, 13]. On the other hand, is providing better coverage as some of the APs are prohibited to transmit, because of the contention domains. The results for two multi-RAT HetNets are also reported, where in a macro-tier has been overlaid with a Wi-Fi tier and in a femto-tier is overlaid with a Wi-Fi tier . Although the power of tier-2 tier-3, the considered density for tier-3 tier-2 which reduces the MAP and, hence, improves the SIR coverage; that is why, all configurations which include tier-3 of Wi-Fi RAT are providing better coverage as compared to those with tier-2.

Figure 6: Comparison of numerical results with simulated ones for single-tier and two-tier Wi-Fi RAT only.
Figure 7: Numerical results for SIR coverage, under various network configurations, obtained through (18).
4.2. Rate Coverage

Under this section, in the following theorem, we derive the rate coverage probability of a randomly located user.

Theorem 11. The probability that a randomly located user, in a network setting as defined in Section 2, receives a rate greater than some defined threshold is approximated bywhere denotes expected load under the serving AP and

Proof. The proof simply follows from [14]; however, for readability the details are included in Appendix D.

Remark 12. The rate coverage is function of four parameters including rate threshold , average load under serving AP , MAP , and bandwidth . Under cellular RAT, the relation of rate coverage with the mentioned parameters can be explained with the help of the following expression

where . It must be clear that the rate coverage of a user under cellular RAT is directly proportional to and of the tier, whereas it is inversely proportional to and . In case of Wi-Fi RAT, by using an approximation , we getSimilar to cellular RAT, the rate coverage under Wi-Fi RAT is inversely proportional to and , and it is directly proportional to . Moreover, the rate coverage is indirectly proportional to the product and at the same time directly proportional to the negative exponent of it. For lower values of , the negative exponential effect dominates, and, therefore, the rate coverage increases. On the other hand, as the term approaches unity, and, hence, the rate coverage starts declining.

Remark 13. The rate coverage under Wi-Fi RAT is inversely proportional to and directly proportional to the negative exponent of it; please see (27). Therefore, for lower values of the term in denominator of (27) dominates, and, hence, the rate coverage improves. As , the term approaches unity, and, hence, the rate coverage starts declining. Thus, in either case the tiers operating at low power levels provide better rate coverage as compared to high power tiers. Equivalently, we can also conclude that the rate coverage increases as a function of sensing threshold .

In Figure 8 the numerical results obtained through (25) have been plotted by considering network configurations similar to those of Figure 7. It must be noted that, in Figure 7, the SIR coverage was slightly affected by the changes in configuration as compared to the rate coverage, in Figure 8, which is significantly varying for various network configurations. The reason behind such a result is the dependence of rate coverage over four different parameters, as clear from Theorem 9 and Remark 12. Moreover, for all those configurations the rate coverage is high which include tier-3 of Wi-Fi RAT, because of its high density and lower power of transmission; please see Remarks 12 and 13 for further details. In Figure 9 the rate coverage for different network configurations has been plotted and the results are in accordance with Remarks 12 and 13. The rate coverage increases for standalone cellular RAT; however, for Wi-Fi RAT it initially increases and then it starts declining. Similarly, under multi-RAT case as the user association with Wi-Fi RAT exceeds the cellular RAT, the rate coverage starts declining. Moreover, the rate of low power Wi-Fi tier is better than high power tier which is in accordance with Remark 13.

Figure 8: Numerical results for rate coverage, under various network configurations, obtained through (25).
Figure 9: Rate coverage as a function of Wi-Fi RAT density or, in case of standalone cellular RAT, it is function of cellular RAT density.
4.3. Stable Queue

Most of the existing studies assume a saturation model for traffic which do not capture the randomness introduced by the temporal domain. In few of the recent works [24, 25, 28], both the temporal and spatial domains have been analyzed by exploiting tools from queuing theory and stochastic geometry. In [25, 29], the conditions for a network to be stable have been derived by assuming a dominant and a modified system. In [28], the probability for a user queue to be unstable has been derived by assuming a Poisson and a uniform distribution for arrival rate of packets, where PPP and Poisson cluster process (PCP) have been used for the distribution of APs across a given region. In all of the aforementioned works, single-tier cellular RAT and a downlink channel have been assumed. As the PPP realization is random and irregular, there are some APs with good and others with poor transmission environment, resulting in some users near APs with good experience and others at the edge under outage [25]. In [30], the outage probability has been derived as a function of distance from a user to the tagged AP, and it has been shown that the outage increases, as the distance increases. By exploiting the given concepts mainly from [25, 29, 30], we derive the stable queue probability for a user under an AP of a given RAT.

In this section, for simplified analysis, we follow a different set of assumptions [24, 28]. We assume standalone single-tier cellular and Wi-Fi RAT, and an interference-limited scenario; i.e., , and . A non-saturated traffic model has been considered, where packets arrive at a user with probability during a time-slot. Further, we assume that represents the probability that an AP is active during a time-slot. For avoiding interacting queues problem, similar to [24, 29], we assume a dominant and a modified system. Under a dominant system, the interfering APs have full buffers and transmit continuously, i.e., , whereas, under modified system, the interfering APs are active with probability ; the packets not delivered successfully are, hence, assumed to be dropped. With the aforementioned assumptions, the following theorem provides the probability that a user queue is stable.

Theorem 14. The stable queue probability of a user under a single-tier cellular or a Wi-Fi RAT, with a packet arrival rate of , is given by respectively, where

Proof. e Appendix E.

Remark 15. From the given condition (E.5), for a stable queue of a user under an AP of Wi-Fi RAT, it is clear that the MAP for an AP must be greater than the arrival rate of packets during a time-slot. Hence, (30) is valid only when .

By assuming a dominant and a modified system, the numerical results for stable queue probability have been reported in Figures 10, 11, and 12 against different parameters of interest. The dominant system in each result is providing a lower bound, whereas the modified system is providing an upper bound [24, 25]. It must be clear from Figure 10 that the stable queue probability for a user under Wi-Fi RAT is slightly better than cellular RAT when the packet arrival rate is low. As increases, decreases and eventually approaches zero when , which is in accordance with Remark 15; please see Figures 10 and 12 for clarification. Moreover, it must also be noted that the decay in as a function of is faster as compared to in Figure 11 and in Figure 12, which is in agreement with the results reported in [29]. as a function of AP density is constant for cellular RAT, because, under interference-limited scenario with , the SIR coverage becomes independent of density of the RAT [4]. On the other hand, under Wi-Fi RAT first decreases because decreases as a function of . After that it increases slightly as the probability of distance between a user and its tagged AP approaches zero; hence, the second indicator function in (29) becomes active; as all other factors are constant, thus increase in results in an increase in . Finally, when the distance between a user and its tagged AP is , the very first indicator function in (29) becomes active, and, hence, starts declining and finally approaches zeros as .

Figure 10: Stable queue probability as a function of packet arrival rate , by assuming a dominant and a modified systems.
Figure 11: Stable queue probability as a function of SIR threshold , by assuming a dominant and a modified system.
Figure 12: Stable queue probability as a function of AP density, by assuming a dominant and a modified system.

5. Numerical Results and Discussions

Under this section, various numerical results for different performance metrics have been discussed. An interference-limited scenario with has been assumed for all RAT-tier pairs . The parameters have been carefully chosen by considering dense HetNet scenario [18, 31] and summarized in Table 2. In general, if not specified, the parameters mentioned in Table 2 have been used for all the results reported in this paper.

The association probability, as a function of Wi-Fi RAT density, for multi-RAT HetNets has been plotted in Figure 13. Initially, most of the users are associated with cellular RAT and as the density of Wi-Fi RAT increases, the user association increases. For a two-RAT scenario, each with single-tier as assumed for Figure 13, the AP density at which the association probability of Wi-Fi RAT becomes equal to the cellular RAT can be obtained by the following relation:For the case when , in Figure 13, the power of Wi-Fi and cellular tier is the same; i.e., . However, in order to get equal association, i.e., , the required as , and this is evident from (31). In Figure 14, the SIR coverage of two different HetNets has been analyzed against Wi-Fi tier density . When , most of the users are associated with cellular RAT, as clear from Figure 13, and the overall SIR coverage of multi-RAT HetNet becomes equal to the single-tier cellular RAT , which is function of the chosen thresholds only . According to the results of Theorem 9 as increases, the association of users with Wi-Fi RAT increases and, hence, the coverage. On the other hand, according to [14], the SIR coverage keeps on decreasing and at last it meets ; as the same thresholds are used (i.e., ), , each denoting the SIR coverage of standalone cellular tiers (macro, femto) and the Wi-Fi tier . It is because of the fact that the framework given in [14], for multi-RAT HetNets, does not capture the effect of different channel accessing schemes. Thus, addition of a new RAT is simply another cellular RAT which operates on a different pool of resources; hence, it does not cause interference to existing RATs. The proposed framework, in this work, captures the effect of both the contention-free and the contention-based channel accessing schemes; therefore, it provides generalization and ease of analysis for various network configurations.

Figure 13: Association probability as a function of Wi-Fi RAT density.
Figure 14: SIR coverage as a function of Wi-Fi RAT density when overlaid with macro- or femto-tier.

In Figure 15, the SIR coverage has been analyzed against sensing threshold and . By increasing , the SIR coverage decreases, because of the increase in density of active APs . It must also be noted that, after a certain sensing threshold, the SIR coverage becomes almost constant as , . Similarly, in Figure 16 the rate coverage has been analyzed against and . Initially, the rate coverage improves by increasing because the density of active APs increases, and, hence, the average load per AP decreases. After a certain limit, it becomes constant as . Please see Remark 13, for an alternative and detailed description of the results reported in Figure 16. The rate coverage has been analyzed against users density and bandwidth of Wi-Fi RAT, in Figure 17, which shows that the increase in greatly affects the rate coverage. Apart from that, as increases, the rate coverage decreases because the average load per AP increases.

Figure 15: SIR coverage as a function of sensing threshold and AP density.
Figure 16: Rate coverage as a function of sensing threshold under various user and AP density.
Figure 17: Variation in rate coverage as a function of bandwidth of Wi-Fi tier and AP density.

In Figure 18, an interesting result has been reported by keeping the density ratio of users and APs constant. Although the average load per AP has been kept fixed, the rate coverage declines as the density increases and the sensing threshold decreases. This is due to the fact that the rate coverage depends on four factors which include both the average load and the MAP of a serving AP. By increasing the AP density and reducing the sensing threshold, under a constant load, the MAP decreases; hence, the overall rate coverage declines. Please see Remark 12 for further details. Under such situations, increasing the number of non-overlapping channels can improve the rate coverage.

Figure 18: Rate coverage against constant user to AP density ratio when femto-tier is overlaid with Wi-Fi tier.

6. Conclusion

In this paper, we have proposed a unified framework for SINR and rate coverage analysis of multi-RAT HetNets by considering different channel access schemes. By assuming a multi-tier Wi-Fi RAT, we have derived the MAP for a random and a tagged AP, where the results show that the MAP for a typical AP approaches that of a tagged AP as density of Wi-Fi RAT approaches . It has been shown that, by increasing the density of Wi-Fi RAT, the SIR coverage of multi-RAT HetNet increases and gradually approaches unity. Moreover, multi-RAT HetNets, specifically with small cell tiers, provide better SIR coverage; however, as the user association with Wi-Fi RAT increases, the rate coverage starts declining. We have also derived the stable queue probability of a user under cellular and Wi-Fi RAT by assuming a non-saturated traffic model. The results show that the stable queue probability of a user under cellular RAT is better as compared to Wi-Fi RAT, when packet arrival rate is high. Although Wi-Fi RAT provides better SIR coverage, it is hard to maintain the stability of a queue, as the medium access probability of an AP is less than unity. This result suggests that the un-bounded increase in the density of Wi-Fi RAT cannot bring significant improvement in users experience; hence, care must be taken while planning the deployment of Wi-Fi RAT.

Recently, research on ultra-dense small cell networks has received significant attention. Various tools and techniques like multi-slope path loss models, LOS and non-LOS channels, and different shadowing effects have been used to provide new insights. However, such investigations are limited to single-tier, single-RAT scenario. A straightforward extension of the proposed work is to incorporate such tools for the analysis of multi-RAT HetNets. Another potential area for future work is the incorporation of queuing theory evaluating the impact of traffic variations on the performance bounds of multi-RAT HetNets.

Appendix

A. Proof of Lemma 1

The proof is an extension of existing studies [10, 15]. The MAP of an AP is the Palm probability that its medium access indicator is 1. Given the timer of a typical AP , the MAP can be derived aswhere follows from small-scale fading which is exponentially distributed with mean unity and the fact that the received signal strength from APs with timers less than is of concern, follows from Slyvniak’s theorem and the probability generating functional (PGFL) of homogeneous PPP, and, finally, we get (11) by deconditioning with respect to “”, where .

B. Proof of Lemma 5

Association of users based on the maximum average received signal strength has been considered in this work. Given that the tagged AP, belonging to the tier of Wi-Fi RAT, is located at , then the MAP can be given aswhere , follows from deconditioning with respect to , and is based on PGFL of PPP and cosine rule; the PPP has been translated in such a way that the tagged AP is located at origin; for further details please refer to emma 2 in [26]. As shown in Figure 19, due to multi-tiers and association based on the maximum average received signal strength, it is possible that the tagged AP is not the nearest one. However, it is the closest among APs of the tier to which it belongs to. Thus, distance, from a user to the tagged AP, has been defined for properly locating the exclusion region around the user which does not include any other AP. This completes the proof and we get the final result (15).

Figure 19: Illustration for the scaling of distance in order to obtain the radius of circle around the user, when there is not any interfering AP.

C. Proof of Lemma 8

For simplification here we drop the notation “'' which is used to denote the Wi-Fi RAT. By following the given definition for cumulative interference under Section 2.2.2, we getwhere follows from the independence of and due to PPP assumption for the set of interfering APs, where , is obtained using PGFL of PPP, and is obtained through Laplace transform of exponential random variable with unit mean. By assuming , the simplified expressions are obtained. Moreover, for compact representation, a general expression given in [14] has been used asHence,whereandWe have approximated the distance to nearest interfering AP, as given in (C.7) and (C.8), by using simple yet effective set of assumptions. For better illustration, let us assume a two-tier scenario as shown in Figure 20, where “’’ represents the mean sensing radius for respective tiers and “'' denotes the distance from a user to the tagged AP. Here, the mean sensing radius has been obtained by using (12). Due to contention domains, we assume that not any AP is allowed to transmit within an approximated region of mean sensing radius around the tagged AP, which provides a lower bound on the expected number of contending APs, as discussed under Remark 4. The approximation is reasonable as the nearest active AP can severely degrade the signal by causing excessive interference. Further, as clear from (C.6), based on the distance from a user to the tagged AP, two different cases have been considered, where the mean sensing radius of APs with minimum power level is exploited as a reference. Due to -tiers, the tagged AP may not be the nearest one; however, it is the closest among APs of the tier to which it belongs to. That is the reason we are using as a reference for defining two cases in (C.6).

Figure 20: Illustration for approximated distance to the nearest interfering AP, under multi-tier Wi-Fi RAT.

In part of Figure 20, a user is associated with an AP of tier having minimum power of transmission such that . Within approximately distance around the tagged AP, there cannot be any other active AP. Therefore, the nearest interfering AP of any tier is at least distance apart from the user. Further, in part a user is associated with an AP of a tier having higher power of transmission such that . Under such situation, due to differences in power levels, the APs of tiers with power less than the tagged AP can be closer to the user. Therefore, by exploiting the tier as a reference, a generalized formula for approximating the distance to nearest interfering AP of any tier has been obtained as . When interfering AP belongs to the tier, the expression simplifies to . Furthermore, for , assuming that , the expression simplifies to which is approximately equivalent to the nearest interfering AP, as clear from part of Figure 20. It must be noted that the given formula is generalized enough and applicable to part as well.

If the distance between a user and the tagged AP , then we assume that the distance to nearest interfering AP is simply function of association [14] and is given by (C.8). This approximation provides an upper bound on interference as some of the interfering APs, within expected sensing region of the tagged AP, may not detect its presence due to random fading effects. Hence, the supposed approximations are tight and provide an upper bound on interference for Wi-Fi RAT and this completes the proof.

D. Proof of Theorem 11

As defined in (7), the probability that a typical user receives a rate greater than some defined threshold from the tagged AP iswhere , is given in (8), and is the load under serving AP. It must be noted here that the rate coverage is function of rate threshold , load under serving AP , transmission probability , and bandwidth of the AP. By increasing or and decreasing or , the rate coverage improves. However, in case of Wi-Fi RAT higher and lower cannot be achieved at the same time. As for higher , lower density of Wi-Fi RAT is required, whereas for lower , higher density of RAT is required. By using emma 3 of [32], the probability mass function (PMF) for number of users, other than the typical user, under a tagged AP can be given aswhere is a constant, and the load under serving AP is given as . By following a procedure similar to [14], we use an approximation , where the expected load under a serving AP is given as . Finally, simplification of (D.7) completes the proof.

E. Proof of Theorem 14

By assuming that single user is connected to each AP of a RAT [25], the conditional SIR, or equivalently the service rate, of a typical user at distance from the tagged AP has been defined in [30] asPlease note that is a random variable, as it is conditioned on a particular PPP realization ; therefore it can be analyzed through a statistical distribution [24, 25]. In order to obtain a simplified solution, by following an approach similar to [28], we approximate the service rate for cellular RAT by (E.2) and for Wi-Fi RAT by (E.3); however, the presented work can be extended by following the given approaches in [24, 25].Assuming that packet arrives at a user with rate during a time-slot, then, on average, for a queue to be stable under a cellular RAT, the minimum required service rate is given byand, for Wi-Fi RAT, it is given byThis implies that, under cellular RAT, when the distance between a user and its tagged AP obeys the relationthen the queue is stable given that the packet arrival rate is . Thus, by exploiting the void probability [4], we obtain the probability that the distance between a user and its tagged AP is less than asor, equivalently, it can be interpreted as the probability that the queue of a user under cellular RAT is stable; as it is within a critical distance from the tagged AP. For Wi-Fi RAT, depending on the distance with respect to expected sensing radius of APs , the distance to the nearest interfering AP changes and, hence, the service rate. Thus, by using (E.3) and (E.5), we obtain the following relation:where when ; otherwise . Further, by exploiting the void probability [4], the stable queue probability of a user under a Wi-Fi RAT, depending on the distance with respect to expected sensing radius of APs , can be given asAs a result, with the help of indicator function, we obtain the final expression (29).

Data Availability

The data used to support the findings of this study are included within the article.

Conflicts of Interest

The authors declare that they have no conflicts of interest.

References

  1. Cisco, “Global mobile data traffic forecast update, 2016-2021,” 2017.
  2. J. G. Andrews, S. Buzzi, W. Choi et al., “What will 5G be?” IEEE Journal on Selected Areas in Communications, vol. 32, no. 6, pp. 1065–1082, 2014. View at Publisher · View at Google Scholar
  3. A. Aijaz, H. Aghvami, and M. Amani, “A survey on mobile data offloading: technical and business perspectives,” IEEE Wireless Communications Magazine, vol. 20, no. 2, pp. 104–112, 2013. View at Publisher · View at Google Scholar · View at Scopus
  4. J. G. Andrews, F. Baccelli, and R. K. Ganti, “A tractable approach to coverage and rate in cellular networks,” IEEE Transactions on Communications, vol. 59, no. 11, pp. 3122–3134, 2011. View at Publisher · View at Google Scholar · View at Scopus
  5. H. S. Dhillon, R. K. Ganti, F. Baccelli, and J. G. Andrews, “Modeling and analysis of K-tier downlink heterogeneous cellular networks,” IEEE Journal on Selected Areas in Communications, vol. 30, no. 3, pp. 550–560, 2012. View at Publisher · View at Google Scholar · View at Scopus
  6. D. B. Taylor, H. S. Dhillon, T. D. Novlan, and J. G. Andrews, “Pairwise interaction processes for modeling cellular network topology,” in Proceedings of the 2012 IEEE Global Communications Conference, GLOBECOM '12, pp. 4524–4529, December 2012. View at Scopus
  7. A. Guo and M. Haenggi, “Spatial stochastic models and metrics for the structure of base stations in cellular networks,” IEEE Transactions on Wireless Communications, vol. 12, no. 11, pp. 5800–5812, 2013. View at Publisher · View at Google Scholar · View at Scopus
  8. N. Deng, W. Zhou, and M. Haenggi, “The ginibre point process as a model for wireless networks with repulsion,” IEEE Transactions on Wireless Communications, vol. 14, no. 1, pp. 107–121, 2015. View at Publisher · View at Google Scholar · View at Scopus
  9. Y. Li, F. Baccelli, H. S. Dhillon, and J. G. Andrews, “Statistical modeling and probabilistic analysis of cellular networks with determinantal point processes,” IEEE Transactions on Communications, vol. 63, no. 9, pp. 3405–3422, 2015. View at Publisher · View at Google Scholar · View at Scopus
  10. H. Q. Nguyen, F. Baccelli, and D. Kofman, “A stochastic geometry analysis of dense IEEE 802.11 networks,” in Proceedings of the 26th IEEE International Conference on Computer Communications, INFOCOM '07, pp. 1199–1207, IEEE, May 2007. View at Scopus
  11. G. Alfano, M. Garetto, and E. Leonardi, “New directions into the stochastic geometry analysis of dense CSMA networks,” IEEE Transactions on Mobile Computing, vol. 13, no. 2, pp. 324–326, 2014. View at Publisher · View at Google Scholar · View at Scopus
  12. R. Giacomelli, R. K. Ganti, and M. Haenggi, “Outage probability of general ad hoc networks in the high-reliability regime,” IEEE/ACM Transactions on Networking, vol. 19, no. 4, pp. 1151–1163, 2011. View at Publisher · View at Google Scholar · View at Scopus
  13. H.-S. Jo, Y. J. Sang, P. Xia, and J. G. Andrews, “Heterogeneous cellular networks with flexible cell association: a comprehensive downlink SINR analysis,” IEEE Transactions on Wireless Communications, vol. 11, no. 10, pp. 3484–3494, 2012. View at Publisher · View at Google Scholar · View at Scopus
  14. S. Singh, H. S. Dhillon, and J. G. Andrews, “Offloading in heterogeneous networks: modeling, analysis, and design insights,” IEEE Transactions on Wireless Communications, vol. 12, no. 5, pp. 2484–2497, 2013. View at Publisher · View at Google Scholar · View at Scopus
  15. Y. Li, F. Baccelli, J. G. Andrews, T. D. Novlan, and J. C. Zhang, “Modeling and analyzing the coexistence of Wi-Fi and LTE in unlicensed spectrum,” IEEE Transactions on Wireless Communications, vol. 15, no. 9, pp. 6310–6326, 2016. View at Publisher · View at Google Scholar · View at Scopus
  16. C.-H. Liu and H.-C. Tsai, “On the limits of coexisting coverage and capacity in multi-RAT heterogeneous networks,” IEEE Transactions on Wireless Communications, vol. 16, no. 5, pp. 3086–3101, 2017. View at Publisher · View at Google Scholar · View at Scopus
  17. J. G. Andrews, X. Zhang, G. D. Durgin, and A. K. Gupta, “Are we approaching the fundamental limits of wireless network densification?” IEEE Communications Magazine, vol. 54, no. 10, pp. 184–190, 2016. View at Publisher · View at Google Scholar · View at Scopus
  18. B. Yang, G. Mao, M. Ding, X. Ge, and X. Tao, “Dense small cell networks: from noise-limited to dense interference-limited,” IEEE Transactions on Vehicular Technology, 2018. View at Google Scholar · View at Scopus
  19. M. Ding, D. Lopez-Perez, G. Mao, and Z. Lin, “Performance impact of idle mode capability on dense small cell networks,” IEEE Transactions on Vehicular Technology, vol. 66, no. 11, pp. 10446–10460, 2017. View at Publisher · View at Google Scholar · View at Scopus
  20. B. Yang, G. Mao, X. Ge, M. Ding, and X. Yang, “On the energy-efficient deployment for ultra-dense heterogeneous networks with NLoS and LoS transmissions,” IEEE Transactions on Green Communications and Networking, vol. 2, no. 2, pp. 369–384, 2018. View at Publisher · View at Google Scholar
  21. H. Elshaer, M. N. Kulkarni, F. Boccardi, J. G. Andrews, and M. Dohler, “Downlink and uplink cell association with traditional macrocells and millimeter wave small cells,” IEEE Transactions on Wireless Communications, vol. 15, no. 9, pp. 6244–6258, 2016. View at Publisher · View at Google Scholar · View at Scopus
  22. G. Ghatak, A. De Domenico, and M. Coupechoux, “Coverage analysis and load balancing in HetNets with millimeter wave multi-RAT small cells,” IEEE Transactions on Wireless Communications, vol. 17, no. 5, pp. 3154–3169, 2018. View at Google Scholar
  23. F. Boccardi, R. W. Heath, A. Lozano, T. L. Marzetta, and P. Popovski, “Five disruptive technology directions for 5G,” IEEE Communications Magazine, vol. 52, no. 2, pp. 74–80, 2014. View at Publisher · View at Google Scholar
  24. Y. Zhong, T. Q. S. Quek, and X. Ge, “Heterogeneous cellular networks with spatio-temporal traffic: delay analysis and scheduling,” IEEE Journal on Selected Areas in Communications, vol. 35, no. 6, pp. 1373–1386, 2017. View at Publisher · View at Google Scholar · View at Scopus
  25. H. H. Yang and T. Q. Quek, “SIR coverage analysis in cellular networks with temporal traffic: a stochastic geometry approach,” 2018, https://arxiv.org/abs/1801.09888.
  26. X. Wang, T. Q. S. Quek, M. Sheng, and J. Li, “Throughput and fairness analysis of Wi-Fi and LTE-U in unlicensed band,” IEEE Journal on Selected Areas in Communications, vol. 35, no. 1, pp. 63–78, 2017. View at Google Scholar · View at Scopus
  27. M. Haenggi, “Mean interference in hard-core wireless networks,” IEEE Communications Letters, vol. 15, no. 8, pp. 792–794, 2011. View at Publisher · View at Google Scholar · View at Scopus
  28. Y. Zhong, G. Wang, R. Li, T. Han, X. Ge, and T. Q. Quek, “Effect of spatial and temporal traffic statistics on the performance of wireless networks,” 2018, https://arxiv.org/abs/1804.06754.
  29. Y. Zhong, M. Haenggi, T. Q. S. Quek, and W. Zhang, “On the stability of static poisson networks under random access,” IEEE Transactions on Communications, vol. 64, no. 7, pp. 2985–2998, 2016. View at Publisher · View at Google Scholar · View at Scopus
  30. H. ElSawy, A. Sultan-Salem, M.-S. Alouini, and M. Z. Win, “Modeling and analysis of cellular networks using stochastic geometry: a tutorial,” IEEE Communications Surveys & Tutorials, vol. 19, no. 1, pp. 167–203, 2017. View at Publisher · View at Google Scholar · View at Scopus
  31. D. López-Pérez, M. Ding, H. Claussen, and A. H. Jafari, “Towards 1 Gbps/UE in cellular systems: understanding ultra-dense small cell deployments,” IEEE Communications Surveys & Tutorials, vol. 17, no. 4, pp. 2078–2101, 2015. View at Publisher · View at Google Scholar · View at Scopus
  32. S. M. Yu and S.-L. Kim, “Downlink capacity and base station density in cellular networks,” in Proceedings of the 11th International Symposium on Modeling and in Mobile, Ad Hoc and Wireless Networks (WiOpt '13), pp. 119–124, IEEE, 2013.