Abstract
Cellular networks are extensively modeled by placing the base stations on a grid, with relays and destinations being placed deterministically. These networks are idealized for not considering the interferences when evaluating the coverage/outage and capacity. Realistic models that can overcome such limitation are desirable. Specifically, in a cellular downlink environment, the full-duplex (FD) relaying and destination are prone to interferences from unintended sources and relays. However, this paper considered two-hop cellular network in which the mobile nodes aid the sources by relaying the signal to the dead zone. Further, we model the locations of the sources, relays, and destination nodes as a point process on the plane and analyze the performance of two different hops in the downlink. Then, we obtain the success probability and the ergodic capacity of the two-hop MIMO relay scheme, accounting for the interference from all other adjacent cells. We deploy stochastic geometry and point process theory to rigorously analyze the two-hop scheme with/without interference cancellation. These attained expressions are amenable to numerical evaluation and are corroborated by simulation results.
1. Introduction
Currently, the demand for the high capacity and low latency is dramatically increasing, corresponding to advancements in communication devices such as smartphones. It is expected that the wireless traffic volume of these communication devices will have a 1000-fold increase over the next decade which will be driven by the expected billions of connected devices by 2020 to access and share data anywhere and anytime [1]. The bottom line is that the technology has to move to higher frequency bands of millimeter wave to adapt the 5G future mobile communications, where a very large bandwidth in frequency band is available. With rapid increase in the number of connected devices, some challenges appear such as the capacity shortage, cost, and cochannel, and intercell interference especially at high dense network is rapidly increased [2]. Hence, the fifth-generation (5G) wireless systems should be able to support the ultradense networks to adopt the exponential growth in mobile users and high data demand. To satisfy the quality of service (QoS), it is expected that the 5G network deployment will be much denser compared to that of 4G. However, as the network density increases, the interference will degrade the system’s performance especially at cell edges (dead zone). One of these efficient solutions that allow the 5G network to meet its QoS requirements is to use relay to improve the ultradense network performance (capacity enhancement and coverage extension). Cooperative communication is an alternative way to achieve spatial diversity and multiplexing gain through Multiple-Input Multiple-Output (MIMO). MIMO attracted a lot of attention due to its potential for interference mitigation and capacity increase. FD relaying equipped with MIMO is able to perform spatial self-interference suppression [2–5]. The FD relaying looks to be an alternative solution to satisfy the high capacity demands of these wireless systems. The FD communication has attracted considerable attention of many researchers and it is expected to be integrated into the 5G wireless systems. Most used cooperative protocols are the amplify-and-forward (AF) and the decode-and-forward (DF); the DF cooperative protocols can operate in either a half-duplex (HD) or full-duplex (FD) mode. Multiplexing loss occurred when implementing the HD mode in the DF protocol; this is attributed to the fact that, in the first time/frequency slot, the relay has to wait for the source’s message and then forward the message in the next time slot to the destination. However, FD cooperative protocols can overcome the HD spectral loss via simultaneous transmission from source to relay and from relay to destination; this enables frequency reuse, higher throughput, and lower transmission delay [6]. However, there are two main issues hardening the implementation of FD system in cellular network: the self-interference, that is, the relay’s signal leakage from its transmit and receive antennas, and the fact that the simultaneous transmission creates intercell interference [7].
However, cellular networks are usually modeled by placing the sources on a grid (with a regular shape) or arranging them on a line or circle as in the Wyner model with the relay and destination being either deterministically or randomly placed across the network to determine the signal-to-interference-and-noise ratio (SINR). The resulting SINR is complex and depends on multiple random variables. Hence, this fails to account for the randomness in the cellular network distribution [8] and the intercell interference [9]. Such models, however, are highly idealized and not tractable; hence, complex system level simulation is used to evaluate the outage/success probability and ergodic capacity. In order to reduce the dependence on simulations, the closed-form SINR was derived using stochastic geometry [10].
Since characterizing the SINR by the grid model and Wyner model is obviously not practicable, recently stochastic geometry has emerged as a powerful tool to model and quantify the capacity, interference, and success probability in cellular networks that are verified to be approximate to the actual networks [11]. The use of the Poisson point processes (PPP) model simplified the analysis and provided insight into the operation of the network in the form of scaling laws. Base station, relay, and destination parameters (e.g., path-loss exponent and transmit power) become the sign of the node in the PPP. Recently, under homogeneity condition, it was shown that the source positions are agnostic to the radio propagation; this makes the received power at the relay from any population of sources as if it generated from PPP, distributed sources [12, 13]. This justifies the modeling assumption of PPP sources and allows computing some metrics performance such as the success probability and Ergodic capacity [14, 15], while [16] derived an upper bound for the success probability. Obtaining full diversity order using distributed space-time codes is detailed in [17, 18], but a distributed space-time code requires precise signaling and very tight coordination among the relays, which increases the complexity and overhead in the system. This paper is motivated by the benefit of MIMO two-hop system, such as performance benefits and reducing implementation complexity.
Researchers have dealt with the interference in many different ways. For instance, an active method known as analog cancellation, to cancel the interfering signal at the receiving antenna, utilizes additional radio frequency (RF) chains; and there is another active cancellation method known as digital cancellation, where the RSI is removed in the base-band level after the analog-to-digital converter [19]; another simple passive method is known as antenna separation, where the RSI is attenuated due to the path-loss between the transmitting and receiving antennas on the FD node. Further, reducing the FD interference using directional antennas is analyzed in [20]. References [21, 22] quantify the impact of self-interference of a heterogeneous network consisting of FD and HD nodes. They conclude that the capacity can be maximized by operating all nodes in either HD or FD compared to their mixtures, whereas [23] considers a single-cell setting and [22, 24] consider multicell setting. A cellular system comprising an FD source and HD destination has been illustrated in which the throughput gain is analyzed via extensive simulation [25]. In [26], the numbers of base station antennas and users antennas are increasing with a fixed ratio, while the capacity grows by the number of users and SNR. In this paper, however, we use a stochastic geometry tool to characterize the randomly distributed performance of FD MIMO relay nodes and derive bounds for the probability of successful transmission and ergodic capacity with/without interference cancellation. Finally, numerical results corroborate the theoretical findings with baseline scheme.
Next, in Section 2, the network system model is described. The SINR characterization is derived in Section 2.3; the success probability and the ergodic capacity are given in Section 2.4 and Section 2.5, respectively. Section 4 depicts the numerical results, and Section 5 offers the conclusion.
Throughout this paper, boldface lowercase letters (e.g., ) represent vectors, and boldface uppercase letters (e.g., ) represent matrices. is the Frobenius norm. stands for conjugate transpose, and denotes the expectation operator, while denotes the magnitude and the trace of a matrix is denoted by ; is the determinant; is the diagonal matrix with diagonal components ; is the -by- identity matrix; is the -by- zero matrix.
2. System Model
2.1. System Setup
In a downlink relay-assisted cellular network, consider a multiple independent FD dual-hop relaying system. The th source with message equipped with antennas communicates with destinations receiving antennas through relay set , with messages , equipped with and transmitting and receiving antennas, respectively, as in Figure 1. The relay uses DF protocol; hence, decodes the message from and then reencodes the message before sending it to . The relays are equipped with FD capabilities, where the reception and transmission of relay signal happen simultaneously. The main challenge is that relay’s transmitted signal is coupled with its receiver chain, causing relay self-interference (RSI) via channel and source causes interference to other relay known as source-relay interference (SRI) via channel in addition to the interference from the sources to the destinations causing source-destination interference (SDI) via . Space division multiplexing is applied so that the two hops are separated.
The locations of sources, relays, and destinations are assumed to be stationary independently marked PPP with intensity , , and on , respectively. In order to investigate the performance of a random access wireless network, consider a classical receiver located at the origin. Conditioning on the node’s event lying at the origin does not affect the statistics of the rest of the process as an outcome from Palm probabilities of a Poisson process. Furthermore, the statistics of received signal at the intended receiver are observable by any receiver due to the stationarity of Poisson process [10].
2.2. Channel Model
The downlink cellular network is comprised of sources and , relays and, and destinations and. Previous studies on cellular networks assumed that the sources and relays are positioned regularly as grid model. However, in practice, this is not true and there are some random characteristics. We applied homogeneous PPP to the spatial distribution of the sources and relays to remedy the model as in [11, 15]. Each destination is served by the nearest relay or destination. This means that the cell area of each source or relay constitutes a Voronoi tessellation [27] as in Figure 2, whereas, in grid model in Figure 3, the sources are located on the centers of hexagonal model, while the relays and destinations are distributed in each cell uniformly. This model does not lead to a tractable model; thus, it is evaluated via Monte Carlo simulations. When a destination attempts to access the sources, it chooses to connect directly or through relay in the Voronoi cell. Let the source transmit the message to the th relay ; receives and transmits simultaneously on the same frequency causing RSI through the channel if for the same th relay and receives IRI through the channel if from the other th relay , in addition to interference from other source . and are transmitted signals from and nodes respectively.
The propagation through two-hop wireless channel is subject to path-loss attenuation modeled as and for distances and , respectively, with path-loss exponent > 2. The sources and relays transmit with power and power . For such channel, the sources cover large areas, while relays cover small areas with separation distances and given if and if . The interfering relays constitute the marked process , with denoting the location of the transmitting relay node. Hence, the relay receives interference power from itself, , and from the th interfering relay and source as and , respectively. The channels , , , , and are the channel gain matrices from source-to-relay, relay itself, interrelay, relay-to-destination, and source-to-destination, respectively. The power constraints on transmit signals are , , and . and are received signals at and nodes. All channels experience Rayleigh fading with parameter . The entries of each matrix are independent and identically distributed (i.i.d.) complex Gaussian variables with zero mean and variance . and are independent circularly symmetric complex Gaussian noise vectors with probability distributions and , respectively, and are uncorrelated to and .
2.3. SINR Characterization
In this section, we formulate the signal-to-interference-plus-noise ratio (SINR), which is needed in the next section to analyze the probability of successful transmission. In doing so, we study the SINR of the SR hop and the SINR of the RD hop.
2.3.1. SR Hop
Consider the th source node located at the origin and, according to the stationarity of and building on Slivnyak’s theorem [10, Ch. 8.5], we hence express the statistics of relay-received signal , with being the distance of from the origin. Therefore the received signal at the th relay node contains the desired signal plus interference and noise ratio, RSI, and SRI, as given as follows:The resulting SINR at the relay is obtained aswhere is the overall interference (RSI and SRI) at th relay; that is,For simplicity, we assume that this hop is interference-limited and, therefore, .
2.3.2. RD Hop
Consider the th relay node located at the origin and, according to the stationarity of and building on Slivnyak’s theorem, we express , with being the distance of from the origin. Therefore the received signal at the th destination node contains the desired signal plus interference and noise ratio, as shown as follows:The resulting SINR at the destination iswhere captures the interference from other source to destination for the RD hop:For simplicity, we assume that this hop also is interference-limited and therefore . Furthermore, the mean link distance of and is and , respectively, such that .
2.4. Success Probability
One of our performance measures is the success probability, which, defined as the success probability of message transmission from the source to destination over the two hops, is given by a joint complementary cumulative distribution function (CCDF) of and . We formulate it by , where is the given SINR threshold. Due to independent sampling of a point process, there is no correlation between the two hops.where and .
2.4.1. SR Hop
Successful SR hop transmission occurs ifthe given threshold satisfied a target SINR; the probability of successful transmission for a relay iswhere (9) is obtained by conditioning and its complementary cumulative distribution function is denoted by . Next, we provide the success probability analysis of the SR hop.
Theorem 1. Let the interfering relays transmitters form a Poisson process of intensity around relay receivers. The success probability of the SR hop iswhere is the Laplace transform of the interferences .with
Proof. Refer to Appendix B.
The result in Theorem 1 provides a fundamental limit on the SR hop and its performance in an interference-limited scenario.
2.4.2. RD Hop
Successful RD hop transmission occurs ifNext, we provide the success probability analysis of the RD hop.
Theorem 2. Let the relays transmission form a Poisson process of intensity around destinations. The success probability of the RD hop iswhere is the Laplace transform of the interference :where and are already defined in (13) and (12), respectively.
Proof. Refer to Appendix C.
Corollary 3. The Laplace transform in (16) is bounded by , wherewith and already being given in [21, 28] as follows:
This corollary will help us in proving the interference bound of success probability of RD hop; this can be done by substituting in (15).
2.5. Ergodic Capacity
The other performance measure is the ergodic capacity; the goal of FD relay is to increase the network capacity; for the relay served by source, the following theorems give the outage probability (CDF of SINR) and the SR hop capacity.
Theorem 4. The probability that falls below a given target level is known as outage probability, which is the CDF of of the SR hop denoted by and given byand similarly for the RD hopHence, the SR hop capacity can be shown asand the RD hop capacity isThe total channel capacity is denoted by
Proof. Refer to Appendix A.
For the single-antenna case, the following corollary shows that the FD relay outperforms the HD relay in terms of capacity.
Corollary 5. Consider the SR hop where relay and destination nodes are having only single antenna . Letting , the FD relay capacity is lower-bounded byon the other hand, the HD relay capacity is upper-bounded bytherefore, as long as the following condition holds:
Proof. The proof is obtained by combining Corollary 3 and Theorem 2.
It is evident that increasing the density beyond the threshold does not compensate for the additional RSI, even if we use twice the FD capacity; this is due to FD simultaneous transmission and reception. Hence, the HD relay is optimal in this case.
2.6. Minimum Throughput Gain
We demonstrate the feasibility of FD relay over its counterpart, the HD relay, for the SR hop capacity only, but similar approach can be applied to the RD hop as well since the total capacity follows (23). We define a performance metric called the minimum throughput gain, which shows the worst case FD relay performance over HD relay in terms of capacity as follows:when , the FD relay outperforms the HD relay.
3. Interference Cancellation
This section considers the interference cancellation at the MIMO receiving nodes, that is, for SR hop at the relay and for RD hop at the destination. We present a low complexity spatial interference cancellation scheme known as partial zero forcing (PZF) [29, 30]. During the SR hop and RD hop, PZF can be applied at the relay and destination to cancel and interferences, respectively, while using the remaining degree of freedom to boost the intended received signal.
3.1. SR Hop: Canceling RSI and SRI
It is beneficial to increase the capacity by canceling the interference (RSI and SRI) from nearby nodes in an interference-limited scenario. Let us denote the points of , in an increasing order from the source, that is, ; the relay applies PZF to cancel its interference. From (3), the overall interference can be rewritten as
For the SR hop, with PZF at the relay, the success probability is given next.
Theorem 6. The success probability of the SR hop with PZF at the relay iswhere is the Laplace transform of the interferences:
Proof. See Appendix A and follow similar steps to those in [30].
3.2. RD Hop: Canceling SDI
It is beneficial to increase the capacity by canceling the interference (SDI) from nearby nodes in an interference limited scenario. Let us denote the points of , in an increasing order from the source, that is, ; the destination applies PZF to cancel its interference. From (6), the overall interference can be rewritten asand its corresponding success probability is given below.
Theorem 7. The success probability of the RD hop with PZF at the destination iswhere is the Laplace transform of the interferences:
Proof. See Appendix A and follow similar steps as those in [30].
4. Numerical Results
Several Monte Carlo simulations are presented to validate the proposed scheme and averaged over 10000 number of channel realizations. Rayleigh fading assumption results in a convenient coverage probability form in terms of the Laplace transform of the interference, which is easier to characterize than the probability density function of the interference [10]. The sources and relays are transmitting with W and W with their corresponding distances m and m, respectively. The SINR threshold is dB and the path-loss exponent is , with 100 destination nodes, 100 source nodes, and 100 relay nodes, with interfering source and relay nodes. We assume that, in this open-access case, destinations (users) are able to connect to any source (base station). Open access requirements often are not satisfied, particularly deploying Wi-Fi in the multi-RAT (radio access technology) femtocells network [31].
Figure 4 shows the ergodic capacity in (23) of , and , and , , antenna configurations. From (21) and (22) it is obvious that increasing the number of antennas employed at the relay and destination adds a positive term to the ergodic capacity; however, the total capacity is limited by the weakest hop (23). It can be observed that the ergodic capacity scales linearly with the number of antennas.
Figure 5 shows that the success probability in (7) increases with increasing the number of transmitting antennas at the source across the whole SINR. Deploying multiple antennas at the relay allows reducing the effect of interference, hence obtaining a substantial SINR gain. Figure 6 indicates that the success probability decreases with an increase in the SINR threshold. It compares the traditional grid model constituted by a Voronoi tessellation (see Figure 3) to the random PPP source model. Considering and , it can be observed that the case of is almost overlapped with , for . As expected, the grid model provides high coverage area across the whole SINR. A small gap has been observed when considering the SNR = 10 and SNR → ∞, due to the noise effect in dense cellular networks, which are known to be interference-limited. This validates the assumption that the noise can be ignored in interference-limited scenario.
Figure 7 presents a crucial issue in FD relay communications by evaluating the performance of multiple receive antennas in canceling the self-interference. The capacity gain is plotted against the self-interference attenuation with dB and . We observed that , even for moderate self-interference attenuation values ≤4, ≤8, and ≤10 dBs for , , and , respectively.
For the SR hop, the relays are equipped with two receive antennas and deploy PZF to cancel either the RSI or the SRI. Figure 8 shows the success probability of the SR hop against the density and compares it with the multiantenna UDN model [30] and the grid model. Note that the grid model is best suited for higher densities, whereas our proposed model performs slightly better than the multiantenna UDN model, due to the use of both antennas at source and relay for array gain to exploit the MIMO capability in boosting the intended received signal.
5. Conclusion
In a MIMO relay cellular network, RSI, SRI, and SDI will likely be the main obstacle reducing the capacity. This paper, however, developed a tractable model to analyze the success probability and ergodic capacity expressions for the random nodes, using tools from stochastic geometry. It is more realistic than the grid-based models and requires an extensive comparison with real base station deployments to validate its accuracy. We showed the effect of PZF in mitigating the interferences including the self-interference to demonstrate the feasibility of FD technology even for moderate values of self-interference attenuation. Future work probably uses this model extensively due to the interference modeling from neighboring nodes.
Appendix
A.
Assume that the typical source is located at the origin o, since the relay always chooses its nearest source to access; the cumulative density function (CDF) of is obtained by Then, the probability density function (PDF) of is as obtained SINR in (2). The CDF of SINR regarding is given byUsing the fact that with simplifies toNow, Let us evaluate the Laplace transform for the interference building on the fact that the typical relay is served by the nearest source with distance . Firstly, the Laplace transform of the interference of relay can be obtained by following the same steps as in [10, Page 125]:Using the fact that with simplifies toApplying the probability generating functional of PPP with intensity for function [32]: Applying the probability generating function to yieldsPlugging (A.8) into (A.3) and assuming that the distance between and is fixed, (19) and (20) are proven.
B.
Let us define as the PDF of . The integration of using the CCDF transforms it asNote that the power of desired signal is distributed as ; the probability of success is given byAssume that and, using Laplace transform, the success probability can be expressed using the transformation of :Finally, using the CCDF , the transformation of is given bywhere (B.6) is obtained by using Laplace transform property . Apply the moment-generating function of and Gamma distribution to [211, Theorem 1] to reach (11).
C.
The success probability of the RD hop is shown as follows:where is defined in (6). The desired signal is distributed as ; the proof of (15) is similar to Theorem 1. On the other hand, again we use the moment-generating function of distribution to provide the Laplace transform of [10, Page 125] and resort toHence, we prove (16).
Conflicts of Interest
The authors declare that they have no conflicts of interest.