Analytical SIR for Self-Organizing Wireless Networks

In a wireless communication environment characterized by dynamic channels, high influence of interference, bandwidth shortage and strong demand for quality of service (QoS) support, the challenge for achieving optimum spectral efficiency and high data rate is unprecedented. One of the bottlenecks in achieving these goals is modeling of the propagation environments [1]. The general aim of the work described in this paper is to assist in the derivation of the statistical properties of the SIR in a self-organizing wireless system, where network planning is minimal, without recourse to Monte Carlo simulations.

In a traditional system capacity studies, the pdf of the SIR has been determined through time-consuming Monte Carlo simulation or by only accounting for either the large-scale path loss [2] or multipath propagations [3], which are incomplete for studying realistic system deployment scenarios. This is primarily due to the complicated integrals involved in the derivation of the pdf of the SIR. Moreover, these studies usually assume strict hexagonal cell layout in order to simplify the calculation. The authors in [3] calculate the capacity of Nakagami multipath fading (NMF) channels assuming that the carrier-to-noise ratio (CNR) is gamma distributed. This assumption neglects the effects of shadowing and large- scale path loss. This paper presents an "exact'' pdf derived from a model which is more closely related to a realistic deployment scenario.

With the results provided here, it is possible to calculate more precise capacity figures. Furthermore since the new path loss model takes into consideration the interaction of the large-scale path loss with the small-scale fading in which the cells are irregular shaped and arbitrarily positioned, this derivation is particularly suited to study the overall system performance of self-organizing networks. Self-organizing networks can be independent infrastructureless ad hoc networks or they can also be an extension to cellular networks, where different self-organizing mechanisms, such as intelligent relaying and adaptive cell sizing, are used to enhance coverage or capacity which are the two most important factors in wireless system planning [4]. The study of coverage and capacity relies on channel quality information. The channel quality can be captured by a single parameter, namely the received SIR. The SIR between two communicating nodes will typically decrease as the distance between the nodes increases, and will also depend on the signal propagation and interference environment. Hence modeling the SIR on the assumption of the strict hexagonal cellular structure and the well-known path loss model that ignores the small-scale fading would not be applicable to self-configuring systems. Therefore analytical derivation of the pdf of SIR is a crucial step in constructing efficient system design.

Tellambura in [5] uses a characteristic function method to calculate the probability that the SIR drops below some predefined threshold (probability of outage) under the assumption of Nakagami fading. Zhan [6] also uses a similar characteristic function approach to derive outage probability for multiple interference scenario. These papers give a significant advantage in reducing the computational complexity involved in solving multiple integrals in SIR computation. But, a major shortcoming of these and other similar papers [7] is that, only the small-scale fading or large-scale fading is considered in analytically deriving the SIR statistics.

To the best of our knowledge, there has not been any work done to analytically derive the pdf of the SIR using the three mutually independent phenomena: multipath fading, shadowing and path loss together.

The rest of this paper is organized as follows. In Section 2 the system model considered is presented and in Section 3 the analytical derivation is described in detail. Section 4 provides the numerical and the simulation results. Section 5 concludes the paper.

For simplicity the cell layout used to derive the pdf of the SIR assumes circular cells, as shown in Figure 1, with maximum cell radius instead of hexagonal cells. The cells are randomly positioned resulting in potentially overlapping cells. Randomly positioned cells model an important network scenario, which lacks any frequency planning as a result of self-configuring and self-organising networks, cognitive radio and multihop ad hoc communication. A receiver experiences interference from transmitters within its accessibility radius, . Due to propagation path loss, a transmitter outside the accessibility region incurs only a negligible interference. Since the aim is to model a realistic interference limited environment, the receiver accessibility radius is taken to be much greater than the cell radius, that is, . The dashed line in Figure 1 represents the interference link between transmitter, , and receiver, while the solid line shows the desired link between transmitter and receiver and vise versa. Throughout the paper omni directional antennas with unity gains are considered. The pdf is calculated assuming one interfering user. The results obtained can be extended to multiple interfering users by using laguerre polynomials to approximate the multiple integration resulting from the multiple interfering users. The analytical derivation of SIR for multiple interference is under study.

In an interference limited environment, the received signal quality at a receiver is typically measured by means of achieved SIR, which is the ratio of the power of the wanted signal to the total residue power of the unwanted signals. let and denote the transmit and received power respectively. Let denote the path gain and is the link gain between the interfering transmitter and the receiver . For the purpose of clarity, unless otherwise stated, a single subscript , or specifies the node, and a double subscript such as specifies the link between node and node . A node is any entity, mobile station (MS) or base station (BS), that is, capable of communicating. For a single interfering user depicted in Figure 1:

Assuming fixed and constant transmit powers, , (1) simplifies to:

where , are the path losses between transmitter and receiver and and respectively.

Like the gain parameter , the loss parameter incorporates effects such as propagation loss, shadowing and multipath fading.

The generalized path loss model for the cross-layer environment is given by

where is an environment specific constant, the constant corresponding to the desired link while corresponds to the interference link. The distance is a constant and is a random variable, is the path loss exponent, is the random component due to shadowing, and is a random variable modeling the channel envelop.

The commonly used path loss equation [2] only accounts for the large-scale path loss with regular cell deployment scenarios, which is incomplete for studying self-organizing networks. The new path loss model proposed in this paper takes into consideration the interaction of the large-scale path loss as well as the small-scale fading. This model is particularly important in studying the performance of self-organizing self-configuring networks.

For the interference scenario described in the system model, the path loss for the desired path and the path loss between the interfering transmitter and the receiver (interfering link) are

where is the path loss model for the desired link and is the path loss model for the interfering link. models the distance between the interference causing transmitter, , and the victim receiver . and are the path loss exponents, and are Gaussian distributed random variables modeling the shadow fading with each zero mean and variances and respectively, and and are the channel envelope modeling the channel fading. For the purpose of clarity, the time and frequency dependencies are not shown. The channel envelope is assumed to follow the Nakagami- distribution. Nakagami distribution is a general statistical model which encompasses Rayleigh distribution as a special case, when the fading parameter , and also approximates the Rician distribution very well. In addition, Nakagami-m distribution will also provide the flexibility of choosing different distributions for the desired link and interfering link, such as the Rayleigh for the channel envelope of the desired link, and Rician for the interfering link, or vice versa.

Using (3) and (5), the SIR can be given as

From (7), the SIR has six random variable components, , , , , and . In order to analytically derive the pdf of the SIR, the pdf of the individual components and also their ratios and products need to be determined first.

The following two formulas provide the basic framework for the analysis and will be used throughout the derivation. Given two independent random variables and the pdf of their product where is

Given two independent random variables and the pdf of their ratio where is

In case of an environment that is is not interference limited, the (signal to interference and noise ratio) SINR is required to fully describe the communication channel. SINR can easily be found by modifying the SIR equation given in (1):

where is the random variable modeling the Gaussian noise with mean and a standard deviation of . By applying the generalized path loss equation in (4), SINR at the receiver is given by:

where the pdfs of the individual random variables are given in the previous section. Let which are derived in the previous section. The pdf of , , is given as

where .

The next step in the derivation is to find the pdf of the path loss of the desired link by utilizing (9) and (17). Let be the random variable denoting the path loss of the desired link. The pdf of is given as

where .

The pdf of the path loss of the interference path denoted by the random variable is give as

where

In order to find the pdf of the interference plus noise, , it is assumed that interference is independent of noise. The pdf of the sum of two independent random variables and , each of which has a probability density function, is the convolution of their individual density functions

therefore the pdf of , is given by:

where . Utilizing (9), the pdf of the SINR is given by

For the special case where the noise approaches zero, the pdf of the noise is represented as delta function or also known as, a unit impulse function, around zero. Therefore (30) can be rewritten as

Thus

by the definition given in (9), the given in (33) is the pdf of the SIR( . Therefore, when the noise approaches to zero, the pdf of the SINR given in (31) reduces to the pdf of SIR given in (22).

This sub-section has presented the pdf of the SINR as an extension to the pdf of the SIR. To validate the analytically derived SINR pdf, it is important to show that the core derivation, SIR derivation, is valid. The next sub-section validates the derivation through comparative numerical simulations of the SIR. The results presented were obtained using the adaptive Simpson quadrature numerical integration of the SIR.

Monte Carlo simulations are carried out in order to validate the analytically derived pdf results. Figures 2 and 3 show plots of the pdf of the SIR for different scenarios. The results presented in Figures 2–4 show that the analytical pdf is in good agreement with the Monte Carlo simulation. The parameters used for the shadow fading, channel standard deviation and path loss exponents reflect a realistic deployment scenario for users moving at a speed of 25 to 40 km/h [10]. All simulations assume a channel envelope with a Nakagami-m distribution with different m parameter, which corresponds to different fading scenario. These parameters are summarized in Tables 1, 2, and 3.

Figure 2 depicts three different plots depending on the . As the cell radius increases there is a significant cell overlap leading to high mean value of interference which in turn leads to lower SIR mean value. Therefore, as the ratio of the cell radius to the accessibility radius approaches to one, the pdf is skewed towards smaller SIR. These plots show that the node with the lowest cell radius, , has the highest SIR mean.

Figure 3 shows the effect of different environments on the pdf of the SIR. The figure presents plots from an ad hoc free space outdoor deployment with line of sight scenario on the desired link, and , to the most severe non-line-of-sight scenario of obstructed indoor (in building) environment, and . The radius of the cell, , has been set to 100 m, which is considered a good configuration example for ad hoc networks. The accessability radius, is assumed to be 500 m. The results illustrate that the node with the best line-of-sight (LOS) link, and , has the highest mean SIR value and the biggest variance or spread. While the node with the most obstructed inbuilding environment, exhibits the lowest mean and the smallest variance or spread of all. These can be attributed to the higher interference contribution of interfering node in NLOS link than those in LOS condition.

Figure 4 present the cumulative density function of the SIR. The simulation parameters are summarized in Table 3. From Figure 4 it can be observed that for a target SIR of 25 dB, being a reasonable assumption for 64-QAM modulation, the probability that the SIR exceeds the target SIR in the most severe non-line-of-sight scenario of obstructed indoor (in building) environment is about 10% resulting in a high outage probability enforcing the use of lower order modulation schemes. On the other hand, for the link with best LOS condition of outdoor free space environment the probability that the SIR exceeds the target SIR is 85% allowing the use of higher order modulation. Therefore from the results in Figure 4, it can be deducted that the analytical work presented in the paper can be used in determining the boundaries for varying the modulation order. A similar work of determining the boundaries for adaptive modulation was presented by Goldsmith et al. and M.-S. Alouini [11] assuming Nakagami distribution thus ignoring the shadowing effect, the pdf presented here can be used to extend the results presented in [11].

The main contribution of this paper is the derivation of the pdf of the SIR in a self-organizing wireless system, where network planning is minimal, without recourse to Monte Carlo simulations. The derivation was carried out using a generalized path loss model that accounts for both large and small- scale path loss. The use of Nakagami- distribution for the fading channel gives the flexibility to use Rayleigh or different channel fading models for the desired and interfering links. The results obtained show excellent agreement with the Monte Carlo based results. The SIR derivation was in turn used to derive the pdf of the SINR. The SINR derivation is important in non-interference limited environment. These derivations can be further used in applications where the knowledge of SIR is necessary, such as link adaptation algorithms and cognitive radio design. The analytical derivation of the pdf from a single interferer in this paper lays a solid foundation to calculate the statistics from multiple interferers.