Ray-optical propagation model

Even though highly accurate models of radio signal propagation exist, these mo- delling and simulation environments are of considerable computational complexity and are therefore unsuitable for the incorporation into real-time protocols, particu- larly on resource-constrained platforms such as MANET nodes. We have therefore proposed a simplified ray-optical signal propagation model in [113] which takes into accound the position of nodes as well as topographical information, but does not in- corporate a comprehensive model of physical effects. We have implement the model as a module of NS-2, facilitating an easy integration of our model into NS-2. The core part of our ray-optical propagation model was implemented in the master thesis of Reidt [113]. Improvements regarding efficiency and accuracy were performed in this thesis as well as the validation of the model against real test data. We use the model for the simulations in this thesis; the implemented scenarios are defined in Section 3.2.

3.1 Modelling the physical layer

3.1.2.1 Background

Hoppe et al. [67] introduced a ray optical propagation model that takes into account both reflection and deflection effects on buildings. This model requires the pre- processing of the environment, which by far exceeds the computational capabilities of mobile devices. Pre-processing the data on a powerful server and then storing it on the mobile device is infeasible due to the size of the pre-processed data. The use of this model is therefore restricted to the use of powerful computers or to devices that allow the storage or sending of huge amounts of data is possible. The accuracy of this model was verified in [130] and [111], showing the potential of the approach to model radio propagation by a ray optical model. The model was further extended by Hoppe et al. for the use in indoor environments in [66] and [110].

Dhoutaut et al. [44] propose the use of the Shadowing-Pattern Model to simu- late radio wave propagation in VANETs where packet losses occur frequently. This model takes into account most possible types of disturbances while keeping a low computational cost and allowing the easy tuning of any particular disturbance inde- pendently of all others. The model is probabilistic and therefore especially useful for VANETs, where disturbance effects are highly correlated with the density of cars, and where typical characteristics of streets allow similar configuration of the model for most VANET scenarios. In Tactical MANETs however, it is difficult to estimate the required configuration parameters of the model. Furthermore, the probabilistic approach cannot take disturbances into account, such as the interruption of signals by buildings in a city.

As noted above, current signal propagation models are typically optimised for high accuracy and they are not intended for use in a resource-constrained environ- ment in which computations must be performed within a near-real-time interval. However, in the following we briefly review several models which are widely used

3.1 Modelling the physical layer

and which partly form the basis for the ray-optical model as described in Section 3.1.2.2. The models discussed here are typically suitable for describing propagation over arbitrary distances and at frequencies ranging from 1 MHz to 40 GHz unless noted otherwise.

The Free Space model [51] assumes a line-of-sight connection between sender and receiver node without consideration for other influences. Based on these as- sumptions, the model calculates the power transmitted by the direct line-of-sight connection between sender and receiver. The equation for calculating the power P for a distance d is qualitatively given by P (r) _{∼ 1/d}2_{. The Two Ray Ground}

model [51] is a direct extension of the Free Space model which also takes ground reflection of radio waves into consideration [8]. It is based on the assumption of horizontally polarised radio waves, and the power P at distance d is qualitatively given by P (r)∼ 1/d4_.

The Shadowing model [51] used in the NS-2network simulator includes line-of- sight components and time-dependent parasitics and scattering. The equation for calculating the receiving power is qualitatively given by P (d) _{∼ 1/d}β _{· X, where} 0 < β ∈ R provides a configurable parameter for adjusting the parasitics, and where X is a random variable modelling scattering.

The COST Walfish Ikegami [8] model considers obstacles such as buildings in the vertical plane and effects such as multiple diffraction over rooftops between the transmitter and the receiver node. The transmitter node is assumed to be 4 metre to 50 metre above the ground and the distance between nodes needs to be at least 20 metre. This model is therefore mostly constrained to environments in which the transmitter is located on a rooftop or similarly elevated terrain feature.

3.1 Modelling the physical layer

3.1.2.2 Model

In the ray-optical propagation model introduced by Reidt [113], a 2D ray-tracing approach is used to develop a simplified but efficient radio propagation model. Ac- cording to [90] and [56], this approach is defensible under three main conditions:

1. The used frequency band is beyond 1 GHz.

2. Considered surfaces are large in comparison to the wavelength.

3. The surface structures of individual terrain features are approximately con- stant.

The first condition is satisfied for the ISO 802.11 (a/b/g/h) series of standards (which use bands from 2.4 GHz to 2.5 GHz and 5.15 GHz to 5.85 GHz, respec- tively). The appropriate wavelength of approximately 10 cm at these frequencies is substantially smaller than the topographic objects such as buildings. Further- more, the model as specified by Reidt [113] provides only uniform surfaces and does not include additional modifiers such as surface textures. Therefore, it also satisies condition (iii). A further simplification for efficiency is the restriction to vertical surfaces. This simplification allows to store a 2D instead of a 3D map and to use a 2D instead of a 3D raycasting algorithm. In the following section we validate the accuracy of the model of Reidt. A detailed description of the model can be found in [113].

3.1.2.3 Evaluation and analysis

Based on the implementation of our propagation model in NS-2, we now discuss re- sults on the quality of the approximation achieved by the model, as well as empirical data on the performance of the model.

3.1 Modelling the physical layer

3.1.2.4 Evaluation of calculations

The analysis in [113] has shown the consistency of reflection and deflection factors of the ray-optical propagation with data found in the literature [56]. Beyond this theo- retical evaluation, we now report validation results from actual field measurements based on two experiments as reported in [4].

Scenario with obstacles Following the illustration of reflection and deflection factors, which form the main part of the calculations, we next compare the re- sults of our model to measurements from two scenarios. The first scenario in Fig- ure 3.2(a) shows a building and two nodes, representing the sender (dotted circle) and the receiver (crossed circle) [4, Section 3.2.9]. Both sender and transmitter are portable computers equipped with standard 802.11 (a/b) network interfaces. While the sender has a fixed position, the receiver moves away, following the line parellel to the rectangle (building) in this scenario.

4.4 m 1 7 m (a) 4.4 m 2 2 m (b) 5.7 m 5 m

Figure 3.2: Test scenarios.

The power values are measured in the Received Signal Strength Indicator (RSSI) [11], so that all power values originally had to be measured in RSSI and transformed to dBm. Unfortunately, there is no standard for transforming RSSI into dBm or

3.1 Modelling the physical layer

mW. Typically each card manufacturer defines its own relation between RSSI and dBm. This circumstance could be the reason for the almost constant difference of 10 dBm between the measured, and the calculated power values in Figure 3.3 and Figure 3.4. Another reason for this difference could be the ground in the simulation scenarios. While the Two Ray Ground model assumes level ground, the ground surface in the experiments was somewhat uneven and covered with vegetation. As shown in [4], there is a gap of almost 10 dBm between measurements on concrete surfaces as opposed to grass; the reason for this is the different permittivity of concrete and grass. While grass is absorbing much of the transmitted power, concrete and similar substances are reflecting most of it. Moreover, the transmitting power of the sender with a maximum transmitting power of 100 mW was not explicitly defined in [4]. However, we based our calculations on a transmitting power of 100 mW, and compensated for the 10 dBm gap; this gap does, however, indicate the desirability of choosing basic propagation parameters carefully and may indicate a need for incorporating ground permittivity in our constrained model.

−90 −80 −70 −60 −50 −40 −30 −20 0 10 20 30 40 50 distance [m] RSSI [dBm] calculated RSSI

RSSI (area with obstacles) mean RSSI

RSSI in free space

Figure 3.3: Test series 1.

3.1 Modelling the physical layer

the calculation for the first scenario. Each black cross represents one of the mea- surements, which were done in a distance of 1 m, and the red squares are the mean power values for one distance measured in metres. Additionally, the stars show the results of a measurement done under the same conditions but without any obstacles. The calculations, which were performed using the ray-optical propagation model, are illustrated by the dashed line. Apart from the gap of 10 dBm described above, the curve of the calculated values provides a good fit for the measured values. Owing to simplifications in our model, it is not possible to take interference effects into account. Thus, the curve of the calculated values shows a very smooth behaviour, whereas the measurements show interference patterns, most prominently caused by ground reflection.

Deflection scenario While the propagation in the first scenario was dominated by the direct line-of-sight and the reflection on the ground as well as on the building, the second scenario illustrates the deflection on a house corner (Figure 3.2(b)). While the sender has a fixed position, the receiver is moving behind the building, following the line parallel to the building as before.

As already seen in the first scenario, the curve of the calculated values shows a very smooth behaviour. However, calculated values of our model show a good fit to measured values. After 5.7 m the receiver loses its line-of-sight connection to the sender, resulting in a significant decrease of the receiving power.

The data of the investigated scenarios indicates a high degree of fidelity achieved by our constrained model compared to field measurements. However, we observed that parameters of the underlying Free Space model and Two Ray Ground model need to be chosen carefully. Additional parameters for a future improvement of the model are discussed in Section 8.2.2.

3.1 Modelling the physical layer −95 −90 −85 −80 −75 −70 −65 −60 −55 −50 −45 0 5 10 15 20 distance [m] RSSI [dBm] RSSI mean RSSI calculated RSSI

Figure 3.4: Test series 2.

3.1.2.5 Computation Periods

All computations were performed on a Pentium Centrino 1.7 GHz processor with 1 GB of main memory. It should be noted that the resources required for our model including shape file handling do not exceed 5–10 MB depending on the complexity and size of the terrain model.

Figure 3.5 shows the result of a single calculation, which was performed with the help of our iNSpect extension [93]. The scenario shows a 600 m× 600 m square of the centre of London and contains 180 faces and 25 nodes. Such calculations are to be performed on PDAs or other mobile resource constrained devices to improve routing strategies. Table 3.1 lists computation periods based on the scenario described above. Although current PDAs perform at 20–40 % of the performance levels of our test system, improvements in equipment and ongoing optimisation of our algorithms and implementation will significantly reduce the computation times exhibited by our proof of concept model. Results on single routes, however, can already be used effectively for improving existing routing strategies.

3.1 Modelling the physical layer

Figure 3.5: Connectivity between nodes.

Table 3.1: Computation periods.

Description Time [sec]

Power transmitted between two single nodes 0.01

Multihop route with 3 hops 0.025

Multihop route with 5 hops 0.0375

Multihop route with 7 hops 0.05

Connections between all nodes in Figure 3.5 0.6

3.1.2.6 Summary

Based on earlier work [113], we have validated the accuracy and computational efficiency of our ray-optical propagation model which is especially suitable for urban environments. Our evaluation shows the efficiency of our radio propagation model while still obtaining good approximative results. The propagation model is used to simulate urban MANET scenarios, as defined in Section 3.2. Furthermore, it can be implemented on power constrained mobile devices to add valuable information about the connectivity to network protocols, e.g., facilitating elaborated routing protocols and more accurate intrusion detection systems.