Radio wave propagation through vegetation: Factors influencing signal attenuation



[1] The paper describes an extensive wideband channel sounding measurement campaign to investigate signal propagation through vegetation. The measurements have been conducted at three frequencies (1.3, 2 and 11.6 GHz) at sites with different measurement geometries and tree species. The data have been used to evaluate current narrowband empirical vegetation attenuation models and study the prevailing propagation mechanisms. Evaluation of the modified exponential decay (MED), maximum attenuation (MA) and nonzero gradient (NZG) models show that on a site by site basis, the NZG model gives the best prediction of excess attenuation due to vegetation. The MA model has been found to be the worst of the three models. The studies have shown that the measurement site used to obtain the NZG model parameter values given in International Telecommunication Union (ITU) [2001] is influenced by metal lampposts and passing traffic, and thus was based on corrupted data. The results show that the leaf state, measurement geometry and vegetation density are more important factors influencing signal attenuation than tree species or leaf shape. Generally, the 11.6 GHz signal was attenuated much more than the 1.3 and 2 GHz signals by vegetation in-leaf, but the differences in attenuation were not significant in the out-of-leaf state. A successful excess attenuation model due to vegetation must consider the measurement geometry and vegetation descriptive parameters as well as any contributions from ground reflection and/or diffraction over the top or round edges of the trees.

1. Introduction

[2] The reduction of cell size and basestation antenna heights in cellular networks has forced the telecommunication sector and spectrum licensing authorities to investigate the impact of vegetation on radiowave propagation. This knowledge will assist in optimizing spectrum utilization and enhancing the quality of services provided. The study of propagation through vegetation is challenging due to variations in vegetation density, measurement geometry, and vegetation composition. In addition, vegetation is prone to environmental effects, such as wind, that can introduce dynamic variations in the channel signature. The lack of extensive measurement campaigns to aid and verify model development means that empirical models have, historically, been based on small amounts of narrowband data and are often biased towards artifacts at the measurement site [Vogel and Goldhirsh, 1993; Seville and Craig, 1995; Al-Nuaimi and Stephens, 1998]. Analytical methods used to model radiowave propagation through vegetation often either grossly over-simplify the vegetation medium or require the use of numerical analysis to provide solutions to intractable formulations [Brown and Curry, 1982; Tavakoli et al., 1991; Tamasanis, 1992; Rikte et al., 2001].

[3] This paper presents the results of an extensive measurement campaign and the evaluation of current empirical models. The measurement sites are verified using wideband analysis to identify any artifacts and to study the predominant propagation mechanisms. During the investigation, signal propagation through different tree species, both in-leaf and out-of-leaf, and different measurement geometries were studied and the curves of the measured excess signal attenuation are presented. Although not fully discussed in the paper, the study also involved narrowband measurements for frequencies in the range 1 to 18 GHz.

2. Experimental System and Data Processing

[4] The investigation was conducted using a wideband channel sounder, the details and design principles of which are described in Austin et al. [1997]. The sounder has previously been used to characterize various channels, and details can be found in Ndzi et al. [2001] and Savage et al. [2002]. The sounder transmits a 31.25 MHz bandwidth pseudorandom Gaussian noise (PRGN) sequence simultaneously at 1.3, 2 and 11.6 GHz. However, only one frequency can be received at any one time. The noise floor of the receiver is −90 dBm. The received signal is averaged over 128 channel responses (all measured in 1 ms), thus improving the signal-to-noise ratio by 21 dB. The fast channel sampling enables dynamic variations in the channel, for example due to wind, to be measured.

[5] The power at the output of the transmitter, the antennas used and their beam widths are summarized in Table 1. The antennas were mounted on variable height masts. During the measurements, the transmitter was set up in a fixed location while the receiver, integrated in an experimental vehicle, was moved from one measurement point to the next. The transmitter and receiver were synchronized using a 10 MHz phase reference from global positioning system (GPS) receivers or connected to a common reference by a cable. At some measurement sites, the experiments were conducted in collaboration with Rutherford Appleton Laboratories (RAL), Didcot.

Table 1. Beam Widths of Antennas Used and Transmit Power
Frequency, GHzTX Beam WidthRX Beam WidthTX Power
1.3 (yagi loop)18°18° (70° with RAL)17 dBm
2 (horn)40°70°22 dBm
11.6 (horn)18°20°16 dBm

[6] Measurements were carried out at various sites, as listed in Table 2. At each site, the transmitter was located in a clearing in front of the vegetation. Figure 1 illustrates the geometries investigated and the measurement set up. A calibration file was measured at every site with the receiving antenna at the front edge of the vegetation, corresponding to zero vegetation depth as marked by the letter ‘C’ in Figure 1. At each measurement position, the receiving and transmitting antennas were aligned. All of the measured channel frequency responses were calibrated to obtain the channel transfer functions. A single spectral line within the wideband spectrum was extracted from the transfer function and the power of this line was used to emulate a narrowband signal. Figure 2 shows the discrepancy between the power of the extracted spectral line and the average power across the bandwidth. The free space loss, calculated for each position, was removed from the measured signal power and the resulting excess attenuation was normalized to the level received at zero vegetation depth (position ‘C’ in Figure 1). The vegetation depth was measured from the front edge of the vegetation to the receiver antenna position, along the direct path from the receiver to the transmitter. The extraction and use of only a single frequency power allowed the database generated from the study to be used in the evaluation of narrowband channel models.

Figure 1.

Measurement geometries. (a) Line of trees, (b) into vegetation, (c) through vegetation, and (d) edge of vegetation.

Figure 2.

Comparison between a single extracted spectral line power and the power averaged over the bandwidth.

Table 2. Geometry of the Measurements Conducted
SiteTree Species (Scientific Name)GeometryLeaf ShapeLeaf StateTree HeightTX Antenna HeightRX Antenna HeightDistance to Edge
Ravelin ParkSycamore (Acer pseudoplantanus)IntoLobeIn10 m3 m2.5 m & 7.5 m13.8 m
RALHorse Chestnut (Aesculus hippocastanum)SingleLobeIn8–9 m3 m5.3 m26.2 m
RALSycamore (Acer pseudoplantanus)EdgeLobeIn and Out20–25 m3.5 m2.5 m & 7.5 m11.4 m
RALSycamore (Acer pseudoplantanus)ThroughLobeIn20–25 m3 m5.3 mApprox 115 m
QECPLawson Cypress (Chamaecyparis lawsoniana)ThroughNeedlesIn15 m3.5 m2.5 m & 7.5 m18.1 m
QECPCommon Beech (Fagus sylvatica)IntoOvalIn15–20 m3 m2.5 m & 7.5 m31 m
J.J.NurseriesSilver Maple (Acer saccharinum)LineLobeIn and Out7–8 m3 m3 m10.8 m
J.J.NurseriesCommon Lime (Tilia x Europaea)LineOvalIn and Out7–8 m3 m3 m4.4 m
J.J.NurseriesLondon Plane (Plantanus x hispanica)LineLobeIn and out7–8 m3 m3 m3.5 m

[7] The channel impulse responses were estimated from the measured transfer functions using the singular value decomposition prony (SVD-P) algorithm [Hewitt et al., 1989; Lau et al., 1991; Lam, 1995]. Given the transmitted signal bandwidth of 31.25 MHz, the delay resolution using Fourier transformation would be 31 ns. However, detail accuracy test results have shown that the SVD-P algorithm is able to successfully recover two rays that are separated by less than 5 ns with a signal to noise ratio of 24 dB. The technique has been proven to work very well with signal to noise ratios greater than 20 dB [Shen, 1995]. This high resolution enables a detailed study of the prevailing propagation mechanisms. The wideband channel characterizing parameter, root mean square (RMS) delay spread, has been calculated for all the measurements. RMS delay spread gives an indication of the time spread of the received signal, enabling an assessment of the channel quality for communication [Rappaport and Sandhu, 1994].

3. Measurement Sites and Geometries

[8] Measurement sites were selected based on accessibility, tree type and measurement geometry. Four geographical sites were chosen for the measurements: Ravelin Park - Portsmouth (single tree and wedge of trees), Queen Elizabeth Country Park (QECP) - Hampshire (a managed forest), RAL - Didcot (a line and wedge of trees) and J. J. Nurseries - Malvern, (a plantation of rows of immature trees), see Table 2. The measurements were conducted on eight different tree species. However, in-leaf and out-of-leaf measurements at the same site were only conducted on four tree species. To investigate the effect of different parts of the vegetation on signals, some measurements were conducted at more than one receiver antenna height.

[9] Four main measurement geometries were investigated; into forest, line of trees, along edge of forest, and through the forest, as illustrated in Figure 1. At each site, the transmitter was located outside the vegetation and the receiver was moved along the lines as shown in Figure 1. The receiver was positioned inside the vegetation and measurements taken at various points along the direction of signal propagation for all measurement geometries, except for the through vegetation geometry, as shown in Figure 1c.

4. Vegetation Attenuation Models

[10] The generation of an accurate model, either empirical or analytical, requires input parameters that are difficult to acquire. These parameters include any combination of the following: height of vegetation, leaf state, vegetation density, trunk size, leaf size, and canopy height [Welch and Lemark, 1995]. Some of these parameters are difficult to characterize and quantify in an understandable, easy to measure and relevant manner for use in a practical (engineering) model. Three models have been evaluated using the database of measurements; the modified exponential decay (MED) model, the maximum attenuation (MA) model, and the nonzero gradient (NZG) model [ITU, 2001]. Since the empirical models are generated from previous measurements, they model the combined effects of the different propagation modes, not only the through vegetation mode. It should be noted that these are not the only models that can be found in literature, there are also analytical models that attempt to model only the through vegetation signal. Among the analytical models that have been used and reported in literature are; the geometric and uniform theory of diffraction (GTD/UTD) [Matschek and Linot, 1999; Sachs and Wyatt, 1968], radiative energy transfer (RET) [Al-Nuaimi and Hammoudeh, 1994; Ishimaru, 1978] and full wave solutions (FWS) [Didascalou et al., 2000] models. These models are mathematically and computationally intensive and have not been evaluated in this paper. However, it has been reported that the RET model offers the best results for attenuation and scattering due to vegetation and can be applied to a wide range of frequencies and geometries [Qinetiq, 2002].

4.1. Modified Exponential Decay Model

[11] This model was first proposed by Weissberger [1982], and a modified version was included in the International Radio Consultative Committee (CCIR) [1986] recommendations. It has been used to fit data from a variety of experiments, each resulting in different values of the fitted parameters [Weissberger, 1982; COST 235, 1996; Al-Nuaimi and Stephens, 1998]. Different parameter values have been proposed depending on the leaf state. The model is fitted to the measured data to estimate the values for X, Y and Z, in equation (1); where f is the frequency in megahertz (MHz) and d is the vegetation depth in meters. The advantage of the model lies in its simplicity, but it has a major drawback in that it does not take into account the measurement geometry or the propagation mechanisms.

equation image

4.2. Maximum Attenuation Model

[12] The maximum attenuation model involves the use of the maximum excess attenuation (Am) measured and the initial gradient (R) of the excess attenuation curve as input parameters to equation (2). The calculation of the initial gradient is prone to errors because of variations in the data. The dependency of this model on fitted parameters from experimental measurements limits its application, as the parameters are often biased toward the measurement geometry and/or methodology. Another limitation of this model is that it has a fixed final attenuation gradient. The MA model is described by equation (2), where d is the vegetation depth in meters.

equation image

4.3. Nonzero Gradient Model

[13] This model was proposed by Seville and Craig [1995] to overcome the zero final gradient problem associated with the maximum attenuation model, and it is the current ITU-R model for frequencies above 5 GHz [ITU, 2001]. As with the MA model, the NZG model requires input parameters to equation (3) that have been estimated from measured data. These include the initial gradient (R0) and the final gradient (R) of the attenuation curve, and the offset of the final gradient (k). The model also suffers from the problems associated with using values of parameters obtained from fitting curves to experimental data. However, the ITU recommendations do suggest parameters to be used in the model based on frequency and the minimum width of illuminated vegetation. An extension of this model that takes into account antenna beam width and frequency has also been proposed [Seville, 1997]. However, the proposed model has proved to have several inaccuracies [Stephens, 1998].

equation image

5. Evaluation of Attenuation Models

[14] The three models introduced in Section 4 have been fitted to the measured data and the estimated values of the parameters of the models are given in Tables 3, 4, and 5. The mean square error (MSE) of the fittings is given in Table 6. To validate the importance of different measurement and vegetation description parameters, the data was divided into four categories based on, measurement geometry, tree species, leaf shape and leaf state. The models were also fitted to the whole database of measurements, disregarding differences in leaf state and geometry, but taking into account the frequency. The values obtained from fitting to the whole database are presented in Tables 3–6 in the row labeled “All.” The rows labeled “Line-in” and “Line-out” refer to the measurements carried out on a line of trees in-leaf and out-of-leaf, respectively. Since the MA and NZG models do not have explicit frequency dependent terms, the fittings were carried out for each frequency, as stated in brackets in Tables 4, 5, and 6.

Table 3. Fitted Parameters for Fitted Modified Exponential Decay Model
Table 4. Fitted Parameters for Maximum Attenuation Modela
 Am (11.6)Am (2)Am (1.3)R (11.6)R (2)R (1.3)
  • a

    Frequency given in parentheses.

Out-of leaf35.1237.6645.
Table 5. Fitted Parameters for Nonzero Gradient Model
 R (11.6)R (2)R (1.3)R0 (11.6)R0 (2)R0 (1.3)k (11.6)k (2)k (1.3)
Table 6. Mean Square Error From Fitting Data to Models
 MEDMA (11.6)MA (2)MA (1.3)NZG (11.6)NZG (2)NZG (1.3)

[15] Figures 3–5 illustrate a comparison between the fittings of the models to data for particular data sets. The parameters of the models used in Figure 3 have been estimated from all the data irrespective of geometry or leaf state. Figure 4 shows the fitting of the models using values of parameters estimated from all the data measured in the in-leaf state, taking into account frequency. Figure 5 shows the fitting using parameter values obtained from a data set that takes into consideration the frequency and measurement geometry. Figure 3 shows that the MA model underestimates attenuation at short vegetation depths (<35 m) and overestimates at large depths. The MED model provides a consistently close fit at depths greater than 8 m. Analysis of the fitting of all of the models to all of the measurements conducted showed that overall, the NZG model provides a close fitting at all depths. As Figures 4 and 5 are the same measurement data, it shows that including the measurement geometry results in a much better fit to the measured data.

Figure 3.

Fitting of the models to experimental data at 1.3 GHz from measurements on London Plane using parameter values estimated from all the experimental data at 1.3 GHz.

Figure 4.

Fitting of the models to experimental data at 11.6 GHz from in-leaf measurements on Sycamore (edge geometry) using parameter values estimated from data not grouped according to vegetation geometry at 11.6 GHz.

Figure 5.

Fitting of the models to experimental data at 11.6 GHz from in-leaf measurements on Sycamore (edge geometry) using parameter values estimated from data grouped according to vegetation geometry at 11.6 GHz.

[16] When the data were grouped based on the measurement geometry, better fittings were achieved, in general, for all the models. This shows that measurement geometry is an important consideration for model development. The increase in accuracy is thought to be due to the fact that the propagation mechanisms depend on the measurement geometry. Better fits were also obtained for measurements that were conducted in the out-of-leaf condition. When the measurements were conducted in the out-of-leaf condition there were no leaves to attenuate the signal and no variations in the signal induced by changes in leaf density. This was especially noticeable at 11.6 GHz for measurements conducted on lines of trees. When the data was grouped based on the leaf shape there was no improvement in the accuracy of the fittings, indicating that the leaf shape has little influence on radiowave propagation through vegetation at the frequencies investigated.

[17] When the models were fitted to individual measurements, the NZG model gave the most accurate prediction of excess attenuation. This was expected as the NZG model has been developed based on individual site measurements. The values obtained at one site may not be used to predict attenuation at another because they encompass propagation anomalies that may not exist at both sites. Fitting the models to combined data from different sites will average out any anomalies and make the model parameter values more generic. When the models were fitted to the data sets listed in Table 6, the NZG model performed generally better than both the MED and the MA models by 1 dB and 2 dB on average, respectively, except for the measurements conducted at 11.6 GHz. At 11.6 GHz the best fit was obtained with the MED model. The MA model performed consistently worse than both the NZG and MED models.

6. Propagation Mechanisms and Wideband Analysis

[18] The current ITU-R recommendation [ITU, 2001] does not take into consideration all possible mechanisms involved in radiowave propagation through vegetation. The propagation mechanisms that can exist in the presence of vegetation include diffraction, reflection and scattering [Tamir, 1977; Seker, 1989], as illustrated in Figure 6. Detailed analysis of the measured channel responses, in both time and frequency, complemented by knowledge of the measurement set up can be used to identify these mechanisms. The identification of these mechanisms and knowledge gained from channel response analysis are essential to ensure that the received signal is dominated by through vegetation scatter.

Figure 6.

Possible propagation mechanisms in the presence of vegetation.

[19] Assuming that the transmitting antenna illuminates just the vegetation, at short vegetation depth the signal is dominated by a coherent wave, with little scattered power [Schwering et al., 1988; Ulaby et al., 1990; Al-Nuaimi and Hammoudeh, 1994]. The vegetation medium acts as randomly distributed scatterers and at greater depths the radiowaves become incoherent with random amplitudes and phases. Propagation is dominated by scattering and the signal level decrease with vegetation depth is slower than when the signal is coherent. This is exhibited in the signal attenuation against vegetation depth curve by a dual gradient characteristic. This represents the ideal case where only the effects due to vegetation exist. In practice, however, more than one propagation mode often exists. While the proposed empirical models represent attenuation by vegetation as a bulk effect encompassing all the propagation modes, analytical techniques often focus on the vegetation medium alone [Brown and Curry, 1982; Tavakoli et al., 1991; Tamasanis, 1992; Rikte et al., 2001].

[20] In the investigation, a case study of the prevailing propagation mechanisms involving detailed measurements and simulations were carried out on the sycamore wedge at RAL. The measurements were undertaken using a narrowband system for frequencies in the range 1 to 18 GHz at antenna heights from 5 to 19 m. The canopy height at this site was approximately 20 to 25 m. These studies were conducted to assess the impact of ground reflection, over-the-top and edge diffraction on the measure signal. Detailed simulations using equations in ITU-R Recommendation 526 for edge diffraction and equations in ITU-R Recommendation 527 for ground reflection were also carried out. The results from this study showed that over the top diffraction could only be received for receiver antenna heights of 17 m and above. However, edge diffracted components could potentially dominant the received signal at 1.3 and 2 GHz at all antenna heights when the receiver was positioned close to the edge of the wedge of trees. The edge diffracted components could not be avoided in the final wideband measurements because smaller vegetation depth could only be achieved close to the edges due to the shape of the plot of trees. The study also revealed that at small vegetation depths the ground reflected components were significant at all frequencies whereas at the larger vegetation depths they were only identifiable at 1.3 and 2 GHz.

6.1. Lines of Trees

[21] The measurements in this category include all those at J. J. Nurseries (on London Plane, Common Lime and Silver Maple trees) and Fermi Avenue (Horse Chestnut), see Table 2. The line of trees had clearance on both sides. Figures 7 and 8 show examples of excess attenuation curves for the two leaf states, both exhibiting dual gradient slopes. Using Silver Maple as an example, the figures show that higher excess attenuation was observed at 11.6 GHz than at 1.3 or 2 GHz for the in-leaf condition. In the out-of-leaf condition less excess attenuation was measured at 11.6 GHz whereas a general increase in excess attenuation was recorded at 1.3 and 2 GHz. The strong through vegetation component at 1.3 GHz and 2 GHz observed at JJ Nurseries is due to the fact that most of the branches were smaller than the wavelength of these signals. However, at 11.6 GHz, the branch dimensions are comparable to the wavelength of the radiowaves and the signal was significantly scattered by branches and attenuated by the leaves. In out-of-leaf condition the scattered components at 11.6 GHz suffered very little absorption resulting in less excess attenuation. The smaller excess attenuation at 2 GHz compared to 1.3 GHz can be associated with the larger beam width of the 2 GHz antenna which is able to receive more paths than the 1.3 GHz antenna. The differences in excess attenuation between the two leaf conditions can be explained by considering the prevailing modes of signal propagation, which can be deduced with the aid of the estimated channel impulse responses, shown in Figures 9 and 10.

Figure 7.

Silver Maple in-leaf excess attenuation for line of trees geometry (receiver antenna height: 3.5 m).

Figure 8.

Silver Maple out-of-leaf excess attenuation for line of trees geometry (receiver antenna height: 3.5 m).

Figure 9.

1.3 GHz Silver Maple in-leaf estimated impulse responses (vegetation depth 18 m).

Figure 10.

1.3 GHz Silver Maple out-of-leaf estimated impulse responses (vegetation depth 18 m).

[22] Figures 9 and 10 illustrate the channel impulse responses measured at 18 m vegetation depth over a period 4.3 s for in-leaf and out-of-leaf conditions, respectively, at 1.3 GHz. The time delay axis has been normalized to the strongest received component and each received signal path is represented as an impulse. Figure 9 shows that there were four defined signal components reaching the receiver antenna and very few relatively insignificant scatter components within 150 ns time delay. However, Figure 10 shows that there was one strong path and numerous small scatter components.

[23] In the in-leaf condition, signal propagation in the line of trees geometry was significantly influenced by diffraction around the edges and top of the trees, especially at 1.3 and 2 GHz. Some of the diffracted signals arrived too close, in time, to the component that has propagated through the vegetation and hence could not be resolved by the SVD-P algorithm. In Figure 9, the signal impulse at 0 ns is the through vegetation component that is thought to be influenced by diffracted and ground reflected paths. A distinct diffracted signal component with time delay around 10 ns is also shown in Figure 9. Reflected and scatter components off adjacent lines of trees were received at 40 ns and 100 ns time delay. The scattering effects of leaves on the through vegetation signal and the influence of diffracted paths at 1.3 GHz can be seen by comparing the components at 0 ns for in-leaf (Figure 9) with that from the out-of-leaf (Figure 10) condition. The out-of-leaf impulse response shows the first received component with a stable power (amplitude), emphasizing the lack of influence from diffracted paths that can induce temporal variations in the received signal, for example, due to wind.

[24] While diffraction, scattering and reflections were the propagation modes observed at J. J. Nursery, the measurements conducted at another site (Fermi Avenue) demonstrated the importance of site selection in propagation studies. The line of Horse Chestnut trees was located adjacent to a busy road lined with metal lampposts. Detailed analysis of the measured data showed that the received signal suffered from strong reflections off the lampposts and passing traffic. The combination of these factors resulted in negative excess attenuation at certain points. The wideband analysis and the results obtained at this second site (Fermi Avenue) are important as the parameter values recommended in ITU [2001] were based on measurements conducted at this site, and so are contaminated by artifacts of the measurement site.

6.2. Through Vegetation Geometry

[25] In this geometry, the transmitter and the receiver were located outside the vegetation as shown in Figure 1c. This is the scenario most likely to be encountered with practical systems. The measurements in this category include those carried out on Lawson pine and the sycamore wedge at RAL. The receiver antennas were positioned close to the edge of the vegetation to discriminate against over the canopy diffracted signal components. For this geometry, two main characteristics of a group of trees are important. In a group of trees, trees in the middle or away from the edge have few branches that are often found close to the canopy. However, trees at the edges have numerous branches and leaves from ground level to the top of the canopy.

[26] Figure 11 shows the excess attenuation curves for the measurements conducted on the sycamore. The excess attenuation curves show that the dual slope characteristics is less defined and it exhibits an almost linear increase with vegetation depth, with a peak value of around 30 dB at 11.6 GHz. By comparison the results from the Lawson pine measurements had a steep initial gradient at all the frequencies before rapidly leveling off. For these measurements attenuation at two receiver heights was investigated. The results show that, overall, higher attenuations were measured at 7.5 m antenna height than at 2.5 m. This could be expected due to increased branch and leaf density at greater heights.

Figure 11.

Sycamore in-leaf excess attenuation for through vegetation geometry (receiver antenna height: 7.5 m).

[27] The differences in the excess attenuation levels of the pine and sycamore can be explained by differences in tree densities. The pine forest was densely planted while the sycamore was sparse. After the initial slope of the excess attenuation curve for the Lawson Pine measurements, the signal reached the receiver by means of forward scattering. The trunk density ensured that all the frequencies were attenuated similarly. However, in the sycamore, the low trunk and branch density meant that 1.3 and 2 GHz signals suffered very little attenuation. For the through vegetation geometry, an increase in the received signal is expected as the antenna height approaches the top of the canopy. The dominant propagation mode in this case would be diffraction over the top of the trees. Detailed investigation, through measurements and simulations, showed that the absence of the steep initial gradient in the excess attenuation against vegetation depth curve for the sycamore measurements was due to edge diffraction.

6.3. Edge of Forest

[28] Only one site with this geometry was investigated, the Sycamore wedge at RAL. The measurements were conducted for both in-leaf and out-of-leaf conditions. This geometry allowed investigations to be conducted to greater vegetation depths and also at relatively high trunk and branch density. The measurements were taken at two receiver antenna heights along the edge of the wedge of trees.

[29] The excess attenuation curve for in-leaf measurements, Figure 12, exhibits dual slope characteristics. The peak attenuation for 11.6 GHz in-leaf condition is the largest for all measurements, approximately 60 dB, and has the steepest initial slope. The peak attenuation at 1.3 and 2 GHz is between 30 and 40 dB. The reason for the large attenuation values at this site is due to a high density of branches and leaves. The out-of-leaf measurements show much lower attenuation values, peaking at 35 dB for 11.6 GHz, 20 dB for 2 GHz and 15 dB for 1.3 GHz.

Figure 12.

Sycamore In-Leaf excess attenuation for edge geometry (receiver antenna height: 5 m).

6.4. Into Vegetation

[30] This geometry includes the measurements conducted on Sycamore trees at Ravelin Park and Common Beech trees at QECP. Measurements were conducted at both 2.5 m and 7.5 m receiver antenna heights and only for the in-leaf condition. Analysis of the data showed only minor differences between the measurements at these antenna heights. Figure 13 shows the excess attenuation from QECP beech trees at 7.5 m antenna height. There were not many differences between the three frequencies and no dual slope characteristic was identified. The results also show a linear increase in attenuation with vegetation depth. The large clearance below the high canopy for this particular geometry meant that mainly tree trunks and very few branches were in the line of sight path between the transmitter and the receiver antenna. Compared with other geometries the canopy for these sites were relatively flat and thin. This created suitable conditions for lateral wave propagation along the top of the canopy and in-ward propagation through gaps [Tamir, 1977]. This was more evident in the measurements carried out at QECP where less attenuation was recorded at 7.5 m receiver antenna height than at 2.5 m.

Figure 13.

Beech in-leaf excess attenuation for through vegetation geometry (receiver antenna height: 7.5 m).

6.5. Effect of Wind

[31] Figure 14 shows the effect of wind on the impulse responses of the channel at 2 GHz and Figure 15 shows measurements carried out at the same site on a nonwindy day. The effect of the branches swaying in the wind on the channel response is pronounced at the receiver antenna height of 7.5 m, which was the canopy height for most of the trees. The movement of the branches causes fast variations in the amplitudes and phases of the received components that introduce fast variations in the signal envelope. At 11.6 GHz the effect of wind is more pronounced and does not result in the well defined components found at 1.3 and 2 GHz, but manifest as multiple scatter components. This is because most of the twigs and branches are of a size comparable to the wavelength of the signal, thus the random motion causes scattering. Figure 16 illustrates traces of the narrowband signal measured at different frequencies. It clearly shows that as the frequency increases the temporal variations become very rapid.

Figure 14.

Sycamore 2 GHz estimated impulse responses with wind (receiver antenna height: 7.5 m).

Figure 15.

Sycamore 2 GHz estimated impulse responses without wind (receiver antenna height: 7.5 m).

Figure 16.

Frequency dependency of temporal signal variation.

[32] Wideband channel parameters were calculated for all the measurements. While these results are not presented in this paper, some trends in the parameters were identified. Analysis of the data measured in windy condition showed that on average, the delay spread increased from approximately 12 ns to 16 ns at 2 GHz compared to measurements conducted in nonwindy condition.

[33] Table 7 summarizes the excess attenuation and RMS delay spread at some vegetation depths for 1.3, 2 and 11.6 GHz measurements at a few sites. Generally, the delay spread for in-leaf measurements was greater at 11.6 GHz than results obtained from out-of-leaf investigations. However, this was not the case at 1.3 and 2 GHz where larger values of delay spread were measured in out-of-leaf state than in-leaf, except for London Plane.

Table 7. Estimated Excess Attenuation and RMS Delay Spread at 1.3 GHz, 2 GHz, and 11.6 GHz
Tree TypeLeaf StateParameter EstimateVegetation Depth, m
1.3 GHz2 GHz11.6 GHz
5 m15 m50 m5 m15 m50 m5 m15 m50 m
London PlaneInExcess attenuation, dB514184161651826
  Delay Spread, ns224316141058
 OutExcess attenuation, dB313173152371325
  Delay Spread, ns686210105109
Silver MapleInExcess attenuation, dB2131651619203540
  Delay Spread, ns45939103670
 OutExcess attenuation, dB151625152416152635
  Delay Spread, ns2253715101620
Common LimeInExcess attenuation, dB5121881424223858
  Delay Spread, ns44186121798100
 OutExcess attenuation, dB6121015172081123
  Delay Spread, ns324582010414
Horse ChestnutInExcess attenuation, dB002000200020
  Delay Spread, ns054005404440

7. Discussion and Conclusions

[34] This paper has reported an investigation to study signal propagation through vegetation and generate a database of measurements that can be used to evaluate existing narrowband vegetation attenuation models. The wideband study has shown that different combinations of diffraction, reflection and scatter propagation mechanisms are present at every site. The superposition of signal components propagating through these mechanisms results in significant frequency selective fading, manifested by variations in excess attenuation with measurement positions. The results have also shown that the measured signal levels exhibit spatial and temporal variations due to changes in vegetation density and movement of the vegetation components.

[35] Different measurement and vegetation geometries have been investigated. Of all the geometries, the into-vegetation geometry was found to be the only one that did not exhibit a distinct dual gradient characteristic. The rest of the geometries exhibited similar wideband characteristics and attenuation trends. Excess attenuation due to vegetation has been found to be affected mainly by the vegetation density and measurement geometry for a particular leaf condition. Although it could be deduced from the measurements that over the top diffraction and through canopy downward propagating components minimized the attenuation observed, further studies are required to confirm this assertion. This has implications in satellite communications where the ground station is located within trees.

[36] Three empirical vegetation attenuation models have been evaluated using the measured data. These models include the modified exponential decay (MED), maximum attenuation (MA) and the nonzero gradient (NZG) models. The fittings were carried out on (a) all the data and, data sets divided according to, (b) leaf state (in-leaf and out-of-leaf), (c) leaf shape, and (d) measurement geometry. The results show that (1) the models performed better in out-of-leaf than in-leaf state; (2) the NZG model performs best at 1.3 GHz, although it has been recommended only for frequencies above 5 GHz (the linear slope model proposed in ITU [2001] is considered inaccurate, as dual slope mechanisms have been noted at 1.3 and 2 GHz); (3) the shape of the leaves do not have a significant impact on attenuation, for the three frequencies investigated; (4) the MA model gives the worst fit to data of all the three models evaluated; (5) globally, the MED model gives more consistent results than the MA and NZG models, although higher accuracy could be achieved with the NZG model for some measurements; and (6) the measurement and vegetation geometries are a very important factor in the modeling and the determination of attenuation due to vegetation.

[37] The MED model has proved to be more consistent in predicting attenuation for a wide range of scenarios. In the ITU [2001] recommendations, into-vegetation geometry is separated from all other geometries, and the MA model is recommended for calculating the excess attenuation. However, the MA model was not found to perform distinctly better than the other models in this investigation. Based on the results from the measurements, it is proposed that a linear model will generate a better fit for this environment, although more studies are required.

[38] Figure 17 shows the fitting to one of the measurements (at 11.6 GHz) using the model parameters estimated from the whole database. The values are given in Tables 3–5 in the rows labeled “All.” The values for the NZG model parameters as recommended in ITU [2001] have also been used to fit the model to the data set. The results indicate that better predictions of excess attenuation are obtained using the parameter values calculated using measurements from different sites. The large mean square error obtained (approximately 20 dB) by using parameter values in the ITU-R has been found to be due to the fact that signal propagation at Fermi Avenue, the site used to obtain the ITU-R parameter values, is significantly influenced by reflections from metal lampposts and passing traffic. Thus the authors advocate the use of different parameter values for NZG or MED model based on geometry and/or leaf condition, as given in Table 5, or the values given in the row labeled “All” if the geometry is not known for frequencies between 1 and 11 GHz.

Figure 17.

Comparison of the fitting of models to measured data using parameter values given in ITU [2001] and values estimated from “all” (Tables 3–5) data at 11.6 GHz.

[39] The study has revealed that there is a strong frequency dependency of signal attenuation by vegetation, with high frequencies experiencing more attenuation than low frequencies. Wind has been found to influence delay spread, but this will have a noticeable impact on a communication system only if the receiver is located very close to or inside the vegetation. The effects at 1.3 and 2 GHz manifest as a variations in the amplitude and time delay of clearly defined multipath components. At 11.6 GHz, however, the wind scatters well defined signal paths (impulses) into random and severely attenuated components, especially in-leaf condition.

[40] Wideband analysis shows that RMS delay spreads were consistently less than 20 ns at 1.3 and 2 GHz for foliage depth less than 50 m. However, larger values were obtained at 11.6 GHz especially at large vegetation depths. There was a general lack of correlation between excess attenuation and delay spread. This means that, although a narrowband model may predict the reception of a strong signal, the signal may be rendered unintelligible for communication purposes by multipath.

[41] The results have emphasized the importance of careful site selection in signal propagation studies to ensure that appropriate measurement geometries are chosen and taken into account in the interpretation of the results. Most importantly, care needs to be exercised to ensure that site specific artifacts do not influence the spatial and temporal characteristics of the channel. The study has shown that a generic vegetation attenuation model must consider all modes of signal propagation by taking account of reflection and diffraction in addition to through-vegetation propagation, where necessary. This has culminated into a study aimed at developing a generic vegetation attenuation model, using the knowledge gained and data obtained from the measurement campaign reported in this paper. The model, while modeling ground reflection and diffraction based on the geometry, uses the radiative energy transfer (RET) technique to estimate the signal propagation through the vegetation medium. Preliminary results of this generic model development study are reported by Qinetiq [2002] and are beyond the scope of this paper.


[42] Special thanks to the U.K. Radiocommunications Agency for sponsoring the project and to the other consortium members; Rutherford Appleton Laboratory, Qinetiq and the University of Glamorgan. We would also like to thank J. J. Nurseries Ltd. of Twyning, the management of Queen Elizabeth Country Park, Hampshire, and all others who kindly granted us access to their property during the measurement campaign.