Channel Measurements and Modeling at 6GHz in the Tunnel Environments for 5G Wireless Systems

,


Introduction
The next generation (5G) of wireless communications will use systems operating from 500 MHz to 300 GHz [1,2].The 3 GHz-30 GHz spectrum is defined as the super high frequency (SHF) band, while 30-300 GHz spectrum is assigned to the extremely high frequency (EHF) or millimeter-wave band (mmWave) [3].Because radio waves in the SHF and EHF bands share similar propagation characteristics, the 3-300 GHz spectrum, with wavelengths ranging from 1 mm-100 mm, can be referred to as the EHF bands [4,5].The EHF bands will be a key component, which can deliver multigigabit-per-second transmission throughputs for various multimedia services for 5G radio systems [6,7].Thus, fundamental knowledge of the EHF channel propagation characteristics including accurate and reliable channel models is vital for developing 5G wireless communication systems.
Although a large number of channel-measured results can provide reliable channel model, the expense of highprecision measurement equipment is very high and the number of observation points is limited in measurement campaigns, which are both great challenges for channel sounding.As an another approach, the shooting and bouncing ray-tracing/image (SBR/IM) method [33] is an effective method, which can extend the sparse empirical datasets and can be utilized to investigate the propagation characteristics of EHF bands.Also, it is able to capture multipath components in time-varying wireless channels and can prove the accuracy and effectiveness of the proposed channel model by comparing simulated channel model with the actual measured channel model.In this paper, we verify that the simulated results provide good agreement with measured results over all the measured paths, as we compared simulations with measurements in the same tunnel environment and distance ranges for verification.
Despite many measurement campaigns were conducted up to now, there are still omitted environments and frequency bands, which require to be intensively investigated.Hence, extensive actual channel measurements and simulations should be performed for various scenarios and significant frequency bands.By analyzing the measured results and simulated results, a standardized channel model should be proposed.The contribution of this paper is fivefold.First of all, channel measurements and simulations in the tunnel environments are performed and analyzed at 6 GHz, which have not been detailed in previous work.Therefore, propagation characteristics are analyzed at 6 GHz in the tunnel environment.Second, the omnidirectional path loss models are characterized based on close-in (CI) free-space reference distance model and the alphabeta-gamma (ABG) model.In addition, the comprehensive parameter table of path loss models including measured results and simulated results for all scenarios is given.Third, the statistical analysis of the RMS delay spread for tunnel scenarios is described using measurement datasets.The fourth part of the contribution is to investigate the different distribution models based on the cumulative distribution functions (CDF) in terms of the parameters of received power, Ricean K-factor, and angle of arrival (AOA).Finally, some important angle of arrival is extracted based on the subtractive clustering algorithm.
The rest of the paper is organized as follows.The measurement setup and environments are described in Section 2. The results and analysis of path loss models are presented in Section 3. Specially, the parameters of received power, Ricean K-factor, and angle of arrival (AOA) are evaluated in Section 3. Finally, conclusions are drawn in Section 4.

Measurement Setup and Environments
2.1.Measurement Setup and Procedure.Measurements were conducted in the frequency domain in which the channel impulse responses (CIRs) were measured and recorded.Figure 1 illustrates an overview of the channel measurement setup.The height of the transmission antenna is 1.8 m, while the height of the receiver antenna is 1.6 m above the floor in the tunnel environments.Table 1 shows the 6 GHz channel sounding system parameters, and more detailed information of the channel sounder and calibration method can be found in [34].respectively.The receiver antenna moves along a straight line.The first to be taken (LOS-1 measurement) was performed at a sampling distance of 1 m.The transmitter (TX1) was placed almost at one end of a long tunnel, while the receiver (RX) was moving along a LOS route (A-B, 19 m), which corresponds to 20 samples as shown in Figure 2(c).The second to be taken (LOS-2 measurement) was also taken at a sampling distance of 1 m.The transmitter (TX2) was placed at the corner, while the receiver (RX) was moving along a LOS route (C-D, 11 m), which corresponds to 12 samples as shown in Figure 2(c).Moreover, the third to be taken (NLOS measurement), the transmitter (TX1) was also placed almost at one end of a long tunnel, while the receiver (RX) was moving along a route (A-B-C-D, 30 m), which corresponds to 31 samples as shown in Figure 2(c).The last to be taken (LOS-3 measurement) was taken at a sampling distance of 0.5 m.The transmitter (TX) was indicated in the tunnel scenario, while the receiver (RX) was moving along a LOS route (A-B, 8 m), which corresponds to 17 samples as shown in Figure 2(d).At each receiver location, measurements were acquired using a circular track with 8 equally spaced local area measurement points separated by 45 ∘ increments.The radius of the circular track yielded 3λ and 2λ separation distance between consecutive points along the circular track in tunnel 1 and tunnel 2 scenarios, respectively.Note that in the measurement, no one was moving in the tunnel, so as to eliminate the influence of moving persons.

Ray-Tracing Simulations.
The shooting and bouncing ray-tracing/image (SBR/IM) method is developed to deal with the radio wave propagation for complex environment.
It can track all the triangular ray tubes bouncing with high accuracy and computational efficiency.If the RX is within a ray tube, the ray tube will have contribution to the received field at RX and the equivalent source (image) can be determined.So SBR/IM method is an effective method which can be used to predict the propagation characteristics at 6 GHz.The ray-tracing simulation performs the method of SBR/IM using the software tool, Wireless InSite [35].In our simulation settings, the ray-tracer accounts for up to 6 reflections, 2 penetrations, and 1 diffraction for each ray.The material properties are frequency-dependent, and the parameters of dielectric constant ε r and conductivity σ in this paper are estimated for the 6 GHz spectrum band based on the material properties in different bands [36][37][38][39].The buildings are assumed to be concrete with the dielectric constant ε r = 6 94 and σ = 0 73S/m, respectively.For each simulated RX point, a discrete equivalent baseband CIR was composed from the simulated rays based on their complex amplitude and delay spread.Furthermore, the ray-tracing simulation performed for the purpose of validation is presented for proper comparison and validation of simulated results and empirical results.This is further discussed in Section 3.

Channel Models and Statistical Analysis
3.1.Path Loss Models.The path loss is a significant parameter which can be applicable to describe the large-scale effects of the propagation channel [40].In this work, two kinds of path loss models are considered based on the measurements and simulations, namely, the close-in (CI) free-space reference distance path loss model [41,42] and the alpha-betagamma (ABG) path loss model [9].
Two well-known models are used to develop omnidirectional path loss models in this paper.First of all, the equation for the CI model is given by where d 0 is the close-in free-space reference distance, f is the carrier frequency in GHz, c is the speed of light, n denotes the path loss exponent (PLE), d is the distance between transmitter and receiver, and X CI σ is a zero-mean Gaussian random variable with a standard deviation σ in dB (shadowing effect).The minimum mean square error (MMSE) is used to calculate PLE and standard deviation.The physical interpretation of n = 1 is a guided wave in one plane, n = 2corresponds to a free-space path loss, and n = 4 corresponds to a situation, where low antenna heights cause the first Fresnel zone to be obstructed [15].
The ABG model is another famous model which can be used to discuss the frequency dependence of path loss.It can be defined as [24] where PL ABG f , d denotes the path loss in dB over frequency and distance; α and γare coefficients showing the dependence of path loss on distance and frequency, respectively; β is an optimized offset value for path loss in dB; and X ABG σ is a zero-mean Gaussian random variable with a standard deviation σ in dB derived from MMSE closed-form optimization.

International Journal of Antennas and Propagation
When used at a single frequency, the ABG model reverts to the existing 3GPP floating-intercept (FI) model with three parameters where γ set to 0 or 2 [5,23].Moreover, to verify the model accuracy, we use the simulated channel parameters to compare them with that of measured results.
Different path loss models have been deduced based on extensive wideband measurements at 6 GHz in terms of LOS and NLOS scenarios.Figures 3(a), 3(b), 3(c), and 3(d) show the scatter plots of all measured and simulated path loss values in dB versus TX-RX separation distance optimized for the CI, CI-opt, and ABG models in both LOS and NLOS environments.As a reference, free-space path loss models are also included.Table 2 summarizes the parameters of path loss models including measured results and simulated results for all scenarios including ABG models and CI models which consist of 1 m CI models and optimized d 0 models (CI-opt models).
From Figure 3, we can see that the values of path loss increase with increasing distance between transmitter and receiver.From Table 2, we can also see that the measured PLEs of CI are 1.72, 1.69, and 1.58 for LOS-1, LOS-2, and LOS-3 paths, respectively.Also, the simulated PLEs of CI are 1.66, 1.50, and 1.74, respectively.The measured PLEs of CI-opt are 1.75, 1.45, and 1.45, respectively.Furthermore, the simulated PLEs of CI-opt are 1.61, 1.55, and 1.71, respectively.The values of PLE are identical for measured CI-opt models in both LOS-2 and LOS-3 paths.These results indicate that the values of PLE are less than the free-space PLE (n = 2), implying that the multipath components (MPCs) from both side walls and floor add up constructively in the tunnel environments, as a guided wave phenomenon.It is seen that the measured PLE and simulated PLE of CI models are 2.18 and 2.20 for NLOS paths, respectively.Also, the measured PLE and simulated PLE of CI-opt models are 2.41 and 2.11 for NLOS paths, respectively, indicating faster signal level degradation over the distance between transmitter and receiver.Obvious explanation for this is that there are different dominating propagation mechanisms due to different structures of the tunnel scenarios.As indicated in Table 2, the standard deviations for CI models vary between 0.53 dB and 3 dB for LOS paths and the standard deviations for CI models vary between 2.74 dB and 4.2 dB for NLOS paths.Also, the standard deviations for CI-opt models vary between 0.38 dB and 2.10 dB for LOS paths and the standard deviations for CI models vary between 1.38 dB and 2.74 dB for NLOS paths.It is interesting to note that the difference of standard deviations for CI-opt models and CI models is always less than 1 dB for majority measurement and simulation sets, which demonstrate that there is no essential difference in standard deviations between CI models and CI-opt models.However, the standard deviations for ABG models vary between 1.06 dB and 8.27 dB for LOS paths and the standard deviations for ABG models vary between 7.41 dB and 7.68 dB for NLOS paths.It is worth noting that the difference of standard deviations for ABG models and CI models is always larger than 1 dB for majority measurement and simulation datasets indicating the higher accuracy and reliability of CI models compared to ABG models.
From these analyses, we can conclude that CI model with the reference distance of 1 m provides more accuracy and stability in LOS and NLOS tunnel scenarios at 6 GHz.These simulated results are in agreement with previous works [15,16,22,32,43,44] which indicate that the path loss is best modeled by a log-distance relationship.
3.2.Received Power Results and Analysis.In wireless communication channels, the signal is transmitted and then undergoes direct reflection, transmission, scattering, and diffraction.Hence, the signal arriving at the receiver is the superposition of the various multipath components [45].All cumulative distribution functions (CDF) of received power including measured results and simulated results are demonstrated for LOS and NLOS as shown in Figure 4.In both LOS and NLOS tunnel scenarios, the ray-tracing predictions agree fairly well with the measured results.There are slight differences between measured and simulated results as shown in Figure 4.They could be attributed to scattering by small objects within the tunnel scenarios and are not expected to influence the performance of wireless communication system.Moreover, the statistical method to determine the distribution of the CDF of received power after regression analysis is needed.Notice that the CDF is close to a normal distribution.The parameters of the distribution are listed in Table 3. μ and σ denote the mean and standard deviations of the normal distribution, respectively.

Statistical Analysis of RMS Delay Spreads.
The mean excess delay and root-mean-square (RMS) delay are two important parameters used to characterize the temporal dispersive properties of multipath channels.The mean excess delay τ m is defined as the first moment of the power delay profile (PDP) [46].
where a k , τ k , and P τ k are the gain coefficient delay and PDP of the kth MPC, respectively.The RMS delay spread (τ rms ) is the square root of the second central moment of the PDP and is defined to be where The RMS delay spread for LOS-2, LOS-3, and NLOS paths is shown in Figure 5.The RMS delay spread values vary between 2.77 and 18.76 ns.From Figure 5, we can see that the correlation between RMS delay spread and the distance between transmitter and receiver is very low.Also, it is seen that the RMS delay spread values in LOS paths are lower than the values in NLOS path due to the difference of structure of tunnel and the small scattering objects.In fact, the presence of obstructions in tunnel environment (compared with the 6 International Journal of Antennas and Propagation    Gaussian distribution.In order to identify the distribution, the Log-likelihood test is adopted.After estimating the maximum likelihood of the distribution parameters, it is, therefore, evident that the Poisson distribution best fits the measured data of RMS delay spreads for LOS scenarios and the Gaussian distribution best fits the measured data of RMS delay spreads for NLOS scenarios as shown in Figure 6.Moreover, it is apparent that 90% of the energy arrived at the RX within 13.2 ns for LOS scenarios and 14.6 ns for NLOS scenarios.The mean, minimum, and maximum of RMS delay spreads are listed in Table 4 for all tunnel environments.
3.4.Ricean K-Factor Analysis.K-factor is a significant parameter in wireless communications because it is able to characterize the type of fading environments [48].Ricean K-factor is the power ratio between the LOS component and the sum of other propagation components [42].K-factor can be calculated as where P k max is the maximum power for LOS path and P k is the total power for all paths.Ricean K-factor (KF) values at different distances between transmitter and receiver for LOS paths are shown in Figure 7.The maximum K-factor for LOS-1 is 12.67 dB, and the mean is 7.67 dB.Specially, the maximum K-factor for LOS-3 is 11.21 dB and the mean is 6.73 dB which are consistent to the values for LOS-1.However, it is seen that the maximum K-factor for LOS-2 is 18.30 dB and the mean is 9.61 dB, which are larger than the values of LOS-1.The difference between them is mainly because tunnel 1 and tunnel 3 are straight tunnel and tunnel 2 is curved tunnel which may result in fewer multipath components.It is seen that the correlation is low between Ricean K-factor and the distance between transmitter and receiver.Moreover, the CDF plots of K-factor are shown in Figure 8.The results lead to an observation that the normal distribution provides good fits to the Ricean K-factor for LOS tunnel scenarios.The statistical parameters for K-factor (expressed in dB) are shown below: 3.5.AOA Analysis.The angles θ A and φ A with reference to the spherical coordinate system give the direction from which the propagation path arrives at receiver point.The mean angle of arrival from which energy arrives at the receiver is defined as where P i is the power carried by ith path and a i is the unit vector in the direction from which the ith path arrives at the receiver.In terms of the angle of arrival, some conclusions are presented.The distribution of AOA in the elevation plane has been researched [49,50].The distribution of AOA in the elevation plane is described as Gaussian [49] and uniform [50].However, there are few reports on the characteristics of distribution of AOA at 6 GHz.The CDF distributions of mean AOA are shown in Figure 9. From Figure 9, we can see that the normal distribution provides good fits to the mean AOA for LOS-1, LOS-3, and NLOS paths.Moreover, the best distribution that fits the measured data is the   5.These parameters presented here can be used for wireless communication system design in indoor environments and can be used for adaptive transmission technologies for 5G wireless network.

Conclusion
In this paper, extensive measurements and characterizations of wideband tunnel channel have been proposed.Channel characteristics such as path loss models, received power, RMS delay spread, Ricean K-factor, and AOA are described and modeled.Based on extensive radio channel sounding campaigns and simulations, it is found that the ray-tracing predictions agree fairly well with the measured results.Comparison with path loss models illustrates the fact that the CI model with the reference distance of 1 m was shown to be the most suitable because of its accuracy and simplicity in tunnel scenarios.The CI path loss models indicated that the PLEs vary between 1.50 and 1.74 in LOS scenarios and between 2.18 and 2.20 in NLOS scenarios.The CDF of received power follows the normal distribution.The normal distribution provides good fits to the Ricean K-factor for LOS scenarios.The Poisson distribution model best fits the measured data of RMS delay spreads for LOS scenarios, and the Gaussian distribution model best fits the measured data of RMS delay spreads for NLOS scenarios.Moreover, the normal distribution and the uniform distribution reasonably fit the AOA well for all tunnel scenarios.In addition, propagation characteristics with the effects of human movement and vehicular motion on wireless channels require further measurement and analysis in the complex environments, which is the next research direction.

Figure 4 :
Figure 4: Cumulative distribution function (CDF) of the received power at 6 GHz in LOS and NLOS tunnel scenarios.(a) LOS-1 and NLOS paths and (b) LOS-2 and LOS-3 paths.

Figure 5 :
Figure 5: RMS delay spread variation with TX-RX separation distance in LOS and NLOS tunnel scenarios.(a) LOS-2 and LOS-3 paths and (b) NLOS path.

Figure 7 :
Figure 7: Ricean K-factor variation with TX-RX separation distance in LOS tunnel scenarios.

Figure 8 :
Figure 8: CDF for Ricean K-factor in LOS tunnel scenarios.

Figure 9 :
Figure 9: CDF for mean AOA in tunnel scenarios.
R X se p ar at io n d is ta n ce (m ) R e c e iv e d p o w e r ( d B m ) AOA (degree) R X se p ar at io n d is ta n ce (m ) R e c e iv e d p o w e r ( d B m ) AOA (degree) R X se p ar at io n d is ta n ce (m ) R e c e iv e d p o w e r ( d B m ) AOA (degree) R X se p ar at io n d is ta n ce (m ) R e c e iv e d p o w e r ( d B m ) AOA (degree) (d)

Table 1 :
Channel sounding system parameters used at 6 GHz.

Table 2 :
Parameters in the ABG, CI, and CI-opt path loss models in tunnel scenarios in terms of LOS and NLOS paths at 6 GHz.Dis.Ran.denotes distance range.Number of data points denotes the number of data points.

Table 3 :
Received power values for all scenarios.

Table 4 :
RMS delay spread values for all scenarios.

Table 5 :
AOA parameters for all scenarios.