Radio propagation measurements and modeling for standardization of the site general path loss model in International Telecommunications Union recommendations for 5G wireless networks

The International Telecommunications Union Radiocommunication Sector (ITU‐R) Study Group 3 identified the need for a number of radio channel models in anticipation of the World Radiocommunications Conference in 2019 when the frequency allocation for 5G will be discussed. In response to the call for propagation path loss models, members of the study group carried out measurements in the frequency bands between 0.8 GHz up to 73 GHz in urban low‐rise and urban high‐rise as well as suburban environments. The data were subsequently merged to generate site general path loss models. The paper presents an overview of the radio channel measurements, the measured environments, the data analysis, and the approach for the derivation of the path loss model adopted in Recommendation ITU‐R P.1411‐10 (2019‐08).


Introduction
In the World Radiocommunications Conference 2015, WRC15, seven frequency bands with different bandwidths from 1.6 to 10 GHz in the frequency range 24-86 GHz were identified for possible allocation for future 5G wireless networks. This has prompted several propagation studies across the world to derive path loss models for the envisaged deployment scenarios and to estimate wideband channel parameters such as r. m.s. delay spread to aid in the design of 5G networks (Keusgen et al., 2014;mmMagic Deliverable 2.2, 2017;Raimundo et al., 2018;Sun et al., 2016;. Due to the spread across multiple bands with different bandwidths and the need to characterize the radio channel in these new bands in preparations for WRC 2019, a number of correspondence groups were formed in three of the working parties of Study Group 3 (SG3) of the International Telecommunications Union (ITU) which provides recommendations on radio wave propagation. Due to the high path loss and blockage in the identified frequency bands, three correspondence (CG) groups were set up to tackle different aspects of the radio channel. CG-3K-6 was set up within working party (WP) 3K to harmonize the path loss models over short urban paths in ITU-R recommendation P.1411. CG-3K-3M-12 was set up to predict clutter loss, jointly between WP 3K and WP 3M which deals with point-to-point and earth-space propagation. The third CG-3J-3K-3M-8 was set up between WP 3K, WP 3M, and WP 3J which deals with propagation fundamentals to model building entry loss.
In this paper, we give an overview of the work carried out by CG 3K-6 to derive the site general path loss model adopted in the recent Recommendation ITU-R P.1411-10 (2019) for two out of the five propagation environments classified in the recommendation. These are (i) urban high rise characterized with tall buildings of several floors each and (ii) urban low rise/suburban with wide streets and building heights with less than three stories making diffraction over roof top likely. Both line of sight (LOS) and non-LOS (NLOS) were measured for two distinct scenarios: (i) below the rooftop scenario where both stations are below the height of the surrounding rooftops, where station refers to either the transmitter or the receiver and (ii) above the rooftop scenario, where one station is above the rooftops of neighboring buildings and the second station is below the rooftops.
In this paper, we start by describing the different measurement equipment including calibration procedures. This is followed by the methodology of data collection and data analysis to derive the path loss model.

Measurement Equipment and Environment
Three different types of equipment covering different frequency bands were used in the measurements. These include multiple band continuous wave (CW) transmissions and wideband measurements using either a dual band pseudo random binary sequence (PRBS) or frequency-modulated continuous wave sounder (FMCW, also known as chirp). The wideband sounders have varying bandwidths ranging from 250 MHz to 6 GHz and the measurements extended over different distances up to 1,200 m. In this section, we detail the equipment used in the measurements, the calibration procedures, and the measurement environment.
Each transmitter and receiver were individually calibrated where the power output of each transmitter was verified via a power meter, and each receiver was connected to a calibrated signal generator to determine its dynamic range as illustrated in Figure 2. A high gain Low Noise Amplifier (LNA) was used with narrowband receivers for the detection of low power signals.
The setup was used to conduct measurements in two scenarios (Sasaki et al., 2018(Sasaki et al., , 2017(Sasaki et al., , 2017(Sasaki et al., , 2015(Sasaki et al., , 2018. For above the rooftop scenario shown in Figure 3, the transmitter was set up at 55 m above ground and the receiver at 2.5 m. For the 66.5 GHz measurements, the half power beam-width (HPBW) of the transmit antenna was 30°while it was 60°for the 2.2-, 4.7-, and 26.4-GHz bands. For all frequency bands, the receive antenna had an omni-directional radiation pattern.
For below the rooftop measurements, both the transmitter and receiver antennas were omni-directional with the transmit antenna set up at 10 m and the receive antenna at 2.5 m. The measured environments are shown in Figures 3a-3b where buildings with multiple stories (about 40-m high) were lined along the road with typical road width of about 30 m.

Wideband Measurement Equipment
Two waveforms were used in the wideband measurements: PRBS and FMCW. The PRBS sounders clock rates were either 250 or 500 MHz while the FMCW sounder had a programmable bandwidth with a maximum bandwidth of 6 GHz in the 50 to 75-GHz band and 3 GHz in the 25 to 30-GHz band.
The 500-MHz PRBS sounder used a sequence length of 4,095 with a 12,500 sliding-factor. Using different Radio Frequency (RF) up converters, it covered the 28-GHz band and the 38-GHz band with 29-and 21-dBm output power, respectively. Figure 4 displays the different modules used in the sounder, which also used a 3-D positioner to mount directional horn antennas while controlling the boresight of the antenna with an accuracy of 1° (Lee et al., 2016).
The sounder is calibrated from back to back (B2B) tests with calibrated attenuators as shown in Figure 5a and from on the air measurements. The resulting B2B impulse response of the sounder, is shown in Figure 5b which following equalization closely approximates the ideal impulse response with about 47-dB peak to noise ratio (Kwon et al., 2015).
To capture the effects of the antenna and cable connections, an open-area calibration along a straight-line road is conducted using an identical setup to that employed in the field measurements. Figure 6 shows that the calibrated open-area path loss follows the theoretical two-ray model where the discrepancies are attributed to the nature of the road in the open area (Lee et al., 2018).
The sounder was used in below the rooftop measurements in the urban high-rise environments of Figures 7a, 7b, and 7c and the urban low-rise environment in Figure 7d. The transmitter height was set either at 10 or 4 m while the receiver height was fixed at 1.5 m. The transmitter antenna with a 30°H PBW was pointed towards the receiver which used an omni-directional antenna. The data were either Similarly, the 250-MHz PRBS sounder which covered the 10-and 60-GHz bands was calibrated from B2B tests and open environment tests. The sounder was used to collect data in urban high-rise and suburban low-rise environments.
The FMCW sounder described in (Salous et al., 2016) was upgraded in order to cover additional frequency bands as identified by WRC15. This was achieved by a programmable local oscillator (ADF5355) as shown in Figure 8 to up convert the Intermediate Frequency (IF) signal in the 2.2-2.9 GHz band to the band  between 12.34 to 18.2 GHz with a maximum bandwidth of 1.5 GHz. Using the new IF unit in conjunction with two RF heads as shown in Figure 9, three of the frequency bands identified by WRC15 are covered; the 25.5-28.5 GHz and 51-57 GHz bands were measured simultaneously, and the measurements were repeated for the 67-73 GHz band along the same route. At the transmitter a two-way switch was used to enable the switching between the two bands and using built in switches in the RF heads, horizontal and vertical polarizations were transmitted using directional antennas with 55°HPBW in the 50-75 GHz band and 33°for the 25-to 30-GHz band. The use of the additional switch at the transmitter enables the identification of the polarization at the transmitter by introducing an off period. Thus, for each band, the sequence of transmission was horizontal, vertical, and two off periods where each period corresponds to one sweep. At the receiver, omni-directional antennas were used for all the bands. The sounder was calibrated from B2B measurements and on the air measurements in an anechoic environment.
The sounder was used in suburban below the rooftop and above the rooftop, LOS, and NLOS measurements in the environment shown in Figure 10 with transmit antenna height either at 3 or 18.2 m, respectively, with the receiver antenna height being fixed at 1.6 m. Figure 11 displays an example of the power delay profile for the co-polar and cross polar transmission at 25.5-28.5 and 51-57 GHz in a LOS scenario. Data were collected continuously over 2 s every 1 min to provide consecutive spatial measurements.   Table 1 gives a summary of all the measured frequency bands, the type of transmission, the distance covered, and the propagation category.

Data Analysis and Derived Propagation Model
To derive the channel model for the categories defined in Recommendation ITU-R P.1411, it was necessary to identify a suitable path loss model approach as well as the spatial samples and the minimum signal to noise ratio to be used in the derivation of the model.

Model Approach
Several path loss models are proposed in the literature which are based either on a single frequency as in equation (1) or on multiple frequencies as in equation (2), where α and γ are the distance and frequency coefficients, d is the 3-D transmitter-receiver (T-R) separation distance in meters, f is operating frequency in GHz, and N(0, σ) is a normal distribution with standard deviation equal to σ describing the large-scale variations of the path loss about the mean over distance. The value of β in dB can be either estimated from the measurements, or fixed to the free space  path loss at 1 m as in the close-in path loss model (Sun et al., 2016). The model in equation (2) is referred to as the alpha, beta, gamma model and can be used across a wide range of frequencies without the need for a different coefficient for each frequency band as in the alpha, beta model of equation (1).
with an additive zero mean Gaussian random variable N(0, σ) with a standard deviation σ (dB).
Since the measurements were collected over a wide frequency range from 0.8-73 GHz, the model in equation (2) was adopted as it provides a single set of four parameters and only requires distance and frequency.

Data Verification
Since the collected data had varying number of spatial data points and different waveforms were used in the measurements, it was necessary to identify a suitable approach for the estimation of path loss, minimum acceptable signal to noise ratio, and the spatial sampling of the data. For the wideband measurements, the data were analyzed to estimate the received power from the area under the power delay profile (PDP) as in Figure 11 following the procedure in Recommendation ITU-R P.1407-6 (2019). For the wideband measurements performed with the FMCW sounder, the data were processed with 2-GHz bandwidth for the PDPs  for all the three bands. For each meter of measurement, five power delay profiles were estimated by averaging 488 impulse responses giving a PDP every 20 cm. For each PDP, the noise floor mean was estimated, and a 3 dB above the mean noise was used to set a noise threshold to ensure that the computed received power did not include noise samples. For CW measurements, a sampling rate of 45 kHz was used giving about 4,000 points of data per meter. The distance between the transmitter and receiver antennas was calculated using global positioning system data.
Since the data sets were collected with different number of spatial samples and with different systems over varying distances, it was necessary to identify a suitable common number of samples per meter to avoid the model being biased by any particular data set as illustrated in Figure 12a which shows the collected data versus distance across the frequency band from 0.8-70 GHz for the LOS below rooftop environment. Two approaches were tested for the decimation of data. The average values or the median value every 1 m were computed from each data set, and the model parameters were estimated. No significant difference was detected in the two approaches and the local average at every 1 m was adopted in the model as illustrated in   Figure 12b. Similarly, 10-and 15-dB signal to noise ratio thresholds were tested and a minimum signal to noise ratio of 10 dB was used for all the data sets. The data were also combined in different groups to identify the frequency range and distance for the estimation of the model parameters. Only vertical to vertical polarization data were used in the estimation of the model parameters as all the data sets except for the FMCW data were collected with single polarization.
The estimated model parameters for the different environments are given in Table 2 as given in ITU-R P 1411-10 tables 4 and 8.

Use of the Channel Model in Monte Carlo Simulations
In network design and sharing studies, Monte Carlo simulations are used to estimate the path loss from the model taking into account the standard deviation from the median value. Due to the high value of σ and the steeper slope of the NLoS model with respect to the free space deterministic path loss model, path loss values can be lower than free space as illustrated in Figure 13 at 70 GHz where the urban low rise/suburban model is shifted from the median by 1%, 5%, and 10% of σ in comparison to the free space path loss. The model gives values below free space loss for distances less than 80 m. Since the measurements did not have any values below free space path loss, a capping approach similar to Recommendation ITU-R P. 2109-0 (2019) model was investigated. The capping method limits the excess path loss with respect to free space loss, L FS , such that no values generated in the simulation fall below free space.
This corresponds to the condition that the difference with respect to free space will not exceed 10log 10 (10 0.1A +1) (dB), where A is a random variable with a normal distribution, N(μ, σ), μ = PL(d,f) − L FS , L FS = 20log 10 (4 × 10 9 πdf/c), and c is the speed of light in meters per second. The capping method is adopted in recommendation ITU-R P. 1411-10.
In Sun et al. (2016) values of path loss coefficients for the alpha, beta, gamma model are given for two scenarios' classified as UMa and UMi which refer to the transmit antenna height as either 25 m above rooftop or 10 m at rooftop. Therefore, according to this classification only the UMa results can be compared with the model reported in this paper. However, the values given in table 3 by Sun et al. (2016) cover the frequency range 2-38 GHz for distances 61-1,238 m, whereas the values presented in this paper cover the frequency range 2.2-66 GHz and for distances from 260 to 1,200 m. Similar classification of scenarios is also adopted

Radio Science
in the more recent 3GPP report "3GPP TR 38.900V15.0.0 (2018-06)" which gives path loss models for frequency ranges from 6 to 100 GHz. For the NLoS UMa scenario, the model is given in equation (3) which has a correction factor for the user-terminal antenna height, h UT , above 1.5 m. Assuming, that the user terminal antenna height is 1.5 m, the model gives values of 3.908 for alpha, 13.54 for beta, and 2 for gamma with sigma equal to 6 dB. PL 0 UMa-NLOS ¼ 39:08log 10 d 3D ð Þþ13:54 þ 20log 10 f c ð Þ−0:6 h UT −1:5 ð Þ dB for 10m≤d 2D ≤5km To avoid the path loss falling below free space, it limits the path loss to the maximum of PL ′ UMa−NLOS ; PL UMa−LOS À Á . This effectively leads to two different path loss coeffcicients; which is avoided in the capping method.
It also gives an optional model as given in equation (4), which assumes free space propagation for the value of alpha, a coefficient of 2 for frequency and has a value of 3 for alpha with sigma equal to 7.2 dB.
PL ¼ 30log 10 d 3D ð Þþ32:4 þ 20log 10 f ð Þ dB (4) The model presented in this paper is the adopted ITU model in its Recommendation ITU-R P.1411-10 (2019) with the approved capping method which aims to provide a model appropriate for the scenarios as defined in the recommendation.

Conclusions
Measurements were performed in different environments in Japan, Korea, and the United Kingdom as classified by Recommendation ITU-R P.1411 to derive a suitable path loss model for 5G wireless networks. The measurements covered frequency ranges between 0.8 to 73 GHz with either narrowband or wideband sounders. Data were systematically collected over typical distances, and the data were classified as LOS and NLOS for below the rooftop and above the rooftop scenarios, and the model parameters were estimated. The model was adopted in Recommendation ITU-R P.1411-9 and the capping approach approved for future updating of the recommendation.