Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Abstract Little is known about millimeter wave (mmWave) Mobile and Wireless Communications Enablers for the
path loss in rural areas with tall base station antennas; yet, as Twenty-twenty Information Society (METIS) [6], Millimetre-
shown here, surprisingly long distances (greater than 10 km) Wave Evolution for Backhaul and Access (MiWEBA) [7],
can be achieved in clear weather with less than 1 W of power.
This paper studies past rural macrocell (RMa) propagation Millimetre-Wave Based Mobile Radio Access Network for
models and the current third generation partnership project Fifth Generation Integrated Communications (mmMAGIC) [8],
(3GPP) RMa path loss models for frequencies from 0.5 to 30 European Telecommunications Standards Institute (ETSI) [9],
GHz adopted from the International Telecommunications Union- and IEEE 802.11ad [10]. The 3rd Generation Partnership
Radiocommunication Sector (ITU-R). We show that 3GPP and Project (3GPP), the global standards body of the wireless
ITU-R RMa path loss models were derived for frequencies below
6 GHz, yet are being asserted for use up to 30 GHz. Until this industry, released its study on channel models for frequencies
paper, there has not been published data to support mmWave above 6 GHz in 3GPP TR 38.900 V14.2.0 (Release 14) [11],
RMa path loss models. In this paper, 73-GHz measurements which has been aligned with channel models for below 6 GHz
in rural Virginia are used to develop a new RMa path loss such as 3GPP TR 36.873 (Release 12) or ITU-R M.2135
model that is more accurate and easier to apply for varying such that the models support comparisons from 0.5 GHz to
transmitter antenna heights than the existing 3GPP/ITU-R RMa
path loss models, and may be used for frequencies from 0.5 to 100 GHz [12].
100 GHz. The measurement system used here has a measurement The 3GPP channel model includes path loss models in [11],
range comparable to a wideband (800-MHz radio frequency using contributions from academic and industrial partners
bandwidth) channel sounder with 21.7-dBW effective isotropic from extensive mmWave measurement campaigns and ray-
radiated power. Measured data verify a new path loss model that tracing simulations [13][21]. Scenarios specified in 3GPP TR
uses a close-in free space reference distance with a novel height-
dependent path loss exponent (CIH model). This work shows 38.900 [11] include urban microcell (UMi), urban macrocell
that the CIH model is accurate and stable, and is frequency- (UMa), and indoor hotspot (InH) for office and shopping
independent beyond the first meter of propagation, and effectively mall [14], [15], [18]. Rural macrocell (RMa) is an additional
models the path loss dependence on base station height in rural scenario that is included in 3GPP [11] and METIS path loss
channels. models [6], although it has not been thoroughly investigated
Index Terms Millimeter wave, mmWave, rural macrocell for frequencies above 6 GHz. In fact, a search of the literature
(RMa), 73 GHz, path loss, channel model, 3GPP, ITU-R,
standards.
and standards reveals only one very limited and unpublished
measurement campaign at 24 GHz that was used to attempt
I. I NTRODUCTION to justify the 3GPP RMa path loss model [22]. The literature
LOS path loss models for use in urban cellular system design
in the low GHz range.
The RMa LOS path loss model equation (1) from ITU-R
M.2135 [23] and 3GPP TR 38.900 [11] is similar to the LOS
path loss model provided in [32], but is slightly different with
an additional height correction factor: min(0.044h 1.72, 14.77),
that adjusts path loss for low building heights. A different
formula is used for calculating path loss after the breakpoint
distance in (1) compared to what appears in [32]. Path
loss after the breakpoint distance, given by P L 2 [dB]
in (1), includes path loss up until the breakpoint distance Fig. 2. (a) LOS breakpoint distance vs. frequency in (2) for default
parameters: h B S = 35; h U T = 1.5 m [1]. This shows that the break-
(P L 1 (d B P )), and an attenuation rate of 40 dB per decade point distance in (2) exceeds 10 km at frequencies greater than 9.1 GHz.
of distance beyond the breakpoint distance. In ITU-R 5D/ (b) Frequency [GHz] and base station height (h B S ) combinations where the
88-E, the LOS path loss equation after the breakpoint distance RMa LOS path loss model in (1) reverts to a single-slope model, as identified
by the shaded area since the breakpoint distance in (1) and (2) is beyond
includes the attenuation rate of 40 dB per decade of distance 10 km (applicability range of model). The mobile height (h U T ) is 1.5 m.
but also includes correction factors for transmitter height,
mobile height, average building height, and street width, which B. 3GPP RMa LOS Path Loss Model Description
were not adopted by ITU-R or 3GPP for RMa LOS path
loss. There is also a noticeable discontinuity between path Upon inspection, (1) is a cumbersome equation without
loss before and after the breakpoint distance in [32]. an intuitive physical meaning [1]. This is clearly seen by
The breakpoint distance was initially developed by Bulling- observing the correction factor min (minimum) functions
ton [29] for far propagation distances (usually many kilome- of average building height, that create an upper bound on path
ters), where the PLE transitions from free space (n = 2) to the loss for large average building heights. The min correction
asymptotic two-ray ground bounce model of n = 4 [25], [30]. factor terms are non-physical curve fitting adjustments. The
More recent work [31], [34] showed that the breakpoint use of average building height h is quite odd, considering that
distance model by Bullington is a good fit to microcellular an RMa scenario does not have tall buildings. The breakpoint
channels, as well. A simulation of the two-ray ground bounce distance used in (1) and (2), however, does have some physical
model compared to FSPL is provided in Fig. 1 for a center basis in representing the distance at which the PLE approaches
frequency of 2.6 GHz, a TX height of 35 m, an RX height of the asymptote of n = 4 [25]. While (1) has not been fully
1.5 m, and clearly shows that the log-distance path loss slope explained in the literature, it is obvious that the first term
changes from n = 2 to n = 4 at the breakpoint distance (d B P ) 20 log10 (40 d3D f c /3) is the theoretical FSPL (in dB) at
near 2.85 km. When used at mmWave frequencies, the break- distance d3D as shown here [16]:
point distance becomes very large, on the order of tens of km 4 d3D f c 109 40 d3D f c
[see Fig. 2(a)]. 20 log10 = 20 log10
3 108 3
The precise RMa LOS path loss model in ITU-R M.2135 (3)
and 3GPP TR 38.900 was not given in [32] and [33], or
any other published material that we could find, leaving us by Friis free space transmission formula for f c in GHz [25],
to conclude that the existing 3GPP RMa LOS path loss model [28], [35].
was never fully confirmed for a rural environment, nor was The breakpoint distance for RMa LOS path loss in (2) has a
it confirmed with propagation data above 6 GHz. For the rest remarkable practical frequency limitation. This is easily seen
of this paper, when the 3GPP RMa model is mentioned, we by using the 3GPP default height parameter settings from
are referring to the 3GPP TR 38.900 (Release 14) standard Table I [11]. Using these default parameters, it is readily seen
for frequencies above 6 GHz [11]. When the ITU-R model is that the breakpoint distance exceeds the defined maximum
mentioned, we are referring to ITU-R M.2135 [23]. usable distance of the 3GPP RMa LOS path loss model
1666 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 35, NO. 7, JULY 2017
(10 km) for frequencies greater than or equal to 9.1 GHz, where f is the frequency in MHz and d is the distance from
as shown in Fig. 2(a). Thus, the 3GPP RMa LOS path loss the base station in km. The model was extended from 450 MHz
model must be assumed to be a single-slope model above to 2200 MHz with an additional frequency extension term not
9.1 GHz (for distances within the applicable range) as given by shown here, but which is given in [37].
P L 1 in (1). Fig. 2(b) displays the region (shaded) for various In the literature [38], [39], the extended version of the
base station heights (h B S ) and frequency combinations where Sakagami model (4) replaces all of the building height terms
the 3GPP RMa LOS path loss model breakpoint distance with the median building height and substitutes the frequency
exceeds the maximum 10 km propagation distance model term with 20 log10 ( f ) [38], [39]. An expansion to account for
limit. Surprisingly, the breakpoint distance portion of (1) is not mobile heights above 1.5 m was adopted from the Okumura-
usable above 32 GHz for any h B S base station height (mobile Hata model [40] and is formulated as: P L Sakagami a(h m )
height of 1.5 m) defined in the 3GPP and ITU-R RMa models, where [41][43]:
leading us to conclude that the LOS breakpoint distance
(2) is not appropriate for mmWave bands in RMa scenarios. a(h m ) = 3.2(log10 (11.75h m ))2 4.97 (5)
Section IV introduces a more reliable, accurate, and robust with h m as the mobile (UT) antenna height in meters. The
single-slope CI path loss model [16][18], [21] that avoids the equation in (5) was used to curve fit path loss for frequencies
breakpoint issue highlighted here, and is much simpler to use above 400 MHz and for UT heights above 1.5 m [37]. The
than the 3GPP/ITU-R RMa path loss models [11], [23]. While combination of expansions and extensions on the Sakagami
the ITU-R references the sub-6 GHz WINNER II channel model explained above, is what appears in the final 3GPP and
model for RMa, it includes traditional LOS path loss models ITU-R RMa NLOS path loss model as shown here [11], [23]:
that were modified with COST231-Hata correction factors for
curve fitting and that do not match 3GPP [36]. P L = max(P L R MaL O S , P L R MaN L O S )
P L R MaN L O S = 161.04 7.1 log10 (W ) + 7.5 log10 (h)
C. 3GPP RMa NLOS Path Loss Model Origin (24.37 3.7(h/ h B S )2 ) log10 (h B S )
The RMa NLOS path loss model in 3GPP [11] is taken + (43.42 3.1 log10 (h B S ))(log10 (d3D ) 3) + 20 log10 ( f c )
directly from ITU-R [23] and originates from work by (3.2(log10 (11.75h U T ))2 4.97); S F = 8 dB (6)
Sakagami and Kuboi [37]. The empirical model in [37] was
where W is the street width, h is the average building height
developed from measurements in metropolitan Tokyo in 1991
(by combining all building height coefficients [39]), h B S is the
at 813 MHz and 1433 MHz in a square shaped area with a total
base station height, h U T is the mobile UT height, and d3D is
length of 270 to 420 meters, and with measurement courses
the 3D T-R separation distance, where all distances and heights
along circles with radii of 0.5, 1, 2, and 3 km, for various
are in meters. Additionally, f c is the center frequency in GHz
measurement distances. Parameters selected for the model [37]
and the log-normal SF standard deviation is set to S F = 8 dB.
include base station antenna height (h b0 ), base station antenna
Applicability ranges for the model are provided in Table I as
height above the mobile station (h b ), building height near the
extracted from [11]. The max (maximum) operator in (6)
base station (H ), average building height (< H >), height of
acts as a strange mathematical patch and is used to solve a
buildings along the street (h s ), street width (W ), and street
model artifact, where the model predicts much stronger power
angle ( ), with all heights and distances in meters and in
at close-in distances in NLOS (say within a few hundred
degrees. The street angle was defined as the angle between the
meters) than what physics dictates for LOS environments. The
line connecting the UT and the base station, and the street from
patch is used to ensure that the estimated NLOS path loss
where the measurement was performed (a maximum angle of
is always greater than or equal to the equivalent LOS path
90 induces 2.07 dB of additional path loss) [37]. Given that
loss at the same distance. This problem was shown to exist
this work was for a dense urban environment (Tokyo) and at
in many other 3GPP-style path loss models [18], leading to
ultra high frequencies (UHF), many of these parameters do not
the optional CI path loss models in 3GPP [11], [16][18].
apply to RMa path loss modeling, since it makes no physical
A footnote for (6) in [11] specifies the applicable frequency
sense to have correction factors for street width and building
range as 0.8 GHz < f c < 30 GHz in 3GPP, although evidence
height in rural environments that do not have large buildings
suggests this is not valid, as is now explained.
or urban canyons.
Using these physical parameters of the environment in
downtown Tokyo, a multiple regression analysis was con- D. 3GPP RMa NLOS Path Loss Model Description
ducted in [37] to simultaneously solve for nine model coef- Three differences between the extended Sakagami (4) and
ficients. The coefficients that minimized the residual variance the eventual 3GPP (6) and ITU-R RMa NLOS path loss
between the model and data were determined based on models are explained here. The first difference is the removal
1000 path loss data points and resulted in the following NLOS of the street angle term: 0.023 . The second change is the
path loss model [37]: modification of the first term from 100 in (4) to 161.04 in (6)
since the 3GPP model is in units of GHz rather than MHz
P L Sakagami = 100 7.1 log10 (W ) + 0.023 + 1.4 log10 (h s )
(FSPL difference at 1 m between 1 MHz and 1 GHz is 60 dB).
+ 6.1 log10 (< H >) (24.37 3.7(H / h b0)2 ) log10 (h b ) The additional 1.04 dB difference was not explained in the
+ (43.42 3.1 log10 (h b )) log10 (d) + 20.4 log10 ( f ) (4) standards, and while it appears in [44], it is not explained
MACCARTNEY AND RAPPAPORT: RMa PATH LOSS MODELS FOR mmWAVE WIRELESS COMMUNICATIONS 1667
there or in the literature. The last difference is the addition combined together and best-fit indicators (e.g. RMSE) were
of 3 in (log10 (d3D ) 3), to account for the fact that d3D not provided, leading one to question the comparison validity.
is in meters as compared to km in [37] (log10 (1000) = 3). The facts outlined here uncover the questionable origins
The use of (6) as a path loss model above 6 GHz for of (1) and (6) and the little empirical evidence to support the
rural environments is highly questionable based on the fact 3GPP RMa NLOS path loss model above 6 GHz. While the
that evidence shows measurements were made at frequencies WINNER II channel model [36] included sub-6 GHz RMa
only up to 1433 MHz [37]. The measurement frequencies used path loss measurements, they were still limited according
to generate (6) are much lower than the specified maximum to [48] and resulted in modifications of the COST231-
applicable frequency (30 GHz), since the original model is Hata path loss models that do not resemble or align with
based on measurements at 813 and 1433 MHz, with an those found in 3GPP or ITU-R [11], [23]. Given the fact
extension for up to 2200 MHz (although this extension is not that other 3GPP models suffered from errors that required
included in the 3GPP or ITU-R model [11], [23]). It is worth mathematical corrections for relatively small T-R distances,
noting that others in the literature have attempted to extend the leading to much higher losses for large T-R separation dis-
Sakagami model based on measurements up to 8.45 GHz in tances [18], [21], we conducted a rural macrocell measurement
urban environments, yet this introduced, even more, correction and modeling study in clear weather at the 73 GHz mmWave
factors and evolutions of the Sakagami model, leading to even band for LOS and NLOS conditions [1]. This empirical
more complicated correction factors that can only be applied in data for RMa path loss above 6 GHz and in the mmWave
urban environments [39], [45]. While physical parameters such bands makes it possible to generate a generic but reliable
as average building height and street width are sensible for RMa path loss model for frequencies above 500 MHz and
urban areas, they are not used in any of the other 3GPP outdoor beyond 100 GHz that could be used by 3GPP, ITU-R, and
(e.g. UMi or UMa) path loss models [11], so they surely others.
do not make sense in a rural model. Even more concerning
is that the NLOS path loss model in 3GPP and ITU-R is III. 73 GHz RMa M EASUREMENTS
based on urban measurements, not the rural areas which are
to be modeled [1], [46], since the extended Sakagami path A. 73 GHz Measurement Equipment
loss prediction formula is unsuitable for areas with extremely Path loss measurements were conducted in Riner and Chris-
low average building height (i.e. rural and suburban areas). tiansburg, Virginia, rural towns in southwest Virginia, USA, at
Omote et al. [46] conclude that (6) is only applicable for the 73 GHz frequency band. The TX was positioned on a porch
average building heights greater than 5 m, so they added a at Professor Rappaports mountain home at a height of 110 m
new parameter to account for this based on occupancy ratio above the surrounding terrain [1]. A narrowband continuous
the ratio between the occupancy area of buildings in a sampled wave (CW) signal was transmitted with a maximum RF
area and the total sampled area [47]. power of 14.7 dBm (29 mW) using a rotatable 7 azimuth
The only effort we could find to validate (6) above 6 GHz and elevation half-power beamwidth (HPBW) horn antenna
in [11] was from a small measurement campaign at 24 GHz with 27 dBi of gain for a total effective isotropic radiated
conducted over limited 2D T-R separation distances between power (EIRP) of 11.7 dBW relatively low EIRP compared to
200 to 500 m [22], even though (6) is specified for 2D currently deployed RMa cellular base stations. The transmitter
distances up to 5 km (and the original Sakagami model used a 5.625 GHz CW tone that was mixed with a 67.875 GHz
in [37] was valid up to 10 km). The work in [22] indicates local oscillator (LO) signal (22.625 GHz frequency multiplied
a reasonable match between the measurements and model x3 in an upconverter) as depicted in the TX schematic in
between 200-500 m, but LOS and NLOS path loss data were Fig. 3.
1668 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 35, NO. 7, JULY 2017
Fig. 5. Sketch of the TX location on the porch of the mountain home, and surrounding areas [1].
These new models are based on the optional path loss models
in 3GPP [11] and the CI models as found in [16][19]
and [21]. Here, we propose a physically-based CI RMa large-
scale path loss model using a 1 m free space reference
distance [16], [18], [50] and develop a new model with
a base station height dependent path loss exponent (CIH).
Both models have a solid physical basis, are proven to be
accurate, reliable, to match measured data well, and are easy
to understand and apply. As noted in Section III, the lack
of spatial lobes [19] in the RMa channel allows the same
path loss model parameters to be used for both omnidirec-
tional and directional antenna systems [25], [28], [52][54].
Fig. 8. Relationship between TX base station height (h B S ) and decrease in
Furthermore, models of this form have been shown to work NLOS path loss for T-R separation distances of 150 m, 500 m, 1 km, 2.5 km,
well for all frequencies from 0.5 GHz to 100 GHz, specif- and 5 km, for the 3GPP RMa NLOS path loss model (6). The UT height
ically in the mmWave bands [13][18], and are already in (h U T ) is 1.5 m.
use in 3GPP [11]. CI models have also proven to have
excellent stability and accuracy when predicting path loss
values for scenarios and distances outside the scope of the such as in the RMa scenario, the distance d may be represented
original measurements used to create the model [18], [21]. by the 2D or 3D distance, as the difference is minuscule.
Although measurements were conducted at a single frequency The first term after the equality sign in (7) models
(73 GHz), evidence in the literature shows that measurement frequency-dependent path loss up to the close-in reference
from 500 MHz to 73 GHz generate accurate path loss mod- distance d0 = 1 m [25], and is equivalent to Friis
els that work well up to 100 GHz [13], [14], [18], [21]. FSPL [25], [35]:
Therefore, we make the reasonable assumption that beyond 4 f c 109
FSPL( f c , 1 m)[dB] = 20log10 c = 32.4 + 20log10 ( f c ) (8)
the first meter of propagation, the path loss exponent may be
frequency independent as this has been proven many times where f c is the center frequency in GHz, c is the speed of
from measurements and models across many mmWave bands light in free space or air, 3 108 m/s, and 32.4 dB is the
in UMa scenarios [11], [13], [14], [16], [18], [19], [21], [54]. FSPL at 1 m at 1 GHz. Thus, (7) is given by:
PLCI ( fc , d)[dB] = FSPL( f c , 1 m)[dB] + 10n log10 (d) +
A. CI Path Loss Model = 32.4 +10n log10 (d) + 20 log10 ( f c ) + ;
CI path loss models have been used for decades for estimat- where d 1 m (9)
ing path loss in a variety of scenarios and environments [25]
and have recently been shown to represent outdoor channels Setting the free space reference distance d0 = 1 m provides
over a vast range of mmWave frequencies with surprisingly a standardized and universal modeling approach for path
good and robust accuracy [16], [18], [21]. The simplest form loss comparison with a single parameter, the PLE [13][16],
of the model, with a 1 m free space reference distance (d0 ), [25], [51], and was approved as an optional model for UMa,
led to its adoption as an optional model for UMa, UMi, and UMi, and InH in 3GPP TR 38.900 [11]. The CI model for
InH scenarios in 3GPP [11], based on numerous experiments RMa (9) requires only a single model parameter, n, also
at mmWaves [13][17]. Thus, it would also seem reasonable called PLE, to describe the distant-dependent average path loss
to consider a CI option for the RMa scenario. We show over distance for a wide range of mmWave bands [11], [16],
subsequently from the measured data that indeed the CI model [18], [21]. The use of 1 m makes sense, because there are
offers a good fit for RMa, with a much simpler expression clearly no obstructions in the first meter of propagation from
than (1) and (6). a transmitting antenna, it models the frequency dependency
The general expression for the CI path loss model is: of propagation in outdoor channels over a vast span of fre-
quencies, and has exhibited consistent accuracy and parameter
d
PL ( f c , d)[dB] = FSPL( f c , d0 )[dB]+10n log10
CI
+ ; stability across numerous scenarios, distances, and frequency
d0 ranges [16], [18], [21].
where d d0 and d0 = 1 m (7)
where d (usually 3D distance) is the T-R separation in m B. CIH Path Loss Model
between the TX and RX, d0 is the close-in free space reference When considering the existing RMa path loss models in
distance in m, n represents the PLE [16], [25], [51], and f c 3GPP and ITU-R (see (1) and (6)), there are clearly terms such
is the frequency in GHz. Shadow fading is represented by as building height and street width that do not make physical
the zero-mean Gaussian random variable with standard sense, yet there are others, such as TX and RX height above
deviation in dB [16]. For large T-R separations (several km) ground, which would be expected to impact path loss in rural
As mentioned in Section II, the subtraction of antenna gains while keeping environments. Since the current 3GPP/ITU-R RMa models
the omnidirectional PLE can be done in RMa channels because of the lack considered TX heights as low as 10 m and as tall as 150 m,
of energy arriving from directions other than the dominant angle. this parameter clearly has much greater range and physical
MACCARTNEY AND RAPPAPORT: RMa PATH LOSS MODELS FOR mmWAVE WIRELESS COMMUNICATIONS 1671
significance than other model parameters (a simulation to requires two optimization parameters, the PLE n, and bt x :
follow shows this). The RX height in the rural scenario, as
specified in 3GPP, ranges from only 1.5 m to 10 m, which is PLCIH ( fc , d,h B S )[dB]= 32.4 + 20log10 ( f c )
h B S h B0
negligible when considering T-R separation distances of many + 10n 1 + bt x log10 (d) + ;
kilometers, suggesting the TX height would be the single most h B0
significant physical parameter to include in an RMa path loss where d 1 m, and h B0 = avg. BS height (11)
model, besides the close-in free space reference distance. As with the CIF model, the CIH model simplifies to the CI
Fig. 8 shows the effect of base station height (h B S ) on model when bt x = 0 (no dependence on base station height),
NLOS path loss for five 3D T-R separation distances (150 m, or when h B S = h B0 . The closed-form solutions for deriving
500 m, 1 km, 2.5 km, and 5 km) using (6), and the average the optimal CIH model parameters n and bt x are provided
decrease in path loss over all of the T-R distances, as a function in the Appendix.
of base station height. The results in Fig. 8 are independent
of frequency from (6), and show that by increasing the base
station height from 10 m to 150 m, the path loss is effectively C. 3GPP RMa Monte Carlo Simulations
reduced by approximately 26 dB and 32 dB for T-R separation To compare the modeling accuracy of the CI and CIH RMa
distances of 150 m and 5 km, respectively, with an average path loss models (9), (11) with the current RMa LOS (1) and
decrease of 29 dB over all of the T-R distances. NLOS (6) path loss models in [11], we used the 3GPP default
Here we extend the CI model to include various base station parameters in Table I and performed Monte Carlo simulations
heights (CIH model), such that the model remains physically for two cases. Case one is for a fixed base station height and
grounded to FSPL at a close-in distance but also models case two is for a range of base station heights.
the PLE dependence on base station height. The CIH model 1) Case OneSimulation for 3GPP Default Parameters:
was inspired by the CIF model introduced in [17] and [18], In case one, 50,000 random path loss samples (with 3GPP
which was shown to accurately model the frequency depen- default parameters from Table I) were generated from (1)
dence of path loss in indoor environments. This model high- and (6), for the following frequencies: 1, 2, 6, 15, 28, 38,
lights the importance of the PLE in being able to model 60, 73, and 100 GHz, resulting in 450,000 samples each
the physical effects in the environment, and encapsulates (50,000 samples 9 frequencies) for LOS and NLOS. Fre-
a fundamental physical basis of the frequency dependence quencies below and above 6 GHz were used for the multi-
due to Friis equation at close-in distances (while avoiding frequency simulation and modeling since the overall frequency
the need for mathematical patches and ensuring remarkable applicability range covers 0.5 GHz to 100 GHz for a majority
accuracy when applied for values outside of the measurement of path loss models in [11]. The CI model (9) ensures a
set [16][18], [21]). seamless path loss model for frequencies from 0.5 GHz to
Work to date used an ad hoc, non-physical basis for mod- 100 GHz, without discontinuities as demonstrated in [13],
eling the impact of TX height [11], [33], [37][39], [43]. [14], [16], and [18]. Each path loss sample was randomly
By incorporating the TX height as an adjustment to path loss, generated for a 2D T-R separation distance ranging between
we postulated that it would be possible to model secondary 10 m and 10 km for LOS and between 10 m and 5 km
path loss effects due to antenna height, just as the CIF model in NLOS,1 along with the corresponding SF values (in dB)
accounts for secondary frequency-dependent effects while from (1) and (6). When simulating (1), path loss samples for
retaining the physics of the primary frequency dependency frequencies above 9.1 GHz were generated from the single-
of FSPL at close-in distances. slope portion of (1), based on the breakpoint distance (2) in
The CIH model uses the same mathematical form as the LOS. From the simulated 3GPP path loss samples for each
CIF model [17] and is given here for d0 = 1 m: environment, CI models that best fit the data were derived
that resulted in the minimum root mean squared error (RMSE)
PLCIH ( f c , d, h B S )[dB] = FSPL( f c , 1 m)[dB] between the model and the data. The CI models derived from
simulated LOS (PLCI-3GPP ) and NLOS (PLCI-3GPP
h B S h B0 LOS NLOS ) path loss
+ 10n 1 + bt x log10 (d) + ; samples, were found to be:
h B0
where d 1 m (10) PLCI-3GPP
LOS ( f c , d)[dB] = 32.4 + 23.1 log10 (d) + 20 log10 ( f c ) + LOS ;
where d 1 m, and LOS = 5.9 dB (12)
where h B S is the RMa base station antenna height in meters, NLOS ( f c , d)[dB] = 32.4 + 30.4 log10 (d) + 20 log10 ( f c ) + NLOS ;
PLCI-3GPP
and h B0 is the default base station height or is taken as
where d 1 m, and NLOS = 8.2 dB (13)
the average of all TX heights from a measurement set. The
distance dependence of path loss is denoted by n (identi- Both the LOS and NLOS CI models in (12) and (13) emphat-
cal to the PLE from the CI model), and bt x is a model ically show that the complicated 3GPP/ITU-R RMa path loss
parameter that is optimized and which quantifies the lin- models in (1) and (6) can be reformulated into succinct and
ear base station height dependent PLE about the average easy to understand equations with nearly identical performance
base station height h B0 . An effective PLE (PLEe f f ) results
1 2D distances were randomly generated per the limits defined in TR
from the scaling of n by bt x andthe TX heights such
38.900 [11] and Table I, from which 3D distances were calculated using
that: PLEe f f = n 1 + bt x h B Shh
B0
B0
. The CIH model only antenna heights and trigonometry, for simulations.
1672 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 35, NO. 7, JULY 2017
VI. C ONCLUSIONS The RMa measurements were also used to derive the
This paper provided an in-depth study on the existing optimal parameters for the CIH model since TX height has
3GPP [11] RMa LOS and NLOS path loss models for frequen- historically been proven to have a large impact on path loss.
cies from 0.5 GHz to 100 GHz and found that no substantial Since the measurements consisted of a single base station
empirical evidence exists to date to support adoption of this height (110 m), we used the simulated CIH models in (14)
model by ITU-R [23]. Given that no work existed in the and (15) along with measured data to derive the optimal CIH
literature, and the questionable parameters used in 3GPP model parameters. Doing this ensured that for the conditions
and ITU standards, this paper describes field measurements of our measurement campaign (h B S = 110 m, and f c =
conducted in rural Virginia and develops new RMa path loss 73 GHz), the CIH models in (21) and (22) revert to the
models that have been verified by field data and are shown to CI models in (16) and (17). It is obvious from the derived
be more accurate as well as much easier to use and understand, CIH path loss models in (21) and (22) that the effective PLE
based on the fundamental physics of radio propagation, and decreases as the base station height increases (bt x = 0.03 in
include a height dependent PLE since the TX height can have LOS; bt x = 0.049 in NLOS), where the average decrease
a considerable effect on RMa path loss. in path loss across all T-R distances, when considering a base
The LOS CI path loss model in (12) derived from simulated station height of 150 m compared to 10 m, is 17 dB. The
path loss samples resulted in an RMSE of 5.9 dB, a good closed-form solution equations to derive optimal CIH model
match with the 3GPP LOS model in (1) which specified a parameters are provided in the Appendix so that others may
SF of 4 to 6 dB. Similarly, the NLOS CI model in (13) that create similar models for RMa path loss.
was derived from simulating the 3GPP NLOS RMa path loss This paper investigated the questionable use of the current
model in (6), resulted in an RMSE of 8.2 dB compared to the 3GPP/ITU-R RMa path loss models for frequencies above
SF of 8 dB specified in 3GPP/ITU-R [11], [23]. The similar 6 GHz. Empirically-based CI and CIH path loss models
performance in LOS and NLOS CI models optimized from were proposed in (16), (17), (21), and (22), that can be
simulated data suggests that a model grounded in the true considered for adoption in 3GPP and ITU-R for frequencies
physics of free space propagation within the first meter could above 500 MHz to beyond 100 GHz for the RMa scenario.
be used as an optional model in 3GPP for RMa path loss [17], The models herein are validated with real-world mmWave
[18], [21]. measurements at the 73 GHz mmWave band, and have the
It was also shown that the CIH model could effectively same mathematical form as the optional UMa, UMi, and InH
model RMa path loss in LOS and NLOS environments for path loss models already in 3GPP [11], and are proven to offer
a large range of base station heights. When optimized to fit superior prediction accuracy when applied to new frequencies,
simulated 3GPP data from (1) and (6) across an array of distances, or use cases [18]. It is suitable to extrapolate the
transmitter heights, the optimized CIH models in (14) and (15) CI and CIH models for 0.5 GHz to 100 GHz from 73 GHz
had comparable performance with RMSE values of 5.6 dB and measurements since many measurements and models across
8.7 dB for LOS and NLOS, respectively, compared to 4 to 6 dB various macrocell scenarios and mmWave bands have shown
and 8 dB SF values for the cumbersome 3GPP models in [11]. that the propagation path loss exponent (PLE) is independent
The CIH models effective scaling of PLE by the transmitter of frequency beyond the first meter of propagation. However,
height a reasonable physically motivated correction factor more measurements for other heights and different frequencies
is, therefore, an accurate option for RMa path loss modeling above 6 GHz and in the mmWave bands are needed to further
from 0.5 GHz to 100 GHz. confirm the accuracy of the CI and CIH RMa path loss models
The CI and CIH model parameters were optimized presented herein.
from real-world measurement data with directional antennas
using [28, Eq. (3.9)] at mmWave frequencies, in clear weather A PPENDIX
and in a rural setting with a TX height of 110 m and RX CIH PATH L OSS M ODEL PARAMETER D ERIVATION
heights from 1.6 m to 2 m. The derived CI models from mea- The closed-form solutions for obtaining the CIH path
surements resulted in a LOS PLE = 2.16 ( = 1.7 dB) and a loss model parameters are found by solving for the optimal
NLOS PLE = 2.75 ( = 6.7 dB), indicating how models with parameters that minimize the SF standard deviation, i.e., the
one parameter can faithfully estimate path loss in a much sim- minimum mean squared error (MMSE) between the model
pler form than 3GPP-style models. The solid physical basis of and simulated or measured data. The CIH path loss model
the CI path loss model is an important feature as it models true parameter derivation is identical to the CIF model derivation
free space propagation in the first meter, compared to the 3GPP in [17] and [18], except that here, the frequency parameter is
models that contain numerous and odd correction factors. The replaced by a base station height parameter h B S and default
two models for RMa path loss in (16)(17) and (21)(22) (or average) base station height h B0 .
may be used for frequencies from 500 MHz and up to at least The CIH path loss model in (10) is rearranged in the form:
100 GHz. It is noteworthy that the narrowband CW measure-
ment system used here is equivalent to a wideband channel PLCIH ( f c , d, h B S )[dB] = FSPL( f c , 1 m)[dB]
sounder with 21.7 dBW EIRP of RF transmit power over
nbt x
800 MHz of bandwidth and with 190 dB of dynamic range + 10 log10 (d) n(1 bt x ) + h B S + CIH ;
h B0
(10 dB greater than in [49]), which is still a relatively small
transmit power compared to todays cellular base stations. where d 1 m (23)
1676 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 35, NO. 7, JULY 2017
[24] Federal Communications Commission, Spectrum Frontiers [49] G. R. MacCartney, Jr., and T. S. Rappaport, 73 GHz millimeter
R&O and FNPRM: FCC16-89, Jul. 2016. [Online]. Available: wave propagation measurements for outdoor urban mobile and backhaul
https://apps.fcc.gov/edocs_public/attachmatch/FCC-16-89A1.pdf communications in New York City, in Proc. IEEE Int. Conf. Com-
[25] T. S. Rappaport, Wireless Communications: Principles and Practice, mun. (ICC), Jun. 2014, pp. 48624867.
2nd ed. Upper Saddle River, NJ, USA: Prentice-Hall, 2002, ch. 4, 5. [50] J. B. Andersen, T. S. Rappaport, and S. Yoshida, Propagation mea-
[26] S. Sun et al., Propagation path loss models for 5G urban micro- surements and models for wireless communications channels, IEEE
and macro-cellular scenarios, in Proc. IEEE 83rd Veh. Technol. Commun. Mag., vol. 33, no. 1, pp. 4249, Jan. 1995.
Conf. (VTC-Spring), May 2016, pp. 16. [51] G. R. MacCartney, Jr. et al., Path loss models for 5G millimeter
[27] G. R. MacCartney, Jr., M. K. Samimi, and T. S. Rappaport, Omni- wave propagation channels in urban microcells, in Proc. IEEE Global
directional path loss models in New York City at 28 GHz and Commun. Conf. (GLOBECOM), Dec. 2013, pp. 39483953.
73 GHz, in Proc. IEEE 25th Int. Symp. Pers. Indoor Mobile Radio [52] A. I. Sulyman et al., Radio propagation path loss models for 5G cellular
Commun. (PIMRC), Sep. 2014, pp. 227331. networks in the 28 GHz and 38 GHz millimeter wave bands, IEEE
[28] T. S. Rappaport, R. W. Heath, Jr., R. C. Daniels, and J. N. Murdock, Commun. Mag., vol. 52, no. 9, pp. 7886, Sep. 2014.
Millimeter Wave Wireless Communications. Upper Saddle River, NJ, [53] S. Sun et al., Synthesizing omnidirectional antenna patterns, received
USA: Prentice Hall, 2015. power and path loss from directional antennas for 5G millimeter-wave
[29] K. Bullington, Radio propagation at frequencies above 30 megacycles, communications, in Proc. IEEE Global Commun. Conf. (GLOBECOM),
Proc. IRE, vol. 35, no. 10, pp. 11221136, Oct. 1947. Dec. 2015, pp. 17.
[30] M. J. Feuerstein et al., Path loss, delay spread, and outage models [54] A. Sulyman et al., Directional radio propagation path loss models
as functions of antenna height for microcellular system design, IEEE for millimeter-wave wireless networks in the 28-, 60-, and 73-GHz
Trans. Veh. Technol., vol. 43, no. 3, pp. 487498, Aug. 1994. bands, IEEE Trans. Wireless Commun., vol. 15, no. 10, pp. 69396947,
[31] K. L. Blackard et al., Path loss and delay spread models as functions Oct. 2016.
of antenna height for microcellular system design, in Proc. IEEE 42nd
Veh. Technol. Conf., vol. 1. May 1992, pp. 333337.
[32] Proposed Propagation Models for Evaluating Radio Transmission Tech-
nologies in IMT-Advanced, Int. Telecommun. Union, document 5D/88-E,
Jan. 2008.
[33] S. Ichitsubo et al., Multipath propagation model for line-of-sight street George R. MacCartney, Jr. (S08) received the
microcells in urban area, IEEE Trans. Veh. Technol., vol. 49, no. 2, B.S. and M.S. degrees in electrical engineering from
pp. 422427, Mar. 2000. Villanova University, Villanova, PA, USA, in 2010
[34] V. Erceg et al., An empirically based path loss model for wireless and 2011, respectively. He is currently pursuing the
channels in suburban environments, IEEE J. Sel. Areas Commun., Ph.D. degree in electrical engineering with the NYU
vol. 17, no. 7, pp. 12051211, Jul. 1999.
WIRELESS Research Center, New York University
[35] H. T. Friis, A note on a simple transmission formula, Proc. IRE,
(NYU) Tandon School of Engineering, Brooklyn,
vol. 34, no. 5, pp. 254256, May 1946.
NY, USA, under the supervision of Prof. Rappaport.
[36] P. Kysti et al., WINNER II channel models, Eur. Commis-
He is a recipient of the 2016 Paul Baran Young
sion, IST-WINNER, D1.1.2 V1.2, Feb. 2008. [Online]. Available:
Scholar Award from the Marconi Society. He is also
http://projects.celticinitiative.org/winner+/WINNER2-Deliverables/
[37] S. Sakagami and K. Kuboi, Mobile propagation loss predictions for the recipient of the 2017 Dante Youla Award for
arbitrary urban environments, Electron. Commun. Jpn., vol. 74, no. 10, Graduate Research Excellence in Electrical and Computer Engineering from
pp. 1725, Jan. 1991. the NYU Tandon School of Engineering ECE Department. He has authored
[38] Y. Ohta et al., A study on path loss prediction formula in microwave or co-authored over 30 technical papers in the field of millimeter-wave
band, IEICE Tech. Rep. A P2003-39, Mar. 2003. (mmWave) propagation. His research interests include mmWave propagation
[39] K. Kitao and S. Ichitsubo, Path loss prediction formula in urban area test and measurement system design, and channel modeling and analysis for
for the fourth-generation mobile communication systems, IEICE Trans. fifth-generation communications.
Commun., vols. E91B, no. 6, pp. 19992009, Jun. 2008.
[40] M. Hata, Empirical formula for propagation loss in land mobile radio
services, IEEE Trans. Veh. Technol., vol. 29, no. 3, pp. 317325,
Aug. 1980.
[41] T. Fujii, Path loss prediction formula in mobile communication
An expansion of SAKAGAMI path loss prediction formula, IEICE Theodore S. Rappaport (S83M84SM91
Trans. Commun. (Japanese Edition), vol. J86-B, no. 10, pp. 22642267, F98) received the B.S., M.S., and Ph.D. degrees
Oct. 2003. in electrical engineering from Purdue University,
[42] T. Fujii, Delay profile modeling for wideband mobile propagation, in West Lafayette, IN, USA, in 1982, 1984, and 1987,
Proc. IEEE 60th Veh. Technol. Conf., (VTC-Fall), vol. 1, Sep. 2004, respectively. He founded major wireless research
pp. 3842. centers with the Virginia Polytechnic Institute and
[43] T. Fujii and T. Imai, Prediction formula of path loss for wideband State University (MPRG), The University of Texas
DS-CDMA cellular systems, IEICE Tech. Rep. RCS97-236, 1998. at Austin (WNCG), and NYU (NYU WIRELESS)
[44] T. Fujii, Y. Ohta, and H. Omote, Empirical time-spatial propagation and founded two wireless technology companies
model in outdoor NLOS environments for wideband mobile communi- that were sold to publicly traded firms. He is an
cation systems, in Proc. IEEE 69th Veh. Technol. Conf. (VTC-Spring), Outstanding Electrical and Computer Engineering
Apr. 2009, pp. 15. Alumnus and a Distinguished Engineering Alumnus from Purdue University.
[45] K. Kitao and S. Ichitsubo, Path loss prediction formula for urban and He is currently the David Lee/Ernst Weber Professor of Electrical and
suburban areas for 4G systems, in Proc. IEEE 63rd Veh. Technol. Conf., Computer Engineering with the New York University Tandon School of
vol. 6. May 2006, pp. 29112915. Engineering, New York University (NYU), Brooklyn, NY, USA, and the
[46] H. Omote, Y. Sugita, and T. Fujii, High accurate path loss prediction Founding Director of the NYU WIRELESS Research Center. He also holds
formula by using occupancy ratio for mobile radio propagation, in Proc. professorship positions with the Courant Institute of Mathematical Sciences
10th Eur. Conf. Antennas Propag. (EuCAP), Apr. 2016, pp. 12. and the School of Medicine, NYU. He is a highly sought-after technical
[47] S. Kozono and K. Watanabe, Influence of environmental buildings on consultant having testified before the U.S. Congress and having served the
UHF land mobile radio propagation, IEEE Trans. Commun., vol. 25, ITU. He has advised more than 100 students, has more than 100 patents
no. 10, pp. 11331143, Oct. 1977. issued and pending, and has authored or co-authored several books, including
[48] D. S. Baum et al., IST-2003-507581 WINNER, final report on the best seller Wireless Communications: Principles and PracticeSecond
link level and system level channel models, Eur. Commission, docu- Edition (Prentice Hall, 2002). His latest book Millimeter Wave Wireless
ment IST-WINNER, D5.4 V1.4, Nov. 2005. [Online]. Available: http: Communications (Pearson/Prentice Hall, 2015) was the first comprehensive
//projects.celtic-initiative.org/winner+/DeliverableDocuments/D5.4.pdf text on the subject.