Modeling and Link Budget Estimation of

Directional mmWave Outdoor Environment for 5G

Sheeba Kumari M Sudarshan A. Rao Navin Kumar
Dept. of ECE BigSolv Labs Pvt Ltd., Dept. of ECE
VTU Bangalore, India. Amrita School of Engineering
Bangalore, India dr.sudarshan.rao@ieee.org Amrita Vishwa Vidyapeetham
sheeba.kumarim.in@ieee.org Bangalore, India
navinkumar@ieee.org

Abstract— Narrow pencil beamforming is a key enabling derived by steering the narrow beam towards the receiver, it
technique for millimeter wave (mmWave) systems to achieve necessitates a realistic characterization of the channel’s
high link quality. Employing highly directional narrow beam directional properties, preferably using models that can
antennas will deliver unique requirements on mmWave channel perform real time prediction of channel characteristics.
modeling. In this paper, we propose a low complexity mmWave
channel model to analyze the performance of line of sight (LOS) A few channel models based on measurements have been
and non-line of sight (NLOS) directional outdoor links in several proposed for mmWaves at certain carrier frequencies and
reference cases. This ray tracing based model integrates the deployment scenarios in [1], [3-6] and references therein. As
antenna geometry of a highly directional antenna in 3D plane. these models make diverse assumptions on channel parameters
The effects of specific attenuation and foliage losses are also such as the line of sight (LOS) distance, number of dominant
incorporated to enhance the model accuracy. Simulation results paths, cluster size and the like, there might be certain
show that the deterministic channel fading gain varies as a variations in the estimated model parameters and capacity
function of antenna location, height and beamwidth in addition analysis [1]. Furthermore, most of the reported statistical
to its variations against distance. The model is validated against models are either distinctly omnidirectional [3] or derived
standard channel models and measurements, at frequencies and from omnidirectional measurement set up [4] or even
deployment scenarios reported in the literature. Furthermore, converged to omnidirectional formulations [5], with a view to
link budget estimates and signal-to-noise ratio (SNR) results are preserve the existing sub-6 GHz modelling approaches.
evaluated. The proposed model, unlike many other models,
However, these omnidirectional model formulations might be
describes the channel for any mmWave frequency, antenna
beamwidth, transmitter/receiver height and location by just
inconclusive in analyzing directional links [1], [5] as they fail
changing these parameters in the simulator. The study highlights to attribute the spatial filtering of directional antennas, wherein
the need to integrate directional antenna effects and deployment a limited number of paths aggregate to the total received
parameters in the channel model to have accurate performance signal. Measurements and model descriptions using directional
predictions. antennas can be found in [6] and related works, for example,
the directional close-in (CI) reference distance model that
Keywords—5G; mmWave; ray tracing; path loss model; link estimates directional path loss exponents (PLEs) and the
budget estimation directional beam-combining model. However, the former is
defined for fixed input specifications and the latter needs the
I. INTRODUCTION number of strongest beams (in range of 1-4) as a model input
Wideband millimeter wave (mmWave) spectrum is being in order to compute the received signal strength.
proposed for the fifth generation (5G) mobile networks to The existing models pose additional limitations, that is,
offer very high data rates [1-2]. While the high frequency of lack of measurements (and model parameters) at ‘non-popular’
mmWaves can increasingly contribute to the massive multiple frequency bands and nonexploitation of both elevation and
input multiple output (MIMO) and network densification azimuth angle dimensions [2]. We also identify that there is a
paradigms, several challenging issues need to be solved to need to report the mmWave link budget to assist practical
leverage this technology [1]. Most importantly, the channel deployments. Even though [7] estimated the budget of a 60
should be accurately modeled to develop any such techniques GHz link using a popular path loss (PL) model for urban
that can adapt to the unique propagation characteristics of microcell (UMi) LOS and non-LOS (NLOS) scenarios, it is
mmWaves. required to estimate the link budget for other mmWave
The mmWave systems will likely employ high gain frequencies as well as for the channel statistics predicted by
antennas at both the transmitter (TX) and receiver (RX) to other known models.
overcome the large signal attenuation arising from higher free In this work, we propose a deterministic directional
space path loss (FSPL), absorption loss and building channel model for mmWaves in 3D UMi street canyon (SC)
penetration loss [1]. Advanced beamforming (BF) and beam scenario to evaluate the link performance in several reference
steering techniques with narrow beams will be exploited to cases. UMi SC, compared to open square, allows to evaluate
achieve high link quality, signal-to-noise ratios (SNR) and cell the worst case outdoor PL by generating maximum number of
coverage distances [2]. While the required BF gains can be

978-1-7281-0546-8/19/$31.00 ©2019 IEEE 106

strong reflected multipath components (MPCs). The proposed review 3 popular UMi SC outdoor models, aiming to assess
model, based on ray tracing, integrates the antenna geometry how they recreate the directional channel characteristics.
of a highly directional antenna in 3D plane thereby exploiting
directional transmissions for reducing the ray tracing 1) NYUSIM PL Model: NYU developed CI and ABG
complexity. The effects of frequency dependent specific models for both directional and omnidirectional cases at 28,
attenuation and of foliage losses are also incorporated to 38, 60 and 73 GHz [5-6]. The differences between various
enhance the model accuracy. Our overall objective is to cases are established by the choice of model parameters, which
compare this simulator with existing models to identify its were determined through minimum mean square error
benefits, limitations and use cases. To this end, we develop a (MMSE) fit to measurement data. While both these models
simplified step procedure to obtain the direction-dependent could be used to characterize mmWave channels, the channel
large scale PL values for both LOS and NLOS links by simulator-NYUSIM reported in [5] employed CI model due to
attempting to preserve the popular statistical models’ its ease of applicability, reduced parameters and model
expression, which allows for easier comparison. To the best of stability. Though measurements were conducted using
the authors’ knowledge, such investigation is not found in directional antennas, the model preserved the traditional
available literature. Results show that the channel fading gain omnidirectional form by synthesizing omnidirectional received
powers. To generate a directional PL model, these
varies as a function of antenna location, height and beamwidth
in addition to its variations against distance. Through omnidirectional MPCs should be weighted with the desired
simulations, we establish that a simple inclusion of directional directional antenna pattern anew.
antenna boresight gains into an omnidirectional PL model 2) 3GPP PL Model: The 3rd Generation Partnership
cannot produce directional models. The proposed model, Project (3GPP TR 38.901) [3] proposed channel models to
unlike many state-of-the-art models defined for fixed input characterize links of frequencies 6–100 GHz, which are
specifications, describes the channel in 3D plane for any extensions of the well-established sub-6 GHz omnidirectional
mmWave frequency, antenna beamwidth, transmitter/receiver PL models. A dual slope breakpoint model was used for LOS,
height and location by merely changing these parameters in while the ABG model with a user equipment (UE) height
the simulator. Link budget estimates and SNR results are also dependency term was adopted for NLOS. The LOS model
obtained and compared. The significance of the proposed work reduces to CI model when the TX-RX separation distance is
can be appreciated by assessing some of the popular modeling smaller than the break point distance. Therefore, the
methodologies in mmWaves, their assumptions and model applicability of a dual slope model in mmWave small cells is
parameters which are briefly discussed in Section II. uncertain since their breakpoint distance exceeds the cell
Rest of this paper is organized as follows. Section II radius (~150-200 m inter site distance), even with the smallest
describes state-of-the-art mmWave channel models. The base station (BS) and UE heights [1], [10].
proposed directional PL model is presented in Section III 3) MiWEBA PL Model: A quasi deterministic channel
while the results and discussion are presented in Section IV. model based on ABG model was proposed by MiWEBA [4],
The concluding remarks are included in Section V. which was derived from measurements conducted at 60 GHz
across 5 to 50 m separation distances using omnidirectional
II. STATE-OF-THE-ART PATH LOSS MODELS antennas. To model the effect of directional antennas and that
The models based on measurements are generally derived of spatial filtering, each of the omnidirectional measurement
from either the one parameter CI PL model [5-6] or the alpha- channel impulse responses (CIR) was filtered on the delay
beta-gamma (ABG) PL model based on three parameters [6], axis, thereby attenuating those multipaths that are far away
with modifications relevant to the mmWave requirement. For from the LOS.
instance, the CI mmWave model uses a lower reference
To sum up, most of these models accurately predict the PL
distance d0 of 1 or 5 m since mmWave systems typically
with simple equations, yet their consistency of model
employ low height base stations and small cell sites.
parameters between varying specifications is a concern [1].
According to the CI model, the PL at any 2D separation
Also, the values of model parameters n, α, β and σ, reported in
distance d is [5], [6],
vast literature, are for very precise input specifications, i.e., the
values estimated for a frequency and deployment cannot be
PLCI (d) [dB] = FSPL (d0, fc) + 10 log10 (d) + Xσ,CI (1)
reused for another. Hence, the researchers who adopt these
models for system analysis cannot readily extract the exact
where, fc is the carrier frequency, is the PLE and Xσ,CI is a
values that best fit their input specifications. Moreover, the
zero mean Gaussian random variable with σ dB standard
existing models are either omnidirectional or antenna-
deviation. FSPL ( , ) = 20 log (4 ⁄ ) , where λ is the
independent and hence do not include the antenna beamwidth.
wavelength. The ABG model on the other hand characterizes The need to have an accurate model simulator that estimates
the mmWave channel using a best fit line to the measured PL for any input specifications and that includes directional
data, specified as [6], antenna effects for each contributing MPC, motivates us to
develop a ray tracing model for an identical scenario.
PLABG (d) [dB] = α + 10β log10 (d) + Xσ,ABG (2)
III. PROPOSED DIRECTIONAL PATH LOSS MODEL
where α is the floating y intercept, β is the path loss slope as a
In this section, we derive the proposed deterministic
function of distance and σ is the standard deviation of zero
directional model, wherein the model expression is carefully
mean Gaussian random variable, Xσ,ABG. We now briefly

107
formulated by starting from the standard statistical where GL(φ,θ) and Gi(φ,θ) are the products of transmit and
omnidirectional expression to provide for an easy comparison. receive antenna gains for the LOS and the ith reflected rays
As a typical case, we choose a downlink channel scenario respectively, at their corresponding azimuth angle φ and
where the BS is transmitting signal at mmWave carrier elevation angle θ. In this work, we use a highly directional
frequency fc to the UE located at a distance d. horn antenna design that generates the radiation pattern (and
The omnidirectional received power for the standard the gain at any angle pair (φ,θ)) for a given half power
statistical CI model defined in (1) [5-6] is rephrased as: beamwidth in elevation (HPBWE) and in azimuth (HPBWA)
planes [9]. N is the number of resolvable reflections that fall
within the antenna beam area. The 3D path lengths, dL for
, ( , )= . , (3)
LOS and di for all N paths, and their respective angles are
evaluated as functions of d, ht and hr using ray tracing
where Pt is the transmitted power, , = 10 , ⁄ is the technique. As ray tracing is implemented in three dimensions,
statistical fading gain that characterizes the large scale fading reflections from both ground and side walls are considered.
and is log normal. Here, do is assumed to be 1 m. As stated in Reflection coefficient on the ith reflected path Ri will be the
[1], CI models do not consider the frequency dependency of perpendicular reflection coefficient for wall reflections and
received power beyond the first meter of propagation. Hence, parallel reflection coefficient for ground reflections if the TX
even though (3) is a function of fc, it does not specify any and RX antennas are vertically polarized [8]. Ka, the
mmWave specific absorption losses that are functions of both coefficient of exponential absorption (say, 0.0016loge10 for 16
frequency and distance. In other words, the absorption loss dB/km at 60 GHz) is evaluated for each mmWave frequency
which is a function of d is being accounted for only at 1 m. using the ITU-R approximate model, which is integrated into
Furthermore, there is no reference to antenna heights ht/hr or the channel model [9].
beamwidths in the model expression, implying that n and σ
values were deduced to best fit the antenna height, beamwidth Equation (5) can be modified to generate an expression
and carrier frequency specified in that study. analogous to the statistical model expression in (3) as:
.
1) Deterministic omnidirectional modelling: On the , , , (, ) = (, ) (, ) . ,
contrary, deterministic ray tracing models can characterize the
(6)
actual effect of the environment. We now define the same
channel scenario deterministically and the received power can where , = |ℎ | is the deterministic directional fading
be hence modified as: gain that characterizes the fading phenomenon due to all
resolvable specular reflections falling within the antenna
, ( , )= . , (4) beam area. It may be noted that Gt(φ,θ).Gr(φ,θ) is equal to
GL(φ,θ) and its value is the product of transmit and receive
where, , = |ℎ | is the channel fading gain owing to antenna gains at boresight, if the beams are aligned through
the paths arriving at the RX from all angular directions and is a beam steering. To compare the two models, (3) should be
function of antenna height and position. Ray tracing for the multiplied with (or (5) should be eliminated with) GL(φ,θ).
omnidirectional case is hence complex as there is a need to
aggregate every possible ray arriving at the receiver. And the ℎ , the sparse CIR normalized with respect to ideal LOS
link can be given by:
computation of , generally demands efficient
commercial ray tracing tools and powerful computing units. 
ℎ = 1 + ∑ (7)
2) Deterministic directional modelling: A mmWave where, ai and ψi are the relative gain and phase of each
outdoor channel that employs narrow-beam directional resolvable reflected ray, computed as:
antennas is spatially sparse with a few specular reflected rays,
mainly because the outdoor communications do not readily (,)
rely on diffraction and diffused scattering [2] and as only a = (,)
(8)
limited number of the reflected MPCs fall within the narrow
beamwidths of TX/RX antennas. By exploiting this, we devise and  = (2 ⁄ )( − ). Clearly, ai and ψi are functions of
a low complexity ray tracing model formulation based on d, ht, hr, fc and the antenna gain corresponding to each path.
image technique. The resulting received power, as a function Accordingly, the deterministic directional PL as a ratio of TX
of 3D antenna geometries, can be expressed as [8], [9]: power to RX power is obtained from (6) as:
/
, , (, ) =
(,)
+ [ ] = FSPL ( , ) − [ ]− [ ]+

[ ] − , [ ] (9)
/
(,)
∑ (5)

108
 Fig. 1. Proposed channel model Fig. 2. 2D top view of simulation
illustrating model inputs and scenario illustrating multiple reflected
outputs. rays.
.
where FSPL ( , ) is evaluated at dL. [ ] = 10 log Fig. 3. Path loss evaluated using six ray model at 60 GHz for LOS and
is the specific attenuation due to oxygen absorption in NLOS UMi SC scenario.
mmWave frequencies. A further attenuation of Lfol [dB] =
0.2fc0.3Df0.6 is added to (9) to incorporate the effects of foliage,
where Df is the foliage depth in meters [2]. A major benefit of
the proposed model is that it only calculates those rays that
reach the receiver antenna beam area thereby reducing the
overall computations. Also, it estimates the PL variations for
a change in any input specification shown in Fig. 1, where, ds
is the width of the street and dw is the distance from the
TX/RX to the reflecting wall.
IV. RESULTS AND DISCUSSION
We first present the steps in PL evaluation by invoking the
spatial filtering phenomenon of directional antennas and the
high attenuation properties of mmWaves, and then provide the
major results.
A. Evaluation of Directional Path Loss
Fig. 2 shows the UMi SC propagation geometry considered
in our simulation and illustrates how the direct and reflected
components aggregate at the receiving node. A narrower
beamwidth typically leads to six or less significant rays
aggregating at the UE (ground reflected ray is considered but
not shown in the figure due to limitation of dimensionality).
The model generation consists of three major steps as stated.
i) The first step evaluates the received power for free space Fig. 4. Simulated directional fading gain CDFs of the proposed model
at distance dL after integrating the antenna gain GL(φ,θ) and for various UE locations, UE heights and antenna beam widths.
the Ka value of the selected mmWave carrier frequency, as
dw, ht and hr. The PL values can be evaluated using (9) or as a
described in Section III. To simulate access links, the antenna
heights ht and hr are differently selected. Also, an estimated ratio of , to Pt, to which the foliage loss is further
correction in θ is provided to align the beams for maximum added.
received signal strength, if beam steering is enabled. For B. Channel Characterization of MmWave Directional Link
simulating NLOS case it is assumed that there is no clear path
between the TX and RX antennas and no beam steering. First, we analyze a 60 GHz backhaul channel for antenna
placements on lampposts at 7 m from the ground, i.e. for both
ii) The model next determines the path lengths, di, of all
ht and hr. Fig. 3 depicts the PL values evaluated over distance
reflected rays based on the TX-RX placement. The angle of
using (9). It is observed that a direct LOS ray, a ground
arrival (AoA) in both azimuth and elevation planes of each
reflection, two first order and two second order wall
reflected path is analytically evaluated from the model
reflections aggregate at the RX for a directivity of 21.6 dBi
geometry through ray tracing. If AoA falls within the antenna
realized using antenna half power values of HPBWA = 15° and
beamwidth, the received signal strength due to that reflected
HPBWE = 13°. For this simulation settings, the link is purely
ray is determined by placing in the angle dependent antenna
LOS (without any reflected components) up to a separation
gains and the frequency dependent absorption losses, or else,
distance of 30.7 m as the narrow beam area in both azimuth
the ray is discarded.
and elevation planes ensures that the reflected rays are
iii) Finally, the total received signal power, , is received outside the receiver beamwidth up to this distance,
estimated by combining signal powers due to free space and i.e., no diversity. The distance at which reflection begins to
all available reflected components. Its value is a function of contribute as a propagation mechanism is obviously a function
frequency, directivity and the deployment parameters d, ds,

109
 TABLE I. UMI SC OUTDOOR PATH LOSS MODELS WITH DEPLOYMENT AND MODEL PARAMETERS

Freq. Tx Height Rx Tx /Rx HPBW Tx /Rx Range

Item Env. Model Parameters
(GHz) (m) Height (m) (0) Gain (dBi) (m)

LOS 31≤ d ≤102 nCI = 1.9, SFCI = 1.1 dB

28 7/17 1.5 10.9 24.5
NLOS 61≤ d ≤162 nCI = 4.5, SFCI = 10 dB
Measurement
[6] LOS 48≤ d ≤54 nCI = 2.2, SFCI = 5.2 dB
2
73 17 7 27 NLOS 59≤ d ≤181 nCI = 4.7, SFCI = 12.6 dB
(Access)
NLOS 59≤ d ≤181 nCI = 4.7, SFCI = 12.7 dB
LOS 10≤ d≤5000 nCI = 2.1, SFCI = 4 dB
3GPP [3] 6-100 10 1.5-22.5 Omni 1
NLOS 10≤ d≤5000 nCI = 3.19, SFCI = 8.2 dB
MiWeba [4] 60 6.2 1.5 H = 18, V = 14 19.8 NLOS 25≤ d≤50 αABG =82.02 dB, βABG=2.36
LOS 1≤ d ≤30 ndet = 2.0, σdet = 0 dB
V = 13,
This Work 28 7 7 21.6 LOS 31≤ d ≤60 ndet = 1.98, σdet = 1.6 dB
H = 15
NLOS 61≤ d ≤200 ndet = 2.15, σdet = 3.42 dB

of ht, hr, dw and d of the model geometry. The PL for access Table I compares the n and σ values estimated for the
scenario is likewise analyzed by setting a UE height of 1.5 m. proposed model with the model parameters estimated in
Existing study [5-6] had validated their simulations for a popular UMi SC models [3-4] and measurement studies [6],
minimum d of 30 m since the UEs located near BSs are for a selection of input specifications reported in the
generally power controlled in the near field. However, the literature. The backhaul analysis conducted at 73 GHz shows
dimension of mmWave antenna array systems is normally in only minor variations in the results from access and is hence
the range of 100-300 mm for BSs or even below 100 mm for omitted. For our proposed model, we analyze the PL
UEs [10]. Hence, the far field distance for a downlink scenario discretely for 3 distinct ranges and then apply MMSE fit
will begin at 1.9 and 4 m for 28 and 60 GHz respectively, separately to each window to yield the model parameters. As
motivating us to analyze the PL distribution from 1 m itself. stated earlier, the propagation is FS up to 30.7 m, attributing
The study is further extended to evaluate the n values of the to n = 2. We consider the distance, d, from 31 to 60 m as LOS
proposed model through MMSE fit to CI model that is based and beyond 61 m as NLOS identical to [6]. For the NLOS
on measurements. That is, by computing (n ) = access scenario of ht = 7 m, hr = 1.5 m, n shows small
∑ [ ( )− ( )] for M arbitrary receiver variations where as σdet = 4.41 dB. Also, ndet = 2.32, σdet =
locations and solving for the mean squared error. From Fig. 3, 18.8 dB when the TX height is increased to 17 m as in [6].
it can be observed that the estimated value of n for NLOS case For 73 GHz (not provided in the table), the n value increases
is large compared to LOS, the value increasing as the to 2.05 for LOS and 2.74 for NLOS, illustrating an increase in
difference between the TX and RX antenna heights increases. PL for higher frequencies analogous to the existing models.
The NLOS value of 2.74 reduced to 2.18 when the antenna
Fig. 4 shows the cumulative distribution of fading gains HPBWs are changed to 7°, showing a PL improvement of 5.6
, . To simulate for varying d, the other parameters are set dB/decade. The gain terms were excluded from (9) while
as HPBWE = 13°, HPBWA = 15°, ht = 7 m and hr = 1.5 m. For deriving these results to bring in similarity with the standard
the hr varying case, the gains at LOS and NLOS are evaluated models which commonly include antenna gain factors in the
at 50 m and 200 m respectively keeping a fixed ht, HPBWE and link budget. Nevertheless, it is observed that the n value
HPBWA. We observe that varying the UE locations and heights increases, for all cases, when the gain factor is included in the
have minimal impact in case of LOS transmission for all PL slope (which is appropriate in directional channel
mmWave frequencies which is not true for the NLOS case. analysis). The increase is prominent in the NLOS case as
Approximately 35% of the simulated fading gains are above -1 Gt(φ,θ) and Gr(φ,θ) are different from the boresight gains
dB for both 60 and 73 GHz, yet the fading gain at 60 GHz when the antenna major beams are unaligned. For instance, a
shows better performance for their lower values because of the 7° HPBW must provide a boresight gain of 27 dBi, but it
heavily attenuated higher order components. Irrespective of drops to 23 dBi for a direct ray if the beams are unaligned.
the chosen beam width, the fading gain at 60 GHz is always Repeated simulations of the proposed model were
greater than 1 dB as compared to its distribution lying in the -5 performed for varying inputs and the major highlights are
to -1 dB range for 73 GHz. This implies that besides the now presented. The PLE and SF values observed for 28 GHz
known increase in FSPL with frequency, even the fading gains LOS channel are very low. The NLOS PLE increased with
for higher and lower frequencies add differently to the total frequency as the waves with shorter lengths are largely
PL. The directional PL is hence a function of all deployment attenuated on reflection from the ground and side walls. In
parameters even beyond their influence captured at d0 as general, all the model parameters are observed to increase
stated in [5-6]. with frequency. Also, the PL variations in dB/decade of
C. Analysis of Path Loss Model Parameters distance reduced with increasing beamwidths, implying that

110
 TABLE II. COMPARISON OF UMI SC DOWNLINK BUDGET USING DIFFERENT 5G MMWAVE PATH LOSS MODELS

Link Budget Measurement [6] 3GPP Model [3] MiWeba [4] This Work
Frequency (GHz) 28 73 28 60 28 73
Bandwidth (MHz) 800 800 100 250 800 800
Pt (dBm) 30.1 14.6 35 15 30.1 14.6
Gt (dBi) 24.5 27 1 15 24.5 27
EIRP (dBm) 54.6 41.6 36 30 54.6 41.6
Environment LOS NLOS LOS NLOS LOS NLOS LOS NLOS LOS NLOS LOS NLOS
Range (m) 50 200 50 200 50 200 50 200 50 200 50 200
PL (dB) 94.2 170.3 108.2 170 99.2 139.2 108.3 122.5 96.5 123.4 103.3 142.9
Pr (dBm) -39.6 -115.7 -66.6 -128.4 -63.2 -103.2 -78.3 -92.5 -41.9 -68.8 -61.7 -101.3
Gr (dBi) 24.5 27 1 15 24.5 27
Pr (dBm) with BF -15.1 -91.2 -39.6 -101.4 -62.2 -102.2 -63.3 -77.5 -17.4 -44.3 -34.7 -74.3
Thermal Noise (dBm) -85 -85 -94 -90 -85 -85
Noise Figure (dB) 7 7 9 10 7 7
Pn (dBm) -78 -78 -85 -80 -78 -78
SNR (dB) 62.8 -13.2 38.4 -23.5 22.8 -17.2 16.7 2.5 60.6 33.7 43.3 3.7

the omnidirectional PLEs will be smaller compared to the fully adaptive BF where the stronger paths should be predicted
directional PLEs. The existing models however do not capture to determine the beam steering direction. The work continues
this dependence on antenna heights and gains. in studying the effect of diffuse scattering on directional PL
values in our extended work.
D. Link Budet Estimation
Finally, we present the practical downlink budget REFERENCES
evaluation for mmWave outdoor channels in table II. [1] T. S. Rappaport, Y. Xing, G. R. MacCartney, A. F. Molisch, E. Mellios, J. Zhang,
Accordingly, the standard models were first simulated with “Overview of Millimeter Wave Communications for Fifth-Generation (5G)
Wireless Networks-With a Focus on Propagation Models,” IEEE Transactions on
parameters set identical to the literature to evaluate their PL Antennas and Propagation, vol.65, no. 12, pp.6213-6230, Dec 2017.
values (CI model generated values identical to (6)). As seen,
[2] I. Hemadeh, K. Satyanarayana, M. El-Hajjar, L. Hanzo, “Millimeter-Wave
the SNRs calculated by standard models for few NLOS [3], [6] Communications: Physical Channel Models, Design Considerations, Antenna
are very low. This indicates that an increase in transmit power Constructions and Link-Budget”, IEEE Commun. Surveys Tuts., vol. 20, no. 2,
or a decrease in coverage distance or channel bandwidth is pp. 870-913, Apr.–Jun. 2018.
necessary for successful reception. The proposed directional [3] 3GPP, “TSG RAN TR 38.803 v14.1.0 Study on new radio access technology:
model provides PL identical to the measurements [6] for LOS, Radio Frequency (RF) and co-existence aspects,” 2017. [Online]. Available:
http://www.etsi.org/deliver/etsi_tr/138900_138999/
validating the model. For NLOS case, we obtain a lower PL
138912/14.01.00_60/tr_138912v140100p.pdf [Accessed: 20-Apr-2018]
relative to other models. It is likely due to the dependency of
[4] MiWeba, “WP5: Propagation, Antennas and Multi-Antenna Technique; D5.1:
fading gain on antenna beamwidths. It is observed that the Channel Modeling and Characterization,” Jun. 2014. [Online]. Available:
values for NLOS are larger than the LOS. Either values are http://www.miweba. eu/wpcontent/uploads/2014/07/ MiWEBA_D5.1_v1.011.pdf.
higher for higher frequencies, all well agreeing with the [Accessed: 20-Apr-2018]
theoretical intuition, thus ensuring realistic SNR values. [5] T. S. Rappaport, S. Sun, and M. Shafi, “Investigation and Comparison of 3GPP
and NYUSIM Channel Models for 5G Wireless Communications,” IEEE 86th
V. CONCLUSION AND FUTURE WORK Veh. Technol. Conf. (VTC Fall) , pp. 1–6, Sep. 2017.
A directional mmWave PL model for UMi SC is proposed [6] T. S. Rappaport, G. R. MacCartney, Jr., M. K. Samimi, and S. Sun, “Wideband
Millimeter-Wave Propagation Measurements and Channel Models for Future
to emphasize the significance of integrating antenna effect in Wireless Communication System Design,” IEEE Trans. Commun., vol. 63, no. 9,
channel modelling. The channel fading gain that determines pp. 3029–3056, Sep. 2015.
the total PL is observed to be a function of UE location, height [7] Joongheon Kim, Liang Xian, Ali S. Sadri, “60 GHz Modular Antenna Array Link
and antenna beamwidth. The downlink budget estimates and Budget Estimation with WiGig Baseband and Millimeter-Wave Specific
SNRs are evaluated. The results obtained from our custom ray Attenuation,” Hindawi International Journal of Antennas and Propagation Volume
tracing based channel model simulator are shown to agree with 2017, Article ID 9073465.
existing measurement studies and those predicted by popular [8] Sheeba Kumari M., S.A.Rao, N. Kumar, “Characterization of mmWave Link for
Outdoor Communications in 5G Networks,” IEEE International Conference on
models. The model allows to accurately evaluate large scale
Advances in Computing, Communications and Informatics (ICACCI), pp. 44-49,
PL due to specular reflections for all mmWave frequencies, 2015.
antenna heights, gains and link distances. It is extremely useful [9] Sheeba Kumari M., S.A.Rao, N. Kumar, “Outdoor Millimeter-Wave Channel
in extending mmWave channel study to system simulations. Modeling for Uniform Coverage Without Beam Steering,” Lecture Notes of the
The proposed angle-dependent mmWave channel model can Institute for Computer Sciences, Social Informatics and Telecommunications
be used to characterize the direction dependent mmWave Engineering, vol 218. Springer, 2017.
transmission in specific applications. For instance, it is suitable [10] Qualcomm, “Mobilizing 5G NR Millimeter Wave: Network Coverage Simulation
in a channel with fixed antenna geometry in place of antenna Studies for Global Cities,” Oct. 2017. [Online]. Available:
https://www.qualcomm.com/documents/white-paper-5g-nr-millimeter-wave-
steering, to avoid the complex signal processing algorithms for network-coverage-simulation [Accessed: 20-Apr-2018]
spatial signature estimation. Also, it is useful in channels with

111