1

5G Coverage, Prediction, and Trial Measurements

Tristan Curry Robert Abbas
Macquarie University Macquarie University
[email protected] [email protected]

Abstract—When planning a 5G network in the sub-6GHz metro area.

bands, similar cell planning techniques to LTE can be applied. The focus of this paper is to present a 5G NR link
Looking at the Australian environment, the n78 band (3.3-3.8GHz budget and corresponding prediction, and compare this against
arXiv:2003.09574v1 [eess.SP] 21 Mar 2020

TDD) is approximately 1GHz higher than the 2.6GHz band used

in existing LTE networks. As a result, the coverage footprint measured NRSRP and throughput data from a live 5G NR site
can be similar, and therefore co-locating 5G NR (New Radio) on to verify the accuracy of the link budget and prediction models
existing LTE base stations is a common strategy for initial net- for planning a 5G network.
work rollout. Any difference in coverage can be compensated by
beamforming gain, less downtilting, or increasing the gNodeB’s
transmit power. This paper presents an initial link budget for II. BACKGROUND
data services, provides a coverage prediction, and measurements
for a 5G NR NSA (Non Stand Alone) trial radiating at 3.5GHz In this section we detail some of the relevant work that has
with 60 MHz bandwidth. The coverage prediction is generated been done in both academia and industry. We note that there
using RF planning tool Atoll, which is then compared to coverage are very few live trial results which have been published in
measurements from the trial. These findings can be used to help the n78 band for mobile networks.
plan a future 5G network in the Sydney metro area or similar
environment. In [2] the authors present a paper comparing link budgets
for the 2.1 GHz band for LTE and the 3.5 GHz band for NR.
Index Terms—5G, New Radio (NR), 5G Cell Planning, Forsk They conclude that in a direct comparison the 3.5 GHz band
Atoll, CrossWave, 5G Coverage Prediction, Link Budget, Beam
Selection, 5G Coverage Measurements, 5G Coverage Perfor- has a number of extra losses in both indoor and outdoor envi-
mance, gNodeB. ronments, however when using beam-forming the associated
gain can overcome these losses. They deduced that coverage
from the 3.5GHz band was better in outdoor environments, and
I. I NTRODUCTION indoor in older buildings, but only 0.9dB different in buildings
5G has support for spectrum ranging from existing mi- built in a modern style.
crowave bands as low as 600MHz up to mmWave bands In [5], the authors provide an analysis of TCP performance
(>30GHz band). With higher carrier frequencies comes in- over 5G NR. They explore the effect of the NSA architecture
creased bandwidth but higher penetration losses, so planning on overall TCP throughput. They deduce that due to the
considerations such as inter-site distance and base station differing latency performance of NR and anchor LTE layers,
architectures used will change accordingly. In November 2018, there is degradation in throughput in spite of any increased
the Australian Communications and Media Authority (ACMA) spectrum bandwidth. They also deduce that the nature of LTE
auctioned off spectrum in the 3.6GHz band [1] in anticipation and NR mean that TCP is susceptible to head-of-line blocking
of 5G networks using this band. The recommendation to use by regular data traffic. This effect de-prioritises TCP ACKs
this band is mirrored by the 3GPP [8]. It is anticipated that and reduces overall throughput by reducing the congestion
most Mobile Network Operators (MNO) in Australia will use window. When testing 5G NR using TCP based methods such
the n78 band in NR frequency range 1 (FR1) for initial 5G as Ookla Speedtests, these effects should be taken into account.
network roll-out so therefore this trial was undertaken as a There have been 5G trials conducted around the world
study of how 5G NR behaves in these bands. by operators and vendors, Telstra and Ericsson in Australia
To determine the minimum received signal level at the demonstrated a 5G call using an over-the-top service on an
user equipment (UE) required at cell edge, a link budget NSA 5G NR RAN and EPC core using the n78 band at the
was applied. This minimum receive level was then used to Gold Coast Commonwealth Games in April 2018 [4]. In Cali-
create a coverage prediction, with contour lines corresponding fornia, Nokia and Keysight have demonstrated the capabilities
to different NR Reference Signal Received Power (NRSRP) of Keysight’s Nemo network measurement equipment to help
levels. The coverage prediction was generated using the 5G optimise and verify the performance of Nokia’s 5G NR RAN
NR Atoll RF simulation suite and the CrossWave propagation equipment [6].
model. These were applied to high resolution raster (2m pixels) We base our understanding and research of NR on the 3GPP
digital terrain maps (DTM) and clutter data, and 3D vector Technical Reports (TR) and Technical Specifications (TS)
building and vegetation data in the trial area. This coverage documents. The current release (3GPP release 15) details the
prediction was then compared to measured data and the specification for 5G NR [8], [9] and outlines the requirements
accuracy of the predictions was determined. This information for NR standalone (SA) and non-standalone (NSA). Due to
can be used to plan contiguous 5G coverage in the Sydney equipment availability and maturity, initial deployment will
 2

Fig. 1: NR Coverage Prediction using the Atoll Network Planning tool and CrossWave Propagation Model

have the NR NSA architecture, utilising LTE on E-UTRA A. gNodeB Physical Configuration
band 1 (2.1GHz FDD) for the anchor layer. This anchor
layer provides signalling to allow the UE to set up an NR TABLE I: 5G NR Link Budget
bearer when in dual connectivity mode and thus connect to
the gNodeB. This arhcitecture can be seen in Fig. 2.

Fig. 2: 5G NR Non-Standalone (NSA) Architecture

III. L INK B UDGET AND C OVERAGE P REDICTION

Industry best practice for cell planning starts with deter-
mining an appropriate receive level at the UE. This level
in conjunction with coverage prediction tools will determine
where the cell edge is. A link budget provides the basis for
what service level the network is designed for and outlines the The gNodeB was built in a typical suburban area in Sydney.
gains and losses in the mobile communications channel. This The physical site parameters were based on the 3GPP channel
information allows a final received power level to be calculated model seen in [8] with the antenna center line (ACL) height
which corresponds to the desired service level. We based the raised up to 27.77 m above the ground due to space constraints
channel parameters on the 3GPP NR channel model [8], and on the headframe, and 3◦ degrees electrical downtilt to reduce
have chosen system bandwidth, MIMO schemes, and heights backlobe interference between sectors and increase coverage.
based on what NR RAN equipment was available at the time As there were no neighbouring gNodeB’s, 0◦ mechanical
of the trial. A summarised 5G NR link budget is shown in downtilt was used to maximise coverage.
table I. This link budget suggests we will achieve 200Mbps at The antennas used were passive antennas with 8 antenna
a UE received NRSRP level of -90.62 dBm. elements. In the NR specification coverage is provided by
Based on 3GPP recommendations [8], the 3.5GHz band is Synchronisation Signal Block (SSB) beams that sweep across
preferred for initial NSA 5G networks and the ACMA has the width of the antennas’ coverage envelope. The antenna
followed this recommendation in Australia by auctioning off used swept through 8 SSB beams in a round-robin manner
more spectrum in the 3.4-3.6GHz bands for mobile commu- from index 0 to 7.
nications use [1]. We have therefore used 60MHz bandwidth The terrain in the area varied in elevation from 50.69 m
centered at 3.5GHz in licensed spectrum for this trial. to 125.65 m above sea level, with clutter primarily being
 3

comprised of small shopping precincts, trees, housing devel-

opments, and industrial areas.

B. Coverage Prediction
Based off our link budget minimum receive level at the UE
we have generated a coverage prediction in Forsk Atoll using
the CrossWave propagation model. CrossWave is a pseudo-
raytracing model which emulates raytracing when used in
conjunction with high enough resolution raster or vector clutter
data. We used 2m resolution raster clutter data with outdoor to
indoor propagation enabled. The coverage prediction for the
site can be seen in Fig. 1.
We note that we expect to achieve 200Mbps DL throughput
in the orange area (-100dBm < NRSRP < -90dBm) based on
the link budget.

IV. M EASUREMENTS AND R ESULTS

A. Testing Methodology
We conducted 2 tests during this trial. The first was a
drive test using a Rohde & Schwarz NR scanner connected
to a PCTEL omni-directional antenna magnetically mounted
to the roof of the car. This antenna has an average of 3dBi
gain in the range 698-3800MHz. The scanner logged data
points in Single Input, Single Output (SISO) mode over more Fig. 3: Coverage prediction route as per measurement route: (a)
than 30km in surrounding streets. The second test was a above, Scanner NRSRP measurements, (b) below, delta of measured
stationary throughput test using a 5G UE (User Equipment) to predicted NRSRP values.
set up in the window in the back of the car. The UE utilised
2x2 Multiple Input, Multiple Output (MIMO) and speed tests
were conducted using a TCP based speedtest service on a
smartphone connected to the UE over 5 GHz WiFi. We
collected 32 samples in the one location.
From the prediction, we determined the drive test area and
drove both near the base station and towards the cell edge.
Key data collected with the scanner included:
• Timestamp (ms).
• UE GPS (Global Positioning System) Coordinates.
Fig. 4: Drive Test NRSRP Measurements
• SSB Beam index.
• NRSRP (dBm) per beam.
• NRSRQ (NR Reference Signal Received Quality) (dB)
per beam. To see the effect of slow fading only, we apply a general
form of Lees method [3]. To compute the local averages of the
Key data collected with the 5G UE included:
signal envelope we used the recommendation from Lee for an
• Downlink Throughput (Mbps).
averaging window 2L = 40λ which corresponds to a distance
• Uplink Throughput (Mbps).
D = 342.86cm, minimum number of samples per average
• Latency (ms).
window N = 36, and a distance between samples d = 0.8λ =
• NRSRP (dBm).
6.86cm [3] with a 3500MHz carrier (f = 3500M Hz, λ =
8.57cm). These values are chosen to ensure when the moving
B. Measured Drive Test Results average of the signal envelope is calculated, the fast fading
The results were obtained over 2 days. Measured NRSRP component becomes unity, and therefore does not affect the
values along the drive test route can be seen in Fig. 3 (a). slow fading component. The normalised signal envelope after
NRSRP level with respect to distance is summarized in this process shows only the slow fading of the signal and can
Fig. 4. The NRSRP fluctuations close to the gNodeB can be seen in Fig. 5. This data can be used to tune the CrossWave
be attributed to fast fading as a result of higher multipath model using the built in tool which utilises curve fitting to
fluctuation as the UE moves in the urban canyon areas near adjust parameters specific to the model to the measured and
the site. This fluctuation can be seen to be above 20dB in normalised signal envelope.
areas close to the site, but this variation reduces further from Fig. 6 shows the NRSRQ and SINR values for each beam
the site as slow fading becomes the dominant fading. from sector 1 at a location in the drive test route. We can see
 4

Fig. 5: Normalised NRSRP Measurements over Drive Test Distance

that NRSRQ and SINR have varying values for each of the 8
beams. Beam index 6 was directed at the measurement location Fig. 7: Speedtest and Latency Results at -91dBm ± 1dB NRSRP
and had the highest NRSRQ and SINR values, and beam index
2 had a dominant reflected path which gave a high NRSRQ and
SINR value at the measured location. Adjacent beams were not V. C ONCLUSION AND F UTURE W ORK
pointing directly at the measurement location and therefore Based on our analysis and application of LTE cell planning
less SSB power was received at the scanner. This corresponds techniques, we conclude they can be applied in initial 5G NR
to a reduction in signal strength, and therefore NRSRQ and network rollout in the n78 band. This trial demonstrated that
SINR. The theoretical relationship between NRSRQ and SINR 200Mbps was possible at -91dBm ± 1dB NRSRP using a link
is: budget analysis and trial measurements. The coverage predic-
1
SIN R = 1 tion and measured values after using Lee’s method aligned
n×N RSRQ − x
well and give confidence in the prediction for further network
Where: n = number of subcarriers, and x = per antenna planning in similar environments. In future we suggest to apply
subcarrier activity factor. the same analysis to more locations and environments, and
From this relationship it can be seen that NRSRQ does not perform speedtests at more NRSRP levels and locations with
fluctuate as highly as SINR for this range of values as they an FTP speedtest methodology that doesn’t have the variability
lie in the non-linear region in the SINR to NRSRQ graph. associated with over-the-top speedtest services. With the focus
on fixed wireless access (FWA) in the United States and
Australian markets, we suggest research into benchmarking
the penetration of the n78 band through different clutter
types and materials found in the Australian environment to
generate more accurate link budgets. We recommend testing
the performance of horizontal network handover in a mobility
cluster of sites, as this will also influence the density of sites
when planning a 5G network.
Fig. 6: NRSRQ and SINR values per beam.
R EFERENCES
[1] Australian Communications and Media Authority, "3.6 GHz band spec-
C. Measured Throughput Test Results trum auction results", Australian Communications and Media Authority,
ACMA, 2019. Available: https://www.acma.gov.au/theACMA/3-6-ghz-
Fig. 7 represents an average throughput comparison between band-spectrum-auction-results [Accessed on 9 March 2019]
downlink, and uplink, and latency at an NRSRP of -91dBm [2] B. Halvarsson, A. Simonsson, A. Elgcrona, R. Chana, P. Machado and H.
± 1dB. The sample size n = 32 > 30 allows us to assume Asplund, "5G NR Testbed 3.5 GHz Coverage Results," 2018 IEEE 87th
Vehicular Technology Conference (VTC Spring), Porto, 2018, pp. 1-5.
normality from the Central Limit Theorem (CLT). [3] De La Vega, D., Lopez, S., Matias, J., Gil, U., Pena, I., Velez, M., . . .
Summary statistics can be seen in Table II. These results Angueira, P. (2009). . Generalisation of the Lee Method for the Analysis
align with what we expected from the link budget, with a for the Signal Variability. IEEE Transactions on Vehicular Technology,
58(2), 506-516.
throughput of 200Mbps at approximately -91dBm. The high [4] Ericsson, Telstra and Intel Get Multi-vendor 5G Commercial Network
variance in latency is due to outliers reaching 300ms. These Data Call Over Licenced 3.5GHz Spectrum. (2018). Wireless News, July
skew the average and can be seen when looking at the median 20, 2018.
[5] Jheng, C., Nuggehalli, P., & Chouhan, M. (2018). "Analysis of TCP
value of only 19.5ms. We believe this is due to IP routing Performance over 5G NR". IEEE Wireless Communications, 25(3), pp4-6.
issues in the internet causing packet loss and therefore high [6] Keysight, Nokia collaborate to verify 5G coverage in live test network.
roundtrip times (RTT). (2018). EE-Evaluation Engineering, 57(11), 5.
[7] M. Shafi et al., "5G: A Tutorial Overview of Standards, Trials, Chal-
TABLE II: Speedtest Summary Statistics lenges, Deployment, and Practice," in IEEE Journal on Selected Areas
in Communications, vol. 35, no. 6, pp. 1201-1221, June 2017.
[8] "NR; Base Station (BS) radio transmission and reception (Release 15)",
3GPP TS 38.104 V15.5.0, April 2019
[9] "Study on channel model for frequencies from 0.5 to 100 GHz (Release
15)", 3GPP TR 38.901 V15.0.0, June 2018.