Performance Study of Various Relay Nodes in 5G

Wireless Networks
Jianghong Luo, Ashwin Sampath, Navid Abedini, Tao Luo
Qualcomm Technologies, Inc.

Abstract—This paper studies performance of various types These different types of repeater/relay nodes have
of relay nodes in a 5G wireless network: conventional amplify- different implementation complexities and performance
forward repeaters, (semi-)smart/smart amplify-forward tradeoffs. A number of studies on relay nodes can be found in
repeaters with different levels of side information, and half- the literature [5-9], where theoretical performance and bounds
duplex/full-duplex decode-forward relay nodes with and
without spatial reuse. End-to-end effective signal to interference
on amplify-forward and decode-forward relay operation were
and noise ratios (SINRs) and achievable rates are derived for derived. But in these studies, some practical constraints for
these different types of relay nodes. Performance and amplify-forward operation, e.g. the constraint on the
complexity tradeoffs are discussed with a simulation over a maximum amplification gain for stability, and noise figure
Manhattan topology setting. Over-the-air (OTA) test results difference between amplify-forward and decode-forward
corroborates the findings in this paper. operation, were not captured, and in addition, the performance
difference between conventional amplify-forward repeater
Keywords—conventional amplify-forward repeater, smart versus smart amplify-forward repeater as well as impact of
(network-controlled) amplify-forward repeater, decode-forward different levels of side control information to smart repeater
relay node, integrated access and backhauling (IAB), 5G wireless
network, spatial reuse.
have not been studied. For decode-forward relay nodes, which
can be half-duplex or full-duplex, the impact of spatial reuse
I. INTRODUCTION in case of multi-user scheduling has not been studied before.
In this paper, we present an in-depth analysis of performance
The 5th Generation (5G) of mobile networks promises to
tradeoffs for different types of repeater/relay nodes that
provide high data rate services due to availability of large
address these points.
spectrum in the very high frequencies and adoption of large-
scale Multiple Input and Multiple Output (MIMO) antenna An outline of this paper is as follows. Section II describes
systems. However, high pathloss and blockage sensitivity at the system model for different types of repeater/relay nodes
high-frequency bands are obstacles for broader coverage. For and derives end-to-end Signal to Interference and Noise Ratio
example, the coverage of a 5G base station at a Frequency (SINR) and achievable rates. Section III presents simulation
Range 2 band (FR2=24.25GHz-52.6GHz) may be limited to results with mixed base stations and relay nodes in a
around few hundreds of meters, while the coverage of a Manhattan grid deployment setting. Section IV provides main
typical Frequency Range 1 band (FR1<7.2GHz) macro-cell findings from OTA tests conducted in indoor and outdoor
can be much larger. Signal blockage, at higher bands, will environments. Conclusions are drawn in Section V.
further reduce an effective coverage area of a base station.
Therefore, a cost-effective way, e.g. using wireless relay II. SYSTEM MODEL
nodes to extend coverage for network densification, is A. System Model for Decode-forward Relay node
attractive for 5G networks, especially at FR2 and higher
bands. The 3rd Generation Partnership Project (3GPP) has had System model for a decode-forward relay node with
various study and work items in recent years to provide downlink (DL) operation is shown in Figure 1, where
support of different types of relay nodes in a 5G network: • S: signal with unit power.
• Integrated access and backhauling (IAB), a Layer-2-based • 𝑃𝑇1 : transmission power of gNB.
decode-and-forward relay solution, was first introduced in • 𝑃𝑇2,𝑚𝑎𝑥 : maximum transmission power of relay node.
Release-16 [1] and further enhanced in Release-17 and 18. • ℎ1 , ℎ2 : channel states including array beam forming gains
for backhaul (BH) and access (AC) links respectively.
• Conventional repeaters which simply amplify-and- • 𝑛1 , 𝑛2 : interference and noise with Gaussian distribution
forward received signal were specified in Release-17, with the with zero mean and variance 𝜎12 , 𝜎22 for BH and AC links
corresponding radio frequency requirements in [2]. respectively.
• Smart repeaters (also called network-controlled repeaters) • 𝑆𝐼𝑁𝑅𝐵𝐻 , 𝑆𝐼𝑁𝑅𝐴𝐶 : SINRs for BH and AC links
that perform amplify-and-forwarding operation with side respectively.
control information provided by the network were studied and • 𝛽𝐵𝐻 , 𝛽𝐴𝐶 : fraction of time-domain resources allocated for
specified in Release-18 [3]. BH and AC links respectively.
• The UE-to-Network Layer 2/Layer 3 relay via side-link Based on this system model, we have
was introduced in Release-17 [4] for proximity-based services
in 5G systems. 𝑦 = ℎ1 √𝑃𝑇1 𝑠 + 𝑛1 ,
 𝑧 = ℎ2 √𝑃𝑇2,𝑚𝑎𝑥 𝑠 + 𝑛2 ,
and the BH and AC SINRs can be calculated accordingly,
𝑃𝑇1 ∙|ℎ1 |2 𝑃𝑇2,𝑚𝑎𝑥 ∙|ℎ2 |2
𝑆𝐼𝑁𝑅𝐵𝐻 = , 𝑆𝐼𝑁𝑅𝐴𝐶 =
𝜎12 𝜎22

The decode-forward relay node can operate based on one

of the following two modes:
• Full-duplex decode-forward mode (FDDF): in this
mode, BH and AC links can operate at the same time Figure 2: System model for amplify-forward repeater in DL
with 0 < 𝛽𝐵𝐻 ≤ 1, 0 < 𝛽𝐴𝐶 ≤ 1. Let 𝐶(𝑆𝐼𝑁𝑅) denote
the capacity achieved at SINR. It can be shown that the • 𝛿𝑁𝐹 : Noise-figure relative to decode-forward relay node,
maximum achievable rate is: which captures the noise-figure difference between the RF
𝐶(𝑆𝐼𝑁𝑅FDDF ) = 𝑚𝑖𝑛(𝐶(𝑆𝐼𝑁𝑅𝐵𝐻 ), 𝐶(𝑆𝐼𝑁𝑅𝐴𝐶 )) (1) chains of amplify-forward and decode-forward node
• Half-duplex decode-forward mode (HDDF): in this depending on the actual implementation. In later
mode, BH and AC links are time-division-multiplexed simulation, we use 𝛿𝑁𝐹 = 1𝑑𝐵.
with 𝛽𝐵𝐻 + 𝛽𝐴𝐶 ≤ 1. It can be shown that the optimum
𝛽
resource allocation is achieved when 𝐵𝐻 =
𝐶(𝑆𝐼𝑁𝑅𝐴𝐶 )
, • 𝑓𝐵𝐹 ≤ 1: beamforming loss factor relative to max array
𝛽𝐴𝐶 𝐶(𝑆𝐼𝑁𝑅𝐵𝐻 ) gain over AC link:
and the resulting achievable rate is given by:
𝑎𝑐𝑡𝑢𝑎𝑙 𝑎𝑟𝑟𝑎𝑦 𝑔𝑎𝑖𝑛
1
𝑓𝐵𝐹 ≔ ≤ 1.
𝑚𝑎𝑥 𝑎𝑟𝑟𝑎𝑦 𝑔𝑎𝑖𝑛
𝐶(𝑆𝐼𝑁𝑅HDDF ) = 1 1 =
+
𝐶(𝑆𝐼𝑁𝑅𝐵𝐻 ) 𝐶(𝑆𝐼𝑁𝑅𝐴𝐶 ) For a conventional repeater, without knowledge of UE’s
𝛼. 𝐶(𝑆𝐼𝑁𝑅FDDF ) (2) direction, a fixed broad beam is typically used for AC link
where 𝛼 ∈ (0.5,1) and depends on the relative values of to cover all possible directions, and thus the beamforming
loss 𝑓𝐵𝐹 < 1; while for a smart repeater with knowledge
𝐶(𝑆𝐼𝑁𝑅𝐵𝐻 ) and 𝐶(𝑆𝐼𝑁𝑅𝐴𝐶 ). If they are the equal, 𝛼 = 0.5;
of UE’s direction (provided by the network), a narrow
if one value is far larger than the other value, 𝛼 ≈ 1. The
beam can be formed toward UE without any beamforming
corresponding end-to-end effective SINR can be obtained
loss, i.e. 𝑓𝐵𝐹 = 1. Note that this beamforming loss factor
by using the inverse function of capacity.
is only for AC link. For BH link, it is assumed that the
The system model for decode-forward relay node with relay node is stationary, and a narrow beam can always be
uplink (UL) operation is similar to DL, except that the 1st formed over the BH link during deployment.
hop is AC and the 2nd hop is BH and resulting achievable
rate can also be represented by (1) and (2). It can be shown that the final received signal 𝑧 at the UE
side is represented as:
𝑧 = √𝑓𝐵𝐹 ∙ ℎ2 ∙ √𝐺 ∙ (ℎ1 ∙ √𝑃𝑇1 𝑆 + 𝑛1′ ) + 𝑛2 .
In order to compare the performance of amplify-forward
repeater with decode-forward relay node, the above system
model is normalized and rewritten, equivalently, as:
1 1
𝑧̃ = 𝑆 + ∙𝑛
̃1 + ∙𝑛
̃,2
√𝑆𝐼𝑁𝑅𝐵𝐻 ⁄𝛿𝑁𝐹 √𝑓𝐵𝐹 ∙𝑓𝑃 ∙𝑓𝑛 ∙𝑆𝐼𝑁𝑅𝐴𝐶

where (𝑛̃,1 𝑛
̃)2 are normalized interference and noise with unit
Figure 1: System model for decode-forward relay node in DL variance, and (𝑆𝐼𝑁𝑅𝐵𝐻 , 𝑆𝐼𝑁𝑅𝐴𝐶 ) are SINRs of BH and AC
links for decode-forward relay node as shown in Section II.A.
B. System Model for Amplify-forward Repeater Parameters 𝑓𝑃 and 𝑓𝑛 are defined below:
For the amplify-forward repeater, a unified system model • 𝑓𝑃 ≤ 1: power loss factor resulting from having finite gain
is established for both conventional and smart repeaters with in the repeater that sometimes prevents achieving the
DL operation, as shown in Figure 2. maximum target transmission power 𝑃𝑇2,𝑚𝑎𝑥 . The power
Here, the amplify-forward repeater is characterized by an loss factor depends on the repeater’s received power
𝑃
amplification gain 𝐺 and following parameters: 𝑃𝑦1 = 𝜎12 ∙ (𝑆𝐼𝑁𝑅𝐵𝐻 + 𝛿𝑁𝐹 ). If 𝑃𝑦1 ≥ 𝑇2,𝑚𝑎𝑥 , there is
𝐺𝑚𝑎𝑥
• 𝐺𝑚𝑎𝑥 : max amplification gain. The repeater adjusts the no loss 𝑓𝑃 = 1; otherwise, 𝑓𝑃 < 1.
amplification gain 𝐺 to achieve a target transmission 𝑃𝑦1 ∙𝐺𝑚𝑎𝑥
power 𝑃𝑇2,𝑚𝑎𝑥 , unless it is limited by 𝐺𝑚𝑎𝑥 . 𝑓𝑃 ≔ 𝑚𝑖𝑛 (1, )=
𝑃𝑇2,𝑚𝑎𝑥
𝜎12 ∙(𝑆𝐼𝑁𝑅𝐵𝐻 +𝛿𝑁𝐹 )∙𝐺𝑚𝑎𝑥
𝑚𝑖𝑛 (1, ).
𝑃𝑇2,𝑚𝑎𝑥
• 𝑓𝑛 ≤ 1 : Noise-forwarding loss factor. Note that the
transmission power of amplify-forward repeater includes
both signal part and noise part. This loss factor captures
the ratio of signal power over total transmission power of
relay node.
|ℎ1 |2 𝑃𝑇1 𝑆𝐼𝑁𝑅𝐵𝐻
𝑓𝑛 ≔ = ≤ 1.
𝑃𝑦1 𝛿𝑁𝐹 +𝑆𝐼𝑁𝑅𝐵𝐻
′ ′
Let 𝑆𝐼𝑁𝑅𝐵𝐻 = 𝑆𝐼𝑁𝑅𝐵𝐻 ⁄𝛿𝑁𝐹 , 𝑆𝐼𝑁𝑅𝐴𝐶 = 𝑓𝐵𝐹 ∙ 𝑓𝑃 ∙ 𝑓𝑛 ∙
𝑆𝐼𝑁𝑅𝐴𝐶 . It can be shown that the end-to-end effective DL Figure 4: Extension to N-Hop network
SINR for amplify-forward node is given by:
1 C. Benefits of Side Control for Smart Repeaters
𝑆𝐼𝑁𝑅𝐴𝐹,𝐷𝐿 = 1 1 , (3)
+
𝑆𝐼𝑁𝑅′𝐵𝐻 𝑆𝐼𝑁𝑅′𝐴𝐶
Comparing with a conventional repeater, a smart amplify-
forward repeater has side information provided by the
and the resulting achievable rate is 𝐶(𝑆𝐼𝑁𝑅𝐴𝐹,𝐷𝐿 ). network to improve the performance over the conventional
repeater. In this section, we discuss the impact of two types
Similarly, the system model for amplify-forward repeater of side information for smart repeater: TDD DL/UL
node with UL operation can be seen in Figure 3. configuration information and the scheduled beam
information associated with scheduled UE for AC link.

Figure 3: System model for amplify-forward repeater in UL

For UL, the 1st hop is AC link and the 2nd hop is BH link.
The noise figure 𝛿𝑁𝐹 is associated with 1st hop, loss factors
(𝑓𝑃2 , 𝑓𝑛2 ) are associated with 2nd hop, and the beamforming Figure 5: Stability concerns for amplify-forward repeater
loss factor 𝑓𝐵𝐹 is always associated with AC link. The
resulting UL effective end-to-end SINR is: A conventional repeater, without knowledge of TDD
DL/UL configuration of the system, may turn on two
1 amplification chains for both DL and UL directions in every
𝑆𝐼𝑁𝑅𝐴𝐹,𝑈𝐿 =
1 1 slot as shown in Figure 5, regardless of whether a slot is a DL
′′ + ′′ slot or an UL slot. With both amplification chains on, the
𝑆𝐼𝑁𝑅𝐵𝐻 𝑆𝐼𝑁𝑅𝐴𝐶
′′ ′′ transmitted signal will loop back to the receiver side via
wherein 𝑆𝐼𝑁𝑅𝐵𝐻 = 𝑓𝑃2 ∙ 𝑓𝑛2 ∙ 𝑆𝐼𝑁𝑅𝐵𝐻 , and 𝑆𝐼𝑁𝑅𝐴𝐶 = same-side coupling between two chains (the red loop) and the
⁄
𝑓𝐵𝐹 . 𝑆𝐼𝑁𝑅𝐴𝐶 𝛿𝑁𝐹 . other-side coupling within each chain as shown in Figure 5
The effective SINR and achievable rate calculation can be and may lead to unstable oscillation, if the amplification gain
extended to a general N-hop network, shown in Figure 4, as exceeds a maximum gain limit 𝐺𝑚𝑎𝑥 . The gain limit 𝐺𝑚𝑎𝑥
follows: depends on the coupling matrix between transmitter and
receiver. Note that the same side coupling with aligned beam
• the end-to-end achievable rate for full-duplex decode- directions is much stronger than the other side coupling
forward relay nodes with N hops is 𝑁ℎ𝑜𝑝𝐶𝐹𝐷𝐷𝐹 = without aligned beam directions.
𝑚𝑖𝑛(𝐶1 , 𝐶2 , … 𝐶𝑁 ), with 𝐶𝑖 being the capacity of the ith
hop. If the TDD DL/UL configuration can be provided to a
• the end-to-end achievable rate for half-duplex decode- smart repeater, the smart repeater only needs to turn on one
forward relay nodes with N hops is 𝑁ℎ𝑜𝑝𝐶𝐻𝐷𝐷𝐹 = chain based on whether the slot is a DL slot or an UL slot. In
this case, there is only other-side antenna coupling, which is
𝑚𝑖𝑛(𝐶𝐻𝐷𝐷𝐹,2 , 𝐶𝐻𝐷𝐷𝐹,3 , … 𝐶𝐻𝐷𝐷𝐹,𝑁 ) , with 𝐶𝐻𝐷𝐷𝐹,𝑖 =
1
much weaker than the same-side antenna coupling.
th
1 1 being the capacity of the i two-hop system with Therefore, a smart repeater with knowledge of TDD DL/UL
+
𝐶𝑖 𝐶𝑖−1 configuration can operate at a much higher maximum stable
the optimum resource allocation as shown in (2). gain than a conventional repeater. In our OTA experiment
• the end-to-end SINR for amplify-forward repeaters with (Section IV), it is observed that the gain limit 𝑮𝒎𝒂𝒙 can be
N hops can be expressed recursively as 𝑒𝑓𝑓𝑆𝐼𝑁𝑅𝑁 = improved by up to 20dB with TDD information.
1
1 1 .
+
𝑒𝑓𝑓𝑆𝐼𝑁𝑅𝑁−1 𝑆𝐼𝑁𝑅′𝑁
Another side information is the scheduled beam direction
for AC link. For a conventional repeater, due to lack of such
information, a fixed broad beam is used on the AC link to
cover all potential beam directions with a smaller array gain,
i.e. 𝑓𝐵𝐹 < 1. But for a smart repeater, if scheduled beam
 information can be provided by the network, the smart that there are buildings along the sides of streets and avenues
repeater can form a narrow beam toward the scheduled UE and the wireless signals are diffracted by the building.
with the maximum antenna array gain, i.e. 𝑓𝐵𝐹 = 1.
Table 1: Simulation parameters
Note that the TDD DL/UL configuration can be semi-
static, which does not change in the time scale of scheduling Parameter Value
slots; while the scheduled beam information can be dynamic, Manhattan grid, 84 gNBs, 156 Relays
which changes in time scale of scheduling slots and has larger Topology inter-(avenue, street) distance = (200, 80)m;
control signaling overhead. In order to understand how much (avenue, street) width = (14, 8)m
performance improvement can be achieved with different • gNB: 16 × 4 per sector
Antenna
• Relay: 4 × 1 per BH sector; 16 × 4 per AC sector
levels of side information, in section III, we consider two (Azi x Ele)
• UE: 2 × 1
types of smart repeaters: semi-smart repeater with only • (fc, BW)=(28, 0.8)GHz, power per PA=7dBm
TDD-awareness, and smart repeater with both TDD- • Pathloss exponent
awareness and scheduled beam information. Channel
o BH: 2 if distance<200m; 3.2 o.w.
o AC: 2 if distance<30m; 3.2 o.w.
Comparing with an amplify-forward repeater, a decode- • Shadow fading: 8dB (AC link), 4dB (BH link)
forward relay node has larger implementation complexity • Knife-edge diffraction [10], reflection not modeled
and latency, because it needs additional digital components Relay [𝐺𝑚𝑎𝑥,𝐴 ]𝑑𝐵 = 50𝑑𝐵, [𝐺𝑚𝑎𝑥,𝐵 ]𝑑𝐵 = 70𝑑𝐵, [𝛿𝑁𝐹 ]𝑑𝐵 =
to decode the received packet and then encode and transmit Param 1𝑑𝐵
to the next hop. Azimuth:
• For AC links of conventional repeaters: fixed broad
III. SIMULATION RESULTS beam.
Beamformer
• For other cases: constant phase offset (CPO) beam
A. Simulation Assumptions steering
Elevation: max array gain of CPO beams.
We consider a deployment of gNBs and relay nodes in a Link- One-hop RX power
Manhattan grid as shown in Figure 6, where gNBs are placed association
at intersections along every even street, and relay nodes are • Round-robin;
placed at intersections along every odd street. Each gNB has Scheduler • spatial reuse between direct and indirect UEs can be
four sectors, covering east, west, north, and south directions. used for decode-forward relay scheme.
Inter-cell • Based on azimuth beam pattern + fixed elevation gain
Each relay node connects to one gNB, from which it receives Interference • No interference between AC and BH links
the strongest signal via one of its two BH sectors pointing to We designed a broad beam pattern for the conventional
north or south directions and provides service to the UEs repeater with 16 azimuth antenna elements that fit the
along the adjacent street via its two AC sectors pointing to Manhattan grid geometry. This beam pattern has a
east and west directions. It can be seen that if relay nodes are beamforming loss factor of [𝑓𝐵𝐹 ]𝑑𝐵 ≤ −8𝑑𝐵 relative to a
not deployed, there will be no coverage on odd streets, except smart repeater and a decode-forward relay node with the
in the areas close to the intersections. maximum array gain.
We consider the following cases for performance
comparison:
• noRepeaterRelay: no repeater or relay nodes.
• conventionalRepeater: conventional repeater with a
single broad beam and[𝑮𝒎𝒂𝒙,𝑨 ]𝒅𝑩 = 50dB;
• semi-smartRepeater: TDD-aware repeater, that uses
the same broad beam as conventional repeater but with a
higher amplification gain limit [𝑮𝒎𝒂𝒙,𝑩 ]𝒅𝑩 = 70dB;
• smartRepeater: TDD-aware repeater with scheduled
beam information that can point its beam to a scheduled
UE, with a gain limit [𝑮𝒎𝒂𝒙,𝑩 ]𝒅𝑩 = 70dB;
Figure 6: Manhattan grid deployment
• halfDuplexRelay-SpatialReuse, halfDuplexRelay-
We consider a simulation area of 2000 × 2000 meters NoSpatialReuse: half-duplex decode-forward relay
with a total of 84 gNBs and 156 relay nodes. 840 UEs (10 node with and without spatial reuse scheme.
UEs per gNB) are randomly dropped outdoor and along the • fullDuplexRelay-SpatialReuse, fullDuplexRelay-
streets and avenues. For a UE, the serving node (a gNB or a NoSpatialReuse: full duplex decode-forward relay node
relay node) is determined as a node from which the received with and without spatial reuse scheme. In these two
power at the UE is the largest. If the serving node is a gNB, cases, the self-interference for full-duplex decode-
the UE is called a direct UE of the gNB, otherwise (when the forward relay node is modeled, assuming 130dB
serving node is a relay node, which is connected to a gNB via isolation between Tx and Rx.
BH link), the UE is called an indirect UE of the gNB. Detailed
simulation parameters are shown in Table 1. It is assumed
 The scheduler follows a simple round-robin scheme,
where for each sector of a gNB, the gNB selects and serves
UEs out of all its associated direct or indirect UEs one-by-
one over consecutive slots. An advanced scheduler, with
spatial reuse, can be enabled for cases with decode-forward
relay nodes as shown in Figure 7. When an indirect UE is
scheduled by a gNB via a decode-forward relay node in a slot,
there may be a time period within the slot that the BH link is
not used in parallel with the AC link for the indirect UE, e.g.
for the case with half-duplex operation, or for the case with
full-duplex operation when the BH link with larger rate
finishes its TX earlier than the AC link. During that time
period, the gNB can schedule another direct UE while the
relay node still serves the indirect UE on the AC link.
Figure 8: CDF of effective DL SINR for direct and indirect
UEs

Figure 7: Example of spatial reuse for decode-forward relay

node

For the cases of conventional repeater or semi-smart

repeater, the repeater is assumed to be always on due to lack
of dynamic scheduling information, which will always
generate interference to other links. For the cases of smart
repeater or a decode-forward relay node, the repeater/relay
node can be turned on or off based on whether an associated Figure 9: CDF of DL spectrum efficiency for indirect UEs
indirect UE is scheduled by the gNB, to reduce interference
to other links.
B. Simulation Results
Under the simulation assumptions shown in Section III.A,
we evaluate the DL performance for various cases and the
distributions (CDF) of various metrics are shown in Figure 8-
Figure 10. Note the number of indirect UEs connected via
repeater/relay nodes are different for different cases, the
percentage of indirect UEs are (32%, 38.5%, 45.7%)
respectively for cases with (conventional repeaters, semi-
smart repeaters, smart repeaters as well as decode-forward
relays).
Figure 8 presents the CDF of end-to-end effective SINR
of direct and indirect UEs. While all types of relay nodes can
Figure 10: CDF of DL sector throughput
improve the SINR performance compared to the case without
relay nodes, smart repeaters and full-duplex decode-and- Figure 10 shows the CDFs of sector throughputs
forward relays offer the best performance. including both direct and indirect UEs. As expected,
Figure 9 shows the CDFs of achieved rates over smartRepeater outperforms the other analog-and-forward
scheduled slots for all indirect UEs. smartRepeater’s repeater solutions. It further provides a better overall sector
performance is only slightly worse than the fullDuplexRelay, throughput compared to the half-duplex decode-and-forward
and much better than the other solutions. It can further be relay with no spatial reuse, and almost a similar performance
observed that the advanced scheduler schemes, with spatial to that of the full-duplex decode-and-forward relay with no
reuse, has not much impact on the indirect UE’s performance. spatial reuse. Advanced scheduling schemes, employing
They may indeed slightly worsen the performance, due to spatial reuse, allow for better resource utilization at the
more interference caused by the direct UEs. network side and hence an improved sector throughput for
both half-duplex and full-duplex decode-and-forward relays.
IV. OVER-THE-AIR TESTS duplex/full-duplex decode-forward relay node with and
To validate the analytical and simulation results presented without spatial reuse for multi-user scheduling. The practical
in the previous sections, over-the-air tests, at 28GHz carrier constraints and factors, e.g. the maximum amplification gain
frequency, were conducted using repeaters (with various for stability of amplify-forward repeaters, the noise figure
level of available side information), in both indoor and difference between amplify-forward repeater and decode-
outdoor environments (Figure 11). The repeater has two units forward relay node, self-interference and inter-cell
for the BH and AC links respectively, where each unit interference, etc., are considered in the analysis and/or
comprises two arrays of 16 × 8 antenna elements. To test the simulation.
conventional repeaters (without TDD information), the two The end-to-end effective SINRs and achievable rates for
arrays on each side are simultaneously active to forward various repeater/relay nodes were analytically derived and
signals in both UL and DL directions. For TDD-aware extended to a general N-hop network. The system-level
repeaters, only one array per unit is active in each slot, studies were also conducted using a Manhattan grid
depending on the TDD DL/UL state of the slot. deployment. The performance of different types of repeaters
was further evaluated for a repeater prototype
implementation in 28 GHz and using different OTA test
setups. We discussed performance and complexity tradeoff
between these types of relay nodes. The conventional
repeaters have the least complexity and latency but achieve
the lowest data rate, the full-duplex decode-forward relays
can achieve the highest throughput at the cost of higher
complexity and latency. For the half-duplex decode-forward
relays, the half-duplex penalty can be compensated partly via
spatial reuse in multi-user scheduling. The smart amplify-
forward repeaters with different levels of side information
strike a good balance between performance and complexity.
ACKNOWLEGMENT
Figure 11: OTA test setup: (a) outdoor-to-indoor coverage
The authors would like to thank Juergen Cezanne (Senior
extension, and (b) outdoor.
Director, Technology) of Qualcomm Technologies, Inc. for
The repeater stability test reveals that TDD-awareness many fruitful discussions, especially on beam design, as well
allows driving the repeater’s maximum amplification gain at as the Qualcomm millimeter wave prototyping team for
a higher level before repeaters entering an instable region. developing a repeater implementation and conducting the
The maximum gain depends on the environment (e.g., the OTA tests.
distribution of near-by clutters) and repeater’s beamforming REFERENCES
configuration. Compared to the conventional repeaters
[1] 3GPP TR 38.874 “Study on Integrated Access and Backhaul (Release-
(without TDD information), TDD-aware repeaters can 16)”, V16.0.0, Dec. 2018.
provide up to 10dB and 20dB higher gains in indoor and [2] 3GPP TS 38.106, “NR Repeater Radio Transmission and Reception
outdoor environments respectively. This, in turn, results in up (Release-17)”, V17.6.0, Sept. 2023.
to 100% improvement in DL throughput in the indoor test [3] 3GPP TR 38.867, “Study on NR Network-Contolled Repeaters
setup. (Release 18)”, V18.0.0, Sept. 2022.
Without AC link beamforming information, a fixed broad [4] 3GPP TS 23.304 “Proximity based Services (ProSe) in the 5G System
(5GS) (Release-18)”, V18.3.0, Sept. 2023.
beam of 30-degree beam-width is used. On the other hand,
[5] G. Liu, F. R. Yu, H. Ji, V. C. M. Leung, and X. Li, ‘‘In-band full-duplex
narrow beams of 6-degree beam-width are used, when the AC relaying: A survey, research issues and challenges,’’ IEEE Commun.
link beam info is available. Adaptive AC link beamforming Surveys Tuts., vol. 17, no. 2, pp. 500–524, 2nd Quart., 2015.
offers up to 7dB extra array gain compared to the solutions [6] J. N. Laneman, D. N. C. Tse, and G. W. Wornell, “Cooperative
with fixed AC beam. diversity in wireless networks: Efficient protocols and outage
behavior,” IEEE Trans. Inform. Theory, vol. 50, no. 12, pp. 3062–
All types of repeaters demonstrate great improvements in 3080, Dec. 2004
the DL throughput and range, compared to the scenarios [7] M. R. Souryal and B. R. Vojcic, “Performance of amplify-andforward
without repeaters. In outdoor test setups, smart repeaters and decode-and-forward relaying in Rayleigh fading with turbo codes,”
outperform conventional repeaters with up to 40% higher in Proc. ICASSP 2006, vol. 4, pp. 681-684, 2006.
offered DL throughput. [8] G. Levin and S. Loyka, “Amplify-and-Forward Versus Decode-and-
Forward Relaying: Which is Better?”, International Zurich Seminar on
V. CONCLUSIONS Communications (IZS), February 29 – March 2, 2012
[9] J. Xin, S. Xu, S. Xiong, H. Xu, H. Zhang, “A Survey on Network
In this paper, we analyzed and compared the performance Controlled Repeater Technology”, IEEE the 8th International
of different types of repeater/relay nodes: conventional Conference on Computer and Communications, Dec. 9-12 2022.
amplify-forward repeater, semi-smart/smart amplify-forward [10] Radio and Microwave Wireless Systems”, University of Toronto
repeaters with different levels of side information, and half- ECE422 Course lecture note, Prof. Sean V. Hum.