Full Duplex Integrated Access and Backhaul for 5G NR: Analyses and Prototype Measurements
Gee Yong Suk, Soo-Min Kim, Jongwoo Kwak, Seop Hur, Eunyong Kim, Chan-Byoung Chae
11 Full Duplex Integrated Access and Backhaul for5G NR: Analyses and Prototype Measurements
Gee Yong Suk,
Student Member, IEEE , Soo-Min Kim,
Student Member, IEEE , Jongwoo Kwak,
StudentMember, IEEE , Seop Hur, Eunyong Kim, and Chan-Byoung Chae,
Senior Member, IEEE
Abstract —Integrated access and backhaul (IAB) frameworksfor 5G new radio (NR) as a cost-effective alternative to thewired backhaul have been investigated by 3GPP. A promisingsolution for this framework is the integration of full duplex(FD) technologies to enhance the spectral efficiency and makean efficient use of the network resources—termed FD IAB.However, FD IAB presents significant technical challenges, suchas self-interference (SI) in the IAB framework, which may castdoubt over the feasibility of FD IAB. In this article, we presenta brief tutorial of the FD IAB framework and its enablingtechnologies. We then numerically evaluate and discuss the link-level SI reduction and the system-level downlink throughputperformance of the FD IAB. Finally, we validate the feasibilityof FD IAB using 28 GHz hardware prototype measurements ofpropagation-domain SI suppression. Our numerical evaluationsand prototype measurements confirm that the FD IAB serves asa promising framework for 5G NR.
I. I
NTRODUCTION
To fulfill the ambitious visions pertaining to future 5Gnetworks, the third generation partnership project (3GPP) hascompleted the standardization of a new radio (NR) accesstechnology called 5G NR [1]. One of the main distinctivefeatures of this standardization is its use of the millimeter-wave (mmWave) frequency band. The large available spec-trum of the mmWave band enables a significant enhancementof transmission speeds, which results in the consideration ofcarrier frequencies up to . . However, owing to thesevere path loss and penetration loss experienced herein, thesystem may suffer from limited coverage and capacity.Analog beamforming techniques have been widely ex-ploited to overcome the path loss and penetration loss issues inthe mmWave frequency band via focusing of the signal powerinto narrow beams [2]. The cooperative operation of multipleantenna elements enables the formation of a highly directivebeam. Thus, the small wavelength of the frequency band alsoattributed to analog beamforming by allowing a large numberof antenna elements to fit into a compact form factor. Ingeneral, the narrow beams not only increase the propagationpath length but also reduce the interference among the var-ious links, thereby providing significant potential for spatialmultiplexing gain.Network densification is another primary approach forenabling extended cell coverage and capacity expansion. Itsobjective is to provide a reliable access channel by reducing G. Y. Suk, S.-M. Kim, J. Kwak and C.-B. Chae are with the School ofIntegrated Technology, Yonsei University, Korea (E-mail: { gysuk, sm.kim,kjw8216, cbchae } @yonsei.ac.kr); S. Hur and E. Kim are with SamsungElectronics Co., Ltd., Korea (E-mail: { s.hur, eunyong.kim } @samsung.com). the inter-site distance, i.e., deploying more cellular basestations (BS) with smaller coverage in a given area [3], [4].However, a significant setback of this approach is that the re-quired large number of BSs and the consequent fiber backhaulinterconnections incur huge capital/operational expenditures(CAPEX/OPEX).Nevertheless, the future network is expected to be highlydense in order to support the high standards of futureapplications such as virtual/augmented reality, internet ofthings, edge computing, and vehicle-to-everything. However,the traditional fiber-backhauling is often an economicallyimpractical solution for carrier operators. In this context,integrated access and backhaul (IAB) technology has beenproposed as a cost-effective alternative to the traditional fiber-backhauled system. In the case of IAB, only a fraction of BSsare connected to the traditional wired infrastructures, whileother BSs relay the backhaul traffic wirelessly [5], [6]. In atypical IAB framework, the access and backhaul links sharethe same frequency spectrum, which results in the resourcecollision problem; thus, resource management is required toresolve this issue. Owing to the simplicity of implementation,many previous studies have incorporated the use of half-duplex (HD) constraints in their frameworks [5], [6], whichwe call HD IAB. In HD IAB, the access and backhaul linksmust make orthogonal use of the given radio resources, be itregarding time or frequency. While this helps to prevent thecollision of the two separate links, it fails to exploit the fullpotential of the given radio resources.Accordingly, another IAB framework, which we denoteas FD IAB, smartly manages the collision of resources andignores the HD constraint by adopting full duplex (FD)technology. FD is an advanced technology, the objective ofwhich is to realize more efficient utilization of the given radioresource by allowing the transmission and reception to occurat the same time/frequency resource block. FD technology notonly doubles the spectral efficiency but also grants more flexi-bility regarding the design of wireless protocols, based on theassumption that the resource collision issue has been appro-priately addressed [7], [8]. The application of FD techniquesto the inter-relaying node infrastructure has been previouslystudied via the concept of an FD relay [9], [10]. However,although these studies laid the foundations for the study ofadvanced FD networks, such as that presented in this article,not enough attention was devoted to the features of 5G NRIAB, i.e., the unique impact of mmWave frequency band andthe strong transmit power from the BS. The majority of thepast analyses and experiments were focused on the sub- a r X i v : . [ ee ss . SP ] J u l Intelligence Networking Lab.
Backhaul
Downlink
Access
Downlink
Access
Uplink
Backhaul
UplinkBackhaul & AccessDownlink Backhaul & AccessUplinkIAB-nodeIAB-donor UECore network internetHD IABFD IAB
Time G A P IAB-nodeIAB-donor UE SI DLIDUMT (a) (b) G A P G A P G A P Fig.1
Fig. 1. (a) IAB system architecture with time-domain partitioning; (b) functional structure of an IAB-node in DL and the sources of interfering signals(downlink). frequency band and a transmission power typically muchlower than that of the next generation Node B (gNB), e.g., Wi-Fi, failing to consider the performance and requirements of FDIAB. As a consequence, the performance and feasibility of FDIAB are yet to be determined. In this article, we aim to analyzethe performance and feasibility of FD IAB by performinga couple of numerical performance analyses and hardware(H/W) prototype measurement, while also presenting a brieftutorial of the framework and its potential.The remainder of this article is organized as follows. InSection II, we present a tutorial of the concept and thepreliminaries of FD IAB. In Section III, we present a coupleof numerical performance analyses of FD IAB at the linklevel and system level. In Section IV, we present some H/Wprototype measurement results to validate the feasibility ofFD IAB. In Section V, we present some remaining technicalchallenges and the conclusion of this study.II.
FD IAB: A C
ONCEPT AND P RELIMINARIES
A. Full-Duplex IAB Framework
Included in the 3GPP NR Release 16, the finalization ofthe work item on IAB is planned for the second quarter ofthe year 2020. Previously, the 3GPP established a study itemon IAB [6], wherein the feasibility of IAB over 5G NR wasassessed. A typical IAB architecture is illustrated in Fig.1,where only a portion of gNBs (IAB-donors) are connectedto the traditional wired infrastructures (core network internet)via fiber, while the rest of the IAB-nodes wirelessly relaythe backhaul traffic. The typical scenario considers an in-band system where the backhaul and access links share thesame frequency spectrum and that the IAB-nodes decodeand forward the backhaul traffic. In this architecture, eachIAB-node possesses two NR functional blocks, i.e., a mobiletermination (MT) and a distributed unit (DU). The MTmaintains the wireless backhaul link with an upstream IAB-node or IAB-donor, while the DU provides the access linkwith the downstream IAB-nodes or user equipment (UE).The lower-layer functions such as radio link control (RLC),medium access control (MAC), and physical (PHY) layerare nested at the distributed IAB-nodes, whereas upper-layerfunctions, such as radio resource control (RRC) and packet data convergence (PDC), are held at the central unit (CU) ofthe IAB-donor.As mentioned previously, an interference problem occursowing to the spectrum sharing of the backhaul and accesslinks. To address this issue, HD constraints, such as time divi-sion multiplexing (TDM), as illustrated in Fig. 1(a), frequencydivision multiplexing (FDM), or spatial division multiplexing(SDM), are often incorporated in the IAB system [5], [6].The fact that the design and implementation of HD IAB isgenerally much simpler than those of its counterpart FD IABalso contributes to the wide adoption of HD constraints.In contrast, FD IAB offers an efficient and flexible systemdesign upon incorporation of the concept of an in-band FDrelay . An in-band FD relay is a network framework in which arelay-node carries the reception and transmission at the sametime and the same frequency resource block, making the HDconstraints unnecessary. The unique characteristics of FD IABis derived from the exploitation of high-frequency band, andhigh transmission power in the system, of which its impactwill be demonstrated in the later section. As illustrated inFig. 1(a), it is apparent that the FD IAB framework has thepotential to demonstrate a spectral efficiency of more thandouble that of HD IAB. Furthermore, the application of FDtechnology can also reduce the end-to-end delay and feedbackdelay, and improve wireless protocol design [10].However, while these may be true, there exists a criticaltechnical challenge regarding self-interference (SI) that mustbe addressed. As illustrated with the downlink example ofFig. 1(b), SI is an undesired interfering transmitted signalfrom an IAB-node’s DU to the UE that is received by thenode’s MT. Owing to the short distance of the interferencesource (DU) from the MT, compared to the distance ofthe desired signal’s source from the MT, SI is a powerfulinterference signal that can result in significant performanceloss. For certain configurations, SI can even deteriorate theperformance of FD IAB worse than that of HD IAB. Anotherpresent issue in the FD IAB is the direct-link interference(DLI), which comprises a backhaul signal along a direct pathto the access link’s UE and bypasses the relay node. Note that,the access link of a DU and UE would be adaptively selectedfrom its beam codebook, whereas the backhaul link of an MT \ Tx power
Rx noise floor
ADC range
Propagation-domain
Digital-domain P o w er Analog-circuit-domain Digital-domain Analog-circuit-domain Propagation-domain mixer
Analog Beam controlRF circuit control
IAB-donorUEWireless channel ~ DAC LOmixer ~ LOADC
LNA
HPA
DUMT
FD IAB-node
Receive bits + SI + AnalogBeam control
Transmit bits RF chainsRF chains
Residual SI
Fig.2
Fig. 2. A block diagram of the SI reduction procedure in an FD IAB-node in downlink scenario. The illustration is partially adopted from [8]. and IAB-donor would have its stable backhaul link through aperfectly aligned fixed beam. Consequently, the experiencedSI at the MT may change accordingly.
B. Self-Interference Reduction Techniques
In this section, with our focus set on FD IAB, we brieflysummarize the existing SI reduction techniques [8]–[10]. SIreduction techniques can be categorized according to theirdomain of realization: propagation-domain, analog-circuit-domain, and digital-domain. The overall procedure of SIreduction for a given downlink scenario is illustrated in Fig. 2.The SI signal first propagates with the Tx power , and thenits power is reduced to a value close to the Rx noise floorby the consecutive SI reduction techniques. This continuousprocedure is analogous to a group of people taking turns todrink a full cup of water. The more one drinks, the lower isthe burden left for the next one.
1) Propagation-domain SI suppression:
SI suppression inthe propagation-domain is likened to the first person drinkingthe water. The SI is suppressed at the propagation-domain,before it reaches the MT, via electromagnetic isolation of thetransmitting and receiving antennas. This isolation is realizedby a combination of path loss, antenna directionality, andcross-polarization [12].A straight-forward method of suppressing the SI is toincrease the amount of path loss experienced by the SIsignal by physically separating the RF front-ends of the Txand Rx by a large distance or placing an electromagneticbarrier between them. The greater the separation distance orthe more absorbent the shielding between them, the greaterthe SI suppression that can be achieved. However, becausethis can easily result in bulky form factor of the equipment, Higher transmit power requires stronger SI reduction performance. De-livering the sufficient SI reduction for high transmit power of a typical BShas been considered as a research challenge itself [10], [11]. the tradeoff between the amount of SI suppression and thephysical size of the network equipment should be consideredin practice.Another approach comprises exploiting the antenna di-rectionality of the RF front-end. By employing the direc-tional transmit and receive antennas that focus their radiat-ing/sensing capability in specific directions, the interferencebetween them can be avoided in advance. Because the designof a highly directive analog beam through the use of anantenna array is universal in the mmWave frequency band,this approach is especially effective for FD IAB. Likewise, cross-polarization is another method of propagation-domainSI suppression. The occurrence of interference is prevented inadvance via the adoption of a set of cross-polarized antennas,which only transmits/receives horizontally/vertically polarizedsignals, at the RF front-ends.
2) Analog-circuit-domain SI cancellation:
The next SIreduction occurs in the analog-circuit-domain. As shown inFig. 2, the canceller circuit taps the outgoing transmit signal,regenerates an estimate of the SI signal, and subtracts it fromthe received signal. There are two approaches for regeneratingthe SI signal: the channel-aware approach, according to whichthe knowledge of the SI channel is periodically acquired andits canceller circuit is actively tuned to mimic it, and the channel-unaware approach, according to which its cancellercircuit is tuned once the system is calibrated. As per thechannel-unaware approach, only the direct-path componentof the SI channel is captured. In contrast, as for the channel-aware approach, the time-variant reflected-path componentsof the SI channel are also captured, which facilitates a closerestimate of the true received SI signal, which adaptivelychanges according to its surrounding environment.In practice, a canceller circuit involves certain costs, suchas additional power consumption, circuit complexity, and eventraining overhead in case it requires channel information.Furthermore, for the adaptive circuit to support multi-stream
IAB-donorIAB-node 1 IAB-node 2 (c)(b)(a)UE grid S I c h a nn e l re s p o n s e ( d B ) Delay ( 𝝁𝒔 ) Direct-pathReflected-path
Propagation paths IAB-node 1 City view3D modeling
Fig.3
Fig. 3. System level simulation of IAB in 28 GHz, 3D modeled environments: (a) Network deployment configuration; (b) 3D modeling of an urban environmentand the realistic channel modeling via ray-tracing; (c) an example of the determined SI channel at IAB-node 1. or wideband signals, greater costs than the channel-unawarecase would be incurred. Thus, it is crucial to design anefficient circuit structure and a tuning algorithm such that thecompromise on the cancellation performance is minimal.
3) Digital-domain SI reduction:
As depicted in Fig. 2, theSI that remains after the propagation-domain and analog-circuit-domain is then subjected to the digital-domain SIreduction, which is likened to the last person drinking thewater. The ease of implementation of complicated signalprocessing is the most significant advantage of the operationsin the digital-domain. However, the performance of digital-domain SI reduction is limited by the dynamic range of ananalog-to-digital converter (ADC). For ideal SI reduction suchthat the residual SI is cut to a value of the Rx noise floor,the magnitude of the residual SI power should be lower thanthe magnitude by which the ADC’s dynamic range is greaterthan the noise floor. In other words, the SI reduction involvedin the propagation-domain and analog-circuit-domain shouldmake the magnitude of residual SI power reach the gray zonein the graph on the right-hand side of Fig. 2.The first approach of the digital-domain SI reduction is digital SI cancellation . The expected baseband-equivalent SIsignal is regenerated based on the estimated residual SI chan-nel and then subtracted from the received baseband digitalsignal. In the case of a multi-stream scenario, transmit/receivebeamforming is another method used for suppressing SIvia digitally weighting complex-valued gains (digital precod-ing/combining) to each of the streams in an adaptive manner.The main hurdle presented by these methods is that the nonlin-earity caused by RF imperfections—e.g., power amplifier (PA)nonlinearity, in-phase and quadrature (I/Q) imbalances, andphase noise at a local oscillator (LO)—should be accountedfor. Appropriate channel models are necessary for taking thesenonlinear characteristics into consideration [13]. III. N UMERICAL A NALYSES AND D ISCUSSIONS
In this section, we numerically analyze and discuss theperformance of FD IAB for 5G NR. This study involves link-level and system-level simulations that are interdependent.The procedure of our analysis is as follows:(i) Set the 3D environment and IAB deployment configura-tions(ii) Compute the channel modeling of various wireless prop-agations in the scenario(iii) Determine system-level decisions such as UE’s cell-connection and beam selection(iv) Perform a link-level simulation of the SI reduction ofanalog-circuit-domain and digital-domain(v) Evaluate the downlink throughput of various networktopologies
A. Network Deployment Configurations
In this study, we consider an in-band IAB scenario withphysically fixed relays over a
120 MHz bandwidth with acenter frequency
28 GHz . As illustrated in Fig. 3(a), weemploy one IAB-donor and two IAB-nodes to support theequally-spaced 4,527 outdoor UEs forming a rectangular gridin the
500 m ×
500 m region via maximum single-hop relaying.The IAB-donor and the IAB-nodes are placed at the top ofthe buildings
130 m ,
126 m , and
99 m in height (approximately33–43 floors high).Illustrated in Fig. 3(b) is the 3D urban model of the down-town area in Rosslyn city, Virginia, USA. Owing to the char-acteristic of the mmWave channel whereby it possesses fewerreflected paths than its sub- counterpart, it is essential toconsider the actual surroundings of the transceivers to realizerealistic modeling of the wireless propagation channel. Thus,we adopt the ray-tracing algorithm coupled with geometricoptics and the uniform theory of diffraction [14] using Wire-less Insite to achieve the precise channel response for ourspecific scenario of interest. The modeling of the channel reflects the impact of cell deployment, radiation pattern of theantenna, antenna separation, and reflection/diffraction fromthe surrounding environment. Thus, the effect of propagation-domain SI suppression with antenna separation and antennadirectionality is considered during the SI channel modeling.We assume that the UE scheduling for multiple accessis performed in advance and that the network supports oneUE at a given time instance. Each UE connects to the cellproviding the best access link in terms of signal-to-noise ratio(SNR). Similarly, for the DU to support its allocated UE,the maximum-SNR beam selection scheme is deployed. Forthe access links thereof, the DUs share a universal codebookthat divides the 120 ◦ in the azimuth plane into eight maindirections and the 30 ◦ in the elevation plane into two maindirections to obtain a total of sixteen beams. The DUs of theIAB-donor and IAB-nodes and the MTs of the IAB-nodesdeploy a shared radiation pattern of a directional antenna,which has a
20 dBi directivity gain with an approximate beamwidth of 12 ◦ , whereas the UEs deploy a uniform dipoleantenna with an isotropic radiation pattern in the azimuthplane. The DUs transmit their signal with a high transmissionpower of
43 dBm . B. Link-Level Evaluation of SI Reduction
After acquiring the residual SI channel subsequent tothe propagation-domain SI suppression, we conduct a link-level SI reduction simulation based on single-stream orthog-onal frequency-division multiplexing (OFDM). A pilot-basedchannel-aware two-tap canceller circuit is considered for theanalog-circuit-domain, and a nonlinear digital SI cancella-tion [13] is considered for the digital-domain reduction.For the analog-circuit-domain SI cancellation, we adopta two-tap RF canceller. The fixed delay values are pre-determined based on the delay of the direct path of eachantenna separation case. For the cases of d = , d = ,and d = . , the two delays of the canceller are set as { , } , { , } , and { . , . } , respectively.Furthermore, the effect of non-linear components in the actualRF circuit board, such as PA (
10 dB gain and .
63 dBm P )and a 14-bit ADC, is considered. In the OFDM, a fast Fouriertransform size of 1024, employing a subcarrier number of 792,and cyclic prefix length of 140 were considered.Fig. 4 presents the averaged SI reduction and its com-position for all the transmission cases in the scenario withthree different antenna separations. For all various antennaseparations, the average total SI suppression is sufficient to cutthe residual SI to a value close to that of the Rx noise floor.It is encouraging to observe that, for the d = and d = cases, the SI reduction of the propagation-domain alonemanaged to safely reach the effective ADC range ( .
24 dB ,with a 14-bit resolution [8]) above the Rx noise floor, whichis the requirement for digital-domain SI cancellation. Theobservation leads to new opportunities wherein active-analog-domain SI cancellation may be unnecessary for specific FDIAB configurations, which provides the network operatorswith freedom of system-design. For the d = . case,although the average propagation-domain SI suppression was Antenna separation ( d ) Rx noise floor14-bit ADC range Propagation-domainAnalog-circuit-domainDigital-domain S I P o w e r ( d B m ) Fig.4
Fig. 4. The simulated result of full SI reduction with different antennaseparations. sufficient to lie within the effective ADC range above the Rxnoise floor, there were specific cases wherein the suppressionwas insufficient, which resulted in significant performanceloss at the digital-domain cancellation. Therefore, analog-circuit-domain SI cancellation techniques have also beenapplied, resulting in a near-optimal performance. Overall, thehuge impact of antenna separation and analog beamforming atthe propagation-domain is distinctive, comprising a significantproportion of the total SI reduction.
C. System-Level Evaluation of IAB Throughput
In this section, we present a downlink throughput perfor-mance of various IAB configurations with antenna separationcases of d = , d = , and d = . . We aim to observethe performance of the FD IAB with full SI reduction, whichindicates the inclusion of the simulation result demonstrated inthe link-level simulation (FD IAB). As a benchmark, we alsoevaluate the network topology of a fibered-backhaul systemwith three wired gNBs, each positioned at the IAB-donor andIAB-nodes’ coordinates (fibered-backhaul); the same topologywith HD constraints and operation (HD IAB); an ideal FDIAB that assumes zero SI (Ideal FD IAB); and FD IABwithout the SI reduction in the analog-circuit-domain anddigital-domain (FD IAB without (w/o) full SI reduction).First, Fig. 5 shows that the FD IAB without full SIreduction suffers from performance degradation owing to theSI, which intensifies as the antenna separation decreases.Note that, in the case of d = . , the performance ofFD IAB became worse than that of the HD IAB. In thecases of d = and d = without full SI reduction,the degradation is evident but not as severe as that for d = . . This is because the backhaul link was solid enoughto withstand the residual SI for the two cases, which can belargely attributed to its perfectly aligned fixed beam betweenthe IAB-donor and IAB-node. It should be noted that the SIdeteriorates the backhaul link’s quality (see Fig. 1(b)), whichforms a virtual upper bound to the access link. Combined with Intelligence Networking Lab.
Throughput (Mbits/sec) C D F DLI lossFD gain
Full SI reduction
Ideal FD IAB
Fibered-backhaulHD IAB
Fig.5
Fig. 5. The downlink throughput CDF of the outdoor UEs in the cases ofdifferent antenna separation. the discrete selection of adaptive modulation schemes at thebackhaul, this bottleneck results in the unsmooth curve of theFD IAB without full reduction.On the other hand, the curve of ideal FD IAB representsthe performance upper bound of the FD IAB systems. Notethat the performance of all considered FD IABs approximatesthe ideal FD IAB through the application of full SI reduction.The observed performance gap between the ideal FD IABand the fibered-backhaul is the performance loss due to theDLI introduced earlier. While the system performance can beimproved by clever SI reduction, there is no straight-forwardmethod at the link-level to mitigate the DLI once the backhaullink is set as fixed, implying that DLI should be consideredat the cell planning stage.IV. P
ROTOTYPE M EASUREMENTS OF
SI S
UPPRESSION
Upon our prior observation where the SI suppression at thepropagation-domain accounts for a considerable proportionin the total SI reduction process, we implement a
28 GHz
H/W prototype for the measurement of the propagation-domain SI suppression to validate the feasibility of FD IAB.The prototype is implemented using the LabVIEW systemdesign software and a field-programmable-gate-array-based(FPGA-based) PXIe software-defined radio (SDR) platform.As shown in Fig. 6(a), the measurement was conducted inan outdoor rooftop of a four floor building. The objective ofthe prototype was to imitate an FD IAB-node in a suburbandownlink scenario with its DU and MT separated by relativeorientation and distance. The Rx below mimics the MT ofan IAB-node, while the Tx above plays the role of the DU.Although the SI is the access link’s signal, not intended toconsider the MT as an Rx (see Fig. 1(b)), we implement asynchronized communication link between the DU and MTto measure the received signal strength as SI.The block diagram in Fig. 6(b) describes the main compo-nents of our 28 GHz prototype. To form a sharp analog beamat the RF front-end, advanced lens antennas with 19.86 dBigain and approximately a beamwidth of 13.4 ◦ are used [15]. The radio heads of the Tx/Rx up-converts/down-converts the input frequency to the target frequency, whereasthe LO/IF module generates the LO and control signals forthe radio heads. DAC/ADC module is equipped with a 14 bitDAC and 12 bit ADC resolution, respectively. The FPGAprocessing module provides high-throughput baseband pro-cessing such as OFDM processing, channel estimation, timesynchronization, and digital down/upconverting (DDC/DUC).The host uses the real-time controller to control the applica-tions.Fig. 6(c) demonstrates the measured propagation-domainSI suppression according to the relative azimuth angle, inthe cases of different antenna separation distance. An aver-age of .
125 dB , .
26 dB , and .
18 dB suppression wasmeasured for the d = , d = , and d = . casesrespectively, owing to the path loss and antenna directionality.Note that, although they are intended to be designed similarly,the environment of the measurement and the simulation inSection III are not identical, and thus, direct comparison oftheir results are invalid. Some main differences between thetwo environments are the number of surrounding buildings,which act as reflectors, and the existence of the metal polebetween the transceivers, which acts as an absorber in themeasurement. The measurement result demonstrates that largeamount of SI suppression can be effectively achieved in FDIAB, especially when compared to the measurement resultsin . band [10], [12]. For example, SI suppressionof . was measured with . separation distance,absorbing shield, and cross-polarized antennas, whereas ourexperiment with . separation of directional antennasshowed an average of .
18 dB suppression. The large amountof reduction attributes to the sharp radiation pattern and highpath loss in the system. Considering that the design of a highlydirective analog beam is universal in the mmWave frequencyband, such an observation reflects the feasibility of FD IABwith significant SI reduction.V. F
URTHER C HALLENGES AND C ONCLUSION
This study was focused on one of the most criticalchallenges—if not the most critical challenge—of the FDIAB framework, i.e., SI reduction. However, there also existsfurther challenges beyond the PHY layer to exploit the fullpotential of FD IAB, which mainly includes: • MAC layer challenges: In the FD IAB framework,compared to its HD IAB counterpart, more devices aresimultaneously communicating at a given time instance,leading to the increase of interfering sources amongdifferent devices. Thus, a careful resource allocation andscheduling is necessary • Network layer challenges: There exists huge potentialto improve the end-to-end delay of the FD IAB systemvia optimized routing algorithms and protocols. Therouting algorithm may even vary according to the targetperformance measure of the applicationIn this study, we investigated the performance and feasibil-ity of FD IAB in 5G NR. First, we addressed a brief tutorialfor the FD IAB framework and its enabling technology. We (a) (b)Transmitter Receiver
DAC ADCOFDM processing OFDM processingConfiguration, control, monitoring
RF front-end (WR-28 pyramidal horn antenna)Radio head (mmRH-3642, 3652)LO/IF (PXIe-3620)DAC/ADC (PXIe-3610, 3630)Multi FPGA processing (PXIe-7902, 7976)Host (PXIe-8880)Chassis (PXIe-1085)
Radio head TxIF upconverter Radio head RxIF downconverter LO GH z I F Analog baseband3.072 GS/s <—> 153.6 MS/s GH z I F Self interference (c)
Satellite viewRx radio unit (head + antenna)Tx radio unit (head + antenna)
Distance (d)
Azimuth angle f ( θ ) Rx baseband processingTx baseband processing
Fig. 6. (a) A set-up of 28 GHz H/W real-time prototype with the test scenario, and its satellite view; (b) a block diagram of the prototype; (c) SI measurementfor different cases. then presented our analyses on the performance of the FDIAB in a realistic network scenario. Numerical performanceevaluations of the SI reduction at the link-level and theperformance of downlink throughput at the system-level wereperformed. Finally, we obtained the
28 GHz
H/W prototypemeasurements of propagation-domain SI suppression to verifythe feasibility of FD IAB. Based on our numerical analysesand prototype measurements, we expect that FD IAB willserve as a promising solution and our results provide insightinto FD IAB system design.A
CKNOWLEDGMENT
This research was in part supported by Networks Business,Samsung Electronics and IITP funded by MSIT under grants2016-0-00208. R
EFERENCES[1]
TS 38.300, NR and NG-RAN Overall Description, Release 16 , 3GPPStd., Apr. 2020.[2] W. Roh, J.-Y. Seol, J. Park, B. Lee, J. Lee, Y. Kim, J. Cho, K. Cheun,and F. Aryanfar, “Millimeter-wave beamforming as an enabling tech-nology for 5G cellular communications: Theoretical feasibility andprototype results,”
IEEE Commun. Mag. , vol. 52, no. 2, pp. 106–113,Feb. 2014. [3] D. L´opez-P´erez, M. Ding, H. Claussen, and A. H. Jafari, “Towards1 Gbps/UE in cellular systems: Understanding ultra-dense small celldeployments,”
IEEE Commun. Surveys Tuts. , vol. 17, no. 4, pp. 2078–2101, Jun. 2015.[4] C.-B. Chae, T. Tang, R. W. Heath, Jr., and S. Cho, “MIMO relaying withlinear processing for multiuser transmission in fixed relay networks,”
IEEE Trans. Signal Process. , vol. 56, no. 2, pp. 727–738, Feb. 2008.[5] M. Polese, M. Giordani, T. Zugno, A. Roy, S. Goyal, D. Castor, andM. Zorzi, “Integrated access and backhaul in 5G mmwave networks:Potential and challenges,”
IEEE Commun. Mag. , vol. 58, no. 3, pp.62–68, Mar. 2020.[6]
TS 38.874, Study on integrated access and backhaul, Release 15 , 3GPPStd., Jan. 2019.[7] D. Bharadia, E. McMilin, and S. Katti, “Full duplex radios,” in
Proc.Asilomar Conf. on Signal, Syst. and Comput. , Aug. 2013, pp. 375–386.[8] A. Sabharwal, P. Schniter, D. Guo, D. W. Bliss, S. Rangarajan, andR. Wichman, “In-band full-duplex wireless: Challenges and opportuni-ties,”
IEEE J. Sel. Areas Commun. , vol. 32, no. 9, pp. 1637–1652, Jun.2014.[9] G. Liu, F. R. Yu, H. Ji, V. C. Leung, and X. Li, “In-band full-duplexrelaying: A survey, research issues and challenges,”
IEEE Commun.Surveys Tuts. , vol. 17, no. 2, pp. 500–524, Jan. 2015.[10] D. Kim, H. Lee, and D. Hong, “A survey of in-band full-duplextransmission: From the perspective of PHY and MAC layers,”
IEEECommun. Surveys Tuts. , vol. 17, no. 4, pp. 2017–2046, Feb. 2015.[11] M. Chung, M. S. Sim, J. Kim, D. K. Kim, and C.-B. Chae, “Prototypingreal-time full duplex radios,”
IEEE Commun. Mag. , vol. 53, no. 9, pp.56–63, Sep. 2015.[12] E. Everett, A. Sahai, and A. Sabharwal, “Passive self-interference suppression for full-duplex infrastructure nodes,”
IEEE Trans. WirelessCommun. , vol. 13, no. 2, pp. 680–694, Jan. 2014.[13] J. W. Kwak, M. S. Sim, I.-W. Kang, J. S. Park, J. Park, and C.-B. Chae,“A comparative study of analog/digital self-interference cancellationfor full duplex radios,” in
Proc. Asilomar Conf. on Signal, Syst. andComput. , Nov. 2019, pp. 1114–1119.[14] C. A. Balanis,
Advanced engineering electromagnetics . John Wiley &Sons, 2012.[15] Y. J. Cho, G.-Y. Suk, B. Kim, D. K. Kim, and C.-B. Chae, “RFlens-embedded antenna array for mmWave MIMO: Design and per-formance,”