Millimeter-Wave Massive MIMO Testbed with Hybrid Beamforming
MinKeun Chung, Liang Liu, Andreas Johansson, Martin Nilsson, Olof Zander, Zhinong Ying, Fredrik Tufvesson, Ove Edfors
MMillimeter-Wave Massive MIMO Testbed withHybrid Beamforming
MinKeun Chung , Liang Liu , Andreas Johansson , Martin Nilsson ,Olof Zander , Zhinong Ying , Fredrik Tufvesson , and Ove Edfors Department of Electrical and Information Technology, Lund University, Sweden Sony Research Center Lund, Swedenfirstname.lastname@ { eit.lth.se, sony.com } Abstract —Massive multiple-input multiple-out (MIMO) tech-nology is vital in millimeter-wave (mmWave) bands to obtainlarge array gains. However, there are practical challenges, suchas high hardware cost and power consumption in such systems.A promising solution to these problems is to adopt a hybridbeamforming architecture. This architecture has a much lowernumber of transceiver (TRx) chains than the total antenna num-ber, resulting in cost- and energy-efficient systems. In this paper,we present a real-time mmWave (
28 GHz ) massive MIMO testbedwith hybrid beamforming. This testbed has a 64-antenna/16-TRx unit for beam-selection, which can be expanded to largerarray sizes in a modular way. For testing everything frombaseband processing algorithms to scheduling and beam-selectionin real propagation environments, we extend the capability ofan existing 100-antenna/100-TRx massive MIMO testbed (below ), built upon software-defined radio technology, to a flexiblemmWave massive MIMO system.
Index Terms —Beam-selection, beamforming, massive multiple-input multiple-out (MIMO), millimeter-wave (mmWave), testbed.
I. I
NTRODUCTION
Massive multiple-input multiple-out (MIMO) is a promisingmulti-user (MU)-MIMO technology where each base station(BS) is equipped with an excess number of antennas, comparedto the number of user equipments (UEs), e.g., a few hundredBS antennas simultaneously serving tens of UEs. The concept ofmassive MIMO has been demonstrated to achieve an order-of-magnitude higher spectral efficiency with practical acquisitionof channel state information (CSI), as compared to conventionalsmall-scale MIMO technology [1], [2]. In recent years, thedevelopment of massive MIMO prototype systems, operatingbelow- , has been carried out for proof-of-concept andperformance evaluation under real-world conditions [3], [4].Another key approach to enhance the network capacity is theoperation in millimeter-wave (mmWave) bands, i.e.,
30 GHz –
300 GHz [5]. It provides an order-or-magnitude more spectrumthan we ever had access to. At the mmWave bands, a large-scale antenna system, i.e., massive MIMO, is imperative toobtain sufficient signal-to-noise ratio (SNR) due to its high free-space path loss (FSPL) [6]. However, there are fundamentaldifferences between the design and implementation of massiveMIMO below- and at mmWave frequencies. The maindifferences are summarized as follows: • The architectures: the small wavelength at mmWave fre-quencies enables a large number of antennas in a small … Baseband/IF Processing
Digital Subsystem Analog Subsystem … … … Front-End Module (FEM) Beam-switching moduleFrequency Converter (FRECON)16-TRx chains
Fig. 1. System overview of our proposed mmWave massive MIMO testbed. physical size. However, the current high cost and powerconsumption of the transceiver (TRx) chains at mmWavefrequencies make a fully-digital processing approach pro-hibitive. Hybrid analog and digital beamforming can bean alternative architecture for mmWave massive MIMOsystems [7], [8]. This architecture has a much lowernumber of TRx chains than the total number of antennas. • The propagation channels: propagation environments havea different effect on smaller wavelength signals. For exam-ple, diffraction, scattering, and penetration losses. It leadsinto different statistics of both small-scale and large-scalevariations [9]. • The baseband processing algorithms: depend on hard-ware, as well as channel characteristics. As comparedwith below- systems, the mmWave system is moresensitive to hardware impairments, such as phase noise,power amplifier (PA) nonlinearities [10], [11]. Thus, base-band processing algorithms for impairment estimation andcompensation are crucial in mmWave systems.Based on these differences, for testing everything from base-band processing algorithms to scheduling in new environments,we extend the capability of an existing 100-antenna/100-TRxmassive MIMO testbed (below ), built upon software-defined radio (SDR) technology, to a flexible mmWave massiveMIMO system. Recently, we have demonstrated this real-timemmWave massive MIMO system at IEEE Wireless Communi-cation and Networking Conference in 2020 [8].In this paper, we provide an overview of our real-time
28 GHz massive MIMO testbed, which includes a hybrid beamformingarchitecture based on beam selection, as illustrated in Fig. 1. Our a r X i v : . [ ee ss . SP ] D ec Fig. 2. An architecture of 64-antenna/16-TRx hybrid beamforming testbed. DC control signals from each SDR, i.e., for TDD and beam switching, aredelivered to FRECON (only TDD switching signal) and FEM (both). For simplicity, the routes for the DC control signals are omitted in this figure. testbed constitutes a flexible platform that supports up to 64-antenna/16-TRx BS, simultaneously serving a maximum of 12UEs using orthogonal frequency division multiplexing (OFDM)in time-division duplex (TDD) mode.II. T
ESTBED A RCHITECTURE
In this section, we overview the architecture of our testbed. Asillustrated in Fig. 1, the proposed testbed is divided into analogand digital subsystems. The physical hardware setup for thedigital subsystem is in part identical with the . LabVIEW on a standard Windows 7 64-bit operating system toconfigure and control the system. LabVIEW provides both hostand FPGA programming. To perform MIMO processing, e.g.,precoding, detection, we use co-processing modules (FlexRIO7976R). Also, a reference clock source (PXIe-6674T) andreference clock distribution network (Octo-Clock) are includedto be able to synchronize the entire BS. Each SDR containstwo TRx chains and a Kintex-7 FPGA. The SDR basicallyperforms local processing on a per-antenna basis, e.g., OFDMprocessing and reciprocity calibration. Also, it plays a role as aninterface to send control signals from the digital to the analogsubsystem, where there are two kinds of control signals. Oneis the signal for TDD switching, the other for beam-selection.These control signals are delivered through a 15-pin general-purpose input/output (GPIO) in each SDR. The analog subsystem includes a 64-element antenna ar-ray, a clock distribution module (ClkDist), frequency convert-ers (FRECONs), and front-end modules (FEMs) for analog-domain beamforming. For reconfigurability and scalability ofthe testbed, we designed the FRECONs and FEMs in a modularway. Each module has a small number of TRx chains, i.e.,two per FRECON and one per FEM. To up/down-convertbetween IF signal from/to the SDR and
28 GHz bands, wedesigned FRECON printed circuit boards (PCBs) that consist ofup/down conversion mixers, filters, driver amplifiers (DAs), low-noise amplifiers (LNAs), and SPDT switches. One FRECONis connected with one SDR. The main role of the FEM is toswitch between four predefined beams, according to the controlsignal from the digital subsystem. The FEM, thus, contains aSP4T switch, and two FEMs are connected to the FRECON.The 64-element antenna array has 16 subarrays. Each subarray,consisting of four antenna elements, plugs in to one FEM whereone antenna element in the subarray is selected for analogbeamforming. In our testbed, the BS and UE, respectively, has acommon local oscillator (LO) for up/down conversion betweenIF and
28 GHz bands. We employ a . -LO (PLDRO-25500-10). To amplify and distribute the LO signal to multipleFRECONs, we design the ClkDist. Fig. 2 illustrates an architec-ture of 64-antenna/16-TRx hybrid beamforming testbed wherethe quantities of units belonging to each subsystem are alsoshown. III. T ESTBED D ESIGN AND I MPLEMENTATION
To perform measurements in a variety of scenarios, sufficientgain of each TRx chain is imperative in designing a testbed.Also, for our proposed hybrid beamforming testbed, beamswitchability is a key design feature. This section elaborates ower Splitter PA Tx Mixer
Tx RF 2
LNALNA
Rx RF 2Rx RF 1Tx RF 1
EmbedAnt 2EmbedAnt 1
Tx RF In
28 GHz
FRECON FEM 1
RF 1RF 2RF 3 RF 4
FEM 2
LO In
DADADADA DA DADA DA Tx RF In
28 GHz
PAPA
LNALNA
Rx RF out
28 GHz
Rx RF out
28 GHz
Rx MixerRx MixerTx Mixer (a)(b)Fig. 3. FRECON and FEM: (a) block diagram with one FRECON and twoFEMs (b) photographs of fabricated FRECON (left) and FEM (right). on the design of our
28 GHz massive MIMO testbed and itsimplementation.
A. FRECON and FEM
MmWave systems are more sensitive to PA nonlineari-ties, compared to conventional systems below- . For theFRECON and FEM, we focused on an appropriate architecturedesign and component selection to reduce the PA nonlinearities.As mentioned in Sec. II, one FRECON has two TRx chains,and connects with two FEMs. A combined block diagramof FRECON and FEM is shown in Fig. 3(a). The FRECONcontains the same eight DAs (HMC383LC4) but has differenttargets. Four DAs between an LO-input port and mixers is foramplifying the . -LO signal, and the other four DAs forthe
28 GHz transmit (Tx) signal. The mixers (HMC1063LP3E)are used for up/down-conversion between IF and
28 GHz bands,where an LO power of more than
10 dBm is required to operateit. That is the reason why a DA for amplifying the LO signalis needed for each mixer. The conversion gain of the mixer isaround −
10 dB . To compensate this power loss and achieve highoutput power, the Tx chain is equipped with two consecutiveDAs. On the other hand, each receive (Rx) chain contains anLNA (HMC1040LP3CE) to avoid compression. In the front-end of the FRECON, there are SPDT switches (ADRF5020) G a i n ( d B ) P o u t ( d B m ) Pout_CW Gain (dB)
Combined Tx gainCombined Tx power (a)
LO (cid:393)(cid:381)(cid:449)(cid:286)(cid:396) (cid:1085)(cid:1010) (cid:282)B(cid:373)(cid:853) RF (cid:393)(cid:381)(cid:449)(cid:286)(cid:396) (cid:882)(cid:1007)(cid:1006) (cid:282)B(cid:373)(cid:853) RF (cid:296)(cid:396)(cid:286)(cid:395)(cid:437)(cid:286)(cid:374)(cid:272)(cid:455) (cid:1006)(cid:1009)(cid:856)(cid:1011) GH(cid:460) (cid:410)(cid:381) (cid:1007)(cid:1004)(cid:856)(cid:1009) GH(cid:460)
EIT f(cid:396)e(cid:395)(cid:437)enc(cid:455) c(cid:381)n(cid:448)e(cid:396)(cid:410)e(cid:396) (cid:1085) S(cid:381)n(cid:455) f(cid:396)(cid:381)n(cid:410) end m(cid:381)d(cid:437)le (cid:396)ecei(cid:448)e m(cid:381)de (cid:449)i(cid:410)h fil(cid:410)e(cid:396) (b)Fig. 4. Measurement results of FRECON and FEM: (a) output power level(left vertical-axis) and Tx gain (right vertical-axis) of combined FRECONand FEM (b) Rx gain of combined FRECON and FEM. for TDD switching .The employment of FEM is to support testing of long-rangecommunications, as well as beam-switching. As depicted inFig. 3(a), one FEM contains an additional PA (MAAP-011246)and LNA, which have a high power gain. Also, a SPDT switchfor TDD switching, and a SP4T switch for beam-selectionare included. The SP4T switch engages with four RF ports,and performs switching or selecting by control signals fromdigital subsystem. For the beam-switching, the isolation betweenpaths in the switches is crucial. The SPDT and SP4T switches,therefore, are designed so that they both have a high isolation in mmWave frequencies. All the components, except the PA,were developed in-house.Photographs of the fabricated FRECON and FEM are shownin Fig. 3(b). Based on the measurements of each module, theTx and Rx gains for the FRECON are around and , respectively. For the FEM, around
14 dB and
12 dB ,respectively, is achieved. Fig. 4 shows a combined TRx gain ofFRECON and FEM. The Tx gain in its linear region is around
22 dB , as shown in Fig. 4(a). It delivers a 1-dB gain compression We integrated two commercial antennas for future work, together withfour radio frequency (RF) ports. Thus, there are a total of six SPDT switchesto control all the RF inputs/outputs of FRECON. The manufactured SPDT and SP4T switches in the FEM yielded around
38 dB and
30 dB isolation, respectively. owerSplitter(1-to-8)
LO out 1
ClkDist
LO In
PAPAPAPAPAPAPAPA LO out 2LO out 3LO out 4LO out 5LO out 6LO out 7LO out 8
Fig. 5. Block diagram of ClkDist (left) and its photograph with a common . -LO (right). Beam 1 Beam 2Beam 3 Beam 4
Fig. 6. Photograph of 64-antenna array (left) and the beam patterns of 16subarrays (right). point (P1dB) of
18 dBm . The measured Rx gain is shown inFig. 4(b). Its maximum gain is . at .
95 GHz . Also, thepower consumption of the implemented FRECON and FEM is . and , respectively. B. ClkDist
The block diagram of the ClkDist and its photograph areshown in Fig. 5. The ClkDist has 1 input and 8 output ports con-necting with the LO and FRECONs, respectively. Since the 1-to-8 power splitter causes a power loss of more than
10 dB , eachpath in the ClkDist is equipped with one DA (HMC383LC4) tomeet the input-power requirement of mixers in FRECON. Thepower consumption of the fabricated ClkDist and the LO is and , respectively . C. Antenna Array
The planar 64-antenna array, consisting of 16 subarrays, isdesigned on a three-layer PCB using two stacked RO4350Bsubstrates. Each subarray consists of × patch antennas witha butler matrix, capable of forming four directional beams.The antenna-element spacing in the subarray is half a wave-length ( λ/ ), i.e., . . The spacing between each subarrayis λ , i.e.,
22 mm . The peak gain of single subarray and 16subarrays, respectively, is . and . . Fig. 6 shows a LOs operating at mmWave frequencies is very sensitive to temperatures.Thus, we adopt a cooling fan to operate our . -LO. Its powerconsumption is added in the LO’s power consumption. photograph of the manufactured 64-antenna array and the beampatterns of 16 subarrays. D. TDD and Antenna Switching
Both FRECON and FEM have an interface, respectively, toreceive DC ( . ) control signals from the digital subsystem,which are connected with the GPIO port. Since the GPIOport plugs in to an FPGA embedded in each SDR, the DCsignal is controllable according to designed blocks in the digitalsubsystem. For the TDD and antenna switching, we implementcontrol units in the digital subsystem, based on the framestructure and baseband functionalities of LuMaMi testbed [4].Since the LuMaMi testbed operates in TDD mode, its controlsignal in the digital subsystem can be exploited to deliverto the analog subsystem. Using a regular beam sweeping inthe Rx mode, channel estimation block computes the channelmagnitudes, and returns the antenna index of the highest channelmagnitude to the antenna-selection control unit. The systemparameters for the developed testbed is summarized in Table I. TABLE IH
IGH - LEVEL SYSTEM PARAMETERS
Parameter Value
Carrier frequency .
95 GHz
Intermediate frequency .
45 GHz
Sampling frequency .
72 MHz
Signal bandwidth
20 MHz
FFT size 2048Antenna-array configuration 64 elementsNumber of TRx chains 16P1dB of each TRx chain
18 dBm
Peak gain of 16 subarrays . IV. I
NITIAL R ESULTS
This section provides initial results on the link-budget cal-culation through over-the-air (OTA) testing. Also, we performan indoor uplink transmission with 16 TRx-chain BS and twosingle-antenna UEs to validate our testbed design.For the link-budget calculation, we used one FRECON andone FEM for transmission and reception, respectively. To clar-ify the Tx and Rx power of an IF signal, a signal genera-tor (E8257D) is connected to the FRECON input of the Tx side,and a spectrum analyzer (FSU50) to the FRECON output of theRx side. Based on this setup, we calculate a measured FSPL forthe distance d between Tx and Rx antennas, and compare withits theoretical number. Effective isotropic radiated power (EIRP)is the hypothetical power radiated by a isotropic Tx antenna inthe strongest direction and defined as EIRP (dBm) = P tx + G ctx − L ctx + G atx (1)where P tx is the Tx power (IF input), G ctx the effective Tx gainof FRECON and FEM, L ctx the cable loss in the Tx side, and G atx the Tx antenna gain. Using the EIRP, the measured FSPLis PL m (dB) = EIRP − ( P rx − G crx + L crx − G arx ) (2) S D e s k Desk U E U E Wall
Desk . m Fig. 7. Indoor measurement setup in a lab including the positions of the BSand two UEs. There is no obstruction between the BS and the UEs.Fig. 8. Uplink constellations for the indoor experiment when using zero-forcing equalizer at the BS. The UE1/UE2 transmit QPSK and 16-QAM,respectively. where P rx is the Rx power (IF output), G crx the effective Rxgain of FRECON and FEM, L crx the cable loss in the Rx side,and G arx the Rx antenna gain. The theoretical FSPL is PL th (dB) = 20log (cid:32) πdfc (cid:33) − G atx − G arx (3)where f is the carrier frequency, and c is the speed of light.From (1)-(3), the link-budget calculation results are shown inTable II. It is observed that the measured FSPLs are quite closeto the theoretical ones. TABLE IIL
INK - BUDGET CALCULATION RESULTS d ( m ) PL th ( dB ) PL m ( dB ) Gap( | PL th − PL m | ) For the indoor test , we used a 16-TRx fully digital beam-forming BS and two single-antenna UEs. The uplink transmis-sion was performed in line-of-sight-like conditions. Fig. 7 showsthe indoor measurement setup including the positions of the BS and UEs. The distance between the BS and the co-located UEswas . . We observed very clear UL constellations. Fig. 8shows captured constellations of the received uplink QPSK (UE1) and 16-QAM (UE2), where a zero-forcing equalizer is usedat the BS. V. C ONCLUSION
Both academia and industry have been making efforts inmeeting 5G requirements. To support 5G, massive MIMO andmmWave have each shown strength and potential. Furthermore,it has been known that they are inseparably connected. RealizingmmWave massive MIMO in practice, however, is still an im-portant issue that must be solved. As a viable solution, we havebuilt the real-time
28 GHz massive MIMO testbed with a hybridbeamforming architecture. In this paper, we have provided anoverview of our real-time mmWave (
28 GHz ) massive MIMOtestbed, with a hybrid beamforming architecture based on beam-selection, and initial results through OTA testing.A
CKNOWLEDGMENT
The authors would like to thank Chris Clifton and KamalK. Samanta at Sony Semiconductor, UK for useful discussionsand for providing the FEM. In addition, this work is carried outwithin the Strategic Innovation Program “Smartare Elektron-iksystem”, a joint venture of Vinnova, Formas and the SwedishEnergy Agency (2018-01534).R
EFERENCES [1] T. Marzetta, “Noncooperative cellular wireless with unlimited numbersof base station antennas,”
IEEE Trans. Wireless Comm. , vol. 9, no. 11,pp. 3590–3600, Nov. 2010.[2] F. Rusek, D. Persson, B. K. Lau, E. G. Larsson, T. L. Marzetta,O. Edfors, and F. Tufvesson, “Scaling up MIMO: Opportunities andchallenges with very large arrays,”
IEEE Sig. Proc. Mag. , vol. 30, no. 1,pp. 40–60, Jan. 2013.[3] C. Shepard, H. Yu, N. Anand, E. Li, Y. R. Marzetta, T., and L. Zhong,“Argos: Practical many-antenna base stations,”
Proc. of ACM MobiCom ,pp. 53–64, Aug. 2012.[4] S. Malkowsky, J. Vieira, L. Liu, P. Harris, K. Nieman, N. Kundargi,I. Wong, F. Tufvesson, V. ¨Owall, and O. Edfors, “The world firstreal-time testbed for massive MIMO: Design, implementation, andvalidation,”
IEEE Access , vol. 5, pp. 9073–9088, 2017.[5] T. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y. Azar, K. Wang, G. N.Wong, J. K. Schulz, M. Samimi, and F. Gutierrez, “Millimeter wavemobile communications for 5G cellular: It will work!”
IEEE Access ,vol. 1, pp. 335–349, 2013.[6] W. Roh, J. Y. Seol, J. Park, B. Lee, J. Lee, Y. Kim, J. Cho, K. Cheun, andF. Aryanfar, “Millimeter-wave beamforming as an enabling technologyfor 5G cellular communications: Theoretical feasibility and prototyperesults,”
IEEE Comm. Mag. , vol. 52, no. 2, pp. 106–113, Feb. 2014.[7] S. Han, I. Chih-Lin, Z. Xu, and C. Rowell, “Large-scale antenna systemswith hybrid analog and digital beamforming for millimeter wave 5G,”
IEEE Comm. Mag. , vol. 53, no. 1, pp. 186–194, Jan. 2015.[8] M. Chung, L. Liu, O. Edfors, and F. Tufvesson, “Millimeter-wavemassive MIMO testbed with hybrid beamforming,”
Proc. IEEE WirelessComm. and Net. Conf. Workshops (WCNCW) , pp. 1–2, Apr. 2020.[9] M. Shafi, J. Zhang, H. Tataria, A. F. Molisch, S. Sun, T. S. Rappaport,F. Tufvesson, S. Wu, and K. Kitao, “Microwave vs. millimeter-wavepropagation channels: Key differences and impact on 5G cellular sys-tems,”
IEEE Comm. Mag. , vol. 56, no. 12, pp. 14–20, Dec. 2018.[10] M. Chung, L. Liu, and O. Edfors, “Phase-noise compensa-tion for OFDM systems exploiting coherence bandwidth: Mod-eling, algorithms, and analysis,” Jul. 2020, [Online] Available:https://arxiv.org/pdf/2007.09628.pdf.[11] M. Chung, H. Prabhu, F. Sheikh, O. Edfors, and L. Liu, “Low-complexity fully-digital phase noise suppression for millimeter-wavesystems,”