Over-the-Air Testing for Dual-Polarized MIMO Devices: Setup Design and Validation

Yong Li,Junli Li,Xiang Zhang,Guiming Wei

1 Key Laboratory of Universal Wireless Communications(Ministry of Education),Beijing University of Posts and Telecommunications,Beijing 100876,China

2 China Academy of Information and Communications Technology,Beijing 100191,China

Abstract: Over-the-air (OTA) testing is considered as the only feasible solution to evaluate radio performances of the fifth-generation (5G) wireless devices which feature two important technologies,i.e.,massive multiple-input multiple-output (MIMO) and millimeter-wave (mmWave).The multi-probe anechoic chamber (MPAC) based OTA setup is able to emulate realistic multipath propagation conditions in a controlled manner.This paper investigates an MPAC OTA setup which is capable of evaluating the performances of 5G base stations as the devices-under-test(DUTs) which are equipped with dual-polarized antennas.Both end-to-end setup and probe configuration for the considered MPAC setup will be elaborated.Furthermore,since building a practical MPAC setup is expensive,time-consuming,and error-prone,an endto-end software testbed is established for validation purpose to avoid technical risks before finalizing an MPAC setup.The architecture of the testbed is presented,which can emulate both the channel profiles perceived by the DUT and the physical-layer behaviors of the considered link conforming to 5G new radio(NR)specifications.Results show that the performances under the emulated channel agree well with those under the target channel,validating the accuracy and effectiveness of the MPAC method.

Keywords: anechoic chamber; dual-polarization;multi-probe anechoic chamber (MPAC); over-the-air(OTA);performance evaluation

I.INTRODUCTION

Together with the evolutions of wireless communication technologies,the radio measurement technology has undergone substantial advancement to cater for the technical features of different generations of wireless systems.The invention of channel emulator made it possible to implement radio measurement campaigns in a controlled and repeatable manner in laboratory environment,which acts as the powerful enabler for the so-called conducted testing.Physical cables are electrically connected between the channel emulator and radio frequency (RF) ports of the device-undertest (DUT) such that the signals transmitted or received by the DUT can experience the desired channel states.Conducted testing has played an important role in measurement campaigns used to assess base stations(BSs)and user equipments(UEs)of the third-(3G)and fourth-generation(4G)wireless systems.

For the emerging fifth-generation (5G) system,two distinguishing technologies,namely,massive multiple-input multiple-output (MIMO) and millimeter-wave (mmWave),are posing great challenges for conducted testing [1].On one hand,the massive MIMO antenna can be composed of up to hundreds of antenna elements; as the RF port of each antenna element needs to be connected to one port of the channel emulator,the required number of channel emulator ports would be rather large,making the setup of conducted testing prohibitively expensive.On the other hand,the DUT needs to be disassembled to access its internal RF ports for conducted testing,but the mmWave radio devices are expected to have integrated antenna and RF components,which cannot provide accessible RF ports.

To address the aforementioned challenges,over-theair(OTA)testing has been deemed as the only feasible solution in 5G era [2,3].The paradigm of OTA testing shifts from the conducted domain to the radiated domain,which uses the antennas of the DUT as the interface to testing signals and therefore the DUT needs not to be disassembled or modified compared with the case of conducted testing.It is thus via the antennas of the DUT instead of its RF ports that the DUT transmits or receives testing signals,as if the DUT were operating in the normal conditions.

Among different alternatives of OTA testing methods,the multi-probe anechoic chamber (MPAC)method is considered as the most feasible one for 5G systems due to its ability of precise control over the spatial profiles of the emulated radio channel,and has attracted intensive research interests from both academia and industry in recent years [4–8].In the MPAC setup,the target channel model can be synthesized by using multiple probe antennas mounted inside the anechoic chamber with proper configurations of the locations and weights of the probes[9–12].Two different methods are applicable under the MPAC setup,including the prefaded signal synthesis (PFS) and the plane-wave synthesis (PWS)method [13,14].Both methods are compatible with the well-established geometry-based stochastic channel(GBSC)models[15–19].As extra complexity and efforts are needed for the PWS method for phase calibration,the PFS method will be focused in our work.

Distinguished from the related literature which typically considered an UE as the DUT,our work in this paper addresses an MPAC setup in which a 5G BS equipped with massive MIMO acts as the DUT,and the downlink performance is focused.The MPAC setup under consideration consists of a 5G BS as the DUT,an anechoic chamber,a number of OTA probes,a channel emulator,and an UE emulator.Compared with the setup for the conducted testing where only the channel emulator and the DUT are involved,building an MPAC based OTA testing setup needs to integrate the functionalities of both the channel emulator and the anechoic chamber,both of which are expensive.Furthermore,both locations and weights of OTA probes should be sophisticatedly configured to ensure the accuracy of the emulated channel;specifically,far more probes will be equipped in the anechoic chamber as the state-of-the-art MPAC setup has evolved from two-dimensional (2D) probe configuration to threedimensional (3D) probe configuration to account for channel dispersions in the elevation domain as well as in the azimuth domain,making probe configuration a huge challenge[20].Another practical challenge lies in the trend that dual-polarized antennas will be extensively employed for both BSs and UEs in 5G systems to achieve low-profile footprint and polarization diversity[21,22],which requires that channel characteristics in terms of polarimetric dimension should also be accurately emulated in the MPAC setup.However,the MPAC OTA setup for dual-polarization has not been elaborated in the open literature.

The major contributions of this paper include:

? Given the fact that the existing literature typically considered single-polarized DUTs1,in this paper,an end-to-end MPAC setup supporting dualpolarized DUTs is elaborated,and the configurations of internal signal paths within the channel emulator are detailed.

? Probe weighting problem is investigated for the MPAC setup supporting dual-polarized DUTs,with emphasis on the impact of antenna field pattern and XPR on probe weighting.

? For the first time,to the best of our knowledge,this paper presents an end-to-end software testbed to validate the MPAC setup,conforming to 5G new radio(NR)specifications.Simulation results under the target channel and the emulated channel are shown to agree well,validating the accuracy and the effectiveness of the MPAC setup.

The rest of this paper is organized as follows.Section II presents channel models of the target channel and the emulated channel under the MPAC setup,with particular emphasis on the configuration of the channel emulator and probe antennas for realistic dualpolarized antennas.The architecture and the work flow of the end-to-end software testbed are presented in Section III.Validation results are provided and discussed in Section IV.Finally,Section V concludes this paper.

Notations: Vectors and matrices are represented by boldface lowercase and uppercase letters,respectively.(·)?and (·)Tdenote the conjugate operator and the transpose operator,respectively.∥·∥denotes the Euclidean norm.〈·,·〉denotes the inner product operator.

II.MPAC SETUP FOR DUAL-POLARIZED DUTS

2.1 Target Channel Model

The GBSC model has the advantages of clear physical interpretation and separation of propagation parameters and antenna effects,which has been widely adopted to create controlled propagation environment for the purpose of performance evaluation.Assuming a 3D geometry,the spatial structure of a GBSC model is illustrated in Figure 1.The channel between the transmitter(Tx)and the receiver(Rx)in the GBSC model consists of a plurality of temporally dispersed clusters,each of which corresponds to one path in the realistic propagation channel.Further,each cluster is composed ofMsubpaths with uniform power,and each subpath represents scattering on a single scatter in a cluster.Let ?Tx,n,mdenote the spherical unit vector corresponding to the departure direction of themth subpath within thenth cluster on the Tx side,which has azimuth angle of departure (AoD)?n,m,AoDand zenith angle of departure(ZoD)θn,m,ZoD;likewise,let?Rx,n,mdenote the spherical unit vector corresponding to the arrival direction of themth subpath within thenth cluster on the Rx side,which has azimuth angle of arrival (AoA)?n,m,AoAand zenith angle of arrival(ZoA)θn,m,ZoA.

For a MIMO system withSantenna elements on the Tx side andUantenna elements on the Rx side,the channel coefficients of thenth cluster between thesth Tx antenna element and theuth Rx antenna element can be given by

where,tdenotes the time,Pnis the average power of thenth cluster.FVTx(?Tx,n,m) andFHTx(?Tx,n,m)are the antenna field patterns of the Tx antenna element with respect to the direction ?Tx,n,mfor vertical and horizontal polarizations,respectively; likewise,FVRx(?Rx,n,m) andFHRx(?Rx,n,m) are the antenna field patterns of the Rx antenna element with respect to the direction ?Rx,n,mfor vertical and horizontal polarizations,respectively.Note that

where,GTx(?Tx,n,m) andGRx(?Rx,n,m) denote the antenna power gain for the direction ?Tx,n,mon the Tx side and the direction ?Rx,n,mon the Rx side,respectively.We assume identical antenna field pattern for either Tx antenna elements or Rx antenna elements in this paper; furthermore,realistic antenna pattern is considered for both Tx and Rx antenna elements rather than isotropic antenna pattern.An,mdenotes the polarization matrix,which can be written as

By decomposing(1)into four terms,we can acquire more insights abouthu,s,n(t) that the dual-polarized channel is composed of two co-polarized terms,i.e.,and two cross-polarized terms,i.e.,as given by

where,

Since the DUT operates on the Tx side in our work,the Tx spatial correlation would be an important performance metric for channel emulation.For thenth cluster,the spatial correlation function between two different Tx antenna elements under the target channel can be given by

As channel coefficients on different polarizations are considered uncorrelated,(3)can be simplified as

where,

2.2 MPAC Setup for Dual-Polarized DUTs

The diagram of the MPAC setup for dual-polarized DUTs is shown in Figure 2.The MPAC setup under consideration is used to emulate a MIMO system withSTx antenna elements andURx antenna elements using the PFS method.The 5G BS DUT is placed within the so-called test area in the anechoic chamber.Testing signals are generated by the DUT and transmitted over-the-air within the anechoic chamber.Assuming thatKdual-polarized probes,each of which consists of two co-located orthogonally polarized antennas,are installed in a 3D geometry within the anechoic chamber,each of theSDUT Tx antenna elements will transmit signals to theKdual-polarized probes.Due to ideal propagation conditions without any scattering,reflection or refraction,no power leakage between polarizations is assumed within the anechoic chamber.Therefore,the signal impinging on the vertically (respectively,horizontally) polarized antenna of thekth probe should be received from the DUT with antenna gain for the vertical(respectively,horizontal)polarizationFVTx(Φk)[respectively,FHTx(Φk)],where Φkdenotes the spherical unit vector of thekth probe with respect to the DUT.Furthermore,the inner product〈dTx,s,Φk〉represents the path difference of thesth Tx antenna to thekth probe due to Tx antenna geometry.

The free-space propagation pathloss between Tx antennas and OTA probes within the anechoic chamber is neglected,and power variation from different Tx antenna elements to any arbitrary OTA probe is also omitted.Therefore,the time-invariant channel transfer function from thesth Tx antenna element to thekth dual-polarized probe can be given by

The signal received by both the vertically-polarized and horizontally-polarized antennas of theKdualpolarized probes will then be independently fed into theKpairs of input ports of the channel emulator via wired cables.It is north noting thatKpairs of ports,i.e.,2Kports,should be available at the input side of the channel emulator when compared to the case ofKsingle-polarized probes which requires onlyKports of the channel emulator.Channel coefficients are generated inside the channel emulator and convolved with the input signals.The coefficients of the internal signal path from thekth pair of input ports to theuth output port inside the channel emulator can be configured as

where,

andwVn,kandwHn,kdenote the real-valued power weights for the vertical and the horizontal paths of thekth probe,respectively,in the PFS method and

By expanding(6),we have

2.3 Emulated Channel Model

By combing the contributions superimposed over-theair from allKdual-polarized probes,the time-variant end-to-end channel coefficient of thenth cluster between thesth Tx antenna element and theuth Rx antenna element in the dual-polarized MPAC setup can be given by

where,

The spatial correlation function between two different Tx antenna elements under the emulated channel can be given by

where,

2.4 Probe Weighting for Dual-Polarized DUTs

As both the objective function and the constraints are separable,(11) can be decomposed into the following two subproblems:

and

where,wVn?[wVn,1,··· ,wVn,K]Tin (12),andwHn?[wHn,1,··· ,wHn,K]Tin (13).It is noted that both (12)and (13) are quadratic programming problems with linear constraints,which can be efficiently solved by convex optimization techniques.Finally,it is worth noting that bothwVnandwHnshould be further multiplied by the power of thenth cluster specified for the target channel model,whenwVnandwHnare applied in the MPAC setup.

III.ARCHITECTURE OF END-TO-END MPAC TESTBED

It is apparent that the MPAC based OTA setup is far more complicated and thus more error-prone than the conducted testing.OTA testing campaigns can be launched only when the overall setup is settled after long preparation time for the integration of a large pool of hardware equipments and the configuration of a lot of system parameters; once any system performance cannot meet the testing requirement,the whole procedures have to start over,leading to undesired or even unaffordable costs.Therefore,the design of the MPAC setup is faced with huge technical risks and economic costs.In order to avoid the risks in designing an expensive and error-prone MPAC setup,software-based validation can be conducted before a practical MPAC setup is established to verify the correctness of its configurations.

To validate the MPAC setup,the essential idea is to recreate the same signal paths in the software testbed as those in the real-world MPAC setup,which mainly includes:1)the signal paths inside the anechoic chamber,2) the signal paths inside the channel emulator,and 3)the configuration of OTA probes.Furthermore,to emulate realistic end-to-end performance testing,the software testbed should be able to model 5G physical layer behaviors in addition to channel-related parameters such that performance metrics can be obtained as benchmark references for practical OTA testing.

3.1 Testbed for the Target Channel

The processing flow of the software testbed for the target channel is shown in Figure 3(a).The baseband signal processing procedures follow the 5G NR specifications[23,24],which employs orthogonal frequency division multiplexing(OFDM)as the underlying multiplexing technique.The architecture of the testbed consists of both the BS as the Tx side and the UE as the Rx side,and the channel module between them.

On the Tx side,one or two transport blocks (TBs)consisting of random bits are generated by the bit source module,according to the number of spatial layers exploited in the given slot,which is the basic unit for data transmission in the 5G NR specifications.The overall baseband signal processing chain consists of the bit-level processing module,the symbol-level processing module,and the MIMO precoding and OFDM modulation module.The bit-level processing module first appends cyclic redundancy check (CRC) bits to the TB(s)for the purpose of error detection which will later be performed on the Rx side,and handles channel coding related processing,such as code block segmentation,low-density parity check (LDPC) encoding,and rate matching [24].Up to two codewords(CWs)will be forwarded to the next module at the output of the bit-level processing module.

Following this,the symbol-level processing module will take care of scrambling,modulation,layer mapping,and resource mapping[23];particularly,the demodulation reference signals(DMRSs)are inserted at appropriate resource elements(REs)for the purpose of channel estimation in the procedure of resource mapping.The symbol-level processing module will output up to 8 layers of spatially multiplexed I/Q symbols.

A third module will first perform digital precoding,which maps the multiple spatial layers onto a number of logical antenna ports,and then OFDM-modulate the digitally-precoded I/Q symbols and concatenate the cyclic prefix(CP)for every OFDM symbol on every logical antenna ports individually.The OFDMmodulated signal is a sequence of data samples represented in the time-domain,which will then undergo analogue beamforming by a simple operation of phase shifter and finally be mapped ontoSTx physical antennas for transmission.The transmitted time-domain signals are convolved by the channel coefficients randomly generated by the channel module,which is implemented according to the cluster delay line (CDL)model specified in [17].Compared with the time delay line (TDL) model also defined in [17],the CDL model is able to characterize the spatial profiles,such as incoming/outgoing directions and angular spreads,of the channel.

A third module will first perform MIMO precoding,consisting of both digital beamforming and analogue beamforming sequentially,which maps the multiple spatial layers ontoSTx antennas,and then OFDMmodulate the precoded I/Q symbols and concatenate the cyclic prefix(CP)for every OFDM symbol on every Tx antenna individually.The OFDM-modulated signal is a sequence of data samples represented in the time-domain,which will then be convolved by the channel coefficients randomly generated by the channel module,which is implemented according to the cluster delay line (CDL) model specified in [17].Compared with the time delay line(TDL)model also defined in[17],the CDL model is able to characterize the spatial profiles,such as incoming/outgoing directions and angular spreads,of the channel.

The baseband signal processing at the Rx side is basically the inverse procedures of those on the Tx side,which takes charge of signal processing for recovering the original signals.Finally,the recovered TB(s)on the Rx side will be compared to the original TB(s)transmitted from the Tx side to acquire link performances such as block error ratio(BLER).System parameters,such as modulation order and coding rate,can be configured for the above signal processing flow to obtain system performances under various transmission modes.

3.2 Testbed for the Emulated Channel

As shown in Figure 3(b),the software testbed for the emulated channel differs from that for the target channel in that the target CDL channel module is substituted by two modules,namely,the anechoic chamber module and the channel emulator module,with remaining modules keeping the same.Basically,the anechoic chamber module is used to emulate the transfer function between theSTx antenna elements and theKdual-polarized probes,which is dependent on the geometries of both Tx antennas and OTA probes in the anechoic chamber and is time-invariant.Only phase rotations due to propagation delay are taken into account,neglecting the free-space propagation pathloss between Tx antennas and OTA probes.The channel emulator module is used to emulate the internal signal paths from theKpairs of input ports,each of which is connected to one dual-polarized probe,to theUoutput ports of the channel emulator,each of which is connected to one Rx antenna element.The time-variant coefficients of the internal paths have been given by(6)in Section II.Furthermore,a probe configuration module is employed in the testbed for the emulated channel to configure power weights of both polarization paths for different probes to generate the desired spatial profile according to (12) and (13).To our best knowledge,the above work is the first reported effort in the open literature to address the design of an end-to-end softwared testbed for the MPAC setup.

The following procedures are performed when running the testbed for the emulated channel:

1.Select the parameters of the channel model to be emulated,such as temporal disperse,departure angles,arriving angles,etc.

2.Define the anechoic chamber such as arrangement of the probes,and obtain the transfer function within the anechoic chamber according to(5).

3.Derive the probe configurations according to(12)and(13).

4.Create channel impulse responses for the channel emulator according to(6).

5.Execute the testbed and collect performance results.

IV.SIMULATIONS AND DISCUSSIONS

In this section,simulation results are provided to validate the proposed MPAC setup.Two sets of simulation campaigns are conducted;one set is based on the target CDL model used to characterize the multipath propagation conditions of the realistic channel,and the other one is based on the emulated channel model combining the effects of both anechoic chamber and channel emulator.More specifically,the target CDL model is composed ofN=24 clusters,and the nominal AoD,nominal ZoD,nominal AoA,and nominal ZoA of theNclusters are specified by the CDL-C model in[17]; each cluster consists ofM=20 subpaths,and the cluster-wise azimuth spread of departure angles(cASD),cluster-wise zenith spread of departure angles(cZSD),cluster-wise azimuth spread of arrival angles(cASA),cluster-wise zenith spread of arrival angles(cZSA) are initially given bycASD=2o,cZSD=3o,cASA=15o,andcZSA=7o,respectively.It is north noting that the Tx side,corresponding to the BS,has narrower angular spreads in both azimuth and elevation domains,due to less scattering surroundings.The XPRs between both polarizations are considered identical,and thus we assumeκV H=κHV=7dB.Given the sparse and specular channel characteristics of the 5G mmWave frequency bands as well as the large dimension of DUT footprint,a sectored MPAC setup is adopted to reduce the required size of anechoic chamber and the required number of probe antennas to allow a more cost-efficient OTA solution[4].Unlike the conventional MPAC OTA setup of the 2D or 3D probe ring structure,the 5G BS DUT is placed at the edge of the anechoic chamber to fully make use of the space of the chamber.Kdual-polarized probes are installed on a probe wall with adjustable locations,whose locations are determined by sophisticated probe selection algorithm which takes into account the angular positions of dominant clusters and rays[10],and the weights of the probes are derived by(12) and (13) discussed in Section II.Specifically,as illustrated in Figure 3(b),the block ofProbe Configurationwill first be implemented to decide probe locations and weights;given the cost constraint of a practical MPAC setup,20 dual-polarized candidate probes are assumed to locate in the direction of the dominant cluster,and convex optimization technique is iteratively employed to decide the best locations ofK=5 orK=8 dual-polarized probes and their respective power weights which can minimize the deviation of spatial correlation between the target channel and the emulated channel.Then,probe configuration will be fed into the block ofChannel Emulatorto generate the sequences of channel coefficients which can synthesize the desired channel spatial profile.The signal sent by the 5G NR-compliant Tx chain will experience ideal propagation environment emulated by theAnechoic Chamberblock and faded propagation environment emulated by theChannel Emulatorblock successively before being processed by the 5G NRcompliant Rx chain to obtain end-to-end BLER performance.

The practical sectored MPAC setup implemented in our work is illustrated in Figure 4.A 5G MIMO BS is considered as the DUT in our work,which has 32 Tx antenna elements.On the other side of the link,the UE emulator has 2 pairs of co-located and dual-polarized(±45opolarized)Rx antenna elements.Therefore,thechannel emulator employed in our work is configured with 2Kinput ports and 4 output ports.

The proposed testbed can serve as an efficient tool to validate the design of the overall MPAC setup.The channel emulation accuracy can be obtained by comparing end-to-end performances under both the target channel and the emulated channel.Apparently,channel emulation with higher accuracy is expected to yield closer performance results.BLER is employed as the performance metric in our experiment for a given set of transmission formats in term of modulation and coding,rank number,and precoding matrix,which are signalled in 5G NR system by modulation and coding scheme(MCS),rank index(RI)and precoding matrix index (PMI),respectively.It is noted that link adaptation is not enabled in the validation stage because it would complicate performance comparison due to varying transmission formats.The link parameters used in the experiments are summarized in Table 1.2

Table 1.Link parameters used in the testbed.

To evaluate the accuracy of the reconstructed channel in the MPAC setup,the SNR-BLER curves are compared under the emulated channel and the target channel.The impact of the number of probe antennas on channel emulation accuracy is investigated.As shown in Figure 5,the SNR-BLER curves between the emulated channel and the target channel have a marginal gap of about 0.1dB for BLER=0.1 when there are 5 dual-polarized probe antennas.When the number of probe antennas is further increased to 8,the performance gap is shrinked negligibly when compared to the case of 5 probe antennas.It is thus obvious that more probe antennas can yield higher channel emulation accuracy,however,at the costs of more expensive system costs,as increasing probe antennas implies increasing the configurations of the expensive channel emulator.Therefore,the proposed testbed can be used to derive a reasonable tradeoff between channel emulation accuracy and system hardware costs.3Accordingly,the spatial correlations under different numbers of probes are plotted in Figure 6.It is expected that the spatial correlation declines when antenna spacing is increased.Particularly,when probe number is 5,the emulated spatial correlation can overlap with the target spatial correlation with only marginal gap; when probe number is 8,the two curves almost overlap completely.It can be observed that the trends in both Figure 5 and Figure 6 can agree very well.

The SNR-BLER curves under different MCSs are plotted in Figure 7.Three MCSs with different modulation orders and coding rates are considered under an MPAC configuration of 5 dual-polarized probes,i.e.,MCS#0,MCS#3 and MCS#14.Marginal gaps between the two sets of curves can be found,which are 0.1dB,0.15dB and 0.2dB for MCS#0,MCS#3 and MCS#14,respectively.This can confirm that an MPAC OTA setup configured with five probes is capable of reconstructing the target channel with satisfactory accuracy.

Finally,as the main task of the MPAC setup is to reproduce the spatial characteristics of the radio channel,which are largely affected by angular spreads of the scattering environment,the accuracy of channel emulation under different angular spreads is verified as well.It is shown in Figure 8 that,as expected,more scattering channel can deteriorate link performance due to degraded analog beamforming gain; it is also shown that,by varying the azimuth spread of departure anglescASD,the SNR-BLER curves all have small gaps of 0.1dB,0.2dB and 0.15dB for BLER=0.1,respectively,for the cases ofcASD=2o,cASD=5o,andcASD=10o,under the MPAC setup configured with five dual-polarized probes.

The above comparison exhibits that the channel characteristics can be accurately recreated in the proposed MPAC setup and therefore similar link performances can be obtained under both channels.This observation can validate the effectiveness of the OTA solution.

V.CONCLUSION

As radio devices with dual-polarized antennas will be widely employed in the emerging 5G wireless system,the MPAC-based OTA testing setup for DUTs with dual-polarized antennas is investigated in this paper.The end-to-end setup of the considered MPAC solution is elaborated,with emphasises on the configurations of both channel emulator and probe antennas for dual-polarized DUTs.Furthermore,a software testbed to validate the MPAC setup is proposed which can compare link performances under the target channel and the emulated channel.Simulation results suggest that the target channel can be accurately reconstructed in the proposed MPAC setup,thus validating the effectiveness of the OTA solution.

ACKNOWLEDGMENT

This work was supported by the National Natural Science Foundation of China under Grant.61971067.

NOTES

1Although the dual-polarized channel consisting of both vertical and horizontal polarizations has already been treated in practical 4G MIMO OTA testing,little public literature has elaborated the MPAC modelling of dual-polarized channel,for example,how will the cross-polarization radio(XPR)affect probe weighting.

2To guarantee correct implementation of physical layer processing modules on both Tx and Rx sides,link performances under the target channel are compared with relevant 3GPP documents with good consistency.

3As already illustrated in Figure.4,five dual-polarized probe antennas are used in our practical MPAC setup to achieve a reasonable channel emulation accuracy at affordable costs.