Spatial Reciprocity of Uplink and Downlink Radio Channels in FDD Systems

Transcription

1 EUROPEAN COOPERATION IN THE FIELD OF SCIENTIFIC AND TECHNICAL RESEARCH EURO-COST SOURCE: INTHF, Technische Universität Wien, Vienna, Austria Nokia Research Center, Helsinki, Finland COST 73 TD() 66 Espoo, Finland /May/3-31 Spatial Reciprocity o Uplink and Downlink Radio Channels in FDD Systems Klaus Hugl, Kimmo Kalliola and Juha Laurila Klaus Hugl Institut ür Nachrichtentechnik Technische Universität Wien Gußhausstraße 5/389 A-14 Vienna AUSTRIA Klaus.Hugl@mobile.nt.tuwien.ac.at Kimmo Kalliola, Juha Laurila Nokia Research Center P.O.Box 47 FIN-45 NOKIA Group FINLAND Kimmo.Kalliola@nokia.com, Juha.K.Laurila@nokia.com

2 Spatial Reciprocity o Uplink and Downlink Radio Channels in FDD Systems Klaus Hugl 1, Kimmo Kalliola and Juha Laurila 1 Institut ür Nachrichtentechnik und Hochrequenztechnik (INTHF) Technische Universität Wien, Vienna, Austria Klaus.Hugl@mobile.nt.tuwien.ac.at Nokia Research Center, Helsinki, Finland Abstract - We investigate the congruence o the directional properties o uplink () and downlink () as seen rom the base station in WCDMA o UMTS by means o spatial channel measurements. Our study illustrates that the directional properties o the mobile radio channel are strongly correlated in uplink and downlink radio channels. Thereore, the utilization o spatial inormation derived during uplink reception or downlink smart antenna transmission is reasonable. 1. INTRODUCTION The application o adaptive antennas in cellular mobile communication systems has raised increased interest during recent years [1]. The introduction o the 3 rd generation (3G) mobile communication systems supporting high-data rate services, e.g. video streaming and multimedia applications will create asymmetric traic in avor o the downlink. Additionally, heavily loaded CDMA systems like UMTS will be downlink intererence limited. As a consequence, the improvements that smart antennas can provide are especially required in the downlink. One way to operate an adaptive antenna base station (BS) without mobile station (MS) eedback in the downlink is to reuse spatial (angular) inormation derived during uplink reception (e.g. []). But a prerequisite or this methodology is that the spatial channel characteristics at the uplink and downlink requencies are similar. Dierent channel sounding campaigns can be ound in literature [3-6] that tried to answer whether this assumption also holds in systems applying requency division duplex (FDD). But the authors o these papers drew controversial conclusions. Thereore, we irst comment on the available literature and then present our measurement results.. STATE OF THE ART The irst investigations o the congruence o uplink and downlink spatial channel characteristics were done by comparing the dominant direction o arrival (DOA) [3]. In the measurement setup the relative duplex distance was limited to <3% (5MHz at 9MHz) which is much less than in W-CDMA ( 1%). Moreover, these results lack statistical reliability because only some stationary transmitter positions were measured. The authors o [3] conclude that the dominant DOA is relatively stable in the considered requency band. Measurements in Aalborg [4] were perormed using a moving transmitter and a duplex distance o 68MHz at a carrier requency o 1.7GHz. The authors averaged the channel over a distance o 4 wavelengths to overcome small-scale ading eects. They applied the SAGE algorithm [7] to extract the APS and correlated the irst central moment o it. The authors concluded that the irst central moment o the APS is very well correlated. We would like to mention that the correlation o the irst central moment o the APS might have one big drawback: several dierent APS shapes give the same center o gravity, as illustrated in Fig. 1. Power Center o Gravity APS #1 APS # Azimuth[ ] Figure 1: The same center o gravity is obtained by two dierent APS (in blue and red color).

3 Moreover, the distribution o the center o gravity in azimuth itsel has a strong inluence on the derived correlation coeicient even i the deviation is the same. I the dominant propagation clusters are located at array broadside (< <3 ) the correlation coeicient will be lower than or o broadside (3 < <9 ). Thus, the center o gravity correlation coeicient is an ambiguous measure or the APS shape correlation in uplink and downlink. Measurements conducted in Bristol [5] were done by perorming wideband measurements and extracting the and requency band by subband iltering. The measurement bandwidth was 1MHz in the GHz band and thereore the maximum possible duplex distance was limited to the measurement bandwidth. Only a ew mobile station transmitter positions were monitored. The author(s) concluded that the APS o uplink and downlink is uncorrelated in cellular mobile communications systems with requency division duplex. We see the ollowing shortcomings o their surprising result. Samples or each transmitter positions were taken within 1s. Averaging over this time period with a constant transmitter and receiver position does not necessarily guarantee independence o small-scale ading. The directions o arrival were extracted using the Unitary Esprit algorithm [8]. These DOAs were then used to create an artiicial antenna array response with a variable number o antenna elements to calculate the APS using the Bartlett beamormer [9]. Strictly speaking, this means the reduction o the spatial inormation to some discrete parameters (namely the discrete DOAs) to create the continuous azimuth power spectrum. In this way, one looses inormation which might have aected the given APS correlation o and. Recently a ollow up measurement campaign was perormed in Bristol [6], now sounding the channel at two carrier requencies perorming dynamic drive measurements. For measurement data evaluation [1] the authors utilized the -D Unitary Esprit algorithm to estimate the DOA and the delays o the dierent multipath components. Moreover, they extracted the complex coeicients (amplitude and phase) o each multipath component. Ater restricting the dynamic range to 15dB, the APS is calculated with a resolution o.5, as the integral in the delay domain o all complex coeicients lying in a.5 wide bin. As a consequence o their APS deinition, the power (complex coeicient) associated with an azimuth bin is strongly dependent on the phase states o delayed multipath components having nearly the same angle o incidence. Thereore, constructive and destructive superposition o the complex phasors in the delay domain aect the calculated APS measure. Finally, they calculated the correlation o their complex valued APS using the angular resolution o.5. As a consequence o their APS deinition using discrete complex valued waves with a.5 grid, the APS seems to be uncorrelated on a irst look although it is not. Let us now investigate where there are still some pitalls. The integration o the complex coeicients in the delay domain or each requency band introduces artiicial ading. These ading coeicients should represent the directional power and will be uncorrelated because o the dierent carrier requencies resulting in APS decorrelation. Thereore, two discrete waves with correctly estimated azimuths 1 and but complex phasors α 1, =α 1, =α, =-α, (1) result in a totally uncorrelated APS as well. The reason or that is only the changing phase due to the dierent carrier requencies also i the power and the azimuth angle o the waves are identical. The resolution o.5 and the discrete wave assumption leads to decorrelation also i the APS rom our point o view is correlated. A single DOA estimate with.5 estimation deviation in and results in a correlation coeicient o ρ=. We also have to keep in mind that the estimation reliability is aected by imperect array and receiver chain calibrations. Thus, it is not surprising that the authors o [6] state the APS to be uncorrelated. We recommend to not use complex coeicients and the discrete wave assumption or APS correlation purpose. In the ollowing sections, we will describe our measurement setup as well as how we estimate the APS and deine its correlation.

4 3. MEASUREMENT SETUP The spatial channel sounding campaign described herein was conducted by Nokia Research Center (NRC) Helsinki in cooperation with INTHF in June 1 in downtown Helsinki. For measurement purpose the NORPPA channel sounder built by Elektrobit AG [11] was applied. In the measurements, a vehicular transmitter mounted on the roo o a measurement van at a height o about.4m above ground level was used. The sounding signal was either transmitted using a vertically polarized monopole or a dummy handset attached to a phantom head. The measurements were simultaneously perormed on an uplink ( =1935MHz) and downlink carrier ( =15MHz) in the UMTS band. A modulated PN-sequence o length 17 and a chip-rate o 5MChips/s was applied as channel sounding signal. We perormed dynamic drive measurements with a target speed o 4km/h in the center o Helsinki along dierent measurement routes. The dynamic drive measurements allow the spatial channel characterization or ull BS sector coverage without the lack o limited statistics. At the receiver/bs position an 8-column uniorm linear array (A) was used, each column consisting o 4 vertically polarized dipoles. Each column o the physical array has a gain o 11dBi and 3dB-beamwidth o 1 in the horizontal (azimuthal) domain and 18 in the vertical plane. The inter-column distance was d=67mm corresponding to.43λ and.475λ. The array was mechanically downtilted by 7. The signals o the 8 antenna columns were collected using the ast RF switching technique o [1]. In the receiver 4-times oversampling was applied resulting in a delay resolution o τ=5ns. A single recorded array channel impulse response h( ) has a dimension o 8 spatial samples (number o antenna columns) versus 58 delay samples h1 (, ) L h1 (57 τ, ) h ( ) = M M. () h (, ) (57, ) 8 L h8 τ A measurement o a ull array channel impulse response at and was perormed every 3.6ms. Thereore, at least 3 channel snapshots per wavelength are available while the transmitter was moving along the measurement routes. The antenna array receiver was placed on six dierent BS positions in the center o Helsinki. In total, more than complex array channel impulse responses along the drive routes o more than 14km were collected. For more details on the measurement setup and the propagation environment (especially photos o the measurement environments and maps including the route description) see [13]. 4. MEASUREMENT EVALUATION In this section we will investigate the correlation o the spatial propagation characteristics o the mobile radio channel at two dierent carrier requencies. In this case, the dierent carrier requencies correspond to a single uplink and downlink carrier o the UTRA-FDD system. The spatial propagation parameters we consider are the dominant direction o arrival (DOA) and the azimuth power spectrum (APS). 4.1 Congruence o dominant / DOA Let us irst consider the congruence or dierence o the instantaneous dominant direction o arrival at the uplink and downlink carrier. We deine the dominant DOA as the direction o maximum output power using the Bartlett beamormer [9] 57 H dom ( ) = arg max a ( ) h( n τ, ), n= (3) where a() and h(n τ) denote the array steering vector o the A and the n-th delay sample o the recorded complex channel impulse response, respectively. The distribution o the dierence o dominant and DOA dom, di (, ) = dom ( ) dom ( ) (4) o all measurement routes in all sectors is given in Fig..

5 . / DOA deviation All Measurements.5 / DOA Deviation Macrocell (Averaged) PDF(/ DOA deviation) PDF(/ DOA deviation) / DOA deviation [ ] Figure : Deviation distribution o the dominant DOA estimated at the and carrier. Obviously, there is a nice congruence o the dominant DOA estimated rom the recorded channel impulse responses at the uplink and downlink carrier. The remaining deviation is also inluenced by spatially selective ading [14]. To mitigate this eect rom the measurement data evaluations, we illustrate the deviation o the dominant DOA at uplink and downlink carrier including small scale averaging over 5 consecutive channel snapshots (corresponding to 18ms) in Fig. 3. PDF(/ DOA deviation) / DOA Deviation All Measurements (Averaged) / DOA deviation [ ] Figure 3: Deviation distribution o the dominant DOA estimated at the and carrier including small scale averaging. Obviously, the small scale averaging reduces the deviation even more. In [15] we have shown, that there is a signiicant dierence in the dominant wave propagation eects or macro- and microcellular BS installations. Thus, we show the / DOA deviation or the BS installations at the Helsingin Energia building rom the rootop (macrocellular) and below the rootop (microcellular) in Fig. 4 and Fig / DOA deviation [ ] Figure 4: Deviation distribution o the dominant DOA estimated at the and carrier including small scale averaging. Only the measurement data o the macrocellular BS installation in Sector 16 o the Helsingin Energia measurements [13] are utilized. PDF(/ DOA deviation) / DOA Deviation Microcell ( Averaged) / DOA deviation [ ] Figure 5: Deviation distribution o the dominant DOA estimated at the and carrier including small scale averaging (microcellular BS installation in Sector 16 o the Helsingin Energia measurements [13]). Again a nice congruence is visible. For the microcellular BS installation at the balcony o the Helsingin Energia building (6.5m below the macrocellular BS installation at the rootop) the shape o the distribution is slightly wider. The reason or that is the larger angular spreading in the microcellular case due to the relatively large aperture o the street canyons that dominate physical wave propagation. For the macrocell the pseudo-los dominates the propagation scenario (shown in [15]) and thereore the small scale averaging is more eective. In general, the dominant DOA in uplink and downlink show only a minor deviation. Thereore, the utilization o the dominant

6 DOA estimated during uplink reception or downlink beampointing purpose is reasonable. 4. Uplink-Downlink APS Correlation Up to now we only investigated the congruence o the dominant DOA estimated at an uplink and downlink carrier in the UMTS band. But the directional behavior o the mobile radio channel contains much more than just a single DOA. Thus, let us in a second step examine the congruence/correlation o the total azimuth power spectrum (APS). In section, we critically discussed the deinitions and results o the same investigations that can be ound in literature [4,5,6]. Here we would like to state how we see and deine the correlation o the azimuth power spectrum o the uplink and downlink mobile radio channel as seen rom a BS antenna array. For estimation purposes o the APS we use the least-squares power estimator (Capon s beamormer or Minimum Variance Estimator [16]) 1 P ( ) = (5) H 1 a ( ) R a( ) applied on the spatial covariance matrix 57 H R = Et h( n τ ) h ( n τ ). (6) n= This Minimum Variance Estimator produces a smooth estimate o the APS. An angular resolution o 1 inside an azimuthal area o [-7,7 ] rom array broadside or the APS P() is considered. The APS is a real and positive valued unction in the azimuth. Thereore, the correlation coeicient ρ o instantaneous APS estimates at the requencies and ρ = P( P ( ) P( ) ) P ( ) (7) is by deinition positive and is only i P ( ) P( ) =. (8) This is a consequence o the non-zero mean value o the APS P ( ). Thereore, we deine the APS correlation mathematically correctly as the covariance o the APS ρ = ( P( ) P( ))( P( ) P( )) ( P( ) P( )) ( P( ) P( )) (9) We do not average the covariance unction over several realizations. Instead we plot the distribution o the absolute value o ρ as shown in Fig. 6. CDF(ρ) 1 / APS Correlation Macrocell Microcell All Measurements ρ Figure 6: Correlation unction o the estimated azimuth power spectrum at the and carrier. Obviously, the correlation between the instantaneously estimated APS realizations at and is rather high. In more than 5% o the cases the correlation is higher than.8 or the macrocellular as well as the microcellular BS installation. Additionally, we plotted the distribution or all measurements taken in June 1. The correlation or the microcellular BS installation is lower than in the macrocellular case. This is again an eect o the street canyon dominated propagation discussed in [15]. The distribution or the macrocellular case described in [13,15] is similar to the distribution o all measurements. The reason is that overall 5 macrocellular but just a single microcellular BS installation was used in this measurement campaign. The distribution o all measurements is dominated or the high correlation values by the macrocellular installations and the lower values are due to the microcellular case. In Fig. 6 instantaneous estimates o the APS are considered. I several dierent angular areas (carrying signiicant power) ade independently, we get decorrelation o the APS due to the uncorrelated ading at the dierent carrier requencies. Thus, we consider the APS estimated out o the smallscale averaged spatial covariance matrix in

7 Fig. 7 (averaging over a time span o 18ms corresponding to 5 consecutive channel snapshots). CDF(ρ) 1 / APS Correlation (Averaging) Macrocell Microcell All Measurements ρ Figure 7: Correlation distribution o the estimated azimuth power spectrum at the and carrier including small scale averaging. The small scale averaging mitigating the instantaneous ading situation at the two dierent carriers improves the correlation even more. In general we can conclude that the APS in and as seen rom a BS antenna array is very similar. 5. SUMMARY AND CONCLUSIONS We have shown by measurement data evaluation that the spatial behavior seen by an adaptive antenna array at the uplink and downlink carrier o W-CDMA is strongly correlated. This is the case also in urban radio environments and despite o large duplex separation o W-CDMA. Thereore, the utilization o spatial inormation derived during uplink reception or downlink beamorming purpose makes sense. As a consequence, adaptive antenna downlink beamorming without terminal eedback is possible also in urban radio environments. REFERENCES [1] A. J. Paulraj, C. B. Papadias, ''Space-Time Processing or Wireless Communications'', IEEE Signal Processing Magazine, No. 6, Vol. 14, pp , November [] A. Kuchar, M. Tangemann, E. Bonek, ''A Real-Time DOA-Based Smart Antenna Processor'', IEEE Transactions on Vehicular Technology, to be published. [3] L. Bigler, H. P. Lin, S. S. Jeng and G. Xu, ''Experimental Direction o Arrival and Spatial Signature Measurements at 9 MHz or Smart Antenna Systems'', Proc. IEEE VTC'95, Vol. 1, pp , Chicago, USA (1995). [4] K. I. Pedersen, P. E. Mogensen, F. Fredriksen, ''Joint-Directional Properties o Uplink and Downlink Channel in Mobile Communications'', Electronic Letters, Vol. 35, No. 16, pp , 5. August [5] B. H. Allen, ''Smart Antennas or High Data Rate FDD Wireless Links'', Dissertation, University o Bristol, Bristol, UK, April 1. [6] S. E. Foo, et. al., ''Frequency Dependency o Spatial-Temporal Characteristics o UMTS FDD links'', COST 73 TD()7, Guildord, UK, January. [7] B. Fleury, D. Dahlhaus, R. Heddergott, M. Tschudin, ''Wideband Angle o Arrival Estimation Using the SAGE Algorithm'', Proc. IEEE ISSSTA'96, pp , Mainz, Germany (1996). [8] M. Haardt, J. A. Nossek, ''Unitary ESPRIT: How to Obtain Increased Estimation Accuracy with a Reduced Computational Burden'', IEEE Transactions on Signal Processing, Vol. 43, No. 5, pp , May [9] M. Bartlett, ''Smoothing Peridiograms rom Time Series with Continuous Spectra'', Nature, No. 161, [1] Personal communication between Klaus Hugl and Sze Ern Foo, January. [11] PROPSound Radio channel sounder home page o Elektrobit AG. Available at [1] K. Kalliola, P. Vainikainen, ''Characterization System or Radio Channel o Adaptive Array Antennas'', Proc. IEEE PIMRC'97, pp , Helsinki, Finland (1997). [13] K. Hugl, ''Spatial Channel Characteristics or Adaptive Antenna Downlink Transmission'', Dissertation, 19p., Institut ür Nachrichtentechnik und Hochrequenztechnik, Technische Universität Wien, January. available at: [14] J. Bach Andersen, K.I. Pedersen, "Angle-oarrival statistics or low angular resolution", Proc. EPMCC 1, Vienna, Austria (1). [15] K. Hugl, K. Kalliola, J. Laurila, ''Spatial Channel Characteristics or Macro- and Microcellular BS Installations'', Proc. COST 73 Workshop on Opportunities o the Multidimensional Propagation Channel, Espoo, Finland (May 9-3, ). [16] J. Capon, R. J. Greenield, R. J. Kolker, ''Multidimensional Maximum-Likelihood Processing o a Large Aperture Seismic Array'', Proceedings o IEEE, Vol. 55, pp , February 1967.