The Journal of China Universities of Posts and Telecommunications December 2008, 15(4): 46–50 www.sciencedirect.com/science/journal/10058885
www.buptjournal.cn/xben
Indoor channel characteristics based on wideband MIMO measurements at 5.25 GHz GAO Xin-ying ( ), ZHANG Jian-hua, ZHANG Ping Key Laboratory of Universal Wireless Communications, Ministry of Education, Wireless Technology Innovation Institute, Beijing University of Posts and Telecommunications, Beijing 100876, China
Abstract
Extensive indoor channel measurements were conducted in Beijing with wideband multiple-input multiple-output (MIMO) sounder at 5.25 GHz. Both line-of-sight (LOS) and non-line-of-sight (NLOS) propagation were measured in the indoor office and hotspot scenarios. On the basis of measured data, statistical channel characteristics are presented in this article, including the empirical path loss (PL) models, three excess delay parameters, circular azimuth spread (CAS), and circular elevation spread (CES). Comparative analysis of different propagation mechanisms in two scenarios is conducted. These values are significant for indoor coverage and technical research of MIMO and orthogonal frequency division multiplexing (OFDM) for the international mobile telecommunications-advanced (IMT-Advanced) system. Keywords channel measurement, wideband, MIMO, path loss, excess delay, circular azimuth spread, circular elevation spread, IMT-Advanced
1
Introduction
According to the basic requirements of the IMT-Advanced system, the radio frequency (RF) bandwidth and carrier frequency reach 100 MHz and 5 GHz respectively. The wide bandwidth and high frequency, together with deployment of multi-antennas, are suitable for indoor scenario to provide high data-rate (maximal 1 Gb/s) transmission. Realistic channel measurements and data post-processing are the most efficient way to understand radio propagation characteristics [1]. To improve the design of the IMT-Advanced system, we conducted wideband MIMO channel measurements in Beijing, China. The purpose of this article is to present the extracted channel characteristics measured in indoor office and hotspot scenarios. In wireless channel, the large-scale effects are important for coverage prediction and interference analysis. The small-scale fading parameters should be considered when designing the advanced transmission techniques in the IMT-Advanced system, such as MIMO and OFDM techniques. Hence, the log-distance PL models, excess delay, and angle statistics in the three-dimensional (3D) scattering environment are studied Received date: 04-01-2008 Corresponding author: GAO Xin-ying, E-mail:
[email protected]
for the LOS and NLOS propagation respectively. To derive the spatial channel parameters from measured channel impulse responses (CIR), the spatial-alternating generalized expectation-maximization (SAGE) [2–4] algorithm is used. Compared with other algorithms [5–6], the SAGE is not limited to antenna array structure and has strong ability of detecting weak waves. Under the assumption of the far-field condition, the CIR propagating from the ith element of transmitter (Tx) to the jth element of receiver (Rx) is modeled as: I
hij (t ;W ) ¦D A exp ^ j2ʌvAt` c2, j ( ȍ2, A )c1,i ( ȍ1, A )T G (t W A )
(1)
A 1
where the Ath wave is characterized by the parameter set {W A , vA ,D A , ȍ1, A , ȍ2,A } as follows: W A is the propagation delay,
vA is the Doppler frequency, D A is the complex weight, ȍ1,A denotes the angle of departure (AoD) at Tx, and ȍ2, A denotes the angle of arrival (AoA) at Rx. The unit vector ȍ is uniquely determined by the azimuth I and the elevation T as ȍ [cos I cos T , sin I cos T , sin T ]T . c1,i ( ȍ1, A ) is the ith element response of Tx array, and c2, j ( ȍ2, A ) is the jth element response of Rx array. I is the total of multi-path components (MPC). The symbol ()T denotes transpose.
Issue 4
GAO Xin-ying, et al. / Indoor channel characteristics based on wideband MIMO…
47
2 Channel measurement description 2.1
Measurement system
MIMO channel measurements were taken by Elektrobit sounder PropSound at the center carrier frequency of 5.25 GHz with 100 MHz bandwidth in Beijing University of Posts and Telecommunications (BUPT), China. The transmitted power is 26 dBm, chip duration is 10 ns, the sampling rate is 200 M samples/s, the length of the periodic pseudo-random binary signals is 511 chips, and the cycle rate is 21.7 Hz. PropSound works in a time division multiplexing (TDM) mode to save the hardware and reduce the effort to calibrate the system. As shown in Fig. 1, Tx uses a dual-polarized omni-directional array (ODA) with 50 elements, and Rx uses a vertically polarized uniform circular array (UCA) with 8 elements.
(a) ODA (b) UCA Fig. 1 Antenna arrays used in channel measurements
2.2
Indoor measurement scenarios
Figure 2(a) illustrates the layout of indoor office measurements. Rx was fixed in ‘room 2’ as the arrow shows; Tx was located at each dot in ‘room 2’ for the LOS and other rooms for the NLOS propagation. In the indoor hotspot scenario shown in Fig. 2(b), ‘Grid A’ was placed in the hall for the LOS propagation. ‘Grid B’ was surrounded by notice boards with glass sheets for the NLOS propagation. Between Rx and ‘Grid B’, there are huge concrete poles, escalators, staircase with aluminum banisters, etc. The height from the ceiling to the ground was approximately 3 m for all locations. During the measurements, both Tx and Rx were mounted on the trolley with antenna heights of 1.5 m and 2.5 m respectively.
Fig. 2
(b) Hotspot Indoor measurement scenarios
3 Analysis of channel characteristics To separate actual MPCs from noise, the dynamic noise floor and delay search threshold should be determined. For the results reported hereafter, the delay threshold is set 15 dB below the peak of power delay profile (PDP). In addition, the peak must be 18 dB above the noise floor to allow a 3 dB margin between the threshold and the noise floor. If this margin is not realized, the data will be omitted from the statistical description. The PL models and delay parameters can be extracted from PDPs directly. 3.1
PL models
PL is defined as the ratio of effective transmitted power to received power, calibrating out the equipment loss, amplifier gain, and antenna gain. Let the PDP denote as p (W ) , then the PL values are calculated in decibels as: § I · L 10lg ¨ ¦ p (W A ) ¸ GT GR (2) ©A 1 ¹ where GT and GR are antenna gains of Tx and Rx, derived from antenna calibration responses. The modeling method used for the large-scale effects is a linear curve fitting the decibel PL to the decibel distance with a random variation. The empirical PL model [7] is given by: L(d ) L0 10n lg d X V (3) where L0 is the intercept, d is the Tx-Rx distance in meters, n is the PL exponent dependent on the specific propagation environment indicating the rate, at which PL increases with distance. X V is a zero-mean Gaussian variable denoting the fading of shadowing with standard deviation V . The values of L0 and n are extracted using the least square (LS) method,
(a) Office
such that the mean square error between the measured and estimated PL values is minimized. It is well known that n equals to 2 in free space assuming isotropic antennas located in a perfectly dielectric, homogeneous, and unlimited environment. In Fig. 3, the empirical PL models of the LOS and NLOS are illustrated in office and hotspot respectively. Compared with the free space, the
48
The Journal of China Universities of Posts and Telecommunications
exponent n of the LOS is 1.37 in office and 1.18 in hotspot, both of which are smaller than 2. It means that the wave-guided phenomenon is available in indoor LOS propagation because of the reflection and scattering of walls, ground, ceiling, etc. Transited from the LOS to the NLOS, PL increases highly, and n of the NLOS becomes 4.06 in office and 4.33 in hotspot. The increment is given as 10 dB at 3.5 m and 26.5 dB at 40 m.
2008
2) The root mean square (rms) delay spread (DS) is defined as the power weighted standard deviation of the excess delay: I
W rms
¦ (W
A
W m ) 2 p (W A ) (5)
A 1
I
¦ p(W
A)
A 1
3) The maximum excess delay W max is defined as the length of the delay profile between two MPCs, which are the farthest apart for the given threshold. The statistics of the above delay parameters are listed in Table 1, including the 10%, 50%, 90% percentiles of cumulative distribution function (cdf) and mean value. The results reveal that W m , W rms and W max of the NLOS are larger than the corresponding values of the LOS; and each parameter of hotspot in either the LOS or NLOS are larger than that of office, which is reasonable because the indoor hotspot scenario involves greater space than office, resulting in more scattering objects within the wider coverage. Table 1 Statistical results of the measured delay parameters Scenarios
LOS
(a) Office Office
NLOS
LOS Hotspot NLOS
Delay/ns
10%
50%
90%
Mean
Wm
12
18
25.5
20.1
W rms
5.6
8.6
13
9.1
W max
45
60
80
63.8
Wm
15
25.5
82.5
39.9
W rms
8
13
24.5
15
W max
50
100
160
105
30
54.5
33.5
Wm
18
W rms
13.5
26
40
26.7
W max
125
240
300
226
Wm
34
55.5
83
57
W rms
25
40
53.5
39.6
W max
180
275
360
270.9
Specifically, W rms provides a measure for the variability of the delay, coherent bandwidth, and frequency selectivity. Figure 4 plots the cdf curves and the fitted log-normal distribution, expressed as:
(b) Hotspot Fig. 3 Indoor PL models
3.2
W rms 10P
Delay parameters
On the basis of the measured PDPs, three statistical parameters are calculated to present the delay dispersion of these MPCs. 1) The mean excess delay is given by: I
Wm
¦W
A
p (W A ) (4)
A 1 I
¦ p(W A 1
A
)
D
XVD
(6)
where X is also a zero-mean Gaussian random variable with unit variance, P D E ^lg W rms ` is the logarithmic mean of the local DS, and V D is the logarithmic standard deviation of the local DS. It’s shown that P D of the NLOS is larger than that of the LOS both in office and hotspot; P D of hotspot is larger than that of office both for the LOS and NLOS.
Issue 4
Fig. 4
3.3
GAO Xin-ying, et al. / Indoor channel characteristics based on wideband MIMO…
The cdfs and fitted log-normal distributions of W rms
(b) NLOS Estimated MPCs in indoor hotspot scenario
Fig. 5
Angular parameters
For MIMO system, angular statistics in spatial domain are more important in affecting spatial correlation and achievable channel capacity. In the indoor scenario, it is noted that elevation in the vertical plane is too significant to be neglected. Yielded from the SAGE implementation, the estimated MPCs of two snapshots are illustrated in Fig. 5, where each ball is uniquely denoted by the five-dimensional parameter set (W A , I1, A , I2, A , T1, A , PA ) , i.e., the excess delay, the azimuth of AoD, the azimuth of AoA, the elevation of AoD and the power. The size of the balls denotes the elevation of AoD. The 2 power of the Ath wave is calculated as PA D A . The 0D stands for the reference direction of Tx or Rx array. It is noteworthy that the detected paths appear in the form of clusters, which are a group of MPCs with similar parameters [8–9]. Compared with the NLOS, the MPCs with higher power of the LOS are more concentrated. Additionally, the size of the ball of the LOS is larger than that of the NLOS, which implies that the elevation of AoD in the LOS propagation is larger than that of the NLOS.
49
To evaluate the angle characteristics at Tx and Rx numerically, the rms angle spread (AS) is calculated as the square root of central moment of power angular spectrum (PAS). To avoid ambiguity of the origin, i.e., the modulo 2ʌ operation, rms AS should be constant no matter what the linear angle shift ' is. Let MA (') be equal to the azimuth
IA ' or elevation T A ' , the rms AS is calculated to be circular as the minimum of: I
Mrms
min Mrms (' ) '
¦[M (') E (')] P 2
A
A
A 1
I
¦P
(7)
A
A 1
over all angle shift ' and E (' ) is defined as: I
E (')
¦M (') P A
A
A 1
I
¦P
(8)
A
A 1
Both MA (') and MA (' ) E (') in Eq. (7), denoted as Z for notational brevity, are normalized as: 2ʌ Z; Z ʌ ° Z ®Z; Zİʌ ° 2ʌ Z ; Z ! ʌ ¯
(9)
The statistical values of the CAS are listed in Table 2, including the log-normal distribution ( PA ,V A ) and the mean
(a) LOS
value. The results show that the CAS in the NLOS case is slightly larger than the CAS in the LOS scenario. One possible reason could be that the omni-direction antenna array was used. As a consequence, in the LOS case, the multi-path effect is still significant because of the reflections and scattering of the ceiling, floor, walls, and other objects in the considered indoor environments. Both in the LOS and the NLOS cases, it is observed that the CAS of AoA is larger than that of AoD. Considering the
50
The Journal of China Universities of Posts and Telecommunications
higher antenna height of Rx, it is conclusive that the CAS increases with the antenna height. It also appears that the CAS in office is larger than that in hotspot, especially in the LOS case. The reason is that the ‘room 2’ in indoor office is blocked with walls all around, where ‘Grid A’ in indoor hotspot is only blocked by the walls along the corridor. Table 2 Statistical results of CAS CAS of AoD
Scenarios Office Hotspot
LOS NLOS LOS NLOS
CAS of AoA
PA
VA
Mean/ (D )
PA
VA
Mean/ (D )
1.63 1.66 1.58 1.61
0.06 0.25 0.22 0.16
42.2 45.4 38.0 40.7
1.73 1.76 1.59 1.71
0.28 0.61 0.18 0.25
53.2 57.0 38.9 51.3
2008
NLOS propagation both in office and hotspot. The PL models show that the exponent of LOS is lower than 2, whereas that of NLOS is larger than 4. The statistical behavior of the excess delay indicates that the temporal dispersion of hotspot is much more prominent than that of office. In the indoor scenario, the CAS increases with the increase in antenna height. In both the LOS and NLOS propagation, the CAS in office is larger than that in hotspot. Furthermore, it is found that the CAS of the LOS is slightly lower than that of the NLOS, whereas the CES of the LOS is larger than that of the NLOS. Acknowledgements
During the measurement, Tx with lower antenna heights can represent the user terminal. Figure 6 shows the cdfs of CES of AoD. It is found that the CES of the LOS is larger than that of the NLOS for both office and hotspot scenarios, which is consistent with the visual inspection of Fig. 5. This observation can be explained by the rich reflection of ceiling and ground in LOS measurements.
Fig. 6
4
The cdfs of CES of AoD
Conclusions
On the basis of the measured wideband MIMO channel data, this article presents the indoor propagation characteristics to reveal the signal features in the large-scale, delay, and spatial domains. Measured scenarios include the LOS and
This work was supported by the Hi-Tech Research and Development Program of China (2006AA01Z258), China Mobile Research Institute.
References 1. Zhang M, Zhang J H, Zhang P, et al. Broadband channel measurement and analysis. The Journal of China Universities of Posts and Telecommunications, 2006, 13(3): 24–28 2. Fleury B H, Tschudin M, Heddergott R, et al. Channel parameter estimation in mobile radio environments using the SAGE algorithm. IEEE Journal on Selected Areas in Communications, 1999, 17(3): 434–450 3. Fleury B H, Yin X, Rohbrandt K G, et al. Performance of a high-resolution scheme for joint estimation of delay and bidirection dispersion in the radio channel. Proceedings of the IEEE Vehicular Technology Conference (VTC’02-Spring): Vol 1, May 6–9, 2002, Birmingham, UK. IEEE, Piscataway, NJ, USA: IEEE, 2002: 522–526 4. Yin X, Fleury B H, Jourdan P, et al. Polarization estimation of individual propagation paths using the SAGE algorithm. Proceedings of the 14th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC’03): Vol 2, Sep 7–10, 2003, Beijing, China. Piscataway, NJ, USA: IEEE, 2003: 1795–1799 5. Roy R, Kailath T. ESPRIT-estimation of signal parameters via rotational invariance techniques. IEEE Transactions on Acoustics, Speech, and Signal Processing, 1989, 37(7): 984–995 6. Schmidt R O. Multiple emitter location and signal parameter estimation. IEEE Transactions on Antennas and Propagation, 1986, 34(3): 276–280 7. Erceg V, Greenstein L J, Tjandra S Y, et al. An empirically based path loss model for wireless channels in suburban environments. IEEE Journal on Selected Areas in Communications, 1999, 17(7): 1205–1211 8. Vuokko L, Vainikainen P, Takada J. Clusters extracted from measured propagation channels in macrocellular environments. IEEE Transactions on Antennas and Propagation, 2005, 53(12): 4089–4098 9. Li K H, Ingram M A, Nguyen A V. Impact of clustering in statistical indoor propagation models on link capacity. IEEE Transactions on Communication, 2002, 50(4): 521–523