MILLIMETRE-WAVE CHANNEL IMPULSE RESPONSE EXPERIMENTAL EVALUATION AND RELATION BETWEEN DELAY SPREAD AND CHANNEL COHERENCE BANDWIDTH Manuel Dinis1, José Garcia2, Valdemar Monteiro2 and Nelson Oliveira2 1
Portugal Telecom S.A., Rua Eng. J. Ferreira Pinto Basto, 3810 - 106, Aveiro, Portugal,
[email protected] 2 Universidade de Aveiro, Instituto de Telecomunicações, 3810 – 193, Aveiro, Portugal
Abstract – This paper presents the experimental set-up description and some results of a channel impulse response measurement campaign. The measurements were performed in the frequency domain, in the 40 GHz band, and the channel impulse response was obtained applying spectral analysis. A wide indoor sports pavilion was selected for the experiments, considering several static discrete positions for the mobile terminal, therefore allowing to characterise the propagation channel in different cell coverage regions. From the channel impulse response, power and time dispersion parameters were calculated and analysed. Furthermore, the channel coherence bandwidth was estimated both from direct frequency measurements and from the delay spread estimations and the results analysed.
performed in order to determine the channel CB, both by direct calculation and by DS estimates. Although in this paper only results regarding Line of Sight (LoS) and no LoS conditions in the downlink direction will be considered, CIR measurements were also performed for the uplink direction and for both Mobile Terminal (MT) receiving channels (diversity reception) in LoS and no LoS conditions. A different type of antenna in the MT was also tried.
Keywords – Channel Impulse Response, 40 GHz, Frequency Domain, Delay Spread, Coherence Bandwidth and Experimental Set-up.
The scenario and the experimental set-up are presented in section II, the data processing procedures used to obtain the CIR and its respective parameters are presented in section III. In section IV the channel CB obtained by direct frequency measurement and from DS calculations are compared to verify and evaluate their relationship. Finally, the most significant conclusions of this work are presented in section V.
I. INTRODUCTION
II. EXPERIMENTAL SET-UP
Mobile Broadband Systems (MBS) are very demanding in terms of spectrum, which is only available in the millimetrewave frequency band. At these frequencies the propagation impairments and the type of environment play a significant role in the systems performance, due mainly to atmospheric absorption, signal attenuation and shadowing. Therefore, from the point of view of the design and development of efficient telecommunication systems, the Channel Impulse Response (CIR) knowledge is extremely important as well as its parameters, like for instance the Delay Spread (DS), and the channel Coherence Bandwidth (CB). In this paper, the experimental set-up used to perform the frequency domain measurements is presented. The measurements were undertaken in an indoor sports pavilion scenario, in the 40 GHz band, using the millimetre-wave front-ends, and some additional equipment designed for the ACTS SAMBA project Trial Platform [1] [2]. The CIRs were obtained using spectral analysis, windowing (Blackman Harris and Rectangular windows) and employing the Inverse Discrete Fast Fourier Transform (IFFT) [3]. Power and time dispersion parameters were calculated from the CIR, e.g. Normalised Received Power (NRP), DS and Average Mean Delay ( τ ), and further analysis was
0-7803-7589-0/02/$17.00 ©2002 IEEE
A. Scenario and Measurements Description The measurement campaign was carried-out in a wide indoor sports pavilion, sketched in Fig. 1 where several static discrete positions of the MT (circles numbered from 1 to 11) and the Base Station (BS) location are also displayed. The MT positions were chosen in order to characterise the propagation channel in all the extent of the pavilions surface (different regions of the cell coverage area and different propagation conditions). According to previous results of the SAMBA field trials [4], the MT positions distribution shown in Fig. 1 are such that positions 1, 2, 3, 7, and 11 are within the cell planned coverage area, positions 4, 8, 6 and 10 are at its boundary and positions 5 and 9 are clearly outside the cell. Due to the small wavelength (λ = 7,83 mm) and in order to avoid measuring for a very particular situation, for instance a very deep fad, seven locations of the MT antenna were considered for each MT position, disposed as shown in Fig. 2. The MT antenna placements are separated by a distance intentionally greater than the signal wavelength to reduce the probability of repeating the same unfavourable measurement condition. The CIR for each MT position is
PIMRC 2002
the IFFT of the mean of the seven frequency responses (averaging done in the frequency domain). Moreover, in order to ensure time invariance for each frequency response measurement, the acquisition was repeated five times, and a mean of these measured signals was performed.
use a unique reference signal, which is provided to the Frequency Synthesisers and the millimetre-wave front-ends, by a high precision and stable Rubidium generator (10 MHz reference signal). A
42.580 GHz
45 m
40 m
10
BS 11
50m
8.260 RF
10m
BS
2
3
4
7
8
X
IF
6m
6m
6m
6m
20 dB Amp
9 6m
6.660
10m
1.6 GHz
Rows of seats made of concrete
1.6 GHz
GHz
6.660 GHz
NA
Frequency Synthesiser Y
X
R F GHz LO
25 m
10m
6
8.260
IF
LO
5
B
MS
GHz
1
42.580 GHz
Frequency Synthesiser
10 m
1 PS2 2 /
O (0,0)
1
X
PS1 2 /
10MHz
Fig. 1. Scenario description and BS and MT location. Rubidium Reference
R = 8mm
10MHz
50m
Fig. 3. Equipment set-up for the measurement campaign III. DATA PROCESSING AND CIR ESTIMATION Fig. 2. Antenna placements for each MT position B. Equipment Set-Up The frequency response measurement set-up is presented in Fig. 3, where one may identify the HP 8753D vector Network Analyser (NA). The NA performs the S21 parameter measurement in a 400 MHz bandwidth, centred on 1600 MHz and using 801 step points. The NA signal is up-converted to 8260 MHz using an external mixer, which is the input frequency of the millimetre-wave front-ends. In the front-ends the signal is amplified and up-converted to 42580 MHz and then radiated using the antennas developed in the SAMBA project [1]. The MT antenna receives the millimetre-wave signal, which is then amplified and downconverted to 8260 MHz in the MT front-end and then again down-converted to 1600 MHz using an external mixer. The signal is then amplified using a 20 dB amplifier, in order to compensate the losses of the 50 m cable, that carries the signal back to the NA. In order to obtain both the amplitude and the phase of the frequency response, it is necessary to
A. Data Processing The data processing procedure was divided in two distinct steps: calculate the global system Transfer Function (TF) (that includes the measurement system and radio channel) and extract the radio channel TF from the global system TF. As described in Section II, 35 measurements were performed for each MT position and therefore the global system TF is obtained calculating first the mean of the five TF and subsequently the mean of the seven points. Fig. 4, illustrates the process of obtaining the radio channel transfer function. The lower block diagram corresponds to the global measurement system and the upper block diagram corresponds to the measurement system with a wave-guide connecting the MT and the BS millimetre-wave front-ends. The radio channel TF, which includes the effects of the antennas in the channel, C(s), is obtained following equations (1), (2) and (3), where HBS(s) and HMS(s) correspond to the BS and MT equipment TFs, G(s) to the wave-guide TF, and X(s), X1(s) and Y(s), Y1(s) are the input and the output signals.
X1(s) HBS(s)
G(s)
HBS(s)
C(s)
Y1(s)
components caused by the attenuation of the Power Spectral Density of the signal after multiplying by the BlackmanHarris window.
Y(s)
Analysing now more in detail the results presented in Table 1 and considering the central path (positions 1 to 5), one may verify that position 1 corresponds to a region in the BS antenna blank zone, leading to a low NRP and high DS and
HMS(s)
X(s) HMS(s)
Fig. 4. Block diagram of the measurement system TFs Y (s ) = H EB (s ) × C (s ) × H EM ( s) X (s )
(1)
Y1 ( s ) = H EB ( s ) × G ( s ) × H EM ( s ) X 1 ( s)
(2)
Y ( s ) Y1 ( s) C (s) = G(s) × / X (s) X 1 (s)
(3)
B. CIR Estimation The CIRs were estimated by multiplying first the radio channel TF (C(s)) both by a rectangular window and by a three terms Blackman-Harris window, which provides 67 dB of secondary lobe attenuation [3], and finally using spectral analysis and the IFFT. The use of a BlackmanHarris window reduces the time resolution when compared to a rectangular window, by a factor of approximately 1.8, although improving the capacity of detecting smaller impulses, due to the lower secondary lobes.
τ values [1] [2]. In Position 2, as expected from previous results in [4], the NRP value is maximum and the time dispersion values are the lowest. For positions 3, 4, and 5 the power of the LoS component decreases gradually, leading to a reduction of the NRP value and an increase of the time dispersion parameters. It is important to notice that, despite position 5 is the one further away from the BS, the NRP value is higher than the one of position 4. Although there is a reduction of the power level of the LoS component, the other multipath components have power levels in the order of the LoS component and the combination of all those components leads to an increase of the NRP value. Table 1 CIR parameters MS Position
NRP (dB)
DS (ns)
τ
(ns)
Rect
BH
Rect
BH
Rect
BH
1
-81.33
-90.31
37.59
51.27
7.66
13.88
With the rectangular window and for the considered bandwidth, the resulting CIR has a time resolution of 2.5 ns. In our analysis, the CIR time resolution of 2.5 ns and the used 801 sample points, were considered sufficient since most of the multipath components could be determined and, although the wide dimensions of the pavilion, the 2000 ns visualisation window allowed a convenient CIR characterisation.
2
-65.69
-71.99
13.67
10.56
2.11
1.83
3
-77.35
-85.25
15.99
19.41
2.13
3.26
4
-91.86
-100.55
45.71
61.48
14.98
32.44
5
-85.41
-91.1
19.96
20.28
4.85
4.95
6
-85.63
-93.6
39.01
49.2
8.57
13.36
7
-86.44
-91.93
35.09
34.87
8.45
9.17
8
-84.93
-91.65
32.53
35.18
8.93
10.08
9
-88.18
-94.43
28.08
27.77
9.09
10.02
Table 1 presents the NRP, DS and τ parameters obtained from the CIR using both a rectangular and a BlackmanHarris window, in LoS and no LoS propagation conditions. It can be easily verified that the CIR parameters have different values if a rectangular or a Blackman-Harris window is used. Concerning the NRP, one may conclude that the values obtained with a Blackman-Harris window are always lower than the ones obtained with a Rectangular window. This fact is due to an attenuation penalty in the Power Spectral Density of the signal resulting from the multiplication of the original frequency response by the Blackman-Harris window. As can be seen from the table, the differences between the NRP values obtained with the two windows vary from 4 to 9 dB and are in the range of the difference between the energies of both windows. There are also differences in the time dispersion parameters but these differences do not follow any special rule or common behaviour. This fact is due to the different windows time resolution and the different amplitudes of the multipath
10
-86.71
-91.79
42.78
38.98
10.33
8.13
11
-87.69
-91.66
37.04
30.99
10.36
7.72
7 no LoS
-93.67
-99.90
80.39
79.91
80.35
97.67
8 no LoS
-96.27
-103.73
65.12
69.27
59.99
75.11
9 no LoS
-101.09
-107.91
59.22
60.97
87.46
83.78
11 no LoS
-94.15
-97.67
51.70
40.93
23.68
17.67
Positions 6 to 9 belong to the lateral path where the radiation pattern of the BS antenna reduces the power of the LoS component and, therefore, the NRP values of all the positions are around –85 dB. This reduction of the NRP for lateral positions, in respect to positions in the central path, can be easily verified comparing the NRP values obtained for positions 7, 3 and 11. Positions 6 to 9 are at the cell boundaries, justifying the low NRP values and the high channel time dispersion values. It is important to notice that the CIR parameters for this lateral path are very similar for all the positions but the NRP value of position 9 is lower due to the reduction of the LoS component power with the
distance to the BS. Another aspect to be noticed is the increase of the DS of position 6 in respect to the remainder positions of the path, which is due to the fact that the MT is in the BS antenna's blank zone. As can be seen, the results for positions 10 and 11 are very similar to the ones obtained for the opposite lateral path. Analysing now LoS and no LoS propagation it is quite clear from the table that there is a significant decrease of the NRP values and an increase of the channel time dispersion in no LoS conditions. Fig. 5 a) and b) show the normalised CIR obtained for position 7 with LoS and no LoS, and it can be verified that there are more significant multipath components in no LoS than in LoS propagation condition. These significant multipath components, particularly the ones with time delays between 100 ns and 300 ns, lead to the considerable increase of the DS and the τ values. In LoS conditions the LoS component clearly dominates the Power Delay Profile (PDP) and the contribution of the other multipath components to the channel time dispersion parameters is therefore considerably lower.
spectral spread (σ0), given by formula (6) as stated in [6] and having ∆f as the frequency lag. Expression (6) can be described as the minimum value that the correlation coefficient can assume, taking into account the value of σ0 which, in this study in particular and in mobile communications studies in general, corresponds to the DS. B0.5 ≈
1 2πDS
(4a)
B0.9 ≈
1 50 DS
(4b)
These relationships are independent of the frequency band of operation but in order to verify their consistency in the millimetre-wave band, comparisons were done between the channel CB obtained by calculating the autocorrelation of direct frequency response measurements and from DS values obtained from the measured CIR. Bc ≥
arccos(c) 2πDS
(5) (6)
R( ∆f ) = cos(2π .σ 0 .∆f )
B. Results
a)
Table 2 shows the values of the channel CB obtained by direct frequency response measurement and its lower bound calculated by expression (5) for levels 0.5 and 0.9. The values presented refer to the 11 MT positions considered in the CIR measurement campaign and, therefore, include results in LoS and some in no LoS propagation conditions. Table 2 CB at level 0.5 and 0.9 obtained by direct frequency response measurement and its lower bound
b)
Fig. 5. CIR for position 7 in: a) LoS; b) no LoS IV. CHANNEL COHERENCE BANDWIDTH AND DELAY SPREAD A. Introduction It is stated in current literature that the channel CB at a certain coherence level c (Bc) and DS are related by (4a) for a coherence bandwidth level of 0.5 and by (4b) for a level of 0.9 [5]. In [6] it was demonstrated mathematically that the relationship is an uncertainty relation of the same kind as Heizenberg's and the differences between these two uncertainty relations are due to the class of functions that they were formulated. The uncertainty relation between DS and CB is translated by the inequality shown in expression (5). The channel CB is lower bounded by a function with
B0.5 B0.5 Meas. by. (6) (MHz) (MHz)
∆
B0.9 B0.9 Meas. by (6) (MHz) (MHz)
∆
MS Position
DS (ns)
1
37.59
46.50
4.23
91%
2.00
1.91
5%
2
13.67
77.50
11.64
85%
12.00
5.25
56 %
3
15.99
57.00
9.95
83%
7.50
4.49
40 %
4
45.71
39.00
3.48
91%
1.50
1.57
5%
5
19.96
70.00
7.97
89%
4.00
3.60
10 %
6
39.01
44.50
4.08
91%
2.00
1.84
8%
7
35.09
49.00
4.54
91%
2.50
2.05
18 %
8
32.53
37.50
4.89
87%
3.00
2.21
26 %
9
28.08
22.00
5.67
74%
2.50
2.56
2%
10
42.78
80.50
3.72
95%
6.00
1.68
72 %
11
37.04
84.50
4.30
95%
2.00
1.94
3%
7 no LoS
80.39
2.00
1.98
1%
1.00
0.89
11 %
8 no LoS
65.12
2.50
2.44
2%
1.00
1.10
10 %
9 no LoS
59.22
3.00
2.69
10%
1.00
1.21
21 %
11 no LoS 51.70
21.00
3.08
85%
1.50
1.39
7%
(%)
(%)
Fig. 5 and Fig. 6 depict the scatter plot of pairs of estimates of channel CB at level 0.5 and 0.9 and the corresponding DS value. Also in both Fig. 5 and Fig. 6 are shown the lines corresponding to the lower bound expression (5). Correlation coefficients obtained from measurement estimates and by the lower bound tend to have the same value as the coherence bandwidth approaches one. This fact is corroborated by the values presented in Table 2, where ∆ for a coherence level of 0.9 is smaller than the one obtained for a level of 0.5. In the scatter plot most of the points representing pairs of estimates are above the lower bound according to (5). Some points are quite on the boundary limit.
The results show the impact of the selected window for the windowing step in the process and the clear CIR parameters degradation on no LoS propagation condition. A Blackman Harris window was selected in order to allow a better visibility of the lowest multipath components with the price of loosing some time resolution. Although the very high frequency band, the analysis in the frequency domain proved to work well with the inherent disadvantage that no motion is allowed while performing the measurements. The calculated parameters are according to the expected and suitable explanations were presented regarding the different behavior in different locations in the pavilion.
100
Coh Bw (MHz)
was an indoor sports pavilion, representative of a possible scenario where MBS will operate. The locations of the MT were selected in order to gather information about the radio channel in different regions of the cell coverage area. The data processing procedures and steps required to obtain the CIR and the power and time dispersion parameters are also described.
10
B0.5 Meas. (MHz) B0.5 by. (6) (MHz) 1 10
Delay Spread (ns)
100
Fig. 6. Pairs of estimates of CB at level 0.5 and corresponding DS
The relationship between DS and channel CB was also investigated and compared with other studies performed by other authors in other frequency bands. Two CB levels were used: 0.5 and 0.9. It was noticed that for CB level of 0.9, the measured figures approach much more the lower limit than for 0.5. Moreover, figures for no LoS condition shown to be much more closer to the lower theoretical limit. The main conclusion was that the estimations using the theoretical lower limit are sometimes very different from the real values.
100 B0.9 Meas. (MHz)
Coh Bw (MHz)
B0.9 by (6) (MHz)
REFERENCES [1]
10 [2]
[3]
1 10
Delay Spread (ns)
100
Fig. 7. Pairs of estimates of CB at level 0.9 and corresponding DS
[4]
V. CONCLUSION
[5]
In this paper the experimental set-up and the required equipment to perform a CIR measurement campaign in the 40 GHz frequency band is presented. The selected scenario
[6]
C. Fernandes and J. Fernandes, “Impact of Shaped Lens Antennas on MBS Systems”, PIMRC’98 - The 9th IEEE Int. Symp. on Personal, Indoor, and Mobile Radio Communications, Boston, Massachusetts, USA, Sec. C7, p. 744-748, Sep. 1998. M. Dinis, J. Fernandes, M. Prögler and W. Herzig, “The SAMBA Trial Platform in the Field”, ACTS Mobile Communications Summit ’99, Sorrento, Italy, pp. 10131018, June 1999. F. J. Harris, “On the Use of Windows for Harmonic Analysis with Discrete Fourier Transform”, Proceedings of the IEEE, vol. 66 No. 1 Jan. 76 José Garcia, Manuel Dinis and José Fernandes, “Performance Evaluation of a Cellular Millimetrewave Mobile Broadband System Demonstrator”, CONFTELE 2001, Figueira da Foz, 23 - 24 Abril 2001 Theodore S. Rappaport, “Wireless Communications Principles and Practice”, Prentice-Hall, 1996 B. H. Fleury, “An uncertainty Relation for WSS Processes and Its Application to WSSUS Systems”, IEEE Transactions on Communications, vol. 44 No. 12 Dec. 96