Field Test Results for Space-Time Coding Parul Gupta, Weijun Zhu, Michael P. Fitz, Heechoon Lee, Daniel N. Liu and ShingWa G. Wong The University of California Los Angeles
[email protected],
[email protected],
[email protected]
Abstract— Multiple antenna radios are recently the subject of much research due to the significant improvements they offer in throughput and reliability in wireless communications. Research in the field of space-time coding has led to the design of modulation techniques offering improved performance. Their performance, although promising in simulations, needs to be studied in real wireless channels due to the inability to accurately model all possible wireless channel conditions that might be seen in practice. With this motivation, a range of experiments were conducted to measure the performance of some of the well-known space-time codes in literature over real channels and under implementation impairments. This paper discusses the testbed setup for the experiments and present the results of field trials conducted at the University of California, Los Angeles.
I. I NTRODUCTION With the growing demand for high-speed wireless services with a limited spectral resource, there is a need to communicate at high spectral efficiencies. Multiple antenna systems have generated a lot of interest in recent times because of the significant improvements they offer in throughput and reliability in wireless communications. Combining space-time coding techniques at the transmitter with appropriate spacetime processing at the receiver can help harvest these benefits. The standard design criteria for constructing good space-time codes were proposed in [1],[2] and subsequent research led to the development of several classes of high performance codes. While analyses based on theory and simulations typically show the performance of these codes under ideal conditions, it is important to study the same in physical channels and in the presence of implementation impairments to understand what is achievable in real systems. With this motivation, a range of experimental performance studies were conducted at the University of California, Los Angeles (UCLA) Wireless Research and Development Lab (UnWiReD Lab) on some of the best known space-time codes in literature, including orthogonal block designs [3], [4], super-orthogonal schemes [5] and layered designs [8]. In the paper, Section II discusses in detail the test setup for these studies including the testbed hardware, the frame structure and the channel estimation mechanism, Section III presents the field test results, Section IV presents the conclusions and outlines future work. II. S YSTEM D ESCRIPTION A. Testbed Hardware The UnWiReD Lab testbed is a fully configurable, real channel time, multi-antenna testbed. It supports more than pairs, each with KHz of useful bandwidth, in the MHz narrowband radio services band set up by the Federal
USB
Mobile Radio
IF signal 10.273MHz 3-TX Up Converter Radio
RS232 Laptop
3-TX Modem
RS232 221.5625 MHz RF
GPS
Transmitter
GPS
10.273 MHz IF signal Parallel 4-RX LinkPort Port Baseband Down 4-RX DDC Laptop Converter Processor Receiver Radio
RS232
USB Mobile Radio
Fig. 1.
Receiver
Block diagram of the
MIMO test system
Communications Commission for spectrally efficient communications. The low signal bandwidth permits programmable implementations. The radios for this testbed also cover the adjacent amateur radio band. Figure 1 shows the block diagram of the narrowband MIMO test system. The information bits are encoded and pulse shaped in the 3-TX modem by two fixed point DSPs. The generated baseband signals are then digitally up-converted to MHz IF signals. The 3-TX radio further up-converts the IF signals to MHz and amplifies it for transmission with a maximum power of dBm. The receiver chain provides a high performance system for narrowband MIMO processing. The received signals are down MHz IF signals by the 4-channel converted from RF to down converter radio and then digitally down converted to baseband by the 4-channel digital receiver. The overall receiver dynamic range is greater than dB. The demodulation is done by two floating point DSPs. The pilot symbols, demodulated data as well as other important test information is transferred to a laptop for data storage and real-time display of test results. This is a valuable capability as it allows more extensive post processing and analysis of data which might otherwise not be possible in real-time. In case of outdoor test, two GPS units are attached to the laptops to provide location information. The addition of the GPS units makes the characterization of performance vs. distance possible.Additional radio units provide the ability to feed back channel information to the transmitter, enabling
300 Symbols
Preamble
300 Symbols
Upto 42 data slots, each frame 300 symbols
Reed Solomon Alamouti Encoded QPSK Control Packet
Fig. 2.
Alamouti 16-QAM
3-Tx BLAST
70 Symbols
Silence Period
"! $#%
Transmission Superframe Structure
feedback MIMO systems. B. Frame Structure The transmitter sends data in second superframes where each superframe contains a preamble and a multitude of spacetime coded frames, as shown in Figure 2. The preamble allows for simple time and frequency synchronization. Each data slot in the superframe can be potentially modulated with a different space-time coding scheme which enables a fair comparison of various schemes on approximately the same set of channels. The number of slots is programmable and is limited by the DSP memory. In the current setup, there are slots. At the end of the frame, there is silence period to measure the noise power.
All coding schemes examined in this paper were proposed for coherent systems. Consequently, accurate estimation of the channel is crucial for reliable decoding. Pilot Symbol Assisted Modulation (PSAM) is employed to achieve simple processing and good performance in high-mobility situations. For the estimator design, the channel gains between any transmitterreceiver pair are assumed to be spatially independent of any other pair. The Rayleigh flat fading assumption is valid given the low signal bandwidth. Also, the channel coefficients are assumed to be constant over a symbol period but vary from symbol to symbol with a correlation given by
(1)
where the zeroth order Bessel function of the first kind is theis Doppler and spread of the channel. 300
P 5P P 5D P 2P 3D P 5DP 2P 3D P 2P 3DP 5D P - - - - 5D P 3D - - - P 5P 1
7
13
19 Edge part
25
(2)
Because of this orthogonality property, MISO channel estimation can be employed at each receive antenna. An FIR Wiener
,- and Doppler fading rate filter optimized for &('*)+ /. % is used for pilot interpolation [2]. The Wiener filter coefficients for these parameters and the corresponding frame structure, are calculated and stored at the receiver and reused for channel estimation in each frame as shown below. Consider the following model for a MIMO system with 021 transmit antennas and 043 receive antennas. If the 5 176 transmit antenna transmits the 98 symbols long frame given by
> ?; < @% %A% ;=< B8 DCFE =; : < (3) 176 receive antenna, HA: I , can be the received signal at the G
written as
C. Channel Estimation
and are inserted uniformly in the center at the rate of one pilot symbols. every In [2], Guey et al. showed that it is optimal to send orthogonal pilot elements on each transmit antenna. For the 2-Tx case, the pilot element is chosen to be
31
37 Uniform Part
Edge part
Fig. 3. PSAM Frame for the 2-Tx ST Codes. P and D represent the pilots and data symbols respectively.
The pilot symbol frame is designed separately for different number of transmit antennas to get good mean-squared error performance for channel estimation [7]. For example, the 2 Tx STC frame, shown in Figure 3, is
symbols long, comprising of 228 data and 72 pilot symbols. Pilots are more concentrated at each end to compensate for the edge effects
+H : I KJ L : IM NO: I (4) J > ,PRQS ;=: T VU ,PRQS ;X: W A% %@% ,PDQS ;?:Y[Z DC is where L:I the B8 B80\1 transmitted symbol matrix, > L TV] I %A%@% L TV] I ^8 %@%A% L Y Z ] I %A%@% L Y Z ] I B8 DC E is the length B80\1 channel gain vector from: all the transmit 176 receive antenna and N_I is the length 8 antennas to the G
complex white Gaussian noise vector. Following this model for the pilot observations, we may write
H+: I `J a+: I M NO: I
:
(5)
with the superscript p indicating pilot instances. H I is the vector of b^ pilot observations closest to the estimation instant where number of taps for the Wiener filter. :aVI is theb9 length 0 is1 the ! vector gains at !of true b channel the data instances. c is the 0 1 Wiener filter coefficient matrix given by
T : : : : > C > f C e A d A d a V I H A I H + I H + I c & &
(6)
Under the stated assumptions on the channel, this is equivalent to
eT : : : : > F C J J > X C A J @
i j A d A d d A h d g + a I V a I V a I V a I M c & & (7) >k C is the noise variance, i is the b9 b9 Identity matrix and & denotes the expectation operator. The channel estimates
can then be calculated as
a :l I c H : I
(8)
III. R ESULTS A suite of indoor and outdoor tests were conducted on the UCLA campus on the system described in Section II. The schemes implemented included Alamouti code [3], 3-TX orthogonal block code [4], Super-orthogonal 3-TX block code [5], and 3-TX BLAST [8] each with QPSK, 8PSK, 16QAM, 64QAM constellations. The transmit power was fixed at -10 dBm for both indoor and outdoor tests and uniform linear antenna arrays with half-wavelength separation were used. For most of the coding schemes, it was possible to demodulate in real-time for each of the possible number of receive antennas, one through four. This allows an examination of how diversity will improve performance for the same transmission and channel characteristics.
Figures 5 and 6 show a comparison of same rate Alamouti and BLAST schemes in indoor stationary environments for different number of receive antennas. It is interesting to note that for a system, the Bit Error Rate (BER) of Alamouti is better than BLAST for both rates and bits per channel use (bpcu). However, when the number of receive antennas is increased to 4, BLAST outperforms Alamouti at high SNR’s. This can be expected since for BLAST schemes with 4 receivers, there are more degrees of freedom to suppress spatial interference as compared to 2 receivers. Alamouti and BLAST in indoor stationary tests
0
10
−1
10
A. Indoor Field Test Results
−2
10
−3
10 BER
For the indoor tests, the receiver was housed in the UnWiRed Lab and the transmitter was placed at several positions on the same floor of the building. A floor plan of the test environment is shown in Figure 4. The black marks show the various positions at which the transmitter was placed. In the performance plots, each point represents a particular transmitter position in a single test which gives a certain value of SNR. For each point, data was collected for at least two hours. As an example, for the 16-QAM Alamouti scheme this would mean roughly 1.65 million bits. The schemes compared were transmitted simultaneously in the same test. This enables a fair comparison as the channel and hardware conditions were identical across the schemes. Some of the kinks in the plots can be explained as manifestations of the complex channel dynamics. For example, tests conducted during the day when there was a lot of activity and people moving around perform much worse than tests performed in quieter (almost constant) channel conditions at night. Also, the channel at a particular transmitter position might be singular irrespective of the average SNR, leading to poor performance.
−4
10
−5
10
2x2 Alamouti 16QAM, R=4 bpc 2x2 BLAST QPSK, R=4 bpc 2x2 Alamouti 64QAM, R=6 bpc 2x2 BLAST 8PSK, R=6 bpc
−6
10
−7
10
0
Fig. 5.
5
10
BER for
15 20 Eb/N0 per Rx Antenna, dB
25
30
35
Alamouti vs. BLAST in indoor stationary channels
Alamouti and BLAST in indoor stationary tests
0
10
2x4 Alamouti 16QAM, R=4 bpc 2x4 BLAST QPSK, R=4 bpc 2x4 Alamouti 64QAM, R=6 bpc 2x4 BLAST 8PSK, R=6 bpc
−1
10
−2
10
−3
BER
10
−4
10
−5
10
−6
10
−7
10
0
Fig. 6.
Fig. 4.
Floor Plan of the Indoor Test Setup (not to scale)
5
BER for
10
15 20 Eb/N0 per Rx Antenna, dB
25
30
Alamouti vs. BLAST in indoor stationary channels
Figures 7 and 8 show the BER comparison for 3-Tx orthogonal [4] vs the 3-Tx Super-Orthogonal block code [5] for 8-PSK and 64-QAM constellations respectively. We see that the Super-Orthogonal codes achieve a higher transmission
3x4 BLAST in a Stationary Test, QPSK
rate simultaneously with better performance as compared to the regular orthogonal block code. Also, since a constant
for higher rates, transmit power translates to smaller & ' )+ this performance gain is even larger for higher constellations.
70 Worst layer Middle Layer Best Layer 60
Number of Bits in Error
50 3−TX block codes in indoor stationary tests, 8PSK, Lr = 4
−1
10
Orthogonal, R=2.25 bpc Super Orthogonal, R=2.75 bpc Super Orthogonal, R=3 bpc Super Orthogonal, R=3.25 bpc −2
10
40
30
20 −3
10 BER
10
−4
0
10
0
100
200
300 400 Frame Index
500
Fig. 9. Number of bit errors across different layers in a stationary indoor test
−5
10
600
700
BLAST
−6
10
6
8
10
12 14 Eb/N0 per Rx Antenna, dB
16
18
Fig. 7. BER for 3-Tx Orthogonal vs. Super-Orthogonal 8-PSK Block Codes.
3−TX block codes in indoor stationary tests, 64QAM, Lr = 4
0
10
Orthogonal, R=4.5 bpc Super Orthogonal, R=5 bpc Super Orthogonal, R=5.25 bpc Super Orthogonal, R=5.5 bpc −1
10
−2
BER
10
−3
For the outdoor tests, the transmitter was mounted atop an 8-story building and the receiver was driven around the UCLA campus within a radius of 1 km on a low speed electric vehicle. Figures 10 and 11 show the comparison of bit error rates of bpcu Alamouti and BLAST schemes for outdoor rate mobile and indoor stationary environments respectively. While in the indoor tests, BLAST outperforms Alamouti, it loses out in the outdoor case. One of the first thoughts that might occur is the accuracy of channel tracking, but the pilot symbol frame structures used for the current tests have been optimized for the high mobility case [7]. Yet, the different frame structures for the 2-Tx and 3-Tx cases might be a contributing factor and we are currently analysing the various possibilities in this regard. IV. C ONCLUSIONS
10
−4
10
−5
10
B. Outdoor Field Test Results
0
Fig. 8. Codes.
5
BER
10
15 20 Eb/N0 per Rx Antenna, dB
25
30
35
for 3-Tx Orthogonal vs. Super-Orthogonal 64-QAM Block
Figure 9 shows the number of bit errors across different layers of a BLAST system in indoor stationary channel conditions. As expected, the worst layer accounts for most of the errors while the best layer is almost always free of errors. This verifies that an ARQ scheme independent across layers might hold the potential of increasing throughput [9]
This paper made an attempt to quantify the performance of several well-known space-time codes over real channels and in the presence of implementation impairments. The preliminary results of some indoor and outdoor field tests conducted on an infrastructure based testbed were presented. These results brought out several interesting observations and laid the foundation for further research on effects on performance in real systems. Future work includes more extensive testing and analysis of data, implementation of more complex coding schemes, and examination of the effect of channel estimation accuracy on performance. R EFERENCES [1] N. Seshadri, V. Tarokh and A. R. Calderbank, “Space-Time Codes for High Data Rate Wireless Communications: Performance Criterion and Code Construction,” IEEE Trans. Inform. Theory, vol. 44, no. 2, pp. 744–765, March 1998.
Error performance in outdoor mobile test
0
University, 2002. [8] G. J. Foschini, “Layered Space-Time Architeture for Wireless Communication in a Fading Environment When Using Multi-Element Antennas , ”Bell Labs Technical Journal, pp. 41-59, 1996 [9] H. Zheng, A. Lozano, and M. Haleem,”Multiple ARQ processes for MIMO systems ,” IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, vol. 3, Sept. 2002, pp. 1023–1026
10
2x4Alamouti, 64QAM 3x4 Blast, QPSK
−1
10
V. ACKNOWLEDGEMENT
−2
10 BER
The work on this testbed has been strongly supported by Analog Devices Inc.
−3
10
−4
10
−5
10
0
Fig. 10. test
5
10
BER for
15 20 25 Eb/N0 per Rx Antenna, dB
30
35
40
Alamouti vs. BLAST in an outdoor mobile
Error performance in indoor test
0
10
2x4 Alamouti 64QAM, R=6 bpc 3x4 BLAST QPSK, R=6 bpc −1
10
−2
BER
10
−3
10
−4
10
−5
10
−6
10
0
Fig. 11. test
5
BER for
10
15 20 Eb/N0 per Rx Antenna, dB
25
30
35
Alamouti vs. BLAST in an indoor stationary
[2] J. C. Guey, M. P. Fitz, M. Bell, and W. Y. Kuo, “Signal Design for Transmitter Diversity Wireless Communication Systems Over Rayleigh Fading Channels,” IEEE Trans. Commun., vol. 47, no. 4, pp. 527–537, April 1999. [3] S. M. Alamouti, “A Simple Transmit Diversity Technique for Wireless Communications,” IEEE J. Select. Areas in Commun., vol. 16, no. 8, pp. 1451–1458, Oct. 1998. [4] H. Jafarkhani, V. Tarokh and A. R. Calderbank, “Space-Time Block Codes from Orthogonal Designs,” IEEE Trans. Inform. Theory, vol. 45, no. 5, pp. 1456–1467, July 1999. [5] S. Siwamogsatham and M. P. Fitz, “Improved High-Rate Space-Time Trellis Codes via Orthogonality and Set Partitioning,” Wireless Communications and Networking Conference, vol. 1, pp. 264–270, 2002. [6] H. Jafarkhani and N. Seshadri, “Super-Orthogonal Space-Time Trellis Codes,” IEEE Trans. Inform. Theory, vol. 49, no. 4, pp. 937–950, April 2003. [7] Z. Liu, “Design and Implementation of Transmit Antenna Diversity in Wireless Communication Systems,” Master’s thesis, The Ohio State