SMART ANTENNA SOFTWARE RADIO TEST SYSTEM Peter J. Green and Desmond P. Taylor University of Canterbury Department of Electrical and Electronic Engineering Christchurch New Zealand Abstract This paper covers the concept, architecture, development and demonstration of a Smart Antenna Software Radio Test System (SASRATS). SASRATS was designed and developed as a functional and flexible system to facilitate the field testing of space-time processing architectures and algorithms. It also facilitates the correlation between theoretical, simulated and measured performance. The SASRATS architecture has the capability of real time implementation of signal processing algorithms. The paper also covers the testing, debugging and calibration of SASRATS prior to commissioning. The paper will cover the field testing and verification of two space processing algorithms. It is used to verify the work done by the authors and others in the field of signal enumeration and blind beamforming.
1. Introduction Space-time processing algorithms developed in most research labs around world are simulated on a computer using programs such as Matlab under controlled conditions incorporating mathematical models that represent real world phenomenon. There is abundant use of abstract mathematics, reliance on computer simulations and almost complete lack of experimentation with real-life data [1]. With a few exceptions, many space-time algorithms are never field tested. Many algorithms may perform poorly or fail completely in the field because the assumptions made in the mathematical models do not accurately represent the real physics of the problem at hand. Sensitivity of the algorithm to deviations from the assumed mathematical model may also lead to instability [1]. Thus the importance of testing the algorithms in the field under real conditions even under limited or reduced scale cannot be further emphasized.
2. SASRATS Architecture Key design specification of SASRATS are operation at the 915 MHz Industrial, Scientific and Medical (ISM) band,
digital ‘software’ radio programmability and critical for spacetime processing, a flexible data bus architecture for fast and timely transfer of data between receivers, digital signal processors (DSP) and personal computer (PC). The bus must also allow for future addition of receivers to the system. The general system architecture of SASRATS is shown in Figure 1. The basic system consists of 5 digital receivers, 2 digital transmitters, 7 monopole antennas and auxilliary components such as power splitters, power supplies and oscillators. A personal computer is used for transmitter (TX) and receiver (RX) programming, DSP code assembly, data acquisition, storage and batch signal processing using Matlab. A common 10 MHz reference oscillator provides synchronization for the whole system. The TX however, can be configured to work independantly using a separate PC, oscillator and power suppply for long distance operation and synchronization software in the receiver must then be used. For beamforming experiments, a set of 5 monopole anten nas spaced apart are mounted on a large metal ground plane. The antennas are separated from the receivers by at least 20 metres of RF coaxial cables to minimize any interaction between the antenna array panel and the receivers. The transmit antenna cables are between 20 and 60 metres long. To minimize EMI, each transmitter and receiver is housed in a rugged metal enclosure and within the each enclosure, the analog and digital circuits are physically separated by a metal shield. The SASRATS RX architecture consists of 5 receivers; one MASTER and 4 SLAVES. The Master receiver has a dedicated DSP for selecting and programming all slave receivers plus running slower real time DSP algorithms. A fast DSP is also present for running beamforming or diversity algorithms. Each of the 4 slaves has a fast DSP for space-time processing algorithms. A critical portion of the RX architecture design is the Special Communication Bus (SCB). Designed with space-time processing in mind, the high speed bus allows for sharing of complex baseband data between all radios, DSPs and PC, programming between master and slave radios and synchronizing the operation of all the receiver digital down converters. The Special
rLsutvrLwvvtyx
d e f g h ig Y Z [ \ ] ^j _k `l ma bn oq` c p PLSQMTN OU VINX P N QPR PWS
¡ ¢ £ ¤ ¥£ ¦§ ¨© ª «¬ ®¯ °
±L²u³´±Lµ´´³u¶
K I L
¿ · À ¸ Á ¹ ºÂ û ¼Ä Ž Æ ¾ÇLÈ É Ê ËÌ Í Î ÏÐ Ñ => ?> @ AB CG DED> FDC Ú Ò ÛÓ ÔÜÕ Ü ÖÝ × Þ Ø Ü Ù ß HIAKJ @LDC à á â ã ä å ã æLç å â è é ê ë ì í îï ð ñ ò ó ô õ í ñ ö ÷ ø ù úû ü ý þÿ
z { | } ~L
EF GH I J FLK M N I O PQ Q K M
¶ · ¸¹º» ¼ ½ ¾¿¾À Á  ¼ Ã Ä Â ½ Á Ä Å Æ ÄÆ Å Ä Ç È É ÊË Ì Í ÎÏ ÐÑ Ò -. / 0 1 2343235 63 ! "# $% & '( ) *+ ) ,
¡¢£¤¢¢¡¥
# ! " $ "
! "#$ % &' ( ) * + ,-, . / , 0 1 2 3 4 5 6 4 78 6 3 9 :
%& '& ( )* +/ ,-,& .,+ 0 *).,
; ? @A B CDE F GH I J KLM MNOMP Q R S T U V T WYX V S Z [
12 32 4 56 7; 8982 :87 < 65:8
\ ]^ _ ` ab c def g hi j k lmn nopnq r s t u v w u xYy w t z {
¦§©¨ª«¬ªª¨©
®¯°±³²´±±°µ
| } ~
Ó Ô ÕÖ×Ø Ù Ú Û¿ÛÜ Ý Þ Ù ß à Þ Ú Ý à á â àâ á à ã ä åæ ç è éê ëì í
7 8 9 : ; ? @ A B C=D ¡
î-ï ð ñ ò ó ô õ öö÷ ø ù ô ú û ù õ ø û ü ý û¿ý ü û þ ÿ
R S TVU W TYX Z [ R S \ ] X]^ _W ` Z \ U W TYX a [ ] ^ b c d c ef c b gh j i c bjk l g g m n o p q r s tYu o o v p w x yjv pzs { w w | }~
L YjjVYY = jz
ceivers are programmed in time division multiplexed mode on the SCB. The data is picked up by the appropriate DSP for processing. Algorithms for space-time processing and synchronization can be performed in real time. Advanced field tests can use the real-time mode after algorithms are verified to work on batch data collected in the batch or pseudo real-time mode on Matlab. A key SASRATS RX design feature is digital downcon¢¤£¥ ) of 70 MHz. The version at an intermediate frequency ( 70 MHz signal is digitised and the process of mixing, decimation and filtering to recover the I and Q data is performed in the digital domain by a dedicated VLSI chip. The chip gives the system flexible ‘software radio’ programming features for tuning, phase correction between receivers, output bandwidth and resolution[2]. Basic to the SASRATS RX architecture is the use of ¢¦£§¥ ¢¦£§¥©¨«ª¦¬ bandpass sampling at [3]. With MHz and Nyquist sampling bandwidth( ¯® )=10 MHz, conventional ¢¤° dictates a sampling frequency of at least 150 MHz. With bandpass sampling, the desired signal is sampled at a rate that meets the Nyquist criterion for the signal’s bandwidth but not for its absolute frequency. The bandpass sampling technique generates multiple images of the original signal all the way to baseband and the digital downconverter is programmed to tune to one of these images. To work prop¢ ° erly, must be within the limits of[4]
Fig. 1. SASRATS Architecture Communication Bus has two ports to interface with various peripherals. The SCB-Serial port allows high speed serial data transfers between the receivers and a high speed digital input-output (DIO) card on the SASRATS PC and/or the enhanced synchronous serial interface (ESSI) ports available on most popular DSPs’. The DIO card allows up to 15 receivers to be operated simultaneously. The SCB-Parallel port is available for wider bandwidth applications, allowing parallel transfer of data to PC or DSP. The parallel port behaves as a memory mapped external peripheral device. The bus also offers flexible output data options for operation in batch, pseudo real-time or real time modes. In batch and pseudo-real time mode, the serial in-phase (I) and quadrature-phase (Q) data from all five receivers are read simultaneously by a high speed digital input card. In batch mode, large amounts of data are stored on hard disk for delayed processing. In pseudo real-time processing mode, a Matlab program collects small batches of data for immediate processing. Pseudo real-time processing is extremely useful for receiver calibration of amplitude and phase between the receivers and especially for the testing of new algorithms running on Matlab. In the real-time mode, serial I and Q data from all re-
¢ £§¥²± ´
¢¶°
³® µ
µ
¢ £§¥¸· ´
³® ·º¹
(1)
where ´ is any positive integer. The bandpass sampling frequency chosen for SASRATS is 40 MHz and allows both the ADC and the digital down converter to process the signal at much lower data rates thus providing the option of using a less expensive converter. On the transmitter side, a key SASRATS TX design feature is real-time baseband processing of I and Q samples. Flexible software allows for programmable modulation, pulse shaping and baud rate options. It is simple to use. The required TX options are choosen, the DSP code assembled and loaded into the DSP and run.
3. Testing and Calibration Critical in beamforming algorithms is the accuracy and precision (stability) of phase measurements of the transmitted signal picked up by the 5 monopole antennas at the receivers. Multipath makes indoor laboratory measurements useless and the only option we have is to go to a wide open unobstructed field which provide better channel characteristics (relatively free from multipath) for the received signal. For field testing, SASRATS is set up in a station wagon and driven to the test site daily. The first five field tests were
À¸Á¸Â
À
»²ÃÅÄ
À
(2)
Á
where is the desired received signal free from mutual À coupling effects and is the received signal containing mutual coupling interactions from neighbouring antennas. A general impedance coupling matrix » can be created for an array of Æ elements where ¹Ê·ÌËÎÍÏÍ ËÑÐ ËØÒÙÍ ËÑÐ ÇÈ È È ¨ » É
.. .
ËÑ×ÜÍ ËÑÐ
ËÎÍÓÒ ËÑÐ ¹Ê·ÚËØÒÛÒ ËÑÐ
.. .
ËÑ×ÊÒ ËÑÐ
ÔÕÔÖÔ ÔÕÔÖÔ
..
.
ÔÕÔÖÔ
ËÎÍÏ× ËÑÐ ËØÒÙ× ËÑÐ
Ý=Þ Þ
ß
Þ
(3)
¹Ê·ÚËÑ×Ü× ËÑÐ
and àâá is the load impedance. A ãåäæã impedance coupling matrix is needed for the SASRATS 5 element linear array. The important aspect is that the mutual coupling compensation is independent of the direction of arrival of the incoming wave provided that the antennas are single-mode elements where the element aperture currents (electric or magnetic) may change in amplitude but not in shape, as a function of radiation direction. The antennas used in SASRATS are monopoles and are single-mode elements and can be compensated using the technique above. As SASRATS is a digital system, compensation using the » ¼¿¾ matrix is an easy process once the coupling coefficients are determined. Compensation is done prior to beamforming and enable subsequent processing to use ideal element signals free from mutual coupling effects. The angle of arrival measurements was carried out on 4 antenna pairs; 1-2, 2-3, 3-4 and 4-5 and compared with the
Figure A 15
Error (degrees)
10 5 0 −5 −10 −15 −60
−40
−20
0
20
40
60
−20 0 20 Actual Angle of Arrival (degrees)
40
60
Figure B 15 10
Error (degrees)
mainly tests to characteristize and calibrate SASRATS. Many unsuspected problems were encountered. A major field problem observed is the significant phase difference between adjacent antenna pairs as the angle of arrival of the transmitter departs from normal to the array. Theoretically, the phase difference between any two sets of adjacent antennas along a linear array of antennas should be identical for a particular angle of arrival of the transmitted signal. Real antenna systems depart from this ideal behaviour. This departure from ideal is caused by interaction between neighbouring antennas. This effect is called mutual coupling. The impedance of an antenna is changed by the electromagnetic coupling of neighbouring antennas and therefore changes its impedance and therefore amplitude and phase. The degree of mutual coupling is also dependent on the type of antenna, their placement position on the array and the distance separating them. Antennas at the ends of the array see the least interaction because there are antennas only to the left or right but not both. The effects of mutual coupling can be compensated by multiplying the received signal by the inverse of the coupling matrix »½¼¿¾ [5] such that
5 0 −5 −10 −15 −60
−40
Fig. 2. Effect of mutual coupling on a 5 Element SASRATS Linear Antenna Array true angle of arriva1. This is done in 5 degree steps between -45 and 45 degrees from antenna normal. Figure 2A shows the error between measured and actual angle of arrivals for the four pairs of antenna with no mutual coupling compensation. The error range of the 4 antenna pairs broadens as the angle of arrival departs from normal and can exceed 10 degrees. Figure 2B shows a much narrower error spread after mutual coupling compensation is used and thus validates the compensation method. Another major field problem observed is the significant drift in absolute phase over changes in ambient temperature. The 5 antennas on the linear array panel are connected via 5 x 20m coaxial cables. The environment( sun, wind, cloud cover) can cause the panel and cable temperature to change significantly, changing its dielectric properties and phase characteristics. A 120 degree phase change over a 5 minute period in adverse conditions has been observed!. The solution is to measure differential phase between two adjacent antennas on the panel. With 5 antennas, 4 differential phases can be computed. This is based on the assumption that all antennas and coax cables ‘see’ or undergo the same variation over temperature. With careful setup, this assumption is found to be true empirically and we managed to get phase variations down to less than 1 degree over the most adverse of field conditions encountered. A less significant field problem is a slow drift in absolute phase measurements over time. This problem was traced to signal generators which provide the clock and local oscillator signals to the system. Even though the generators are driven from the same reference 10 MHz clock, there is observable phase drift (2 degrees per minute) at startup from cold. This is attributed to the phase noise parameter of the signal generators and takes a few hours to stabilize. After a 24 hour warmup, the drift in absolute phase is less than 10
Figure A 7 6
Actual Signals
degrees per hour. Although the absolute phase drift is substantial, the differential phase error performance between two receivers critical to beamforming experiments, is excellent at better than +- 0.3 degree.
5 4 3 2 1
4. Verification of algorithms in the field
0
0
10
20
30
40
50
60
70
80
90
100
60
70
80
90
100
Figure B 6
Estimated Signals
The testing and verification of two different space processing algorithms will be covered. The first is a robust algorithm developed by the authors called the dynamic signal enumeration algorithm (DSE)[6] and second is a blind beamforming algorithm[7] using cyclostationarity.
4 2 Error Trace = Estimate − Actual
0 −2
0
10
20
30
40
50 Trial Number
4.1. Dynamic signal enumeration algorithm Fig. 3. Simulated performance of algorithm to random number of users at nominal SNR (10dB) Figure A
Actual Signals
3 A @ +15 degrees, −56 dBm B @ − 15 degrees, −56 dBm A & B 10 dB SNR
2
A only
−1
A&B
50
A only
B only
No Signal
No Signal
0
A&B
B only
1
0
100
150
200
250
300
All Off
350
400
Figure B 3 Error Spikes Error Spike
Computed Signals
Many high resolution parameter estimation algorithms for array processing critically depend on knowing the number of signals present at the antenna array output. Their performance is dependent on perfect knowledge of this number. These methods assume that the number is known but in practice, this knowledge of the number of incident signals on an antenna array is not known apriori and must be determined. The DSE algorithm not only enumerates the incident signals impinging on a uniform array but is robust to the extent of correlation of the signals in a Rayleigh flat fading channel environment. The details of the DSE algorithm is beyond the scope of this paper but can be found in [6]. The simulated performance of the algorithm is shown in Figure 3. It is simulated using a 12 element linear array model. The elements are equally separated at a distance of and receive 3 independant Rayleigh fading sources and ±èç ±íì ¬ ±å¹ ¹ ¬ ã¶éëê é¤ê ã¶é¦ê é¤ê ãîé 3 coherent sources arriving at ì ¬ and é . The number of signals vary at random from 3 to 6 users but remain in one particular state over 4 trials. The frequency of the signals are all equal to simulate the arrival of the main and reflected signals and also cochannel signals. The number of snapshots per trial is 100 and 100 trials were conducted. Figure 3A indicates the actual number of signals and Figure 3B shows the estimated number of sources and the error between the actual number of sources and the estimate. Performance was at 93 percent correct estimation at 10dB SNR. The field performance using SASRATS working in pseudo real-time mode is shown in Figure 4. Here the 5 ±å SASRAT ¹ ã¶é and antennas receive 2 coherent sources arriving at ¹ ãîé , 10 metres away. The SNR is adjusted to be equivalent to 10dB SNR. The number of signals vary at random from 0 to 2 users. The number of snapshots ( ï ) per trial is 100 and 420 trials were conducted. Figure 4A show the actual transmitted signals and Figure 4B is the number of signals enumerated by the algorithm at the receiver with 4 error spikes visible. The enumeration accuracy is 99 percent
2
1
0
−1
99% Correct Estimation 1% Over Estimation 0
50
100
150
200 250 Trial Number
300
350
400
Fig. 4. Performance of algorithm in field test at 915 MHz with no underestimation. Performance is better than simulated because the sources experience mild Rayleigh fading in the field experiments. 4.2. Adaptive Cyclostationarity Beamforming Beamforming is the process of summing the weighted outputs from the array of antennas using some prescribed criterion of optimality[8] with the objective of optimizing the beamformer response so that the summed output contains minimal noise and interference. The beamforming process is made adaptive to respond to changes that always occur in the real environment using algorithms such as the leastmean squares (LMS) algorithm. Most of these algorithms and criteria of optimality require an explicit reference signal such as a training sequence or a known pseudo-noise
0
code from the received signal. A blind adaptive system on the other hand, generates its own reference from implicit or hidden characteristic of the desired signal. Cyclostationarity is one such implicit characteristic.
ñ
û
þÓÿ
¨
¨óòô õÕö÷ø¶ù¶ú
±üûîýÑþÓÿ ô ÷
(4)
o
Beamforms at 30
−20
Gain ( dB )
A signal with cyclostationarity has the statistical property of correlating with either a frequency-shift or complexconjugate version of itself for certain values of frequency shift. Cyclostationarity avoids the need for training signals, estimates of directions of arrival and knowledge of noise characteristics. Early cyclostationarity beamforming algorithms are slow to converge but an algorithm proposed by Castedo and Vidal[7] overcomes this problem by using cyclostationary signals to generate spectral lines after going through a non-linear transformation. It chooses the coefficients ð which minimizes the cost function
−10
−30
−40 Compesated Uncompensated −50 o
Nulls at 0
−60 −90 −80 −70 −60 −50 −40 −30 −20 −10 0 10 20 30 Angle from normal ( degree )
40
50
60
70
80
90
Fig. 5. Performance of cyclostationary beamforming algorithm in field
þzÿ
ð is the complex array output, and where selected according to the order and frequency of the spectral line generated by the desired signal.
Field results using SASRATS are shown in Figure 5. Two signals; an unmodulated carrier at 0 degrees and a de· ì ¬ degrees are received sired BPSK signal coming from by the antenna array. Using the batch mode of operation, 6000 samples were stored on harddisk and later processed. The algorithm converges and the plots shows a sharp null at 0 degrees and beamforms (0 dB gain) at the desired BPSK signal arriving from 30 degrees. The dotted and continuous line plots show the response of the beamformer without and with mutual coupling compensation. The presence of mutual coupling changes the position and reduces the depth of the nulls. The adaptivity of the algorithm maintains the beamforming performance at the desired BPSK signal with a degradation in the attenuation of the interferer. This result further emphasises the need for mutual coupling compensation for best performance.
5. Conclusions The paper has presented a successful design and development of a novel smart antenna test system with a flexible bus and operational architecture. Also presented are setup and calibration procedures developed to compensate the negative effects of antenna mutual coupling, changing environmental conditions and oscillator quality with the ultimate goal of achieving best phase measurement accuracy and resolution possible. SASRATS has been successfully commissioned and has verified the operation of two array processing algorithms in the field.
6. References [1] Simon Haykin, “Signal processing: Where physics and mathematics meet,” IEEE Signal Processing Magazine, vol. 18, no. 4, pp. 6–7, 2001. [2] Harris Semiconductor, Ed., Digital Signal Processing Databook 1994, Harris Semiconductor, 1994. [3] N.L. Scott R.G. Vaughan and D.R. White, “The theory of bandpass sampling,” IEEE Transaction on Signal Processing, vol. 39, no. 9, pp. 1973–1984, 1991. [4] Harris Semiconductor, Ed., 1996 Wireless Communications Design Seminar Handbook, Harris Semiconductor, 1996. [5] Hans Steyskal and J. S. Herd, “Mutual coupling compensation in small array antennas,” IEEE Transactions on Antennas and Propagation, vol. 38, no. 12, pp. 1971–1975, Dec. 1995. [6] P.J. Green and D.P. Taylor, “Dynamic enumeration algorithm using array antennas,” Acoustics, Speech, and Signal Processing, 2001. Proceedings. 2001 IEEE InternationalConference on, vol. 5. [7] L. Castedo and A.R. Figueiras-Vidal, “An adaptive beamforming technique based on cyclostationary signal properties,” IEEE Transactions on Signal Processing, vol. 43, no. 7, pp. 1637–1650, July 1995. [8] J. Litva and K.Y.L. Titus, Eds., Digital Beamforming in Wireless Communications, Artech House Publishers, 1996.
