Deterministic RF Nulling in Phased Arrays for the

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. XX, NO. Y, MONTH 1999

100

Deterministic RF Nulling in Phased Arrays for the Next Generation of Radio Telescopes

Bart Smolders, Member, IEEE, and Grant Hampson

The authors are with the Netherlands Foundation for Research in Astronomy (NFRA), Dwingeloo, The Netherlands, E-mail: [email protected]. October 4, 1999

DRAFT


101

Abstract A requirement of the next generation radio telescopes for astronomy is its ability to cope with the forever increasing problem of Radio Frequency Interference (RFI). Unlike conventional fixed parabolic receivers used currently in astronomy, the application of phased-array beamforming techniques opens the possibility to spatially null RFI in the RF domain prior to signal digitisation. This paper presents results from the second phased-array experimental demonstrator, the One Square Metre Array, on calibration and RF nulling performances. The approach is to deterministically null known RFI in the RF beamforming domain and adaptively remove the remaining RFI in the digital beamforming domain. A novel array calibration technique called the Multi-Element Phase toggle technique (MEP) is presented, which allows a fast and very accurate calibration of wide-band phased-array antennas. Array calibration is shown to determine the extent to which RFI can be removed by experimental verification of simulated null depths.

Keywords Phased Arrays, Radio Astronomy, RF Beamforming, Array Calibration, Spatial Nulling.

I. Introduction The international radio-astronomy community is currently making detailed plans for the development of a new radio telescope: the Square Kilometer Array (SKA) [1]. This instrument will be two orders of magnitude more sensitive than telescopes currently in use. A clear scientific case has been formulated for the development of this radio telescope in the frequency range of 200 to 2000 MHz with sub-microJansky sensitivity [2]. Using the latest receiver technology the system noise temperature is dominated by the sky and an increase in sensitivity can only be achieved by using a larger telescope aperture. The specified sensitivity then determines that the telescope should have a total collecting aperture of approximately one square kilometer. Additionally, SKA will have to operate in an environment polluted by Radio Frequency Interference (RFI). This is due to the rapidly expanding use of the frequency spectrum for telecommunication purposes. One of the options for the new radio telescope is to use a phased array consisting of more than one hundred million receiving antenna elements. Phased arrays offer the unique capability to suppress unwanted RFI signals effectively by using deterministic and/or adaptive spatial nulling techniques. In addition, multiple beams can be formed simultaneously with

October 4, 1999

DRAFT


102

no loss in sensitivity. This permits several independent observations to be made by different groups of astronomers at the same time. The top-level architecture of SKA and the corresponding distribution of the 32 stations over a region of 300×300 km is shown in Figure 1. Each station is a phased-array antenna consisting of more than a million receiv-

Station 1

2

3

32

50km

300km

Correlator Imaging

(a) Top-level Architecture of SKA

(b) SKA Station Configuration

Fig. 1. (a) The SKA aperture synthesis telescope consists of 32 stations connected to a central correlator and an imaging process. (b) The collecting area of SKA is distributed over 32 telescopes. The central region provides sensitivity and the outlying stations resolution. (Not drawn to scale.)

ing elements connected by a hierarchical beamforming network. Resolution requirements impose an aperture synthesis telescope configuration. The sensitivity is determined mainly by the dense ellipse-like concentration of the telescopes in the central region. The remaining remote stations are distributed over an extent of 300 km which allows sub-arcsecond imaging [2]. In order to be able to detect weak noise signals from distant radio sources it is crucial that man-made RFI signals are suppressed within each station. Each station will comprise of a hierarchy of beamformers operating on both RF and digital signals. Field tests at the Westerbork Synthesis Radio Telescope (WSRT) show that the maximum RFI level from fixed man-made sources (for example TV towers and mobile telephone base stations) at the input of each receiving element of SKA is approximately 70 dB above the expected thermal October 4, 1999

DRAFT


103

noise level [3]. If an Analogue to Digital Converter (ADC) is used after each individual receiving element at least 14 bits would be required, assuming that 2 bits of the ADC are below the thermal noise level. This already shows that before any digital beamforming can occur, the RFI sources need to be suppressed by other means. Frequency filtering is not very effective since a wide instantaneous bandwidth is required in SKA (>200 MHz). Our proposal is to perform analogue RF beamforming with at least 64 receiving elements and apply deterministic RF nulling before the signals are digitised with an ADC and further processed in a digital beamformer. RF deterministic nulling can be used if the location and frequency of the RFI sources are known a priori. Special attention is given in this paper to the variation of the shape of the main beam versus the null location. The latter is of great importance in radio astronomy since very long integration times (up to several hours) are used to increase the signal-to-noise ratio and/or to create a radio map of a part of the sky. This can only be done accurately when the atmospheric and instrumental errors are calibrated [4]. This means that the beam pattern of the telescopes should be known a priori within a certain tolerance. In this paper we present an experimental phased-array system called the One Square Metre Array (OSMA) [5] which was developed to experimentally verify the effectiveness of spatial nulling for radio astronomy applications. OSMA is a scale model with 64 active antenna elements. Measured results of deterministic RF nulling with the OSMA system are presented. It has been found that effective spatial RF deterministic nulling can only be achieved when the amplitude and phase responses of each receiver element is known accurately. A new calibration procedure was developed which allows groups of elements to be calibrated simultaneously by using FFT signal processing techniques [6]. A special characteristic of this technique is that it doesn’t require high isolation between the receiving channels. The proposed ”Multi-Element Phase toggle” (MEP) calibration method is an extension of a method proposed previously [7]. II. Deterministic RF Nulling in SKA Due to the large number of antennas in SKA the physical beamformer implementation will consist of several levels. A possible hierarchical beamforming architecture is shown in Figure 2. Up to the tile level (approximately 100 antennas) beamforming will be done in October 4, 1999

DRAFT


104

analogue circuitry, which is called RF beamforming in this paper. The outputs of the tiles are digitised and further processed by a digital beamforming system. Finally, the output signals from the stations are correlated with each other. By collecting the correlated data within a certain time period, an image of a small part of the sky can be obtained using well-known image processing techniques [4]. element

RF

sub-tile

1

10

Σ αi A i Σ αi A i Σ βiBi

Σ αi A i

Deterministic Nulling

2

tile

10

sub-domain

10

domain

104

sub-station

105

station

106

Digital

Fig. 2.

0

10

3

Σ χiCi

Adaptive Nulling

Σ δ i Di Σ ε iE i Σ φiF i

Hierarchical beamforming architecture of a SKA station. Up to the tile level (102 elements)

analogue RF beamforming with deterministic nulling is implemented. The remaining beamformers are all digital, with the first stage having adaptive nulling.

Spatial nulling in the proposed architecture of a SKA station can be applied in several domains, i.e., the RF and digital domains. In the digital domain it is straight forward to use a sophisticated adaptive beamformer that doesn’t require a priori knowledge of the locations and signatures of the interfering sources. On the other hand, in the RF domain it is far more complex to use some form of adaptivity. Therefore, deterministic nulling will be used in the RF domain in order to suppress fixed and known interfering sources. Deterministic RF nulling algorithms used to determine the optimal complex weights are well-known from literature and will not be repeated here [7], [8], [9]. The basic idea is to subtract cancellation beams from the quiescent beam in such a way that the interfering sources are canceled out. Deterministic RF nulling will only be successful for radio astronomy applications when the number of interfering sources is much smaller than the degrees of freedom. From field tests it was observed that not more than two or three interferers fall into the same IF frequency band of interest [3]. This indicates that deterministic

October 4, 1999

DRAFT


105

RF nulling can be used on a tile level without a serious degradation of the main beam characteristics. The latter will be investigated in Section V. III. OSMA: An Experimental System for the Square Kilometer Array (SKA) As a lead up to the new radio telescope SKA, several prototype phased-array systems were defined and built. The initial stage is called the Adaptive Antenna Demonstrator [10] and followed by the One Square Metre Array (OSMA). The third stage is called the THousand Element Array (THEA) and is currently being developed [11]. OSMA is a planar phased-array receive-only antenna with a mixed analogue and digital adaptive beamforming architecture operating in the frequency range of 1.5 to 3 GHz. The linearly polarised array consists of an 8 by 8 element active centre region surrounded by two rows of passive elements. The total number of antenna elements is therefore equal to 144. Figure 3(a) shows the front view of the OSMA system inside the anechoic room. The 35mm

35mm

35mm 25mm Quarter Wavelength Transformer Antenna Ground Plane Microstrip Line Connector Ground Plane Connector

(a) OSMA Antenna Array

(b) Single Antenna Element

Fig. 3. (a) A photograph of the OSMA array inside the NFRA anechoic chamber. OSMA is mounted on a positioner which can orientate it in two directions, azimuth and elevation, with respect to a transmit horn. (b) A single element of the bow-tie array showing the integrated microstrip balun. The antennas are not manufactured individually, but in multiples of four.

October 4, 1999

DRAFT


106

array is built up from broadband bow-tie elements with an integrated balun printed on RO 4003 substrate with a permittivity of r = 3.38 [12]. The array elements are on a square grid with an inter-element spacing of 75 mm. The array is backed by a ground plane which is rounded at the edges to reduce diffraction effects. A single element of the bow-tie array is shown in Figure 3(b). The bow-tie array can be used over a 3:1 bandwidth with an average scan loss smaller than 1 dB for scan angles less than 50 degrees w.r.t. broadside [13]. The top-level beamforming architecture of OSMA is illustrated in Figure 4 as well as the measurement configuration. The first stage of the beamforming hierarchy is when the 64 active antenna elements are connected to 16 RF beamformer units (RFBF I). Each RF beamforming unit processes data from four antennas. The passive elements are terminated with matched impedance loads. The outputs of the RFBF I units can be connected to both a 16-channel adaptive digital beamforming (ADBF) unit [5] or to a second stage 16channel RF beamforming unit (RFBF II). The OSMA system can be used in two different modes: a RF beamforming mode, or a mixed RF and digital adaptive beamforming mode. Both modes of operation can be used simultaneously, either generating two RF beams or alternatively one RF and two digital beams. In this paper only the RF beamforming mode will be considered. The control and data acquisition of the OSMA system is managed by Matlab in conjunction with an experimental measurement system [14], from which all system devices can be controlled. Matlab provides a powerful interface to control and analyse data from the OSMA system. Graphical-User-Interfaces (GUI) provide real-time control and visualisation of beamforming experiments. The functional lay-out of the RFBF I units is illustrated in Figure 5 where the active components are denoted by LNA (Low Noise Amplifier), TDU (Time Delay Unit) and VAT (Variable Attenuator). The 4-bit switchable TDU units can be used as true-time delays reaching scan angles within a cone of ±40 degrees. Outside this scan region the TDU’s can be treated as phase shifters. In the latter case the maximum scan range is unlimited, at the expense of a maximum instantaneous bandwidth of 200 MHz. The range of the TDU’s is 0.5 ns, which corresponds to a phase range of 360 degrees at 2 GHz. The VAT has a dynamic range of approximately 25 dB and is controlled by an 8-bit digital

October 4, 1999

DRAFT


RFI

Antenna Elements

64

107

RFBF I x16

Transmit

16

16

Front-end Controller

Serial Links

Positioner θ

RFBF II x2

φ

Receiver x16

2

RF Beam Network Analyzer Synthesiser

Fig. 4.

ADBFS x2

Digital Beams

2

to any RF link

GPIB Link

DAQ x2

Top-level beamforming architecture of OSMA including the measurement setup. A network

analyzer and synthesizer generate the required source signals. The 64 array outputs are combined into 16 signals using the first stage RF beamformers (RFBF I). The two identical outputs of this beamforming stage can be used to generate either RF or digital beams. The receiver contains frequency down-conversion and ADC circuitry. The adaptive digital beamforming system (ADBFS) outputs are connected to a data acquisition (DAQ) system.

word. A power combining network containing three two-stage Wilkinson combiners sum the received signals. Next, the output signal is divided into two identical signals in order to allow the system to work in the two different modes described previously. The RFBF I circuits are printed on the same material as the bow-tie antenna elements. A photograph of a complete RFBF I module is shown in figure 6. A digital control board remotely controls the settings of the TDU and VAT units. The RFBF II units are identical to the RFBF I units, however the LNA is no longer required in the RFBF II units. Figure 7 illustrates the measured phase for each of the sixteen TDU settings, in agreement with the expected linear relationship of phase with frequency. This phase relation makes it possible to steer the RF beam of the array without squinting over a large frequency range. IV. Multi-Element Phase Toggle Calibration (MEP) The purpose of our calibration procedure is to correct for all amplitude and phase errors that occur in the analogue beamforming structure of the OSMA system. These are errors

October 4, 1999

DRAFT


LNA

VAT

TDU

LNA

VAT

TDU

108

Σ Σ LNA

VAT

TDU

LNA

VAT

TDU

2

BF Outputs

Σ

Beamformer Module Controller

Fig. 5. Functional layout of the RF beamformer I units (RFBF I). The active components are denoted by LNA (Low Noise Amplifier), TDU (Time Delay Unit) and VAT (Variable Attenuator).

Low Noise Amplifier

8-bit Attenuator

4-bit Time Delay Unit

Wilkinson Power Combiner

RF Beamformer Digital Control Card

Control Connector Power Connector

RFBF Outputs

Wilkinson Power Splitter

Fig. 6.

Photograph of the RFBF I module, consisting of four identical channels containing a LNA, 8-

bit variable attenuator, 4-bit time delay unit, four 2-stage Wilkinson power combiners and a digital control card.

October 4, 1999

DRAFT


109

Phase of TDU settings (deg)

200 0 −200 −400 −600 −800

1

2 3 Frequency (GHz)

4

Fig. 7. Measured phase shifts for each TDU setting (RFBF I module) over a range of frequencies relative to TDU setting 0 (reference channel) at maximum gain setting. The frequency range of interest is between 1.5 and 3 GHz, where the phase is almost linear with frequency.

that occur in the antenna elements and the RFBF I and II units. A schematic diagram of the calibration principle of OSMA is shown in figure 8. Two types of signals can be injected into each array element: (1) external signal generated by a far-field source, and (2) a signal injected through mutual coupling with the calibration elements. Array Element k com

Plane Wave FF H k

Hk

Reference Signal (Far Field)

Test Signal

Calibration Element Signal

cal

Hk Reference Plane

Fig. 8. Calibration principle used in the OSMA system consists of two signal sources: calibration elements and a far field signal source. The desired result is to measure the complex gains of the beamforming system, Hkcom .

The calibration procedure is based on the relationship between the transfer function at the output of element k from one of the calibration elements, Hkcal · Hkcom , and also the transfer function due to one or more incident plane waves, HkF F · Hkcom . The first part October 4, 1999

DRAFT


110

of the calibration procedure is conducted once inside a high-quality anechoic room. Two separate measurements occur; the first using the calibration elements and the other using a far field source. This results in two measured complex signal coefficients per element: ak = HkF F · Hkcom

(1)

bk = Hkcal · Hkcom

(2)

which can then be used to determine the element coefficients, ck , such that ck =

ak HFF = kcal bk Hk

(3)

The second part of the procedure occurs in the real out-door environment where the ˆ com . A far field element electronics coefficients, Hkcom , are likely to change and given by, H k source no longer exists with enough strength to be received by a single element and thus the calibration elements are used. Hence, measure using the calibration elements ˆ com ˆbk = H cal · H k k

(4)

By combining the results from both calibration steps it is possible to relate the measured gains and phases of each element to an incident plane wave, HkF F . Following from equation 3 the following is obtained ˆ com = ˆbk · ck aˆk = HkF F · H k

(5)

The ratio aˆk /ak gives the required complex gain variations of each element. The discussion so far has however excluded how to accurately obtain the complex coefficients ak and bk . The remaining part of this section develops a new technique known as MEP to measure these coefficients. The Multi-Element Phase-toggle (MEP) method is proposed here in order to calibrate groups of elements simultaneously. This is a significant improvement over other techniques which typically calibrate elements individually. Additionally, the technique has the advantage of reducing measurement errors through phase toggling. The method that will be proposed here is an extension of the technique developed by Lee et al. [7]. The number of elements, K, that can be calibrated simultaneously with the MEP method depends on the number of phase states, N, that are available (in OSMA this October 4, 1999

DRAFT


111

is N = 24 = 16.) To determine the offset amplitude, gk , and phase, φk , of K array elements, N measurements are required (one for each phase state.) The elements to be calibrated are set to their maximum gain and the remaining elements to minimum gain, which maximises the SNR. During the measurements the phase shifter setting of the elements that are being calibrated are toggled with a particular step frequency fk . The received signal, S(n) for (n = [0..N − 1]), from each measurement is now given by: S(n) = =

K−1 X k=0 K−1 X k=0

2πnfk gk exp (φk ) exp N 2πnfk ak exp N

!

(6)

!

for n = [0...N − 1]

(7)

where k is the element index, k = [0..K − 1]. The step frequency fk should be an odd number in order to step through all phase states, for example fk = 2k + 1. The complex gains of the K array elements, ak , can be calculated by solving the set of N equations given in equation 7. In the scenario that the phase shifter has a range of 0 to 2π, a Discrete Fourier Transformation can be used to solve for the gains. The required computation time is further reduced when the number of phase states is a power of 2, allowing implementation of the Fast Fourier Transformation (FFT). The complex gain, ak , of element k is then found in bin fk of the FFT spectrum. Once all the complex offsets have been determined for the array, a look-up table can be formed that corrects for the relative amplitude and phase offsets between the array elements. The MEP method has two main advantages. The first being that an average amplitude and phase offset of each element is measured, not just one particular phase setting. The other is that all interfering signals from other array elements (which are not phase-toggling) are located in bin zero of the FFT spectrum. The MEP technique allows bad isolation between the channels whilst not affecting the desired result. The OSMA RFBF I units have 16 phase states between 0 and 2π at 2 GHz. Given that fk is odd and OSMA contains 16 phase states at this frequency, then a maximum of eight elements (N/2) can be calibrated simultaneously. Sixteen measurements are required with eight active elements. The phase-toggle step frequencies used are fk = 1, 3, 5, 7, 9, 11, 13 and 15. Note that linearisation of the phase and amplitude of the TDU and VAT units is used during each measurement. The array is only calibrated for the broadside direction. October 4, 1999

DRAFT


112

Figure 9(a) shows an example of a measured FFT spectrum (amplitude and phase) with eight elements calibrated simultaneously. The other 56 array elements are set to their minimum gain setting (-25 dB down). The power level in bin 0 of the FFT is very high which is due to the other 56 elements in the array. This however does not influence the measured offsets significantly as the spectral leakage (into bin 1) is marginal. Figure 9(b) illustrates the measured amplitude and phase errors after a MEP calibration. The remaining errors

0

15

−5

10

−10

5

Phase Error [deg]

Relative Power [dB]

are less than half a quantisation step of the TDU (22.5◦ at 2 GHz).

−15

0

−20

−5

−25

−10

−30 0

5

10 FFT Bin

(a) Fourier Transform of Measured Calibration Data

15

−15 0

1

2

3 4 H−Plane Module Position

5

6

7

(b) Phase Error After Calibration

Fig. 9. (a) Measured FFT spectrum with 8 elements calibrated simultaneously. The contribution of the other elements fall into bin 0. Frequency is 2 GHz. (b) Measured phase errors after MEP correction (f=2 GHz) are less than ±0.5 LSB.

In the preceding example the MEP technique is used at 2 GHz where the TDU has 16 phase states between 0 and 2π. At other frequencies, above and below this is no longer the case. The complex offsets ak can be calculated by applying a DFT or by standard linear equation solver. The obtained accuracy at frequencies other than 2 GHz is equal to the accuracy obtained at 2 GHz. V. Experimental results from OSMA The following sections present results from the OSMA system. Firstly, antenna patterns are measured and compared with simulations in both the spatial and frequency domains. October 4, 1999

DRAFT


113

Null depths as a function of RFI spatial location are measured and compared with theoretical results. Finally, the main beam variations and directivity losses are measured. A. Measured Antenna Patterns The RF nulling performance of the OSMA RF beamforming system was measured in the frequency range of 1.5 to 3 GHz. The direction of the OSMA array positioner in the H-plane is given by θ and the direction in which a deterministic null is placed is θRF I . Figure 10(a) shows the measured H-plane antenna pattern at f=2 GHz, with and without a null placed at θRF I = 10◦ . The broadside direction (or location of the transmit horn) of the array is at θ = 0◦ . The main beam direction is given by θLOOK , and in this case, θLOOK = −10◦ . An uniform amplitude taper was used for the quiescent pattern, however deterministic nulling changes these amplitudes to position the null. The measured null

0

0

With Nulling No Nulling

−5

−10 −15

−15

Received Power [dB]

Received Power [dB]

−10

−20 −25 −30

−20 −25 −30 −35

−35

−40

−40 −45

Theory Measured

−5

−45

−60

−40

−20

0 θ [deg]

20

40

(a) Measured Patterns

60

−50

−60

−40

−20

0 θ [deg]

20

40

60

(b) Simulated and Measured Patterns

Fig. 10. (a) Measured H-plane antenna patterns with and without a null placed at θRF I = 10◦, θLOOK = −10◦ , f=2 GHz. (b) Measured and predicted H-plane beam patterns for a null placed at θRF I = 10◦ , θLOOK = −10◦ , f=2 GHz. The array is calibrated using the MEP technique for the measured beams, however the simulation uses ideal weights.

depth is more than -43 dB w.r.t. the main beam. Compared with the sidelobe level of the original beam the signal-to-interferer ratio has been improved by more than 30 dB. Figure 10(b) compares the theoretical and measured results in the case of a null placed at October 4, 1999

DRAFT


114

θRF I = 10◦ . The agreement between measured and predicted data is excellent. The frequency dependence of the antenna pattern of figure 10(b) is plotted in figure 11 over the entire frequency range of 1.5 to 3 GHz. The main beam is positioned at θLOOK = −10◦ over the complete frequency range as a result of the using true time delay beamforming. The measured frequency band for which the null depth at θRF I = 10◦ is more than -30 dB w.r.t. the main beam is equal to 320 MHz. This is in close agreement with the predicted bandwidth of 325 MHz according to the following expression [9]: √ 2f0 λ0 S ∆f = πL sin(θRF I )

(8)

where L=0.6 m is the length of the array and S is the null depth below the original sidelobe pattern. 1.5

0 −5

−15

Frequency [GHz]

2

−20 −25 2.5

−30

Received power [dB]

−10

−35 −40 3

Fig. 11.

−60

−40

−20

0 θ [deg]

20

40

60

−45

Frequency response of a measured H-plane beam pattern for a null placed at θRF I = 10◦ at

f=2 GHz, θLOOK = −10◦ . The circle indicates the zone where the deterministic null is placed.

B. Null Depth versus RFI Location The null-depth that can be realised in practical arrays is limited by the number of bits that are used to perform the complex weighting. Other errors, such as offsets or timevariations, will also affect the null depth, but can be minimised by using an appropriate calibration scheme (see Section IV). The average null depth that can be expected [15] is plotted in figure 12(a). From this figure it can be observed that with OSMA (4-bit October 4, 1999

DRAFT


115

quantised phase and 64 array elements) an average null depth of -37 dB w.r.t. the main beam can be made. However, in the principles planes of the array the null depth will be significantly higher (-28 dB), because groups of elements in the array will have the same phase settings. 6

−25

−20 dB −25 dB −30 dB −35 dB

5.5

−30

4.5 Null depth [dB]

Number of phase bits

5

4 3.5 3

−35

−40

2.5 2

−45

RMS value Measured

1.5 1

10

20

30 40 Number of array elements

50

60

(a) Expected Null Depths

Fig. 12.

−50

−60

−40

−20

0 20 Null location [deg]

40

60

(b) Measured Null Depths

(a) Theoretical RMS null depths (single null) obtained with deterministic RF nulling. These

results are obtained using quantised random phase settings, however the amplitude weighting is ideal. (b) Measured null depths w.r.t. the main beam versus null location, f=2 GHz, θLOOK = −10◦ . Dotted line indicates the RMS value of -32 dB.

The measured null depth versus RFI location for a fixed main lobe position is shown in figure 12(b). The main beam is again directed to θLOOK = −10◦ in the H-plane. The RFI location, θRF I , is scanned through the H-plane with a step angle of 5◦ . Note that the nulls are only placed in the sidelobe region of the antenna pattern. The measured RMS null depth in the H-plane is -32 dB, which is better than the expected value of -28 dB in the H-plane for a 4-bit phase quantisation (see figure 12(a).) This is due to additional amplitude and phase errors in the array, e.g. production tolerances and mutual coupling. This smears out the effect of quantisation errors on the antenna pattern and makes them more uniform over the entire hemisphere.

October 4, 1999

DRAFT


116

C. Directivity Loss and Main Beam Variation The measured and predicted directivity loss versus null location is illustrated in figure 13(a) for θLOOK = −10◦ in the H-plane at 2 GHz. The agreement between theoretical and experimental results is very good. The variation of the directivity loss is similar to the sidelobe structure of the original antenna pattern. The maximum directivity loss is approximately 0.5 dB and occurs when the null is placed in one of the first sidelobes of the antenna pattern. 0.6

Measured RMS=0.11 dB Theory RMS=0.12 dB

0.5 0.4

Relative main beam error (dB)

Directivity loss [dB]

0.5

0.4

0.3

0.2

0.1

0.3 0.2 0.1 0 −0.1 −0.2 −0.3 −0.4 −0.5

0

−60

−40

−20

0 20 Null location [deg]

40

60

(a) Measured and predicted directivity loss

−16

−14

−12

−10 θ (deg)

−8

−6

−4

(b) Main beam variations

Fig. 13. (a) Measured and predicted directivity loss versus null location, H-plane, f=2 GHz, θLOOK = −10◦ . (b) Measured deviation of main beam from reference (H-plane, f=2 GHz, θLOOK = −10◦ ) as a function of RFI null location (5◦ step size.)

Related to directivity loss is the variation of the shape of the main beam versus location of the null. This is of importance in radio astronomy since for accurate synthesis imaging calibration of the primary antenna pattern of the various stations is needed [4]. The measured variation of the main beam within the half power beamwidth versus null location is shown in figure 13(b). The measured data was normalised to a reference pattern. For most situations the variation is less than ±0.4 dB within the half power beamwidth (which was measured to be 12◦ for the quiescent pattern.) From figure 13(b) it is clear that the shape of the main beam mainly depends on the location of the ”sidelobe” of the cancellation

October 4, 1999

DRAFT


117

beam that falls into the main beam region of the quiescent beam. Therefore, the shape of the main beam with deterministic nulling can be predicted quite accurately. The latter makes it possible to compensate for this effect in the (digital) back-end electronics. VI. Conclusions The next generation of phased-array telescopes for astronomy faces many new challenges. The general approach to building such an instrument was given, including possible array geometries and beamforming hierarchies. A prototype demonstrator system called OSMA was presented in order to measure and illustrate results from a real phased-array implementation. The phased-array processing is partitioned into two types of processing: analogue (RF) and digital beamforming. The RF beamforming circuitry contains true-time-delays permitting wide band (approximately one octave) operation. It was shown that deterministic nulling at the RF beamforming level improves the rejection of interference up to 32 dB, thus reducing the requirements of analogue-to-digital conversion. The measured beam patterns were shown to have excellent agreement with simulations. The distortion of the main beam due to the placement of deterministic nulls was also investigated. This revealed that the main beam only changes shape by approximately 0.4 dB over all directions. In order for nulls to be placed precisely the system needs to be calibrated accurately. A novel calibration scheme called MEP (Multi-Element Phase toggle method) was introduced which allows fast and accurate measurement of each elements phase and gain offsets. Several advantages exist for this technique, including simultaneous element calibration and improving sensitivity to system noise. Thus, this paper has shown how it is possible to build a phased-array instrument for astronomy which greatly improves on traditional parabolic reflector techniques. The use of array processing algorithms in hierarchical configurations will make it possible for astronomers to obtain cleaner images.

October 4, 1999

DRAFT


118

References [1]

A. van Ardenne and F. M. A. Smits, “Technical Aspects for the Square Kilometer Array Interferometer,” ESTEC Workshop on Large Antennas in Radio Astronomy, The Netherlands, pp. 117–127, WPP–110, February 1996.

[2]

R. Braun, “The Square Kilometer Array Interferometer,” The Westerbork Observatory, Continuing Adventure in Radio Astronomy, pp. 167–184, 1996. Edited by E. Raimond and R.O. Genee, Kluwer, Dordrecht.

[3]

A. Boonstra, “The effect of EMI on the WSRT receive system,” 1th workshop on RFI suppression techniques for radio astronomy, NFRA, Dwingeloo, The Netherlands, 1998.

[4]

A. R. Thomson, J. M. Moran, and G. W. Swenson, “Interferometry and synthesis in radio astronomy,” Krieger Publishing Company, 1994.

[5]

G. A. Hampson, A. B. Smolders, A. Joseph, H. Heutink, A. Doorduin, and K. Dijkstra, “One Square Metre of a Million,” Proc. of 29th IEE European Microwave Conference, Munich, Germany, October 1999.

[6]

G. A. Hampson and A. B. Smolders, “A Fast and Accurate Scheme for Calibration of Active Phased-Array Antennas,” IEEE Int. Conf. on Antennas and Propagation, Orlando, USA, pp. 1040–1043, July 11-16 1999.

[7]

K. M. Lee, R. S. Chu, and S. C. Liu, “A Built-in Performance-Monitoring/Fault Isolation and Correction (PM/FIC) System for Active Phased-Array Antennas,” IEEE Trans. on Antennas and Propagation, vol. AP41, pp. 1530–1540, 1993.

[8]

S. P. Applebaum, “Adaptive Arrays,” IEEE Trans. on Antennas and Propagation, vol. AP-24, pp. 585–598, 1976.

[9]

R. J. Mailloux, Phased Array Antenna Handbook. Artech House, 1994.

[10] M. Goris, G. A. Hampson, A. Joseph, and F. Smits, “An Adaptive Beamforming System for Radio Frequency Interference Rejection,” IEE Proc. Radar, Sonar and Navigation, vol. 146, pp. 1040–1043, April 1999. [11] A. Smolders, THEA system specification.

NFRA, Dwingeloo, The Netherlands, 1999.

Available on

www.nfra.nl/ska. [12] A. B. Smolders and M. J. Arts, “Wide-band antenna elements with integrated balun,” IEEE Int. Conf. on Antennas and Propagation, Atlanta, USA, pp. 1394–1397, 1998. [13] M. J. Arts and A. B. Smolders, “Design and Construction of an Array of Vertically Oriented Bow-Tie Antennas for the Next Generation of Radio Telescopes,” Proc. of 29th IEE European Microwave Conference, Munich, Germany, October 1999. [14] G. A. Hampson, “A Phased Array Measurement System Using Matlab,” Proc. of 2nd Benelux Matlab Users Conference, Brussels, Belgium, Chapter 8, March 24-25 1999. [15] M. I. Skolnik, Radar Handbook. McGraw-Hill, 1990.

October 4, 1999

DRAFT


119

Bart Smolders was born in Hilvarenbeek, The Netherlands, on December 1, 1965. He received the M.Sc. and Ph.D. degrees in electrical engineering from the Eindhoven University of Technology in 1989 and 1994, respectively. From 1989 to 1991 he worked as a microwave designer at FEL-TNO in The Hague. From 1994 to 1997 he worked as a radar system designer at Hollandse Signaalapparaten in Hengelo. Currently, he is project manager at the Netherlands Foundation for Research in Astronomy (NFRA), where a phased-array prototype is being developed for the next generation of radio telescopes (SKA).

Grant Hampson

was born in Toowoomba, Australia, on April 11, 1972. He received

the B.Sc. and Ph.D. degrees in computing from Monash University, Melbourne, Australia, in 1993 and 1997, respectively. His Ph.D. interests were efficient beamforming architectures and their implementation. During 1996 he completed a beamforming ASIC at the Institut de Microtechnique, Universit´e de Neuchˆ atel, Switzerland. From 1997 to 1999 he was a postdoctoral fellow at the the Netherlands Foundation for Research in Astronomy. During this period he worked on multi-beam beamforming algorithms, adaptive processing and calibration algorithms for the Square Kilometre Array.

October 4, 1999

DRAFT