Radio Science

Buckland Park all-sky interferometric meteor radar


  • David A. Holdsworth,

    1. Atmospheric Radar Systems, Thebarton, South Australia, Australia
    2. Also at Department of Physics and Mathematical Physics, University of Adelaide, Adelaide, South Australia, Australia.
    Search for more papers by this author
  • Iain M. Reid,

    1. Department of Physics and Mathematical Physics, University of Adelaide, Adelaide, South Australia, Australia
    Search for more papers by this author
  • Manuel A. Cervera

    1. Department of Physics and Mathematical Physics, University of Adelaide, Adelaide, South Australia, Australia
    2. Also at Intelligence, Surveillance and Reconnaissance Division, Defence Science and Technology Organisation, Edinburgh, South Australia, Australia.
    Search for more papers by this author


[1] A VHF all-sky interferometric meteor radar system has been developed and installed at Buckland Park, South Australia. The radar is portable, allows a wide range of operating parameters, and can also be operated as a boundary layer radar. The analysis techniques have been developed using extensive simulations in an attempt to improve on standard techniques used by previous investigators. The results suggest that although pulse repetition frequencies (PRFs) around 2 kHz allow meteor velocity and deceleration estimation, PRFs around 500 Hz maximize count rate and improve the quality of meteor echo height estimates for this radar. Typical results are presented, indicating the radar obtains annual count rate variation of between 9000 and 14,000 height resolvable underdense meteors per day.

1. Introduction

[2] Radar techniques have been used for meteor observations for over 50 years. While the earliest observations were predominantly for astronomical purposes, such as meteor showers and meteoroid velocity estimation [e.g., Elford, 2001], atmospheric observations were later made by measuring the radial drift velocity of the ionized trail for investigation of mesospheric and lower thermospheric dynamics [e.g., Robertson et al., 1953], and the decay times for investigation of diffusion [e.g., Greenhow and Neufeld, 1955]. In more recent years, measurements of the decay time of meteor echoes have been used for estimating T/equation image, where T is temperature and P is pressure [e.g., Hocking et al., 1997; Cervera and Reid, 2000], temperature fluctuations [e.g., Tsutsumi et al., 1994, 1996; Hocking and Hocking, 2002], and with some assumptions, absolute temperature [e.g., Hocking, 1999].

[3] Recent advances in personal computers and digitization technology have resulted in a suite of instruments used for online meteor observations. Two recent examples are the meteor detection and collection (MEDAC) system [e.g., Valentic et al., 1996] and the all-sky interferometric meteor radar (SkiYMet) [e.g., Hocking et al., 2001]. MEDAC is a low-cost add-on to existing MST radars, which traditionally use transmit beams with narrow polar diagrams. The SkiYMet system is a dedicated all-sky interferometric meteor radar system using an antenna configuration (hereafter the “JWH configuration”) designed to yield unambiguous meteor angle of arrival while minimizing mutual antenna coupling [e.g., Jones et al., 1998].

[4] This paper describes the implementation of a commercial radar system and it's use as an all sky interferometric meteor radar at the University of Adelaide's Buckland Park field station. Section 2 describes the technical and site details of the radar. Section 3 describes the analysis techniques used. Section 4 describes aspects of the radar's operation and presents a selection of initial results, while section 5 presents a summary. Detailed results will be described in a series of papers currently in preparation.

2. Buckland Park Meteor Radar

[5] The Buckland Park meteor radar (BPMR) is located 35 km north of Adelaide (34°38′S, 138°29′E). Transmission, reception and data acquisition is performed using a portable radar system whose specifications are shown in Table 1. The transmitting system consists of three solid state modules producing coded/uncoded Gaussian/square shaped pulses over a wide range of peak powers and pulse lengths. Pulse repetition frequencies (PRFs) up to 20 KHz are possible. The radar can be operated in two modes. In “meteor” mode the peak power is 7.5 kW, and the duty cycle is 5 (8.3)% for uncoded (coded) pulses, whereas in “boundary layer” mode the peak power is 12 kW, and duty cycle is 3 (5)%. The radar can be configured to operate in a chosen mode at any frequency between 30 and 60 MHz. Mode or frequency modifications can be made with minor hardware adjustments. The radar data acquisition system (RDAS) consists of 6 receiving channels, allowing a wide range of receiver bandwidths, gains, and range and time sampling parameters. Radar control and data acquisition is performed by a WindowsNT “acquisition” PC fitted with data acquisition hardware and software acquiring data with 12-bit resolution. The control program provides considerable flexibility in experimental sequencing and scheduling, allowing multiple experiments to be interleaved, and experiments to be configured to run at specified times or intervals. Data acquired by the acquisition PC is transfered to a Linux (Unix) “analysis” PC providing user interface to the radar. Both PCs are accessible through the internet. The radar is designed for remote unattended operation, running continuously except for a 4-s reconfiguration interval every minute. The meteor observations described in this paper were made at 31 MHz. The radar has also operated as a 54.1 MHz Boundary Layer radar in a configuration similar to that described by Vincent et al. [1998] during the Darwin Area Wave Experiment (DAWEX) [MacKinnon et al., 2002].

Table 1. Buckland Park Meteor Radar Specifications
ParameterMeteor ModeBoundary Layer Mode
Peak power, kW7.512
Maximum duty cycle, uncoded pulse %53
Maximum duty cycle, coded pulse %8.35
Maximum PRF, Hz20,00020,000
Frequency, MHz30 to 6030 to 60
Transmit pulse length (HPFW), m100 to 9000100 to 9000
Receiver filter width (HPHW), kHz18 to 40018 to 400

[6] The antenna field is illustrated in Figure 1, and consists of a pair of crossed dipoles for transmission and five pairs of crossed dipoles for reception. Transmission can be performed using Ordinary (O) mode circular or linear polarization. Initial observations (prior to May 2002) used circularly polarized transmission produced using three dipoles aligned 60° apart, with each transmit module feeding a separate dipole through feeder cables producing phases of 0°, 120° and 240°. This configuration was found to result in significant coupling between dipoles, producing transmitter mismatch errors. Subsequently, circular polarized transmission was produced by combining transmit module outputs and splitting the result two ways to feed two perpendicularly aligned dipoles through feeder cables producing phases of 0° and 90°. For linear transmission, the transmitter modules are combined and connected to a single transmit antenna dipole. Receivers 1 and 2 are connected to the individual dipoles for the “center” receiving antennas, and receivers 3 to 6 are connected to the “outer” receiving antennas. The outer receiving antennas can be configured for linearly polarized reception on one of the pair of crossed dipoles, or O mode circularly polarized reception by combining the crossed dipoles with a 90° phasing cable. The receiving antennas use the JWH configuration, being arranged in two perpendicular arms with spacings of 2 and 2.5 λ. The individually accessible crossed dipoles of the center antennas allow Faraday rotation measurements. Initial observations (prior to June 2003) were made using gamma-matched feeds. The feeds for the receiving antennas were later replaced with balanced feeds for improved tunability.

Figure 1.

Antenna configuration for the Buckland Park meteor radar.

[7] The online analysis allows raw data to be analyzed using different parameter combinations, allowing experimentation to optimize data quality and/or count rate. The analysis allows the storage of all raw time series data, or 4-s subsets about each detection, or each detection which has been classified as an underdense meteor. A postanalysis module allows wind velocity estimation using the meteor angles of arrival (AOAs) and radial drift velocities, and the production of result “plot files,” which can be transfered via ftp to remote locations. When the radar is in operation, hourly updated plot files from the radar are available at the Web address

[8] The radar was designed and manufactured by Atmospheric Radar Systems (ATRAD). Similar radars have been installed in Svalbard, Norway [e.g., Hall et al., 2002], Tromso, Norway, and Wuhan, China [e.g., Xiong et al., 2003]. Similar meteor radar capability has also been included in (M)ST radars installed at Wakkanai, Japan, Davis, Antarctica [e.g., Morris et al., 2003], and Kühlungsborn, Germany. At Wakkanai, transmission is performed using a 100 kW valve transmitter connected to the main (144-element) antenna array, with reception using JWH configuration antennas installed within the transmitting array. At Davis and Kühlungsborn, transmission is performed using a 20 kW valve transmitter feeding a single transmit antenna located outside the main (144-element) antenna array, with reception using JWH configuration antennas installed outside the transmitting array.

3. Analysis Technique

[9] In developing the BPMR analysis techniques a statistical evaluation of commonly used techniques for parameter estimation was performed. This evaluation used two simulation techniques. The first involved generating meteor echoes according to Fresnel diffraction theory [e.g., Cervera et al., 1997] to evaluate detection and techniques for estimating receiver phase differences, radial drift velocities and decay times. The second simulated phase errors in phase difference estimates, and was used to evaluate AOA estimation. Results from the second simulation technique will be described in a paper currently in preparation.

[10] The online analysis applies criteria to determine whether the behavior of a meteor echo candidate is consistent with that expected for an underdense meteor echo, as shown in Table 2. Once a candidate is rejected it is excluded from further analysis. The same analysis procedures (with minor threshold adjustments) are also applied for Buckland Park MF radar meteor observations [e.g., Holdsworth and Reid, 2004a]. In this case the background signal is often ionospheric echoes, rather than white noise as is usually the case for VHF meteor observations. This has led to the thorough investigations of techniques to reject ionospheric echoes, providing an added benefit for VHF meteor observations. Although useful information can be obtained from overdense meteor echoes, such echoes are rejected since parameter estimation is often difficult and not suited to routine analysis. However, overdense echoes can be archived for offline analysis if required.

Table 2. Rejection Criteria for Buckland Park Meteor Radar Analysis
0no error; analysis result ok
1SNR < 12 dB
2angle of arrival (AOA) ambiguously determined
3AOA estimate not feasible
4large difference in AOAs obtained from different antenna baselines
5echo at start or end of time series
6echo less than 5 samples long; too short for analysis
7echo rise exceeds 0.3 s
8echo decay time less than twice rise time
9large power level before echo
10large power level level after echo
11poor fit to amplitude for estimation of decay time
12poor fit to CCF phase variation for estimation of radial drift velocity
13height unresolvable echo: not valid height within 70 to 110 km
14height ambiguous echo: more than one possible height within 70 to 110 km
15radial drift velocity or projected horizontal velocity exceeds 200 ms−1
16Oscillatory echo, indicating event most likely not an underdense meteor

3.1. Interference Removal

[11] Coherent interference is often seen as sharp “bursts” throughout all range gates, lasting from a fraction of a second to over 10 s. These bursts obscure actual meteor events, and can be detected as meteor candidates, resulting in excessive analysis time wastage since they may not necessarily be rejected by any criteria prior to underdense echo selection (section 3.9). An interference filter using Rayleigh signal statistics was developed, and has proven successful in rejecting about 99% of interference bursts. In addition, of those bursts that are not rejected, approximately 90% are rejected by criteria described in the following sections. A different form of coherent interference also results from the colocated VHF radar. This interference produces “noise” correlated between receiver channels, producing a 3–6 dB increase in noise level, thereby reducing meteor detectability.

3.2. Phase Offset Correction

[12] The receiver time series outputs are corrected for any known phase offsets in the paths (antenna, feeder cable, receivers) of each receiving channel. Offsets as small as 5° can have significant effects, such as broadening of height and decay time distributions, and diurnal oscillations in the peak detection height associated with changes in AOA with Earth's rotation [Holdsworth et al., 2004]. Initial phase offsets were estimated by feeding the attenuated transmit pulse into the system at various hardware stages. However, since no satisfactory means of measuring the phase offsets introduced by the antennas was found, a technique was developed to determine the phase offsets using meteor echoes [Holdsworth et al., 2004]. This technique is applied routinely offline to ensure current phase offset estimates are in use.

3.3. Center Antennas Combination

[13] If the center receive antennas are configured for linear polarization and the outer receive antennas are configured for circular polarization, it is necessary to combine receiver outputs 1 and 2 to produce an effective antenna with the same polarization as the outer antennas. The center antenna receiver output is hereafter referred to as “receiver C” (i.e., center), implying receiver 1 only if linear receive polarization is used for the outer antennas, and the combined outputs of receivers 1 and 2 if circular receive polarization is used for the outer antennas.

3.4. Detection

[14] Two detection algorithms are available: “noncoherent” and “coherent.” For noncoherent detection, the combined power series is formed at each range by combining the power series over all receivers, and an echo candidate is registered if at least two contiguous samples of the combined power series exceeds a user selected threshold defined by a noise level multiple (default 4). For each candidate, the range, start time (first sample above threshold), peak time, and end time (last sample above threshold) of the three point smoothed power series are recorded, as well as the magnitude of the first lag of the autocorrelation function (FLA) obtained using the combined power series subset between the start and end times. The first lag is used as the zeroth lag contains a noise spike [e.g., Briggs, 1984]. For coherent detection, the complex time series data is divided into small segments (default length 1 s), and cross-correlation functions (CCFs) for receiver pairs C3, C4, C5 and C6 are calculated. The receiver phase differences are estimated for each segment by applying a linear least squares fit to the CCF phases using lags ±1 and ±2. If the RMS phase difference between the fitted and CCF phases are less than a preselected threshold (default 25°) for all CCFs, the fitted zero-lag phase estimates are used to phase calibrate the time series, allowing the complex time series data to be combined to form the coherently combined power series. The detection process then follows that described for noncoherent detection using the coherently combined power series. The noise level for both detection algorithms is calculated by splitting the combined power series into (typically) 4 s blocks, and using the mean value for each section. The choice of a 4 s block is sufficient for underdense meteor echoes not to influence the noise level estimate, except for the case of strong echoes which would exceed the detection threshold anyhow. Initial analysis used the mean noise level over the entire time series, which resulted in missed detections if the noise level changed throughout the record, such as can occur if the colocated VHF ST radars starts or stops acquiring data during the meteor acquisition record.

[15] The CCF phase criterion used in coherent detection constitutes a check for coherent signal, hence the term “coherent detection.” A similar concept has been used in SuperDARN meteor observations, where meteor detection has been applied using autocorrelation functions [e.g., Hall et al., 1997]. Coherent detection increases meteor echo detectability since the echoes are combined coherently, but this comes at the cost of increased computational intensity.

3.5. Echo Range Selection and Multiple Detection Removal

[16] Detection of the same echo over multiple ranges can occur if the sampling resolution is comparable (or smaller) than the observed echo half power full width (HPFW), or if Barker coding is used, where range sidelobes can result [e.g., Nakamura et al., 1991]. The analysis applies an iterative multiple detection algorithm as follows. The maximum FLA is found, and the echo range, start, peak and end times are recorded. The FLAs of this and all other detections within a specified range limit (based on pulse width, receiver filter width, and coding type) of the maximum FLA range, and with start or end times within ±1 s of maximum FLA are reset to zero to avoid “redetection.” This process is repeated until there are no nonzero FLAs remaining. The echo range estimate for any multiple detection is thus assumed to be that producing maximum echo power [e.g., Nakamura et al., 1991]. Simulations have shown the FLA provides a robust estimate of the actual echo range, even when significant (10 times) oversampling is used.

3.6. Phase Difference Estimation

[17] The phase differences of each candidate for receiver pairs C3, C4, C5, and C6 are estimated by applying a linear fit to the cross-correlation function (CCF) phase over lags ±1 and ±2. This is preferred to the zero-lag CCF phase itself [e.g., Hocking et al., 2001], since correlated noise produces a discontinuity at zero-lag. Furthermore, only the time series samples between the candidate start and end times are used to calculate the CCFs. This is preferable to using a 4 s block around the meteor echo [e.g., Hocking et al., 2001], as use of the extra “noise only” time series samples increases the uncertainty in the phase difference estimates.

3.7. Initial Receiver Combining

[18] The candidate complex time series for receivers C, 3, 4, 5 and 6 are combined coherently after removing the phase differences estimated in section 3.6. The meteor start, peak, and end times are reestimated using the combined time series.

3.8. Phase Difference Re-Estimation and Receiver Recombining

[19] The candidate phase differences for receiver pairs C3, C4, C5, C6, 34 and 56 are estimated using the revised start and end times. If the RMS phase difference between the fitted and CCF phase exceeds a preselected threshold (default 20°) for any receiver pair the candidate is rejected. The complex time series for receivers C, 3, 4, 5 and 6 are combined coherently after removing the reestimated phase differences. The candidate peak time, peak power and SNR are estimated from the combined power series, and written to file, along with the revised phase differences and the mean RMS phase difference.

3.9. Underdense Echo Selection

[20] The combined power series is examined to reject all candidates which do not exhibit the sharp rise time and exponential decay expected of underdense meteor echoes. The criteria used are shown in Table 2. Criterion 1 (low SNR) rejects weak echoes from which no information could be usefully determined, and is also useful for rejecting tropospheric and ionospheric echoes. Criterion 5 rejects incomplete meteor echoes, while criterion 6 rejects short-lived underdense meteor echoes from which no useful information can be determined, interference, and impulsive electrical or lightning spikes. Criteria 7 to 10 reject tropospheric, ionospheric and aircraft echoes. Criterion 16 reject aircraft and overdense meteor echoes. This criterion is similar to that used by Hocking et al. [2001], although in our application smoothing (typically three point) is used to smooth any noise fluctuations, a smaller minimum threshold (0.15 times peak power) is used, and the criterion is applied between the peak and end points of the candidate echo, rather than over the 900 ms following the peak. Further protection against overdense echoes is provided by goodness of fit criteria in the decay time (section 3.10) and radial drift velocity (section 3.11) estimation procedures. Further protection against tropospheric, ionospheric and aircraft echoes is provided by the echo height determination described in section 3.13.

3.10. Decay Time Estimation

[21] The decay time is determined by fitting an exponential to the combined receiver power series from 15 ms after the meteor echo peak to the position where the echo falls to the noise floor. The fit start time avoids inclusion of any effects associated with Fresnel oscillations in the fit [e.g., Cervera and Reid, 2000]. The 15 ms value was determined empirically for observations at 54.1 MHz, suggesting use of a larger value for lower frequency radars such as the BPMR. This is further confirmed by the empirical investigations of Roper [1984], who recommend a value of 45 ms for a 32 MHz radar. However, the value of 15 ms used appears adequate for BPMR observations. Candidates are rejected due to low goodness of fit or nonsensical values (criterion 11). The ambipolar diffusion coefficient is also calculated [e.g., Cervera and Reid, 2000]. The decay time, diffusion coefficient, and their error are written to file.

3.11. Radial Drift Velocity Estimation

[22] The radial drift velocity is estimated using the slope from a least square linear fit to the phase of the sum of all CCFs over lags ±1 and ±2 [e.g., Hocking et al., 2001]. The time series from 15 ms after the peak to the end time are used to calculate the CCFs, thereby avoiding inclusion of any effects associated with Fresnel oscillations or the meteoroid velocity during trail formation in the fit [e.g., Cervera and Reid, 1995], and minimizing the effects of noise inherent in using a 4 s block around the meteor echo [e.g., Hocking et al., 2001]. Candidates are rejected due to low goodness of fit (criterion 12) or for radial or projected horizontal velocities exceeding 200 ms−1 (criterion 15). The radial drift velocity and it's error are written to file.

3.12. Angle of Arrival Estimation

[23] The AOA components for each echo candidate are estimated using the phase differences determined in section 3.8 using a technique based on that described by Jones et al. [1998]. This technique involves estimating phase differences for spacings of 2λ (ϕ), 2.5λ (ϕ2.5λ) and 4.5λ (ϕ4.5λ) for each meteor echo. Differencing ϕ and ϕ2.5λ yields the phase difference that would be measured between antennas with 0.5λ spacing

equation image

yielding the unambiguous AOA component ψ′0.5λ, where the prime denotes an “indirectly estimated” phase difference or AOA component. Adding ϕ and ϕ2.5λ yields a second 4.5λ phase difference

equation image

which is used to determine a set of (unaliased) AOA component candidates ψ′4.5λ. The ψ′4.5λ candidate in closest agreement with ψ′0.5λ is then used as the AOA component estimate.

[24] The aforementioned phase error simulations suggests four concepts for improving AOA accuracy that are implemented in the current analysis. First, ϕ4.5λ is used in preference to ϕ′4.5λ. Since meteor echoes are highly specular, the measurement errors of ϕ, ϕ2.5λ and ϕ4.5λ are expected to be comparable. The measurement error of ϕ′4.5λ will therefore be approximately equation image larger than that of ϕ4.5λ. Second, ϕ is used to determine a set of (unaliased) AOA component candidates ψ. The ψ candidate in closest agreement with ψ′0.5λ is found, and the ψ4.5λ candidate in closest agreement to this is then used as the AOA component estimate. This improves AOA accuracy for large measurement errors. Third, comparisons between AOA components for different antenna spacings are performed using differences in sinθ rather than θ. Fourth, wraparound is allowed, i.e., an AOA component of 90° maps back to −90°. Echo candidates are rejected if the AOA component estimate is not feasible (criterion 3) or there is a large difference between AOA component estimates for different spacings (4). The resulting AOA components are written to file. As described earlier, since the phase differences are also written to file, offline AOA redetermination can be applied if the assumed phase calibration is found to be in error, or if improvements in AOA estimation technique can be made.

3.13. Echo Height Determination

[25] Because range aliasing is often used for meteor radars, echo range “dealiasing” is applied by producing a range ensemble

equation image

for each candidate, where R is the detection range, N = Rmax/Ramb, Ramb = c/(2PRF) is the maximum unambiguous range, Rmax is the maximum echo range, and c is the speed of light. The maximum echo range assumed for the BPMR is 520 km, based on maximum expected height and zenith angles of 110 km and 80°, respectively. A height ensemble is then determined using

equation image

where θ is the zenith angle and Re is Earth's radius. If a single height exists within 70 and 110 km this is the assumed echo height. If no heights exist within 70 to 110 km the candidate is deemed “height unresolvable,” and is rejected by criterion 13. If two or more heights exist within 70 to 110 km the candidate is deemed “height ambiguous,” and is rejected by criterion 14. This is sometimes the case for zenith angles exceeding 65° and ranges exceeding 200 km when high PRFs (e.g., 2 kHz) are used. In this case it is sometimes possible to determine the most likely height offline by comparing the decay time with the theoretical estimate, or mean of the existing estimates, at the each possible height. Both the dealiased range and height are written to file.

[26] Range dealiasing is not applied for coded transmission, as the decoded range aliased echo will generally not appear at the correct (aliased) range. For BPMF meteor observations, height limits of 80 and 140 km are used in accordance with the increased height ceiling for MF radar observations [e.g., Tsutsumi et al., 1999].

3.14. Wind Velocity Estimation

[27] Given a set of meteor echo radial drift velocities (Vr) and AOA direction cosines (l, m, n), the 3-dim wind velocity (u, v, w) can be estimated by determining the least squares solution of

equation image

In applying equation (5) using the BPMR, excessively large vertical velocities (>10 m s−1) often result, even at heights where reliable velocity estimates should be expected due to the large number of AOA and radial drift velocities available. Consequently, a 2-dim least squares fit

equation image

is applied, with the implicit assumption that w = 0 [e.g., Nakamura et al., 1991; Cervera and Reid, 1995; Marsh et al., 2000; Hocking et al., 2001].

[28] The data are grouped into height/time bins of 2 km and 1 hour. Although only two meteor echoes are required to produce a velocity estimate using equation (6), least squares solutions using small numbers of echoes can produce suspect winds [e.g., Hocking and Thayaparan, 1997]. We therefore require a minimum of six echoes in each height/time bin to compute a velocity estimate. Further security against suspect wind estimates is provided by using the wind velocity estimate to calculate the projected radial drift velocity for each meteor echo. If the absolute difference between actual and projected radial wind velocities, ΔVr, exceeds 25 m s−1 the echo is rejected, and the least squares fit is repeated. This process is repeated until there are no further echoes rejected, or there are less than six remaining echoes, in which case a wind velocity estimate is not made at the height/time in question.

[29] The choice of minimum number of echoes, outlier rejection technique and threshold was made based on a statistical evaluation of over six months of wind velocity estimates, and are a compromise between maximizing the number of velocity estimates and minimizing the number of outliers. Our selected threshold is lower than the 30 m s−1 threshold justified in terms of gravity wave considerations [e.g., Hocking and Thayaparan, 1997]. The evaluation also included investigation of an alternative method of outlier rejection which rejects meteor echoes where ΔVr exceeds 3 times the standard deviation of ΔVr, as used in imaging Doppler interferometry (IDI) [e.g., Brosnahan and Adams, 1993]. Although this technique has been successfully implemented for Buckland Park MF radar IDI analysis [e.g., Holdsworth and Reid, 2004b], the ΔVr threshold technique has been preferred for BPMR analysis as it produces more valid wind estimates, and less outliers.

4. Results

[30] The BPMR has operated for five observation periods (OPs). The first two OPs (02/04/2001 to 25/04/2001, and 26/07/2001 to 11/09/2001) involved considerable experimentation to determine optimal transmit and receive antenna configuration, and optimal analysis configuration. Results from the first OP are described by Holdsworth and Reid [2002]. The third to fifth OPs were intended for routine data collection with some experimentation to determine experiment parameter configuration for optimizing count rates. This optimization will be described in a future paper. The third OP (12/07/2002 to 23/10/2002) used linear receiver polarization, with receivers 3 to 6 connected to the 28° azimuth dipole of the outer receiving antennas. The fourth OP (24/10/2002 to 24/03/2003) used circular polarization for the outer antennas, leading to an increase in count rate. Estimation of the count rate increase resulting from the polarization change alone is difficult since an analysis upgrade was performed simultaneously. However, we estimate the count rates increased by 10 to 15%. This is because linearly polarized antennas cannot observe meteors along the dipole azimuth at large zenith angles [e.g., Hocking et al., 2001; Holdsworth and Reid, 2002]. The fifth OP (02/06/2003 to present) corresponds to use of the balanced feed arrangement for the receive antennas, with the inner and outer antennas configured for circular polarization.

[31] The experiment parameters for OP 3 are shown in Table 3, and are based on parameters used by other all-sky interferometric meteor radar systems [e.g., Hocking et al., 2001]. The PRF results in a maximum unambiguous range (Ramb) of 75.8 km. Since most meteors echoes occur at heights above 70 km most will be range aliased. In most cases the range ambiguity can be easily resolved, as described in section 3.13. Use of high PRFs allows meteoroid velocity and deceleration measurements [e.g., Cervera et al., 1997; Elford, 2001], and allows the number of coherent integrations to be increased, thereby reducing the noise level and improving meteor echo detectability. Coherent integration can be performed either in data acquisition or analysis. Since online meteoroid velocity estimation is currently not applied, coherent integration is performed in data acquisition to reduce raw data sizes. The particular PRF (1980 Hz) chosen maximizes the maximum (aliased) sampled range within the limits imposed by range aliasing, transmit pulse width, receiver bandwidth, and data acquisition and control. For example, 2 kHz PRF allows a maximum (aliased) range of 64 km, while 1980 Hz allows a maximum of 66 km, leading to a small but significant increase in sampling range, and hence count rate. The minimum sampling range of 6 km avoids the effects of receiver recovery from the transmit pulse, and reduces reception of tropospheric echoes. During the last weeks of January 2003 (midsummer) it was necessary to increase the start range to 8 km. With these range sampling limits it follows that meteor echoes with ranges of RiRamb + [−6, 9.8], i =1, 2. km, are not sampled. The transmit pulse width and receiver filter width used result in an echo HPFW of ≈3.8 km.

Table 3. Buckland Park Meteor Radar Operating Parameters for Observation Period (OP) 3
Frequency, MHz31
PRF, Hz1980
Transmit pulse HPFW, km2
Receiver filter width, kHz20.1
Pulse typeGaussian
Range, km6 to 66
Range sampling resolution, km2
Coherent Integrations16
Effective sampling time, s0.008
Number of samples6800
Acquisition length, s55

[32] With the range sampling limitations in mind, operation using coded pulses with lower PRFs was trialed during the OP 5. This was achieved by interleaving the 1980 Hz experiment with a 490 Hz experiment between 07/07/2003 and 15/07/2003, using the parameters shown in Table 4. The 490 Hz experiment used a 4-bit complementary code with a 2-km pulse width. This PRF is close to 495 Hz, which yields the same theoretical SNR as expected for the 1980 Hz experiment used in OP3. In reality, the theoretical SNR obtained using pulse coding is seldom achieved due to imperfections in the pulse envelope and target motion. Measurements of the frequency spectrum of the single pulse used for the 1980 Hz experiment and the coded pulse used for 490 Hz experiment have revealed the two spectra are remarkably similar. We therefore attribute the differences in the results of the 1980 Hz and 490 Hz experiments solely to differences in the PRF used. Results from this comparison are shown in Figure 2 and Table 5. The 490 Hz experiment fills the range gaps of the 1980 Hz experiment, and produces no height ambiguous meteors. Figure 2 clearly reveals the 1980 Hz experiment range gaps, and associated reductions in the zenith angle distribution. The reduction between 47° and 57° is associated with the range gap 145.6 km to 161.4 km, while the reduction between 63° and 71° is associated with the range gap 221.4 km to 237.2 km. Figure 2 and Table 5 also illustrate that the 1980 Hz experiment yields a large number (2754) of range estimates either side of the (unaliased) unsampled ranges. Such range estimates result from echoes with ranges on the sampling range limits, and from ranges outside the sampling range limits due to the finite range width of the echoes. The echoes with ranges outside the range limits are therefore erroneously assigned ranges on the limits. This effect can be significant if the echo is smeared in range, which can occur if long pulse widths and/or narrow receiver filter widths are used. For instance, for strong meteor echoes smeared over 4 km HPFW, echoes at (aliased) ranges from as low (high) as 0 (72) km can be incorrectly assigned a range of 6 (66) km. This effect suggests the rejection of all detections with echo ranges lying on the sampling range limits on the basis that their assigned ranges can be in error up to 6 km. This concept has been used for all off-line analysis of BPMR data.

Figure 2.

(a) Range and (b) zenith angle distributions obtained using PRFs 1980 Hz (solid line) and 490 Hz (dashed) for 07/07/2003 to 15/07/2003.

Table 4. Buckland Park Meteor Radar Operating Parameters for the “490 Hz Experiment” Between 07/07/2003 and 15/07/2003
Frequency, MHz31
PRF, Hz490
Transmit pulse HPFW, km2
Receiver filter width, kHz20.1
Pulse type4-bit complementary code
Range, km70 to 280
Range sampling resolution, km2
Coherent integrations4
Effective sampling time, s0.008
Number of samples6800
Acquisition length, s55
Table 5. Number of Data Producing Height Resolvable, Unresolvable, and Ambiguous Echoes for 1980 and 490 Hz Experiments, “Assumed 1980 Hz” Experiments, and 490 Hz Height Unresolvable Echoes for the “Assumed 1980 Hz” Experiment
Parameter1980 Hz490 HzAssumed 1980 HzUnresolvable 490 Hz
Height resolvable meteors33,13137,92137,7861623
Height unresolvable meteors7612833896896
Height ambiguous meteors376302044314
Number of data on range limits2754196N/AN/A

[33] The 490 Hz experiment yields a larger number of height unresolvable meteors. Careful examination of the 490 Hz echoes suggests these echoes are valid underdense meteor echoes. Since we have complete confidence in the echo range determination, we believe these height unresolved meteors result from using erroneous AOA estimates. In order to investigate this further the range estimates of the 490 Hz data set were deliberately aliased to simulate the range aliasing that would occur if a PRF of 1980 Hz was used, and the heights were recalculated. The height statistics for this data set (hereafter the “assumed 1980 Hz” data set) are shown in Table 5, revealing a significant reduction in the number of height unresolvable meteors. Table 5 also shows how the heights for the 490 Hz height unresolvable echoes (hereafter the “unresolvable 490 Hz” data set) are redistributed for the “assumed 1980 Hz” data set, revealing 1623 of these echoes become height resolvable when a PRF of 1980 Hz is used. Since the AOAs of the 490 Hz unresolvable meteors are erroneous, it follows that the heights derived from the “unresolvable 490 Hz” data set will also be incorrect: i.e., the 1622 height resolvable meteors from the “unresolvable 490 Hz” data set will have incorrect height estimates as they have been made using incorrect AOA estimates. It therefore necessarily follows that use of a 1980 Hz PRF will increase the number of echoes with incorrect height estimates.

[34] Table 5 indicates the 490 Hz experiment obtains approximately 14% more height resolvable meteors than the 1980 Hz experiment. However, the number quoted for the 1980 Hz experiment includes data with incorrect height estimates, and data with ranges on the range limits. Precluding this data from the statistics indicates the 490 Hz experiment obtains approximately 30% more correctly estimated height resolvable echoes. One further point worth noting is that since the 490 Hz experiment uses pulse coding, range dealiasing is not applied (as described in section 3.13). The maximum sampling range of 280 km therefore represents the largest range from which the analysis will obtain echoes. However, the 1980 Hz effectively samples to much larger ranges, where the likelihood of being unable to ambiguously resolve the echo height is increased. This explains why the 1980 Hz data set yields approximately 1700 more height ambiguous meteors than the “assumed 1980 Hz” experiment, as shown in Table 5. It follows that increasing the maximum sampling range of the 490 Hz experiment (by decreasing PRF) will further increase the percentage of correctly estimated height resolvable echoes well above 30%.

[35] In summary, the 1980 Hz experiment has four significant limitations: (1) unsampled ranges, (2) height ambiguous meteors, (3) range estimates on the (unaliased) range limits could be in error; (4) incorrect heights can be determined for echoes which are height unresolvable at lower PRFs. Thus although the 1980 Hz PRF allows application of meteoroid velocity and deceleration techniques, it produces reduced count rates, and reduces the accuracy of the meteor height estimates. As a result, lower PRFs (≤490 Hz) were adopted for OP 5, as shown in Table 6. The PRF was reduced to 440 Hz to further increase the maximum ambiguous range beyond 300 km, and thereby further increase the count rate. The transmit pulse width and receiver filter width were selected to optimize SNR, and produce an echo HPFW of ≈3.8 km. The annual variation of the daily count rate of height resolvable underdense echoes obtained using these parameters is typically between 9000 and 14000.

Table 6. Buckland Park Meteor Radar Operating Parameters for Observation Period (OP) 5
Frequency, MHz31
PRF, Hz440
Transmit pulse HPFW, km3.2
Receiver filter width, kHz32.0
Pulse type4-bit complementary code
Range, km70 to 310
Range sampling resolution, km1.6
Coherent integrations4
Effective sampling time, s0.009
Number of samples6100
Acquisition length, s55

[36] Typical results from OP 5 are illustrated using data from 24/07/2002, where a total of 12701 height resolvable underdense meteors were observed. The AOA distribution is shown in Figure 3. The majority of echoes are observed at zenith angles between 40° and 70°. The lack of meteors on the zenith is a well established result [e.g., Hocking et al., 2001]. The height and time histograms are shown in Figure 4. The height distribution shows the expected behavior, peaking at 92 km and decreasing to zero at 70 and 110 km. The time distribution exhibits the well known diurnal variation with maximum at dawn and minimum at dusk [e.g., McKinley, 1961]. The AOA, height and time distributions show very good agreement with those expected using the meteor response function of Cervera and Elford [2004]. Comparisons of experimental and meteor response function distributions will be presented in a subsequent paper.

Figure 3.

AOA skymap for 24/07/2003. The dashed concentric circles indicate zenith angles of 20°, 40°, 60°, and 80°.

Figure 4.

(a) Height and (b) time distributions for 24/07/2003.

[37] The decay times are shown in Figure 5, and show reasonable agreement with the theoretical estimate obtained using CIRA model pressure and temperature. The scatter seen is this plot is typical [e.g., Hocking et al., 1997; Cervera and Reid, 2000] and includes variations in the diffusion coefficients due to fluctuations in temperature (and to a lesser extent, pressure), and the number density of metallic constituents within the trail, rather than just statistical errors in the decay time, AOA and range estimates. Zhou et al. [2001] suggest the production of field aligned irregularities within the meteor trail may also introduce such variation. Furthermore, Elford [2001] shows that the diffusion in the direction perpendicular to the field lines is retarded at heights above 95 km, leading to the measurement of large diffusion coefficients for meteors with AOAs perpendicular to the field lines.

Figure 5.

Scatterplot of the natural log of inverse decay time as a function of height for 24/07/2003. The solid line indicates the theoretical value expected based on CIRA model pressure and temperatures.

[38] The distribution of error codes for 24/07/2003 are shown in Figure 6. The most frequent error code is 16 (oscillatory signal), which is responsible for rejection of ≈25% of the number of height resolved underdense meteors observed. The next most frequent error codes are 8, 9, 10 and 13. The accuracy of the receiver channel phase offsets can be determined by considering the total number of meteor echoes rejected on the basis of height/AOA, as indicated by error codes 3, 4, 13 and 14. The percentage of height/AOA rejections for range aliased (nonrange aliased) experiments is usually around 10% (5%). This is because range aliased experiments contain a larger proportion of height ambiguous meteors. The percentage of height/AOA rejections can be used as a reliable fault indicator, in that a sudden increase in the percentage of unresolvable meteor echoes can indicate development of a system fault.

Figure 6.

Distribution of error codes for 24/07/2003.

[39] An example of horizontal winds estimated using the BPMR during March 2003 are shown in Figure 7. No postanalysis outlier rejection is applied in this plot. Almost complete coverage is obtained between 76 and 100 km. The diurnal tide maximizes in late March in Adelaide, and is clearly evident in both velocity components. Further velocity estimates and comparisons with the full correlation analysis winds estimated using the colocated Buckland Park MF radar [e.g., Holdsworth and Reid, 2004a] will be presented in a subsequent paper.

Figure 7.

(top) Zonal and (bottom) meridional velocity components for 11/03/2003 to 17/03/2003.

5. Summary and Future Plans

[40] A new VHF all-sky interferometric meteor radar system has been developed and installed at Buckland Park, South Australia. Initial results are presented, indicating an annual variation of between 9000 and 14000 height resolvable underdense meteors per day. Although we have evaluated various techniques for estimation of each meteor parameter and selected the most accurate techniques for online implementation, evaluation of the analysis techniques is ongoing.

[41] Meteoroid velocity estimation [e.g., Cervera et al., 1997] and meteor temperature analysis [e.g., Hocking, 1999] have been developed for off-line use. The meteoroid velocity estimation makes use of the archived raw data of meteor detections archived by the analysis. Since the meteoroid velocities can be aliased if the effective sampling time (i.e., number of coherent integrations/pulse repetition frequency) is too large, it is necessary to reduce the number of coherent integrations significantly as compared to the values shown in Table 3. The meteor temperature analysis makes use of the decay times (or diffusion coefficients) stored in the analyzed data files. Meteoroid velocity and meteor temperature will be implemented online after further refinement.

[42] A further online analysis possibility is electron density estimation using Faraday rotation measurements of “down the beam” meteors echoes, which are produced by the ionized region at the head of the meteoroid [e.g., Elford, 2001]. Such echoes are detected in multiple range gates, allowing their use for electron density profiling [e.g., Elford and Taylor, 1997]. The application of this technique has to date been thwarted by the limited number of down-the-beam meteors observed. Although this may in part been due to nonoptimal detection procedures for such echoes, and to a lesser extent interference, we suspect all sky systems may be unsuitable for down the beam meteor observations due to the low power per solid angle. We note that most down the beam observations have been made using narrow beam radars which concentrate higher powers into smaller beam widths [e.g., Elford and Taylor, 1997; Zhou et al., 1998].


[43] We would like to thank Graham Elford, Chris Adami, Bruce Johnson, Callum Heinrich, Steven Wawryk, and John Barnes for their contributions toward the work presented in this paper. The Buckland Park meteor radar was supported by Australian Research Council grant 20006300.