Description and demonstration of the new Middle and Upper atmosphere Radar imaging system: 1-D, 2-D, and 3-D imaging of troposphere and stratosphere

Authors


Abstract

[1] The Middle and Upper atmosphere Radar (MUR) was upgraded in March 2004 for radar imaging capability with 5 frequencies across a 1 MHz bandwidth and 25 digital receivers. Although digitization introduces problems of its own, the uniformity of digitization is a great benefit over the analogue system in place before. This increased reliability will help make the new system an important component of long-term atmospheric science programs. We demonstrate 3-D imaging with Capon's method, which can provide information about structure morphology. In addition, we demonstrate an experimental 0.5 μs pulse mode and compare this to Capon method imaging results.

1. Introduction

[2] The Middle and Upper atmosphere Radar (MUR) [Fukao et al., 1985a, 1985b] in Shigaraki, Japan (34.85°N, 136.10°E), is a Very High Frequency (VHF) electronically steered phased array antenna radar operating at 46.5 MHz and used for mesosphere-stratosphere-troposphere (MST) observations. The MUR was upgraded as of March 2004 in order to apply imaging techniques as a natural progression in the development of atmospheric radar signal processing techniques. In particular, high resolution range images of thin layers in the troposphere and stratosphere have been of interest since the MUTSI experiment of 2000 [Gavrilov et al., 2005]. The upgrade consisted of two parts, the first involving upgrades to antenna elements, cabling and transmitter modules, the second consisting of the installation of a digital multireceiver system. The latter added the capability of transmitting with a maximum of 5 frequencies across a 1 MHz bandwidth using the existing 1 μs pulse, and 25 digital receivers with associated filters and a software combiner. The multireceiver system capabilities permit the application of 25-channel angular imaging, also known as Coherent Radar Imaging (CRI) [e.g., Woodman, 1997]; the use of 5 frequencies permits Range IMaging (RIM) [e.g., Palmer et al., 1999; Chilson et al., 2003], also known as Frequency domain radar Interferometric Imaging (FII, used henceforth in this paper) [e.g., Luce et al., 2001]. Simultaneous combination of multifrequency and multichannel processing can be used for three-dimensional (3-D) imaging. The system also received the addition of an experimental 0.5 μs pulse mode.

[3] This paper describes the upgrades to the MUR system, and the components of the new imaging system, and presents experimental results from the troposphere and stratosphere using the new system. The layout of the paper is as follows:

[4] Section 2 reviews and describes signal processing for imaging, presenting an overview of the procedure.

[5] Section 3 describes the new imaging system, detailing upgrades to the transmission and reception signal transfer system, and the new multichannel digital modulator/demodulator with differences to the previous MUR system noted. As the operation of the transmission subsystem, including the antenna, did not change, the reader is referred to papers on the previous MUR system for details [Fukao et al., 1985a, 1985b]. The emphasis in this paper is on the digital receiver subsystem which provides the capabilities for radar imaging.

[6] Section 4 presents simulation results of imaging with the MUR data analyzed in 3-D, that is, simultaneously scanned both in range (1-D) and in angle (2-D). Since the physics of the atmosphere dictates horizontal layering, FII is useful for viewing the vertical structure of backscatter echoes, while combination with CRI to scan in three dimensions is useful for viewing detailed 3-D structure of selected portions of the data. The processing demands of 3-D imaging are of course much higher so it can become a nontrivial task to process large data sets completely in 3-D. However, 3-D processing can be limited to a single horizontal plane or to a single range line in order to obtain results comparable to CRI and FII results, respectively, but making use of the 3-D imaging function.

[7] Section 5 describes the first use of operational 3-D imaging in an experiment carried out using the new MUR imaging system to observe the troposphere and stratosphere, in particular KH billows. Detailed studies and interpretation of phenomena are reserved for future work. The results of single-frequency 0.5 μs pulse observations are also shown in comparison with both single-frequency 1 μs standard mode observations and multifrequency 1 μs pulse Capon imaging.

[8] Section 6 contains the conclusions and summarizes the capabilities of the new MUR imaging system and its possibilities from the results thus far obtained.

2. Radar Imaging and Data Processing

[9] This section reviews the data processing required for radar imaging and describes how this processing is carried out on the new MUR imaging system. Radar imaging and its predecessors, Spaced Antenna (SA) methods [Briggs, 1984; Doviak et al., 1996] and Dual-Frequency Interferometry (FDI) methods [Kudeki and Stitt, 1987; Palmer et al., 1990], as well as combinations of the two [Palmer et al., 1995], have been discussed in the literature.

[10] All methods are based on the concept introduced by Woodman and Guillen [1974], namely that the VHF radar power spectra and cross spectra (variance) are representative of the Doppler distribution of a large number of randomly distributed volume scatterers. Atmospheric scatter has also been observed to have zenith-angle aspect sensitivity, and azimuthal anisotropy with short timescales [Worthington et al., 2001]. Observation methods in both time (correlation functions) and frequency (spectra) domains evolved to extract as much information about the echoes and their statistics as possible, particularly about turbulence statistics. Pulse compression (coding) techniques improved the sensitivity, followed by the use of more frequencies to obtain subrange resolution when bandwidth limitations were reached. Spaced antenna processing for mean horizontal wind derivation and spaced antenna interferometry, also known Spaced Domain Interferometry (SDI) or as Imaging Doppler Interferometry (IDI) for echo arrival angle estimation, were developed but limited by the small number of separate receiver antennas (or subarrays) available on typical MST radars (4 in the case of the previous MUR system). As a matter of course, imaging was introduced to use all the available information about the scatterer distribution statistics possible from the collected signals.

[11] Imaging is mathematically phrased as an inversion technique. Gubbins [2004] discusses inversion from a geophysical standpoint, including discussion of error analysis (carried out by Hysell and Chau [2006] in the ionosphere), while Woodman [1997] gives an excellent conceptual description of general imaging for atmospheric radar observations.

[12] Two-dimensional angular imaging (CRI) and a variety of Postset Beam Steering (PBS) methods were initially implemented on MST radars [Palmer et al., 1998; Hélal et al., 2001], followed by multifrequency one-dimensional FII experiments on MST radars and Ultra High Frequency (UHF) wind profilers [Palmer et al., 2001; Luce et al., 2001; Chilson et al., 2003; Chen, 2004; Luce et al., 2006], and the combination of SA and FII techniques [Yu and Brown, 2004]. As regards full three-dimensional (3-D) imaging of atmospheric echoes, only a theoretical consideration with simulation has been carried out until now [Yu and Palmer, 2001]. The inversion itself uses various techniques of nonadaptive beam-forming (fixed weights dependent on the array geometry or frequency allocation) and adaptive beam-forming (weights dependent on array geometry and data, such as Capon, Lagunas-Gasull, or MEM), or parametric methods such as the MUSIC and Kumaresan and Tuft (KT) methods, where the statistics of the signals are split into signal and noise spaces based on a priori information or assumptions about the data. Capon's minimum variance method is considered extremely robust but with far greater performance than the simpler Fourier phase-focusing beam-forming method [Luce et al., 2006]. In this paper we use Capon's method in presenting results from the new MUR imaging system.

[13] The visibility (coherence) V between different antenna positions D, subscripted by i and j, and corresponding wave numbers k, subscripted by m and n, measured by a scanning interferometer in the far field, is defined as an integral of the scatterer brightness distribution b over the volume of the radar beam at distances given by R and the pointing angle defined by θ (zenith) and ϕ (azimuth), with the entire volume weighted by the three-dimensional aperture and pulse function W:

equation image

where ϕ = −2ΔkR + kmequation image · equation image + Δkequation image· equation image, Δk = knkm, equation image = equation imageequation image, and equation image is the direction unit vector. Estimates of V are obtained from the cross spectra, differentiated in Doppler frequency f, under the assumption that the process is wide-sense stationary. If we include all frequency components, we can use the cross correlations as estimates for the visibility. The cross spectra are related via the Fourier Transform to the cross correlations of the signals. Specifically, for two signals at arbitrary receive antennas and frequencies,

equation image
equation image

their cross correlation is given by

equation image

where the phase term is given by

equation image

assuming that the convention Eequation image for Rij. The exponential term can be written as −2ΔkR + kmequation image · equation image + Δkequation image·equation image. If we use Rij = Eequation image, then we obtain ϕ = −(−2(knkm)R + knequation image · equation imagekmequation image·equation image), which is the negative of the term given by equation (5). This latter term is the form used in papers by Palmer et al. [1998], Yu et al. [2000] and Yu and Brown [2004].

[14] To calibrate the phase information, we start with a matrix x comprised of the various signals equation imageT (transposed to row vectors) at different antenna locations and for different transmit frequencies. We add 2kmRg to the phase of each signal at frequency km to center the remaining phase difference relative to the range of the center of a gate, Rg, using a statistical phase calibration algorithm [Palmer et al., 1990; Kilburn et al., 1995].

[15] Inversion then obtains an estimate of the three-dimensional spatial brightness distribution b as a function of f compatible with the data (we have no estimates for points other than those actually sampled). In practice, the image is one compatible with a smoothed version of the source brightness structure, due to the regularization of the problem. The regularization can be a posteriori as in the CLEAN [Högbom, 1974] and related algorithms used in radio astronomy, a posteriori and a priori such as for the Capon method and MEM methods, or a priori such as with the WIPE method, a regularized form of the CLEAN algorithm trading resolution for inversion stability [Lannes et al., 1997]. The Capon method assumes that the desired brightness is the one where weighting is adjusted to give minimum overall power to reduce sidelobes, with a gain of unity (phase of zero) in the desired pointing direction. As described in Luce et al. [2001] and others, the weighting (equation imagec) for each frequency and pointing direction is then given by

equation image

where V is the matrix of visibility estimates, equation image(θ, ϕ, R) is the steering unit vector in the direction (θ, ϕ) at range R with the phase term ϕ as calculated above, and {·}equation image denotes the Hermitian operator. The output power Pc of the weighted signals equation image = equation imagex is given by E{equation image}, which simplifies to

equation image

Pc can be calculated for any combination of frequency components of the Doppler spectra.

3. New Imaging System Outline

[16] The specifications of the original MUR system of 1982, and upgrades carried out in 1990, are detailed in Fukao et al. [1985a, 1985b, 1990]. In brief, the MUR system was capable of transmitting at 46.5 MHz, had a receiver bandwidth allocation of 1.65 MHz (defined as the bandwidth containing 99% of transmitted power), and on reception the signals of the 25 antenna groups comprising the main antenna could be arbitrarily combined via analogue circuits into 4 receiver data channels. The signal was digitized at the output of the combiner stage at the 5 MHz intermediate frequency (IF). In 1990, provision was made for transmitting with 2 frequencies [Palmer et al., 1990, 1995], and in 2003 with 4 frequencies, spaced at up to 250 kHz on either side of the center frequency, to investigate FII [Hirono, 2004]. However, since the allocated bandwidth was still limited to 1.65 MHz, the transmit frequencies were limited to a maximal 0.5 MHz spacing. During this period, apart from the scientific results obtained, 2-frequency Frequency Domain Interferometry (FDI) and 4-frequency FII was used to test data-processing algorithms, and acted as inputs into the design process for the new MUR imaging system. In the next subsections, the hardware and software components of the new system are described.

3.1. Multichannel Digital Receiver Subsystem

[17] The new MUR imaging system consists of the design and installation of a digital modulator/demodulator and multichannel digital receiver system. The system, shown in Figure 1, consists of the interface to the MUR, the modulator and demodulator units, and the data processing unit which also acts as the radar controller.

Figure 1.

The modulator/demodulator of the new MUR imaging system allows independent processing of signals from each of the 25 antenna groups.

[18] The new MUR imaging system has the characteristics summarized in Table 1. Imaging is enabled by an extended frequency allocation (defined bandwidth containing 99% of transmitted power) which is increased from 1.65 MHz to 3.5 MHz, permitting a 1 MHz bandwidth for imaging when using a 1 μs pulse, and experimentally the use of a 0.5 μs pulse for standard (single-frequency) observations (performance possibly degraded by the time response of the transmission circuits which were not designed for 0.5 μs operation). The over-allocation of bandwidth will hopefully allow expansion of the hardware in the future. A maximum of five frequencies can be chosen between 46 MHz and 47 MHz in steps of 1 kHz, and do not need to be evenly spaced. It is required though that the order of assignment is made from lowest to highest frequency. The implemented frequency is made via the formula N · (150 MHz)/232, where N is an integer such that the implemented frequency is a close match to the desired frequency (in kHz) set by the user.

Table 1. Parameters of the New MUR Imaging System
MUR System ParameterValue
  • a

    Combination of the number of transmit beams and gates is limited to maximally 4096.

  • b

    An experimental 75 m pulse resolution is available (see section 5.2).

  • c

    Receiver channels can be arbitrarily combined using the digital combiners; however, software provides only 25 data recording channels.

Antenna Specifications
Antenna fieldcrossed Yagi antennas
Configurationcircular array of diameter 103 m
Functionelectronic beam steering pulse to pulse
Full antenna beam width3.6° (one-way half-power full width)
Single antenna group beam width18.0° (one-way half-power full width)
Power distributionseparate TR module per group
 
Transmit System
Maximum peak output power1 MW
Duty ratio5%
Operating center frequency46.50 MHz
Maximum number of frequencies5
Maximum number of transmit beams64a
Pulse resolution150 mb
 
Receive System
Allocated bandwidth3.5 MHz
Receiver channels25
Data recording channels25c
Maximum number of range gates128b

[19] The MUR antenna consists of 25 antenna groups, each with 19 crossed-Yagi elements in E–W and N–S directions (total of 475). 19 of the groups are hexagonal in shape and thus have an approximately circular beam pattern, while the other 6 groups lie on the edge of the antenna, giving the complete antenna its circular shape.

[20] Each antenna group has its own combined transmitter and receiver (TR) module. The antenna beam can be electronically steered pulse to pulse, and all subsystems are computer-controlled, thus allowing rapid setting of and switching between complicated observation configurations.

[21] Transmit signals are sent to all 19 crossed-Yagi elements of all or any of the 25 antenna groups (or optionally, to only a single element in a group, for the case of meteor observations) in accordance with the user-defined configuration. For transmission, ordinarily the entire array is used, and for the case of standard mode and range imaging, this is also true for reception in order to maximize the return signal power: the antenna is driven by independent sets of power amplifiers, with one connected to each antenna group. For angular imaging, however, separate signals from the spatially separated antenna groups are required. Postprocessing can then be used to combine signals corresponding to various configurations of receive antennas.

[22] The transmit beam directions can be chosen, as for the old MUR system, in the zenith direction at 1° intervals up to 16° off-zenith, and from there at 2° intervals out to 30° off-zenith as governed by the original MUR antenna pattern specifications. In the azimuth direction the directions can be chosen at 5° intervals for all chosen zenith directions. Each frequency allocation counts as one beam and needs to be specified by a corresponding direction. This limitation impacts the total number of independent beam directions possible (see section 3.1.2). A further limitation is the total shared memory, which may require a reduction in the number of time series data points when the number of coherent integrations is low and the number of beams, frequencies and receive channels is high. This is not as severe a restriction as it may seem, since the MUR system is coherent from record to record.

[23] The operation of MUR observations is controlled by user-defined radar parameters stored, as for the old MUR system, on the data processor which also acts as radar controller (see section 3.1.3 and Table 2). Software written at RISH runs a schedule file containing a list of parameter files, each corresponding to an observation configuration, and including the number of iterations for an experiment or set of experiments. The software sends the parameters from a single parameter file to the TR module for each antenna group over serial connections using a multidrop protocol. At each TR module, the required phase for each Yagi element in that antenna group is stored, for each of the defined beam directions, so that the initial setup need be done only once for each new set of observation parameters. Once the TR modules have been prepared, the radar observation starts, with the controller server sending signals to the digital modulator system. The observation continues for the specified number of transmit pulses and iterations of a single parameter file. Once the number of pulses has been completed, the radar stops and the software on the controller machine sends the parameters from the next parameter file to the transmitter modules. There is no interval between logical records, that is, iterations of the same parameter file, so data can be viewed as a continuous coherent complex time series sequence. However, depending on the complexity of the parameter file, there may be more than one minute of system setup delay between observations with different parameter files.

Table 2. Parameters of the New MUR System Data Processor
Data Processor ParameterSpecification
  • a

    485 GiB for data storage, 50 GiB for the operating system, MUR control and software, configuration files, and temporary storage.

  • b

    A kernel upgrade is not envisaged, and the existing Universal Serial Bus (USB) subsystem is not useful for data transfer, so instead data should be copied across the network to a separate machine.

  • c

    The internal magnetic drive has given problems, thus only the external media is used.

  • d

    Two external drives.

Operating systemGNU/Linux Red Hat 8.0 (kernel v.2.4.18-14)
Main memory4 GiB
Hard disk535 GiB SATAa
Networknine 1000BaseT ports (20 MiB s−1 continuous)
Optical mediaCD-ROM(24X)/DVD-ROM
Floppy drive3.5 in 1.44 MiB
Other portsone parallel, one serial, two USB v.1.1b
Internal magnetic media20 GiB (2.75 MiB s−1)c
External magnetic media160 MiB (16 MiB s−1)d

[24] The interface to the MUR, shown in Figure 2, connects the existing MUR system to the new digital multichannel system. The combiner/divider now passes signals from the 25 individual antenna groups separately, as well as signals from 4 user-defined analogue combinations of the 25 antenna groups, to the new digital demodulator unit. In the following subsections we describe the modulator and demodulator functions and operation.

Figure 2.

The interface between the new MUR imaging system and the existing MUR system adds functionality for the new modulator/demodulator unit, but also keeps the old analogue combiner functionality intact.

3.1.1. Modulator Unit

[25] This section describes the capabilities and operation of the modulator unit, shown in Figure 1. The unit consists of a modulator card, a Unit Master Central Processing Unit (UCPU) card which controls the local operation, a Global Positioning System (GPS) interface card, timing control and distribution cards, all linked by a Versa Module Eurocard (VME) bus. Overall control is by the data processor (see section 3.1.3).

[26] For transmission, the 5 MHz IF modulation signal is generated by a coherent oscillator (COHO) on the modulator card, passed through an analogue filter on the local signal card, and sent to the divider/combiner unit. At the initiation of the experiment, timing is synchronized with a GPS system to 10 μs accuracy (UTC time). The observation start can be delayed from the GPS signal by up to 1920 μs, settable in 128 μs steps. A 41.5 MHz local signal is generated by a stable oscillator (STALO) on the local signal card and sent to the divider/combiner unit. The local signal frequency can be changed at each inter-pulse period (IPP) between 42.00 MHz and 43.00 MHz for FII experiments.

[27] The control signals for controlling the TR modules (IPP signal, start signal, receive (Rx) busy signal), as seen in Figure 2, are sent to the divider/combiner unit. These signals and the associated serial communications lines are part of the old MUR system. Control commands from the data processor for the modulator and demodulator are received, processed and executed for the control of the entire modulator unit and the 29 digital receivers in the demodulator unit. Required clock and trigger signals are also provided via the modulator unit. The base signal is taken from a 10 MHz crystal oscillator, but can be taken from the GPS signal instead.

[28] The modulation signal can use subpulse widths of 1, 2, 4, 6, 8, 16, 32, 64, 96, or 128 μs, with a maximum total pulse length (including no transmission subpulses) of 2048 μs. The maximum number of subpulses is 512. Subpulses are now three-state instead of binary with the addition of a no-transmit state added. This new feature gives more freedom in the pulse modulation, including multipulse techniques for incoherent scatter (IS) observations. The maximum code length (number of consecutive transmit subpulses) is 32. Modulation signals can be looped with IPP up to 64 times, allowing a total of 64 consecutive and different modulation sequences (patterns). The IPP is variable from 200 μs to 65535 μs in 1 μs steps.

[29] Optimal or truncated code sequences [Spano and Ghebrebrhan, 1996a, 1996b, 1996c] can now be used. Signals are switched in frequency at each IPP, so that a truncated code (Spano-Ghebrebrhan code) for each transmitted frequency uses the coherent integrations gathered over effectively identical time periods to add an extra dimension to pulse decoding to remove sidelobes, rather than using only combinations of already integrated pulses. This allows partial forms of the complete (integrated pulse) code sequences to be decoded, permitting the observation starting range to be lowered, but an additional constraint is that the number of coherent integrations required must be equal to two times the (integrated pulse) code sequence length. For each integrated pulse of the complete code sequence, the code sequence for the set of pulses to be integrated is changed. Thus, for the 16-bit Spano-Ghebrebrhan truncated code sequence, a total of 32 different pulse sequences are required. With the old MUR system it was possible to change between only 2 sequences (for example, complementary code sequences).

3.1.2. Demodulator Unit

[30] This section describes the capabilities and operation of the demodulator unit, shown in Figures 1 and 3. In the old MUR system, an analogue combiner with 25 input and 4 output channels was present at the IF stage, so the (maximum of) 4 outputs of this combiner were quadrature-demodulated at 5 MHz IF in analogue, and then digitized at 1 MHz. For the new MUR system, the receiver module for each of the 25 antenna groups passes the combined signal from its attached 19 crossed-Yagi elements at the 5 MHz IF to the demodulator via the divider/combiner unit (Figure 2). 25 individual outputs (one for each antenna group) and 4 outputs from the extant analogue combiners (signals now digital) are connected to the multichannel receiver subsystem.

Figure 3.

Each receive channel of the new MUR imaging system demodulates one antenna group signal.

[31] The demodulator unit, shown in Figure 1, consists of a Block Master CPU (BCPU) card and digital receiver for each of 29 channels, with timing governed by the GPS-locked clock. The IF signal noise level is −69 dBm, with the dynamic range of the receiver 70 dB maximum, both for combined received signals and for that from individual groups.

[32] An IF digitizer (14 bit ADC) samples the 5 MHz analogue IF signal at 20 MHz for each channel, shown in Figure 3, and the output is then phase-detected (separation of I and Q) by a 5 MHz digital sinusoidal wave.

[33] A matched filter is implemented digitally and carries out band-limiting and decimation by a Cascaded Integrator-Comb (CIC) filter (default of 6, maximum 10 stages) followed by a 16-tap Finite Impulse Response (FIR) filter which adjusts amplitude and frequency and outputs a baseband signal at a sampling rate adjusted to the subpulse width.

[34] A notable feature is the option of a small frequency shift to accommodate Radio Acoustic Sounding System (RASS) observations, for which the center Doppler frequency of the received signal can be adjusted downward from 0 Hz to 120 Hz in 0.1 Hz steps.

[35] Pulse compression (decoding up to 32 bits, looping up to 64 times) of the output of the matched filter is carried out, and optionally also coherent integration (1–256).

[36] The maximum number of range gates Ng is 4096, and the maximum number of beams Nb is 256, subject to the limitation Ng × Nb ≤ 4096.

[37] A 0.5 μs pulse experimental mode is provided which utilizes two consecutive pulse sequences with half-gate shifted gate timing on the second sequence in order to obtain gated signals at intervals of 75 m: the gating interval cannot at present be set to less than 150 m. Thus, there is a reduction by a factor of two in time resolution. Although experimental, the results shown in section 5 demonstrate that this mode can be usefully employed for nonimaging observations.

[38] Optionally, after coherent integration a Fast Fourier Transform (FFT) can be performed on 64, 128, 256, 512, 1024, 2048, or 4096 data points. Maximally 3 independent frequency ranges can be specified for storage.

[39] The Local Area Network (LAN) connection is made via a double buffer memory with a one-sided maximum buffer space of 256 MiB. The time series data (or spectral data) are sent over 1000BaseT Ethernet LAN to the data processor in 32-bit floating point format.

3.1.3. Data Processor

[40] The data processor acts as a control computer to set up the TR modules via existing serial control lines, control the operation of the new MUR modulator and demodulator units via File Transfer Protocol (FTP) connections, and can perform further data processing of the received time series (or spectral data). The specifications are shown in Table 2.

[41] Digital combiner software written at RISH, following rules set up in the user-defined observation parameters, combines signals from the 25 antenna groups and 4 analogue-combined channels into a maximum of 29 different channels of arbitrary configuration. If more combinations of channels are required, this is done in offline postprocessing in order to save data-transfer bandwidth. If time series data (as opposed to spectral data) are received, an optional FFT can be carried out; if spectral data are received or calculated in the data processor, incoherent integrations can be carried out. Received and processed data can then be viewed using quick-look software for raw and spectral data, or viewed using other computers connected to the LAN. Received and processed data are stored together with a GPS time stamp on local hard disks (little-endian format) and a tape device. The data on the hard disks are then transferred over a Wide Area Network (WAN) connection to the RISH site in Uji to be stored in the MUR database (in big-endian format) where they are kept accessible online using Redundant Array of Independent Disks (RAID) storage.

3.2. Maintenance Upgrade of Transmission Subsystem

[42] The architecture of the MUR transmission subsystem remains identical to that described in Fukao et al. [1985a, 1985b, 1990]. The TR modules are located on racks in 6 booths set around the main antenna, with serial I/O lines to communicate observation parameters from the radar controller workstation.

[43] The maintenance upgrade to the antenna radiators and transmitter modules involved checking all 475 crossed-Yagi antennas (each with 2 elements, one N–S and one E–W) for designed Voltage Standing Wave Ratio (VSWR) less than 1.5, and replacing the worst 286 elements along with their associated connectors and coaxial cabling.

[44] Mechanical relays for polarization switching were removed from the TR modules permanently, to reduce signal transfer loss and possible damage to the isolation diodes and power transistors owing to poor VSWR and power reflection. Thus, the new MUR system is now fixed with right-handed circular polarization on transmission and left-handed circular polarization on reception, with phase accuracy within 2°. Although polarization might become a requirement in the future, polarization experiments in the past were rare, so the loss of this functionality was considered an acceptable compromise.

4. MUR Imaging Performance

[45] The new MUR performance for 1-D (FII), 2-D (CRI) and 3-D imaging was simulated using the following parameters. For FII, 5 evenly spaced frequencies in a 1 MHz bandwidth around the center frequency were used, the pulse width was set to 1 μs, and the full MUR antenna array was used for transmission and reception. For CRI, only the center frequency was utilized, transmission was carried out with the full array and reception was carried out using the 19 pseudo-circular antenna groups, and the pulse width was set to 1 μs. For 3-D imaging, we used the 5 frequencies and the full antenna array in transmission as for FII, and receive the signals from the 5 frequencies at the 19 pseudo-circular antenna groups as for CRI. FII simulation assumed that a finite number of layers of finite thickness were positioned horizontally at a given relative range from the center of a range gate, while CRI simulation assumed that one or more point targets were positioned at the central range of a range gate at a given zenith and azimuth within the beam width of the MUR. 3-D imaging simulation similarly assumed one or more point targets, but with complete positional freedom inside the range gate volume.

4.1. 1-D Performance

[46] The performance of various signal processing algorithms for imaging (inversion at each scanned position) for the case of FII under conditions of varying SNR and for layers of various thicknesses and separation distance was discussed by Palmer et al. [1999] and Luce et al. [2001]. The Capon method was shown to be robust and produce no known signal-processing artifacts. For SNR lower than 10 dB the Capon method results gradually degraded and approached the performance of the Fourier method.

[47] Performance comparisons of Fourier (left panel) and Capon (right panel) methods for a point target located at the center of a range gate are shown in Figure 4. The brightness was corrected for pulse width. Numerical accuracy limited the peak value of the Capon processing in the case of high SNR. From the top of each panel, the curves shown correspond to SNR of −10, −3, 0, 10, 20, 30, and 40 dB. It can be seen that the Fourier method approaches its best performance at a SNR of approximately 10 dB, while the Capon method shows continued improvement as the SNR increases.

Figure 4.

The simulated performance of equally spaced 5-frequency Fourier (left) and Capon (right) range imaging (FII) using maximal 1 MHz spacing under varying SNR conditions for a point target located at the center of a range gate of size 150 m is shown. From the top, the lines correspond to SNR of −10, −9, −6, −3, 0, 10, 20, 30 and 40 dB.

4.2. 2-D Performance

[48] For CRI, the MUR antenna array was divided into 25 antenna groups for reception, of which the 19 pseudo-circular groups were used for imaging. The array is shown in Figure 5a, while the coarray, or array of antenna spacings (baselines), is shown in Figure 5b.

Figure 5.

(a) The MUR antenna and its 25 groups. The medium-gray groups at the edge of the antenna field are not used for CRI since their noncircular beam patterns make analysis difficult. The pale-gray groups were used to compare the performance of CRI with one-dimensional array imaging along the W–E and N–S directions. (b) The corresponding coarray of the MUR antenna, showing all possible baselines (antenna spacings) for CRI using the 19 pseudo-circular antenna groups for reception. The angle is shown as a bearing, following the convention of 0° towards the north.

[49] The expected performance of the MUR for CRI, using all 19 pseudo-circular antenna groups for reception, is shown in Figure 6. The resolution of the Fourier imaging method is approximately 4.5°, while that of the Capon imaging method is approximately 1.5°. The beam width of the full MUR antenna is of the order of 3.6°, so Capon imaging is capable of resolving targets inside the beam, while Fourier imaging is able only to reduce the effect of the sidelobes.

Figure 6.

The CRI performance of the MUR using with 19 antenna groups used to receive signals, assuming a vertical transmit beam using the entire array is shown (a) for Fourier and (b) for Capon imaging method. The simulation is for a target located at zenith at the center of a range gate. Brightness is in units of dB and the SNR was fixed at 20 dB.

[50] We compared how (2-D) CRI resolution for a given target differs from that provided by 1-D arrays. The upper panel of Figure 7 shows that brightness scanning for CRI is similar for both W–E and N–S directions for both Fourier (blue solid lines) and Capon (red solid lines) methods. The difference with direction is so small that only one line is required, and the lower panel shows how small this difference in brightness for these orthogonal directions is. The difference arises because of the positions of the receive antenna groups: their alignments with respect to the W–E direction are slightly different to their alignments with respect to the N–S direction. An important aspect of using many receiver groups is that for any direction, the errors will become similar, thus image distortion due to differences in resolution are minimized. However, resolution in any specific direction is poorer (approximately 1.5°) than for an alignment of antennas along that direction (1.0°). This is shown in the upper panel of Figure 7, where the dotted lines show the performance of the 5 antennas aligned most closely in the W–E direction, and similarly the dashed-dotted lines show the performance of the 5 antennas most closely aligned in the N–S direction. Blue indicates the Fourier method, red the Capon method. It is evident that for such linear arrays, the performance along the chosen direction is superior to that of the configuration using all 19 pseudo-circular groups, but the performance has a large variance, being markedly degraded for directions away from the chosen baseline orientation, with a worst case in the orthogonal direction where change in interferometric phase with change in target position becomes close to zero (exactly zero on hyperbolas symmetrical about a point on the baseline).

Figure 7.

The CRI performance of the MUR at a SNR of 20 dB along the W–E and N–S axes is shown. The upper panel shows with solid lines the Fourier (blue) and Capon (red) method brightness using all 19 pseudo-circular receive antennas for these directions. Dotted lines show the performance of the 5 antenna groups most aligned along the W–E direction; dashed-dotted lines show the performance of the 5 antenna groups most aligned along the N–S direction. The lower panel shows the minimal difference in brightness between the directions for Fourier (blue) and Capon (red) methods using all 19 pseudo-circular antenna groups for reception. This difference is owing to the misalignment of true W–E and N–S with the actual baseline directions.

4.3. 3-D Performance

[51] The Capon method performance in 3-D varies as a function of the SNR and number of point targets is shown in Figure 8. Figures 8a and 8c show performance for a SNR of 0 dB and 20 dB for a single point target at the center of a range gate, while Figures 8b and 8d show the corresponding performance for two targets set at 50 m above and below the center of the range gate, respectively. No beam correction for CRI is carried out, but corrections for pulse width are made. In the cases shown here, the variation in SNR does not have a major impact on the resolution: it is the number and spacing of targets which is critical, that is, the imaging parameters of the radar system (band width and baseline lengths). Figure 9 shows the same data as Figure 8, but cut in the W–E plane (90° azimuth) and passing through the center of the beam to show more clearly the angular and range resolutions. With the use of all 19 pseudo-circular antenna groups for reception, 3-D imaging is effectively homogeneous across all azimuth angles at the resolution offered by the Capon method. We note that 3-D has lower resolution than 1-D (compare right panel of Figure 4), but this deficit is compensated for by the homogeneous volumetric imaging capability. The mechanism behind the 3-D performance degradation is the relationship between the MUR system wavelength and imaging geometry. For angular imaging, the MUR is not so well-suited, as the smallest possible baseline is not small enough to ensure a high coherence between receiver signals. Thus, the number of useful receiver spacings for relatively high coherence is limited when considering observations of the troposphere and stratosphere. For 3-D imaging, the longer antenna baseline spacings reduce the total coherence even when the frequency spacing alone (at a single receiver antenna group) may have very high coherence.

Figure 8.

The simulated performance of 3-D normalized Capon imaging with the MUR, including pulse width and but no beam width correction, is illustrated for point targets at different SNR, showing the −3 dB surface. Figures 8a and 8c are for a single point target at zenith at the center of a range gate, while Figures 8b and 8d are for two point targets at zenith and offset in range ±50 m.

Figure 9.

The 2-D cut of the simulated performance of 3-D normalized Capon imaging of Figure 8 is shown in the W–E azimuthal plane, with pulse width but not beam width correction. Figures 9a and 9c are for a single point target at zenith at the center of a range gate, while Figures 9b and 9d are for two point targets at zenith and offset in range ±50 m.

[52] The performance of 3-D imaging in the range direction (in a line along the zenith) can be compared with the performance of FII. This is shown in Figure 10, where it can be observed that 3-D imaging displays degraded resolution of approximately 40 m compared to approximately 20 m for FII. However, a benefit of 3-D imaging is that resolution does not change substantially as the target varies from its ideal position at zenith. It is therefore possible to image targets at any position in the range gate volume without significant distortion, albeit with lower resolution than if they were a priori known to be at zenith.

Figure 10.

The 3-D imaging performance of the MUR for a point target along the zenith (Fourier 3-D imaging dotted line, Capon 3-D imaging solid line) is compared with FII, that is, 1-D, Capon method performance (dashed line). The significant difference between FII and 3-D imaging is owing to the long horizontal baseline components used in 3-D imaging.

[53] A key benefit of 3-D imaging, namely the ability to reflect more accurately than FII what lies along a particular radial direction of the radar beam, can be extremely useful for oblique beam pointing, for wide beam radar, and for vertical beam observations where the target layers are uneven over the width of the radar beam, such as occurs for undulating layers due to KH billow s or gravity waves, for example. The extent of the effect of layer variation on the image for a given imaging geometry is a combination of time resolution, horizontal and vertical wind speed, and beam width.

[54] Comparison of 3-D imaging with CRI in the horizontal plane at the center of the range gate showed only an insignificant performance difference between 3-D imaging and CRI. This is expected in view of how important the baseline length effect, mentioned above, is in reducing coherence. Thus, for assumed horizontal layer structures, stationary or moving in the vertical direction, it is advantageous to use FII only, with the full antenna array used for reception; whereas to study possible tilts and associated motions CRI is required. 3-D imaging is required when there is ambiguity about more than one target in the range gate, such that CRI would not distinguish the tilts or azimuthal positions, or when a horizontal layer is inhomogeneous, which would not be apparent from FII images. Use of 3-D imaging is particularly favourable in the presence of aircraft echoes, since the imaging function is narrower in the FII and CRI functions and can thus be used to filter aircraft echoes away from the range gate center, producing data with markedly less interference than either CRI or FII.

[55] The estimated resolution of 3-D Capon imaging for the present processing method is summarized in Table 3.

Table 3. Simulated Performance of Capon 3-D Imaging With the MUR Is Shown Under Varying SNR for a Single Point Target at the Center of a Range Gate at Zenitha
SNR [dB]ΔR [m]ΔA [deg]
  • a

    The values for range resolution (ΔR) and angular resolution (ΔA) are taken from simulation results such as those shown in Figure 9.

40361.38
30381.40
20391.40
10401.40
0411.50
−3421.60
−6441.80

5. Experimental Results

[56] This section gives examples of the application of the new MUR imaging system to 1-D, 2-D and 3-D imaging in the troposphere and stratosphere. The use of 0.5 μs (75 m range resolution) pulse operation is also demonstrated to show its improvement in standard mode (Doppler beam swinging, or DBS) operation over the usual 1 μs (150 m range resolution) pulse operation.

5.1. 1-D, 2-D, and 3-D Imaging

[57] Imaging data in this section were obtained on 26–28 December 2005 using the parameters listed in Table 4. Data were collected continuously, and the sequence of 512-point (32 s) records processed at a time resolution of 8 s (128 points). The same raw data were processed by applying the 1-D, 2-D and 3-D Capon method. For 1-D imaging, the received signal from the full antenna array was used, whereas for 2-D and 3-D imaging signals from the 19 pseudo-circular antenna groups were used. During processing, 64-point oversampling was carried out to smooth the final images. From 1-D images, much information on atmospheric dynamics could be deduced, as shown in Luce et al. [2006, 2007], and areas for further detailed study via 2-D and 3-D imaging identified.

Table 4. Parameters of the MUR Imaging Experiment on 26–28 December 2005, Configuration Exp-Aa
MUR Configuration Exp-A ParameterValue
  • a

    The full capabilities of the hardware were utilized to obtain high time and spatial resolution.

  • b

    Spano-Ghebrebrhan coding.

  • c

    Lower and upper range gate centers.

  • d

    Full antenna, 19 hexagonal antenna groups.

Transmission
Number (direction) of beams2 (Vertical, 10 Northward)
Transmit frequencies [MHz]46.00, 46.25, 46.50, 46.75, 47.00
−3 dB two-way beam width [deg]3.5
Pulse range resolution [m]150
Pulse coding8-bit S-Gb
 
Reception
Number of receive antennas20c
Range sampling [m]150
Number of ranges sampled128
Range covered [km]1.050–20.100d
 
Time Series
Number of coherent integrations16
Number of time series data points512
Record duration [s]32.768

[58] An example of apparent KH billows will be discussed which could be detected in 1-D imaging as well as in SNR RTI (range time intensity, herein the definition is expanded to include power and Capon brightness) plots from a single frequency. We also note the effect of imaging on an aircraft echo which entered into our chosen observation data. Observations will be analysed in more detail in future works, and are presented here to illustrate the operation of the new MUR imaging system. Figure 11a from 28 December 2005 shows the observation data after 1-D Capon method processing. It is interesting how thin structures at the edges of the KH billow structures appear and how the descending sections of the structures, of which five could be distinguished, were resolved. An aircraft echo is visible with characteristic parabolic echo shape between approximately 18:54–18:55 LT. Figure 11b shows the same raw observation data after 3-D normalized Capon method imaging, scanned along the central vertical axis of each range gate.

Figure 11.

The RTI plots of observational data from 27 December 2005 show suspected Kelvin-Helmholtz instability structures. Figure 11a shows FII, that is, 1-D, Capon imaging, while Figure 11b shows 3-D normalized Capon imaging of a region showing Kelvin-Helmholtz instability and aircraft echoes. The aircraft echoes are markedly reduced by 3-D processing.

[59] The complex covariance matrix (visibility matrix when all Doppler components are utilized) using 5 frequencies and 19 pseudo-circular antenna groups had dimensions 95 × 95. Larger matrices display more easily a phenomenon known as ill-conditioning, in which numerical accuracy is lost during the matrix inversion process (this has little affect on the Fourier imaging method where no inversion is required). Normalization of the covariances was used in an attempt to improve the conditioning of the visibility matrix used in the Capon method processing. The brightness in each gate was then unnormalized after the inversion operation. In Figure 11, for later reference, intervals of 10 segments are indicated by a vertical line, and the center of the gate at height 5.55 km is indicated with a horizontal line. This height was used in comparing 2-D and 3-D imaging, since the descending structure passed through this height several times.

[60] In comparing the results of the 3-D processing in Figure 11b with those of the equivalently processed 1-D results of Figure 11a, the 3-D imaging case show slightly different detail in consequence of being able to scan in the angular direction desired in 3-D imaging with a narrow beam while rejecting signals from other angles, rather than covering the equivalent azimuthal plane with a single wider beam as in 1-D imaging. A related useful feature is that the aircraft echo was much reduced in the number of time segments that it appeared. As the aircraft approached the center of the radar beam it became visible also in the 3-D imaging, but disappeared as soon as it moved outside the angular and range resolution of the 3-D Capon imaging function. Thus, the narrow 3-D imaging function can be used to discriminate against aircraft echoes.

[61] The analysis of the KH billow was as follows. To determine the billow orientation, wind and shear was estimated from 1-minute averaged DBS data at 18:24 LT and 19:35 LT. From the DBS data (not shown), the wind components appeared steady over the altitude range 5.30–6.00 km for the duration of the observations shown here, with a meridional (northward positive) component of −15 m s−1 (southward), and a zonal (eastward positive) component linearly increasing from 40–42 m s−1 (eastward) at 5.30 km to 65–70 m s−1 at 6.00 km. This implied a minimum westward shear of between 33–43 m s−1 km−1 over the altitude range 5.30–6.00 km, drifting on a background wind oriented on a bearing between 103–110°. Thus, KH billows were expected to be oriented N–S but drifting in the direction of the wind. The wind speed of approximately 53 m s−1 at 5.50 km range carried a section of the billow completely through the approximately 350 m wide radar beam (at 5.50 km) in about 6.50 s, while our observation was made at 8 s. Therefore, we could not see clearly any single part of the KH billows, but instead observed harmonics of parts of the billows integrated together (smeared angularly and in range) by the time exposure of the observation. The length of the billows, calculated from the wind speed (53 m s−1) and echoes seen in the Capon method processed images (72 s for one billow), was approximately 3.80 km. The height of a billow could be estimated as 500 m or less, leading to a L/h ratio of about 7.60, which agreed well with theory [Hocking, 1987]. Since the length of the billow was 3.80 km, approximately 9% of it passed through the radar beam during each 8 s observation exposure, so that we could expect to observe the general characteristics of descending edges of the billows, and smeared versions of the double-layer structure.

[62] Detailed 3-D views of a KH billow structure are given in Figure 12, where Figures 12a12c show progressively rotated plots of gates 25–40 with different cuts of the data for segment 84, which corresponds to the last visible KH billow structure in the data of Figure 11. The Capon brightness color scale covers a range of −5 dB to +15 dB, as in Figure 13. For this full 3-D processing, range scanning was reduced to 10 m, and angular scanning in a plane at each scanned range was set to 0.25°. In order to reduce the edge effects after correcting for pulse width, we removed the 2 scanned ranges closest to each gate edge, interpolating these instead to obtain smooth plots. Figure 12a illustrates how smearing creates double-layer structures at 5.50 km and 5.90 km moving into and out of the beam, only visible with 3-D processing. Tilt in the meridional direction is visible in Figure 12b, while Figure 12c illustrates better the structure orientation in the horizontal plane. The 3-D plots show that layers in a single gate exhibited curvature in two dimensions across the width of the beam. In such a case, the echo would appear as a much thicker (assumed horizontal) structure in a 1-D image. With the use of 3-D scanning the structures could be resolved at any location.

Figure 12.

The 3-D normalized Capon imaging of data in segment 84 of Figure 11 for the 16 range gates 25–40 covering 4.575–6.975 km is shown in Figures 12a–12c with slightly different slices and at progressively rotated orientations to depict more information about the observed layer morphology. Figure 12a shows clearly the entry and exit of different layers, Figure 12b shows tilt in the meridional direction, while Figure 12c shows horizontal orientation to good effect.

Figure 13.

W–E and S–N cuts of the 3-D normalized Capon imaging for the 4 segments 81–84 of Figure 11, for the 16 range gates 25–40 covering 4.575–6.975 km, are shown. The progression of layer movement into and out of the beam is clearly visible.

[63] A cut vertically in the W–E and S–N planes through the center of the range gates is shown in Figure 13 for segments 81–84 from Figure 11. The DBS wind information was used to determine the orientation of the cuts to be parallel and orthogonal to the horizontal wind vector. In segments 81–84, the single layer appeared to split, possibly indicating the start of the next billow. In segment 84 we could see possible observation of a section of two consecutive rolls in the W–E cut, with the beginning of the next roll starting to appear at 5.50 km in the east section of the W–E figure. In order to reduce the smearing, the time resolution of the plots could be reduced, but coherence would decrease also: better images would be obtained for cases of lower background wind.

[64] The same data as shown in Figure 13 was analysed in the horizontal plane, shown in Figure 14 for a 2-D cut of the full 3-D Capon imaging at the previously selected segments and a range of 5.545 km, as close to the center of the gate at 5.55 km (indicated by a horizontal line in Figure 11) as our chosen 3-D range scanning allowed. At this range resolution, the echo pattern is probably more representative of the variation in tilt of the echo layers (harmonics of the billow structure) than of the orientation of the billow structure itself. We could still observe the passage of the billow through this range position in segments 81–84. To compare, we processed the data also using 2-D Capon method imaging, using a higher angular resolution of 0.1°, the results of which are shown in Figure 15. Here, since the echoes were integrated in range, we could distinguish the passage of the billow as it descended through the chosen range gate in segments 81–84. In both Figures 14 and 15, an orientation perpendicular to the southwest, or a motion to the southwest, appeared discernible. In the smeared images, we could not distinguish between orientation of billows, and drift owing to the background wind, so such an appearance could be expected. Further detailed analysis of the KH billows and their relation to the wind and shear will be carried out in subsequent work.

Figure 14.

The 3-D normalized Capon imaging of the data in Figure 11 is shown in the 2-D plane for 4 segments, from 81–84, at 5.545 km, close to the center of range gate 31 (5.55 km).

Figure 15.

The 2-D normalized Capon imaging of the data in Figure 11 is shown for 4 segments, from 81–84, at 5.55 km which is the center of range gate 31.

5.2. 0.5 μs Pulse Observations

[65] The standard minimal subpulse width of the MUR is 1 μs, giving a range resolution of 150 m in standard mode and imaging observations. In the new MUR system, an experimental mode using 0.5 μs pulse width is also possible, even though no hardware in the TR modules for the individual antenna groups, the transmission lines from the modulator/demodulator to the TR modules, and the Yagi antenna elements were upgraded. Gating remains at 1 μs (150 m) intervals, necessitating two pulses transmitted sequentially with the gating offset by half the pulse width (0.5 μs) to give an effective 75 m gating. Tests were carried out to determine how usable this mode of operation is, given possible hardware bandwidth limits and corresponding pulse shape degradation, and results shown in Figure 16.

Figure 16.

The 0.5 μs and 1 μs pulse SNR in Figures 16a and 16c is compared with 0.5 μs SNR and 1 μs pulse Capon imaging in Figures 16b and 16d. Figures 16a and 16b cover the range 5.50–8.00 km, and Figures 16c and 16d the range 3.00–5.50 km. The 1 μs data (either SNR or Capon brightness) comprise the left and right portions of data shown in each figure part, while the 0.5 μs SNR data comprise the center portions.

[66] A comparison of SNR for 1 μs and 0.5 μs pulse operation on 20 July 2005 for two different altitude ranges, 3.00–5.50 km and 5.50–8.00 km is shown in Figures 16c and 16a, respectively. The 1 μs data comprise the left and right portions of Figures 16c and 16a, with the 0.5 μs data in the center portions, separated by white spaces indicating DBS observation times. The 1 μs data were obtained using slightly different radar configurations, Exp-B and Exp-C on the left and right, respectively (see Table 5 and Table 6). The 0.5 μs data were obtained using configuration Exp-D (see Table 7). Apart from the pulse width, the other major difference between the three configurations is that Exp-B used 4-bit Spano-Ghebrebrhan coding, while Exp-C and Exp-D used 16-bit Spano-Ghebrebrhan coding. This difference caused a difference in SNR between the two portions of 1 μs data. The data were processed using 32 time series points for configurations Exp-B and Exp-C (observation time 16.4 s), and 128 time series points for configuration Exp-D (observation time 13.1 s). The autospectrum was calculated and the moment method used to find spectral parameters, including the sum of signal power plus noise power. The noise density was calculated from the highest gates in which no signal was present, using the method in Sato and Woodman [1982], after which the SNR was calculated.

Table 5. Parameters of the MUR Imaging Experiment on 20 July 2005, Configuration Exp-Ba
MUR Configuration Exp-B ParameterValue
  • a

    The full capabilities of the hardware were utilized to obtain high time and spatial resolution.

  • b

    Spano-Ghebrebrhan coding.

  • c

    Lower and upper range gate centers.

  • d

    Full antenna, 19 hexagonal antenna groups, 1 hexagonal group duplicated accidentally.

Transmission
Number (direction) of beams2 (Vertical, 10 Northward)
Transmit frequencies [MHz]46.00, 46.25, 46.50, 46.75, 47.00
−3 dB two-way beam width [deg]3.5
Pulse range resolution [m]150
Pulse coding4-bit S-Gb
 
Reception
Number of receive antennas21c
Range sampling [m]150
Number of ranges sampled128
Range covered [km]1.500–20.550d
 
Time Series
Number of coherent integrations128
Number of time series data points128
Record duration [s]65.536
Table 6. Parameters of the MUR Imaging Experiment on 20 July 2005, Configuration Exp-Ca
Configuration Exp-C ParameterValue
  • a

    The full capabilities of the hardware were utilized to obtain high time and spatial resolution.

  • b

    Spano-Ghebrebrhan coding.

  • c

    Lower and upper range gate centers.

  • d

    Full antenna, 19 hexagonal antenna groups.

Transmission
Number (direction) of beams2 (Vertical, 10 Northward)
Transmit frequencies [MHz]46.00, 46.25, 46.50, 46.75, 47.00
−3 dB two-way beam width [deg]3.5
Pulse range resolution [m]150
Pulse coding16-bit S-Gb
 
Reception
Number of receive antennas20c
Range sampling [m]150
Number of ranges sampled128
Range covered [km]1.500–20.550d
 
Time Series
Number of coherent integrations128
Number of time series data points256
Record duration [s]131.072
Table 7. Parameters of the MUR 0.5 μs Experiment on 20 July 2005, Configuration Exp-D
Configuration Exp-D ParameterValue
  • a

    Spano-Ghebrebrhan coding.

  • b

    Full antenna, 19 hexagonal antenna groups.

  • c

    Lower and upper range gate centers.

Transmission
Number (direction) of beams2 (Vertical, Vertical)
Transmit frequency [MHz]46.50
−3 dB two-way beam width [deg]3.5
Pulse range resolution [m]75
Pulse coding16-bit S-Ga
 
Reception
Number of receive antennas20b
Range sampling [m]150
Number of ranges sampled128
Range covered [km]1.500–20.625c
 
Time Series
Number of coherent integrations128
Number of time series data points256
Record duration [s]26.2144

[67] The finer resolution and the matched gating interval of the 0.5 μs pulse led to a clearer presentation of the atmospheric echo compared to the 1 μs pulse data. In Figure 16c, the echo structures seen in the 1 μs pulse SNR at 13:26 LT at 4.00 km and 4.50 km compared well to the 0.5 μs pulse SNR at 13:51 LT. The lower layer appeared to rise from 4.00 km at 13:51 LT to 4.20 km at 14:46 LT, from where it connected to the 1 μs pulse SNR at 14:51 LT. In Figure 16a the layer at 13:26 LT around 6.90 km in the 1 μs pulse SNR connected well to the 0.5 μs pulse SNR at 13:51 LT and then appeared to split into two layers at around 14:26 LT, descending and connecting to the 1 μs pulse SNR at at 14:51 LT around 6.80 km and 6.40 km.

[68] A comparison of Capon range imaging brightness, processed from the same time series data as the 1 μs data shown in the left and right portions of Figures 16a and 16c, with the SNR of the 0.5 μs pulse data, is shown Figures 16b and 16d. The brightness was corrected in the lower 3 gates (Exp-B) and lower 15 gates (Exp-C) for loss owing to the abbreviated Spano-Ghebrebrhan pulse code in those gates. Despite the improved resolution of the 0.5 μs pulse, the lack of fine gating created a much coarser representation of the data than the Capon imaging which scanned every 1 m.

[69] The time continuity of layers from start to end of the data depicted appeared quite poor in Figure 16d for the echoes starting at 13:26 LT at 4.00 km and 4.50 km because of fluctuations in altitude, multiple closely spaced layers, and because the strongest echoes were near the connecting edge of one pair of 75 m range gates; in contrast, in Figure 16b we see that the layers starting at 13:26 LT between 6.50–7.00 km and around 7.50 km showed excellent continuity in their variation with altitude. Furthermore, the double layer visible in the Capon brightness from 14:51 LT around 6.70–6.90 km could not be resolved at all by the 0.5 μs pulse SNR.

[70] Gating at offsets of multiples of equation image of the pulse width can be set for 0.5 μs pulse observations, so that with four sequential pulses a gating of 37.5 m could potentially be achieved. This two-times oversampling would leave the pulse resolution at 75 m but improve the smoothness of the data presentation and compare better with Capon imaging.

[71] In comparison to the 1 μs pulse, the 0.5 μs pulse mode offers improved range resolution for large isolated layers which exhibit vertical motion of the order of two gates or more (150 m), because of the effective diagonal motion of the atmospheric structures seen by the radar, owing to the combination of vertical motion and advection through the radar beam. However, when multiple structures are separated by distances of the order of two range gates or less (150 m), the 0.5 μs pulse cannot be expected to be sufficient to resolve these clearly. Only imaging techniques are able to accomplish this task successfully, since Capon imaging with the MUR has been shown to have expected resolutions of the order of a few tens of meters in high SNR conditions [Luce et al., 2001].

6. Conclusions

[72] In this paper, the new MUR imaging system hardware and software components, and their use, have been described, and several results from experiments conducted in the first two years of operation presented.

[73] The primary benefit which the system offers over the previous MUR system is digitization of signals in the transmit and receive circuits from the IF stage downwards, so that a previously difficult set of analogue instability entry points are removed. The dynamic range of the new system was improved so that attenuation for different altitude ranges is no longer necessary. Thus, existing experiment configurations can be carried out with minor modifications, and results obtained are greatly improved in SNR and stability. The use of Spano-Ghebrebrhan (truncated) coding means that separate modes for troposphere and stratosphere are no longer necessary.

[74] The development objective of the new system, for which stable digitized signals were considered vital, was radar imaging, and with the new system the use of a stable 2 MHz transmit bandwidth, provision for 5 transmit frequencies over a 1 MHz bandwidth, and a total of 25 receive channels (corresponding to the 25 antenna groups which together comprise the main antenna), and digital combiners to give a total of (also) 25 stored receiver data channels per transmit pulse, construct a robust imaging system whose results validate the success of the design.

[75] For the first time operationally, 3-D volume imaging using scanning in three dimensions based on FII and CRI principles was implemented, and could clearly detect detailed layer patterns in the troposphere and stratosphere, and possible Kelvin-Helmholtz instability billow structures, obtaining valuable angular information from the interior of the radar beam. The study of the orientation of radar echo structures and their relation to wind and shear should permit more thorough study of the mechanisms controlling the observed structures. 3-D imaging is useful to resolve the nonuniform echoes in the angular as well as the range dimension, although attention needs to be paid to the relationship of time resolution and background wind. 3-D imaging can show thin layers as they appear at any location in range, isolated from the range shift of the layer across the beam width (within the limitations of the imaging function resolution and time resolution of the observation). This is important for radars or observation modes with wider beams, or alternatively, as the vertical fluctuation of the layer increases relative to its thickness. By the same mechanism, aircraft echoes are much reduced in effect. For the interpretation of echoes in terms of layers or more complex structures, smearing owing to drift of the structures on the background wind, and vertical displacement during the observation time, continue to be a problem.

[76] An experimental 0.5 μs pulse mode was also tested successfully and shown to be useful as an alternative to 1 μs DBS observations. However, imaging based on the 1 μs pulse offers much improved resolution for distinguishing thin or closely spaced layers.

[77] Apart from supporting radar imaging, the new multichannel digital receiver system makes it possible to utilize multiple signal-processing algorithms on the same raw data. Thus, for example, SA-configuration wind measurements using correlation-function or structure-function approaches [Praskovsky et al., 2004] could be implemented and the information combined with radar imaging information about layer morphology. The potential for such combination of signal-processing techniques makes the new MUR imaging system a valuable addition to the atmospheric science community.

Appendix A

[78] The parameters for the 4 experimental observations are given in Tables 47. Adjoining records can be processed as a continuous sequence of time series samples. Exp-A, Exp-B and Exp-C are imaging observations using 1 μs pulses, while Exp-D is a standard mode observation using a 0.5 μs pulse width. The three imaging observation configurations all use a 1 MHz bandwidth on transmission, and 19 pseudo-circular antenna groups on reception, in addition to the entire antenna. They are distinguished primarily in their pulse coding, using 8-bit, 4-bit and 16-bit Spano-Ghebrebrhan (truncated) coding, respectively. Exp-D attempts similar observation quality as Exp-C obtains for standard-mode processing at a single frequency, but using a 0.5 μs pulse. Beam directions are given as vertical (V) for the zenith-directed beam, and as north (N), east (E), south (S) and west (W) for beams at an off-zenith angle.

Acknowledgments

[79] The first author was supported during this research by the Japan Ministry of Education, Culture, Sports, Science and Technology (MEXT) via a grant from the Japan Society for the Promotion of Science (JSPS). The MUR belongs to and is operated by RISH and is maintained by Mitsubishi Electric Corporation (MELCO).

Ancillary