An agile high-frequency radar used for ionospheric research

The Super Dual Radar Network (SuperDARN) is an international high-frequency (HF) radar network used to study plasma convection in the upper atmosphere. An agile SuperDARN radar (AgileDARN radar) was deployed to join the network in Jiamusi, China. The AgileDARN radar is a digital phased array radar equipped with an field-programmable gate array (FPGA)-based digital processing unit (DPU). It can operate as a standard SuperDARN radar or a multiple-input-multiple-out (MIMO) radar. Each channel can be controlled and processed separately. Digital beam forming (DBF) is used to make beam steering flexible and beam switching fast. Multiple sub-beam forming for receiving can be implemented on FPGA or computer to improve position accuracy. Additionally, internal calibration and external calibration are carried out to improve the performance of beam forming. The internal calibration is a selfcalibration without any external connections. Calibration signals are produced by the generators in DPU. The gain and phase of transmitters/receivers are read and compared to the based transmitter/receiver. Then the differences are recorded to compensate for the imbalances. During the external calibration, meteor trails are used as calibration sources to detect the imbalance between antennas (including cables). The imbalances are compensated in the same way as the internal calibration. By calibration, the amplitude and phase imbalances between channels are reduced significantly.


Introduction
The Super Dual Auroral Radar Network (SuperDARN) is an international collaborative organization. The SuperDARN radars are high-frequency (HF) radars that operate on frequencies between 8 and 20 MHz. These radars are used to remotely sense the backscatter from decameter-scale magnetic fieldaligned plasma density irregularities in both the E and F regions of the ionosphere (Greenwald et al., 1995). The movements of these irregularities respond to energy coupled into the auroral and sub-auroral regions from the solar wind and magnetosphere. Therefore, SuperDARN data provide scientists with information regarding the Earth's interaction with the space environment. It has been successful in addressing many space weather questions covering the area of ionospheric convection, field-aligned currents, magnetic reconnection, substorms, MHD waves, the neutral atmosphere, and so on (Chisham et al., 2007;Nishitani et al., 2019).
The history of SuperDARN has been introduced detailedly in some pieces of literature (Greenwald et al., 1995;Chisham et al., 2007;Nishitani et al., 2019). The extensive use of coherent scatter radars to study ionospheric convection began with the Scandinavian Twin Auroral Radar Experiment (STARE) in the late 1970s (Greenwald et al., 1978). STARE consists of two pulsed bistatic phased-array radars located in Nalvik, Norway and Hankasalmi, Finland. It was the only instrument capable of providing two-dimensional images of the structure of ionospheric convection patterns at that time. Subsequently, several paired VHF radar systems including the Sweden And Britain radar Experiment (SABRE) and the Bistatic Auroral Radar System (BARS) were developed. Both of these systems were identified as elements of the DARN network. However, the prototype for the current SuperDARN radar is Goose Bay HF radar developed by the Johns Hopkins University Applied Physics Laboratory, which commenced operations at Goose Bay in October 1983 (Greenwald et al., 1985). The initial purpose of the SuperDARN radar network was to observe the plasma convection at both Polar Regions; therefore, all the early SuperDARN radars covered the auroral and polar regions (Greenwald et al., 1995). However, during periods of magnetic storms and substorm expansions, radars at high-latitude were unable to completely map the convection process because the Topical Issue -Space Weather Instrumentation equatorward expansion of the auroral oval used to reach to midlatitude region. The expansion of the SuperDARN network to the mid-latitudes allows the SuperDARN network to trace the plasma convection processes during magnetic storm and substorm activity (Baker et al., 2007). The first mid-latitude radar, the Wallops Island radar, was constructed in 2005, shortly followed by the second mid-latitude radar in Hokkaido, Hokkaido East radar (HOK) in 2006. Until 2018, more than ten SuperDARN radars have been built in mid-latitude regions in the northern hemisphere (Nishitani et al., 2019).
There is another mid-latitude SuperDARN radar established in Jiamusi (China) in 2017. It is developed as an agile SuperDARN radar (AgileDARN) by the National Space Science Center, Chinese Academic of Sciences. The radar extends the coverage of the mid-latitude SuperDARN chain into China. Its data have been contributed to the SuperDARN data stream since 2019. The measurements during storm time show the contribution to the ionospheric convection patterns (Zhang et al., 2020). Supported by the Chinese Meridian Project-II, six additional AgileDARN radars in three sites (Bayinbluk, Siziwangqi Banner, and Longjing) are under construction to cover the mid-latitude region over China and expand the coverage of the mid-latitude HF radar considerably. The significantly expanded coverage of mid-latitude SuperDARN plays an important role in SAPS (sub-auroral polarization stream) research, including dependence of SPS on solar wind and geomagnetic drivers, interhemispheric comparisons of SAPS, evolution, and variability of SAPS, etc. (Nishitani et al., 2019). Meanwhile, the data from the mid-latitude radars have a dramatic effect on the overall convection strength, increasing the cross-polar cap potential and changing the potential contours on the nightside (Baker et al., 2007;Chisham et al., 2007).
The paper is organized as follows. In Section 2, we describe the system configuration and the specific design features in hardware. In Section 3, we focus on the advanced signal processing methods based on the distributed structure. We present the measured data to validate the proposed methods. Finally, the summary and conclusion are given in Section 4.

System description
The AgileDARN radar is built as one part of the Super-DARN network. Its operational concept is essentially identical to other SuperDARN radars. Meanwhile, the radar has some distinctive designs.
The AgileDARN radar is composed of two arrays, 20 individual transceivers, a digital processing unit (DPU), and a computer (shown as Fig. 1). The primary array consists of sixteen twin-terminated folded dipole (TTFD) elements with 12.8 m spacing, designed as the Blackstone radar (Sterne et al., 2011). The interferometric array with four antennas is used to estimate the elevation angle of the incoming signal. The phasing matrix used in the early SuperDARN radar is abandoned. Beam steering and beam switching are implemented by digital beam forming (DBF) (Whittington et al., 2002;Custovic et al., 2012;Deng et al., 2019). Each antenna has its own transceiver, which has a separate transmitter and receiver. The antennas are transmit-receive shared. They switch between the transmitters and receivers by transmitter/receiver (Tx/Rx) switches. Besides, directional couplers and combination switches are added in transceivers to implement internal calibration, see the "calibration circuit" subsection. All the control signals of the switches are provided by the DPU.
The DPU is composed of a direct digital synthesis (DDS) module, a digital signal processing module, a timing controller, 20 analog-to-digital converters (ADCs), and 20 digital-to-analog converters (DACs). It performs signal generation and processing, commanding and control of the system, instrument health monitoring, scientific data acquisition, and formatting. These functions are implemented based on field-programmable gate arrays (FPGAs), which makes the radar flexible. Most of the radar parameters are adjustable to meet different types of observation. Each channel can be controlled and processed separately. Consequently, the radar can operate as a multipleinput-multiple-output (MIMO) radar, besides a standard SuperDARN radar. The twenty separate receivers are allowed each to be recorded and stored independently. Beamforming and calibrations can be performed online or offline as needed. The radar can respond quickly to a request for special observation. Therefore, the radar is called AgileDARN.
The computer is used as a platform on which radar control software can be run. The control software is to take control of the radar to operate as required. Operation parameters and sequences are configured in the software but carried out on the DPU. The two parts communicate with each other via a peripheral component interconnect express (PCIe) bus.

Field of view
The operating frequency of AgileDARN is between 8 and 20 MHz. The normalized spacing (in units of wavelength) varies with frequency. Therefore, the field of view (FOV) of the array becomes narrower with increasing frequency. In practice, HF propagation conditions mean that most ionospheric scatter is detected using frequencies in the range from 10 to 14 MHz . The FOV is not less than 78°at the frequency under 16 MHz. Above 16 MHz, grating sidelobes begin to appear for the outmost beams (Greenwald et al., 1985). The geographic coordinates of the radar site are 46.8°N, 130.5°E. It looks to the northeast with a boresight azimuth of 44°in geographic coordinates. The fields of view (FOV) of the AgileDARN radar (JME) along with the other Northern Hemisphere SuperDARN radars in operation are shown in Figure 2 (Zhang et al., 2020). According to the figure, we can see that the AgileDARN radar has a common viewing area with the Hokkaido pair of radars (HOK and HKW), which would provide a two-dimensional observation of ionospheric convection in this area. Moreover, the AgileDARN radar can be validated by comparing its observations with those from the Hokkaido radars (Zhang et al., 2020).

Calibration circuit
As mentioned, directional couplers and combination switches are used to perform internal calibration. Between antenna and Tx/Rx switch, a low-loss directional coupler is inserted to sample a small amount of power for measurement. The coupler (shown in Fig. 3) has four ports, an input port (P1), a "through" port (P2), a coupled port (P3), and a reflected port (P4). The coupler is a reciprocal device in that the coupled ports can be used to sample a portion of the mainline power, or they can be used to inject a signal into the main signal path. Besides, a directional coupler provides a sample port in both the forward and reverse directions. Based on the characteristics, directional couplers are used to implement internal calibration and voltage standing wave ratio (VSWR) detection in AgileDARN. A combination switch composed of three switches (S1, S2, and S3 in Figure 3) is added to carry out operation state switching control. The operation states of AgileDARN include transmitting, receiving, transmitter calibrating, and receiver calibrating. Transmit signals and reference signals for calibrations are generated in DPU. During ionosphere observation, transmit signals are amplified by transmitters then sent to antennas. Returns received are amplified and filtered in receivers to extract the desired signals. Transmitters and receivers are calibrated separately. The calibration scheme is shown in Figure 3. The signal paths of the transmitter and receiver calibration are presented as orange and blue, respectively. During transmitter calibration, the reference signal is sent to the transmitter and then enter the coupler. Most of the incident power is emitted by the antenna, and a fixed fraction of it is coupled on to P3 for transmitter calibration. When calibrating receivers, the P3 port is used as an input. The reference signal is injected into the coupler from P3. A sample of the incident signal is coupled on to P1 and fed to the receiver by Tx/Rx switch. The signal passing through the receiver and S1 is converted to digital and processed in DPU.

VSWR detection and protection circuit
When AgileDARN operates in transmitting or transmitter calibration mode, the output power of a transmitter is up to 1000 watts. It is necessary to design a VSWR detection and protection circuit to protect amplifiers under severe mismatch. Directional couplers and dual log detectors are used in AgileDARN to detect VSWR (shown in Fig. 4; Rachakonda & Hawkins, 2000). Once a signal is fed to P1, a sample of the incident and reflected signal are coupled to P3 and P4. The power from P4 is proportional to VSWR. The output of P4 is independent of the VSWR. They are fed to the dual log detector. The detector outputs the logarithmic subtraction of the two signals, which is proportional to the reflection coefficient. The voltage comparator triggers when the difference output V diff is more than the reference voltage V Ref , indicating a high VSWR condition. Once a high VSWR is detected, the input of the transmitter is cut off to protect the amplifiers.

R E T R A C T E D 3 Advanced signal processing techniques
The AgileDARN radar is designed as a SuperDARN radar. It can operate as the standard SuperDARN radar, outputting the returns after beam forming. Besides, the radar is a distributed system, and the DPU is an FPGA-based reconfigurable module. It can operate as a multiple-input-multiple-output (MIMO) radar. The twenty separate receivers are allowed each to be recorded and stored independently, which enables postprocessing beam forming, channel imbalance detection, and correction on computer.

Calibration of channel imbalance
Digital beam forming will work correctly, provided that a plane wave incident on the array remains plane following digitization. Implementation departure from an ideal design means that the plane wave requirement will not be achieved exactly and will degrade beam forming performance. The imbalance between channels is an important contributor to the imperfection (Frazer & Abramovich, 2002). Besides, at high frequency, diffraction during ionospheric propagation will cause wavefront distortion (Theurer & Bristow, 2015).
In the Tiger-3 radar, each transceiver utilizes two additional 14-bit ADCs alongside the main receiver 16-bit ADC to measure the phase variabilities between transceivers (Custovic et al., 2012). Experimental results indicate the worst phase calibration accuracy is of the order of 0.153°at 14 MHz (Nguyen et al., 2013). Theurer & Bristow (2015) presented a scintillation correction algorithm based on adaptive optics techniques to reduce the diffraction during ionospheric propagation. Many methods were presented to estimate the effective time delay caused by the difference in the electrical path length that connects the primary array and the interferometric array to the correlator (Chisham & Freeman, 2013;Ponomarenko et al., 2015;Burrell et al., 2016;Chisham, 2018).
In this paper, internal calibration and external calibration are presented to reduce the channel imbalance from antennas (including cables), transmitters, and receivers. The internal calibration is used to remove the amplitude and phase imbalance between transmitters/receivers. The phase imbalance between antennas (including cables) can be reduced by the external calibration.

Internal calibration
The internal calibration is implemented based on the calibration circuit presented in Figure 3. The transmitter calibration and receiver calibration are performed, respectively, but in the same way.
During transmitter/receiver calibration, the reference signal is generated in the DPU and delivered to transmitters/receivers. After traveling through the modules, the signals are sampled, processed, and packed in the DPU, respectively. Then, the radar control software picks up the data to detect the imbalance between channels as follows.
Supposed the reference signal is A 0 e Àj2pf 0 tþju 0 , and the transfer function of the nth channel, H n , is Then the signal after the nth channel, v n , is v n ¼ A 0 g n e Àj2pf 0 tþjðu 0 þu n Þ ; where A 0 and u 0 denote the amplitude and initial phase of the reference signal, respectively. f 0 indicates the operating frequency of the radar. g n and un depict the gain and phase delay

R E T R A C T E D
of the nth transmitter/receiver. With the same input signal, the difference between the output signals represents the imbalance. Given the 1st channel as the reference, the difference between the nth Tx/Rx and the 1st Tx/Rx can be obtained by comparing v n with v 1 .
where Dg n and D/ n are the correction factors for the nth Tx/Rx that compensate for amplitude and phase imbalance, respectively. Since the imbalances between Tx/Rxs have been detected, the next step is to remove these differences by compensation.
The correction for transmitters is implemented on FPGA and executed in real-time. The transmitter correction factors are distributed to the generators, compensating transmit signals for transmitter imbalances. Assuming that two sinusoid signals with the same frequency and 90-degree out of phase are generated: where f 0 is the operating frequency, and A 0 is the amplitude of the transmit signal. s I,n (t) and s Q,n (t) are the real and imaginary part of the complex transmit signal, s n (t), before DBF.
The transmit signal after correction is expressed as The receiver imbalances are corrected in the same way as the transmitter imbalance, whereas the signals compensated are echoes. The correction process can be executed in real-time (based on FPGA) or post-processed on computer. Figure 5 shows the imbalance between the transmitters before (top) and after (bottom) calibration. The data are measured continuously in 24 h. The amplitude and phase differences between 16 transmitters before calibration are~1 dB and~35°, respectively. After calibration, the imbalances are reduced considerably to~0.03 dB and~0.1°. That means the transmit pattern after calibration will be almost the desired pattern (Nguyen et al., 2013). Additionally, there are tiny fluctuations caused by external interference. Figure 6 shows the receiver imbalance before (top) and after (bottom) calibration during the same observation interval depicted in Figure 5. The amplitude and phase differences before calibration are~1.5 dB and~100°, which results primarily from the difference between electronic components. After calibration, the imbalances are reduced significantly to 0.2 dB and 1°. The receiver noise and external interference bring in fluctuation that cannot be removed by calibration. The noise/interference has a greater impact on receiver calibration than transmitter calibration. The reasons are shown as follows.
Given the external interference is (t), the coupling factor of the directional coupler k C , the calibration signal is v 0 (t), the gains of a transmitter and a receiver are G Tx and G Rx , respectively. During transmitter calibration, the signal sent to ADC, v Tx,cal and the SNR of v Tx,cal (t) is During receiver calibration, the signal sent to ADC, v Rx, and the SNR of v Rx,cal (t) is In general, G Tx ) 1 and k C ( 1. From equations (8) to (10), we can find that the impact of the interference, (t), on SNR Tx,cal is far less than on SNR Rx,cal .

External calibration
The channel imbalance is induced mainly by the antennas (including cables) and transceivers. The imbalance from transceivers can be removed by internal calibration. That from antennas should be detected and corrected by external calibration.
The primary array of the AgileDARN radar is composed of 16 TTFD antennas mounted on poles 9.16 m off the ground. The size of the array is about 200 m. For a physically large array, phase correction is often accomplished through a landbased or ship-based transponder. In this case, the transponder used as the calibration device must be in the far-field of the array and located in the one-hop ionospheric propagation footprint of the transmitter (Frazer et al., 2009). The one-hop distance depends on the incident angle and the height of the F2 layer (Budden, 1985). Since the SuperDARN radars have broad (~35°) beamwidth in elevation (Greenwald et al., 1985), the distance distributes within the range from several hundred kilometers to about 2000 km Greenwald et al., 2017), which varies with the time of day and season. For the variation of the one-hop distance, it is difficult and costly to perform external calibration by a transponder frequently. Fortunately, there are dispersive targets in nature, such as meteors, that can be used as calibration sources. Solomon et al. (1998) proposed the Weiss-Friedlander MUSIC-type approach for estimating sensor positions and mutual coupling by echoes from meteor trails.
The echoes from meteor trails can be routinely observed by the SuperDARN radars on most days (Hall et al., 1997). The meteor trails are dispersive, and most of them emerge in the slant range less than 500 km. The signal propagation between the radar and the target can be assumed straight-line. Thus, the echoes can be used to calibrate the phase offset between two arrays of a SuperDARN radar (Chisham, 2018).

R E T R A C T E D
The backscatter from meteor trails is, in general, featured with planar wavefronts, high signal-to-noise ratio (SNR), adequate duration for sufficient snapshots, and the ability to be received in large numbers over the entire HF band (Solomon et al., 1998). Therefore, the meteor trail can be considered as a point scatter and used to calibrate the channel imbalance of the AgileDARN radar frequently. The arrays in AgileDARN are uniform linear arrays. If a plane wave incidents upon an array, the phase at each antenna element depends on the angle of arrival of the wave. Given an array of N elements, the phase obtained by the nth antenna is a A,n (n = 1,2, . . ., N). The relationship between a A,n and a A,1 is (Zhang, 2009) where k = 2p/k, k is the operating wavelength. d denote wave number and the spacing between antenna elements. h B indicates the angle of arrival of the wave. The phases of the

R E T R A C T E D
antennas increase or decrease linearly, which is determined by h B . When the channel imbalance exists, the linearity will be destroyed. The deviations from linear indicate the imbalance. During external calibration, internal calibration is performed in real-time to remove the imbalance between transceivers. The process steps for external calibration are shown as follows: 1. Pick up meteor echoes based on some currently recognized characteristics of SuperDARN meteor echoes, such as slant range, SNR, line-of-sight (LOS) Doppler velocity, spectral width, and altitude (elevation angle) (Solomon et al., 1998). 2. Further, calculate the phases of echoes received by the antenna elements. The least-squares linear fitting method is used to fit a linear model to the unwrapped phase. 3. The slope of the fit line represents in equation (11). The deviations from the line denote the antenna (including cable) imbalance, Da A,n . 4. The correction factors for transmitting channel (including antenna, cable, and transmitter) and receiving channel (including antenna, cable, and receiver) are expressed as where the subscript n denotes the channel index. D/ T and D/ R depict the correction factor for transmitting channel and receiving channel, respectively. 5. The last step is to correct the imbalance. The antenna imbalances are corrected in the same way as the Tx/Rx imbalances. The transmitting correction factors are used to compensate the transmit signals, and the receiving correction factors are used to compensate the echoes.
As the steps introduced above, more than 30 meteor trails are picked up to get the correction factor of antennas. Figure 7 shows the phase correction factors o the antennas in the main array. The phase imbalance is between À10°and 11°.
Further, pick up another set of meteor trails and correct the antenna imbalance by the correction factors. Figure 8 shows the phases of the echoes received by 16 receivers before correction. Figure 8a shows the unwrapped phases of the 16 channels and the fitted straight lines. The variations of the phase have an approximately linear character with a deviation between À15°a nd 20°(shown in Fig 8b). The means of the deviation from the fitting lines distribute between À10°and 10°, with a standard deviation (STD) of~3°.
After phase correction, the unwrapped phases are re-fitted (shown in Fig. 9a). Most of the deviations distribute between À10°and 10°(shown in Fig. 9b). The means (red circles) are reduced significantly to ±1°, with the same STD. That is because the correction factors are fixed in a calibration interval. The deviation caused by the variation of antennas can be resolved by shortening the calibration interval. In practice, most of the residual errors are caused by the clutter from sidelobes or radio frequency interference (RFI) that cannot be removed by calibration.

Flexible receive beam pattern synthesis
As a new generation SuperDARN radar, DBF is adopted to replace a phasing matrix to accomplish beam steering and switching (Salim et al., 2004;Healey et al., 2006;Bristow, 2019;Kennedy, 2019). There are two ways in AgileDAN to accomplish DBF. Transmitting DBF in the generator is implemented based on FPGA. Digital multiple sub-beam forming for receiving can select to be implemented on FPGA or computer (Deng et al., 2019). When AgileDARN operates as a standard SuperDARN radar, it scans 24 directions with a beamwidth of 3.25°. Echo signals are processed by seven sub-beam forming on FPGA (shown as Fig. 10). It needs to mention that when executing seven sub-beam forming, the same set of data are used to synthesized to seven different directions. The 4th sub-beam is steered towards the boresight of transmitting beam. The sub-beam number is fixed and limited by the resource of FPGA. When the radar operates as a MIMO system, the twenty separate receivers are recorded and stored independently. Multiple sub-beam forming is off-line processing on computer and the sub-beam number can set as needed.
The typical SuperDARN radars scan16-24 directions over a field of view of $52°-78°in azimuth, with a beamwidth of 3.24° (Greenwald et al., 1985). The AgileDARN radar scans 24 directions with a beamwidth of 3.25°. By M sub-beam forming, the angular separation of receiving is decreased to 3.25°/M. For the typical beam forming, the direction of arrival (DOA) is considered as the boresight. The maximum position error is up to 1.625°. By M-beam forming, a beam is split into M sub-beams. The power distribution in the beam is obtained to estimate the DOA. In general, the direction with the strongest echo is considered as the DOA with a maximum error of 1.625°/M. The position accuracy is improved by M times, which means the position accuracy is proportional to M. However, the accuracy is limited by the sensitivity of the radar system. When the sub-beam is too narrow, the difference between sub-beams cannot be distinguished. The position accuracy cannot be improved any more. Figure 11 shows the power distribution in FOV with a separation of 0.464°(seven sub-beam forming) and 3.25°(typical beam forming). During transmitting, array scans 24 directions with a beamwidth of 3.25°. The dwell time on each beam is 2.3 s (integration period). For each integration period, 26 seven-pulse sequences are transmitted, then 26 sets of lag

R E T R A C T E D
zero power for all ranges (from 180 km to 2430 km with an interval of 45 km) are averaged to improve the SNR (Ribeiro et al., 2013). Figures 11a and 11b exhibit the power distribution in FOV with a separation of 0.464°and 3.25°, respectively. There are strong echoes appear in the range of~855 km in the two pictures, whereas the difference cannot be distinguished easily by naked eye. Figures 11c and 11d show the power distribution of the echoes from 855 km in azimuth. The peaks occur at À10.91°in (c) and À11.38°in (d), with a difference of 0.47°. That means that the position accuracy has an improvement of 0.47°by seven sub-beam forming. However, the echoes of the seven sub-beams are synthesized by the same data set rather than by seven independent data sets. Consequently, the power distribution in (c) varies discontinuously with azimuth. Additionally, frequencies from 8 to 20 MHz are a portion of the radio band that the radio waves are bent by the ionosphere so that the waves return to the Earth's surface at long distances beyond the horizon. Thus HF over-the-horizon (OTH) radars are designed to detect not only ionospheric irregularities but also aircraft, ship, and weather effects over the ocean (Skolnik, 2003). The conventional beamformers are susceptible to interference signals. The signals from other OTH radars may contaminate the returns of SuperDARN radars when they occur at sidelobes of the array. By contrast, adaptive, or statistically optimum, beamformers can account for interference signals. An adaptive beamformer algorithm chooses the weights based on the statistics of the received data. For example, an adaptive beamformer can improve the SNR by using the received data to place nulls in the array response. These nulls are placed at angles corresponding to the interference signals. As multiple sub-beam forming, adaptive beam forming can be implemented on FPGA or computer, it is more flexible on computer. The adaptive methods can be selected for different requirements.

Summary and conclusion
The AgileDARN radar is the first mid-latitude SuperDARN radar in China. The geographic coordinates of the radar site are 46.8°N, 130.5°E. It looks to the northeast with a boresight azimuth of 44°in geographic coordinates and has a common viewing area with the Hokkaido pair of radars (HOK and HKW). The AgileDARN radar can be validated by comparing its observations with those from the Hokkaido radars.
The radar has an FPGA-based DPU to implement digital signal processing, such as DBF. Besides, it can provide distributed signal processing as a MIMO radar. The 20 separate receivers are allowed each to be recorded and stored independently. The receive beam forming, the detection of the channel imbalance, and the correction of the receiving channel imbalance are post-processed on computer with flexibility. Multiple sub-beam forming is implemented to improve the position accuracy.
The imbalance between channels can degrade beam forming performance. Internal calibration and external calibration are proposed to reduce the imbalance from transmitters, receivers, and antennas. The calibration circuits are designed to accomplish the transmitter and receiver calibration individually. The reference signals for calibration are generated by the generators

R E T R A C T E D
in the DPU. The internal calibration is executed with an interval of 1 min. By the internal calibration, the imbalances between transmitters/receivers are eliminated significantly. During the external calibration, the echoes from meteor trails are used as calibration signals to detect the imbalance between antennas. The calibration interval is determined by the stability of the antenna imbalance. In a calibration interval, the imbalances between antennas are expected to remain stable. Fortunately, meteors can be routinely observed by the HF radar on most days, but the detecting ability of meteors depends dominantly on transmit power, operating frequency, and range resolution.
With the same transmit power and operating frequency, a higher range resolution can detect more meteors. The calibration can reduce effectively the imbalance between channels but has no effect on the fluctuation caused by clutter or radio frequency interference (RFI).