Architecture and signal processing of sky wave over‐the‐horizon radar
1. Introduction
[2] The sky wave over-the-horizon radar (OTHR) is unique in radar family that employs the ionosphere to reflect the radar high-frequency (HF, 3–30 MHz) signal to illuminate the target from the top down (see Figure 1), thus significantly extending the detection range to 1000–4000 km without the limitation of the globe curvature. OTHR is therefore a good tool for strategic early warning and low-flying target detection. Furthermore, OTHR has a wavelength from 10 m to 60 m that approximates to the physical size of most targets. Owing to the effects of resonant scattering, the OTHR has potential advantage in stealth target detection. Besides, the transmitting signal of the OTHR can also resonate the ocean wave of certain wavelength. By analyzing the resonant echoes, the sea-state information including ocean current, wave-height, and surface wind can be obtained [Headrick and Skolnik, 1974; Headrick and Thomason, 1998; Anderson, 1992].
The skywave over-the-horizon radar operates by refracting its beam from the ionosphere, thus significantly extending the detection range to thousands of kilometers.
[3] As early as the end of World War II, the research of OTHR had begun in the United States. However, the techniques available at that time could not suppress the powerful undesired echoes from the ground/ocean. In the mid 1950s, the U.S. Naval Research Laboratory (NRL) firstly employed the long coherent integration technique to achieve the target detection from strong clutter and noise background. After that, the NRL developed the first experimental OTHR in the world - MADRE (magnetic drum recording), with which many successful detection and tracking for aircraft and large fast ships were accomplished. However, its capacity for wide-area surveillance did not exist and the azimuth resolution, frequency range and frequency agility capabilities were also very modest [Headrick and Thomason, 1998]. In 1970, Stanford University developed another experimental OTHR called WARF (wide aperture research facility), whose azimuth resolution was significantly enhanced and the ground/ocean clutter could be suppressed effectively [Barnum, 1986]. By the 1980s, thanks to the development of digital technology and computer processing power, the OTHR had become an operational reality. Three OTHRs (one AN/FPS-118 and two AN/TPS-71) were constructed by the U.S. Air Force and the U.S. Navy. Besides the United States, the research and development of OTHR were also pursued in Australia, France, Russia, the United Kingdom, Canada, and China.
[4] Although OTHR has many advantages, its operating modes (HF band, the ionospheric propagation medium and looking-down mode) bring some great challenges. 1) Due to the solar radiation, the electronic density distribution of ionosphere changes significantly with the hour, day and season. Thus, the real-time ionospheric condition evaluation is required to select the optimal operating frequency. 2) In HF band, the cosmic noise, atmospheric noise (lighting), man-made noise and meteor echoes will seriously affect the sensitivity of radar receiver. In addition, the HF band is well occupied, the OTHR have to select a relatively clear frequency band in real time to reduce the interference from other HF users. 3) In OTHR, the echo signals contain strong undesired returns from the ocean/ground that severely limit the target detection. 4) The ionospheric nonstationarity will impose a phase disturbance on the echo signal and induce Doppler spreading. 5) At some operating frequencies, the transmitting signal may propagate through different ionospheric reflection layers, which produces two problems, as illustrated in Figure 2.
The effects of ionospheric multipath propagation: (a) A single target creates several slant tracks, and inaccurate propagation mode identification may further produce multiple ground tracks for the single target; (b) different targets have the same slant range, which cause the superposition of Doppler spectrum at this slant range.
[5] Due to the above problems, the architecture and signal processing of OTHR are quite different from the conventional line-of-sight radar system. In the following sections, we will describe the architecture and signal processing scheme adopted in the OTHR. In particular, the techniques on interference suppression, correction of ionospheric contamination, and target detection are analyzed in detail.
2. Architecture of OTHR
[6] OTHR can be separated into three facilities: a transmitting complex, a receiving complex and an operation control center. Frequency management system is also included in OTHR and it is integrated in aforementioned three facilities. The simplified block diagram of the OTHR architecture is shown in Figure 3, in which some information is supplemented relative to the figure of reference [Ferraro and Ganter, 1998].
The simplified block diagram of OTHR architecture.
2.1. Frequency Management System
[7] The frequency management system is to provide real-time ionospheric information to determine radar operating parameters (such as operating frequency and the emission band) and predict potential propagation mode. The FMS generally includes 1) backscatter sounder, 2) oblique/vertical incidence sounder, and 3) HF spectrum monitor. The backscatter sounder is to measure the backscatter energy returned from earth surface as the function of radar operating frequency and the group range. The oblique/vertical sounder is to provide the ionospheric structure information including its height and density, and distinguish the single mode or mixed modes of ionospheric reflection path [Earl and Ward, 1987]. When the above two sounders decide the optimal band of the radar operating frequency, the HF spectrum monitor is employed to find an unoccupied channel in that band for operation, so as to reduce the interference from other HF users.
2.2. Transmitting Complex
[8] In order to achieve long-range target detection, the average transmitting power of OTHR should be very high about 200–1200 kW. Therefore, the continuous wave signal with frequency/phase modulation is often adopted.
[9] Since the operating frequency of the OTHR is often adjusted according to the current ionospheric condition, the OTHR generally has several transmitting antenna array. Each corresponds to a certain operating band. Besides, owing to the long operating wavelength and high transmitting power, the OTHR transmitting antenna array is usually of a large size, and to keep high transmitting gain at low launch angles, a ground screen extending several hundreds of meters in front of the antenna is required.
2.3. Receiving Complex
[10] As the OTHR exploits the continuous wave signal, the transmitting and receiving sites have to be separated with the distance of the order of 80 to 160 km to avoid the leakage of transmitting power. Since the wavelength of OTHR is very long, the receiving antenna array is usually over 2 km in length with several hundred elements assembled in line. This extremely long antenna array can provide necessary azimuth resolution to suppress the ocean/ground clutter and the external noise. In addition, each antenna element is configured to a digital receiver, and the received data will be sent to the high-speed signal processor to perform target detection and ionospheric condition evaluation. Since OTHR is a bistatic system, it is also necessary to realize the precise synchronization in time, space, and phase between the OTHR transmitting and receiving site.
2.4. Operation Control Center (OCC)
[11] After the radar echoes are processed in receiving complex, they will be sent to the OCC. Here the target information will be combined to form the track, firstly in radar coordinates (group range and radar azimuth) and then converted to ground coordinates (longitude and latitude). This conversion is called coordinate registration (CR). It requires the knowledge of the ionospheric virtual height at the point of reflection. The ionospheric support data is derived from the ionospheric sounder and updated periodically. Any inaccuracy in determining the ionospheric height and propagation mode will cause an error in CR process. A relatively easy solution is to employ the external reference sources (such as known terrain features, HF beacons, and aircrafts with precise navigation equipments, etc.) to get the coordinate correction factor that is subsequently applied to the target tracks. Another limitation for the CR process is the multipath propagation. As Figure 2a illustrates, the multiple ionospheric reflection layers produce multiple slant tracks for a single target. And inaccurate propagation mode identification will result in multiple ground tracks for the single target. Thus, the multihypothesis track data fusion is needed [Pulford and Evans, 1998; White and Percival, 1998]. Besides providing the ground tracks, the OCC is also responsible for the radar management such as setting the radar parameters based on the type of task and the current ionospheric condition.
3. Suppression of Interference and Noise
[12] Unlike the conventional microwave radars, the OTHR operates in HF band. In this band, the external interference and noise, such as the cosmic noise, atmospheric noise (lighting), meteor echoes, man-made noise and other HF radiating sources are generally 20–40dB greater than the receiver internal noise, and brings great challenges to the OTHR signal processing.
[13] The adaptive beam forming is an effective approach to suppress the powerful external noise and interference. However, in HF band, some transient interference such as lighting and meteor will produce strong radar echoes that can still enter the receiver via the antenna sidelobes. As a result, high-amplitude broadband interference energy will be deposited in Doppler spectrum, which limits the target detection [Barnum and Simpson, 1997]. Thus besides the adaptive beam forming, some further processing is still needed to remove these transient interference.
3.1. Transient Interference Suppression Based on the Wavelet Transform
[14] Compared with the echoes from ocean/ground, the transient interference has very short duration but with strong energy. It usually shows abrupt change in time domain. So we can employ this property and use the wavelet transform to locate the transient interference [Quan and Li, 1999; Quan et al., 1999]. Having obtained the azimuth-range-Doppler map, the range/azimuth cell where the noise floor is much higher than that of other cells is considered to probably include transient interference. Then the wavelet transform is applied to the echoes of that cell to calculate the wavelet coefficients at different scales. The time period when the wavelet coefficient is above the predetermined threshold will be regarded as the time location of transient interference and the echoes in that time period will be removed from original data and then reconstructed. Presently auto-regressive (AR) model has been applied to the data reconstruction.
[15] For this method, it is found that at different wavelet scales, the time-domain location results of the transient is not completely same. So it is necessary to establish thresholds on multiple wavelet scales to ensure the transient interference can be located correctly.
3.2. Transient Interference Suppression Based on Doppler Filter and Inverse Fast Fourier Transform (IFFT)
[16] The energy of ocean/ground echoes concentrates in the Doppler domain while that of the transient interference concentrates in the time domain, thus we can use this property to locate the transient. The detailed processing steps are as follows: first, filter out the ocean/ground clutter in the Doppler spectrum and then perform the IFFT. Because the strong clutter has been removed, the transient interference will be prominent in the time domain. Then the echoes in that time period will be removed from original data and reconstructed. This method is easy to implement. But note that the filtering operation may also remove the partial spectrum of the transient, and affect its future locating in time-domain.
[17] Figure 4a shows the Doppler spectrum of the real OTHR data operating at 7.5 MHz with coherent integration time of 44 s (64-point data). These data are not affected by strong transient interference and include a ship target whose Doppler frequency is −0.45 Hz. Figure 4b shows the Doppler spectrum after artificially introducing the transient interference to this data. The ship peak is buried. Figures 4c and 4d show the transient-removed spectrum by the method mentioned in this subsection, with data interpolation and not, respectively. It is obvious that the noise floor is decreased significantly after removing the transient interference, and the Doppler spectrum with data interpolation is better than that without data interpolation. In fact, when the transient interference has relatively long duration, the target detection performance will highly depend on the quality of data interpolation.
Transient interference suppression: (a) Original Doppler spectrum, (b) transient-introduced spectrum, (c) transient-removed spectrum, with data interpolation, and (d) without data interpolation.
3.3. Transient Interference Suppression Based on Eigen-Decomposition
[18] This method suppresses the strong ocean/ground clutter without employing the Doppler filter, but taking advantage of the approximately uniform statistical characteristics (such as the temporal covariance matrix) of the ocean/ground clutter time series between the neighboring resolution cells in short integration time [Xing et al., 2002]. The detailed processing steps are as follows: first, construct the estimation of temporal covariance matrix R, which is averaged over a set of neighboring resolution cells
where Xl,k is the time sample from a single resolution cell, and L, K is the number of neighboring range and azimuth cells respectively. Though the transient interference may be strong, it only exists in certain range (such as meteor echoes) or azimuth cells (such as lightning strike). Thus in R which is obtained by averaging over a set of adjacent resolution cells, the transient interference is still weaker than the ocean/ground clutter. For this reason, perform the eigen-decomposition of R, and deem the eigenvectors corresponding to the largest r eigenvalues construct the clutter subspace. Then the echo signal X from the resolution cell where the transient exists is projected onto the clutter subspace and get signal X r, thus the output is set as X − Xr. In this output, the ocean/ground clutter has been suppressed. As a result, the transient interference will be prominent in time domain and its location can be easily determined. Finally, the echoes in that time period are removed from original data followed by data reconstruction.
[19] In addition, some research results also indicate that adopting higher operating frequency (>15MHz) can reduce the effects of transient interference.
4. Decontamination of Ionosphere Distortion
[20] After the adaptive beam forming and the transient interference suppression, the noise and interference level will decrease significantly. However, some other contamination mechanisms such as ionospheric nonlinear phase distortion and multipath propagation will cause the spreading of the ocean clutter spectrum, and still bring great challenges for ship detection and sea-state remote.
4.1. Ionospheric Phase Distortion and Multipath Propagation
[21] When the OTHR signals propagate through the ionosphere, they may experience a quasi-random phase contamination due to the ionospheric nonstationarity. Suppose X is the uncontaminated echoes from a resolution cell, the contaminated one can be represented as Y = ejϕ(t)X, where ϕ(t) is the phase contamination function. For a short coherent integration time (CIT), ϕ(t) is approximately linear and only results in a Doppler shift. But when the CIT is longer than 20–30 s, ϕ(t) is often nonlinear and causes the Doppler spreading.
[22] Multipath propagation is another contamination mechanism. As Figure 2b illustrates, at some operating frequencies, the independent echoes from different range cells may propagate through different ionospheric layers and reach the receiver at the same time. As a result, the Doppler spectrum of received signal is the sum of the spectrums of independent echoes [Abramovich et al., 1996; Anderson and Abramovich, 1998], i.e.,
where P is the number of propagation modes, fi0 is the Doppler shift of each mode associated with the respective ionospheric reflection layer. Because different layer imposes different fi0, the Doppler spectrum of the combined echoes appears superposition and spreading, which will seriously affect the ship detection. So for ship detection, it is better to select operating frequency that can achieve single-mode propagation to avoid the spectrum spreading.
4.2. Decontamination of Ionospheric Phase Distortion
4.2.1. Employing a Short Coherent Integration Time
[23] As the above mentioned, for a short CIT, the ionospheric phase contamination is approximately linear which only results in a Doppler shift. This shift can be easily estimated by the symmetry of two Bragg lines. Therefore, employing a short CIT is a simple way to avoid the spectral spreading. However, for short time series, the discrete Fourier transform cannot provide the sufficient Doppler resolution. To solve this problem, the high-resolution spectral estimation techniques can be adopted.
4.2.2. Decontamination Based on Maximum Entropy Spectral Analysis
[24] In the Doppler spectrum of OTHR echoes, the ocean Bragg line is so powerful that can be easily identified, even in its broadened state. Therefore, Bourdillon and Gauthier [1987] proposed to extract the broadened positive (or negative) Bragg peak with a band-pass filter and then perform IFFT. When the temporal Bragg signal is obtained, it will be divided into several time segments of about 10 seconds for each. In this time scale, the ionospheric phase contamination is assumed to be linear. Then the maximum entropy spectral analysis is applied to each segment to accurately estimate the quasi-instantaneous frequency of Bragg signal. These quasi-instantaneous frequencies are further interpolated to get instantaneous frequency on each time samples. Finally, by using the Bragg frequency variation in total integration time, the phase contamination can be compensated.
[25] For this method, note that if the phase contamination varies quickly and appears nonlinear in short period, the maximum entropy spectral analysis may not get the correct quasi-instantaneous frequency estimation of Bragg line, which will lead the decontamination to fail.
4.2.3. Decontamination Based on Eigen-Decomposition
[26] This method exploits the high correlation of the phase disturbance experienced by adjacent range and azimuth cells [Abramovich et al., 1996; Anderson and Abramovich, 1998]. Similar to above method, it also divides the total integration time into several segments, and suppose the phase contamination in each segment is linear. For the i th time segment, construct the temporal covariance matrix Ri using the echoes from adjacent range and azimuth cells, and then perform eigen-decomposition of Ri to get the signal subspace [Si,1⋯Si,r] and noise subspace [Gi,1⋯Gi,M−r]. If the phase contamination is not present, the signal subspace of the (i + 1)th segment should be orthogonal to the noise subspace of the i th segment. That is to say, when the phase contamination is correctly compensated, the following projection function will be minimized
where 
is the phase compensation function, ω = 2πΔ, and TR is the repetition period. Thus the differential frequency Δi between the (i + 1)th and the i th segment is estimated by 
= arg (min μi (Δ)). These differential frequencies
should be integrated to get quasi-instantaneous frequency of each
segment. Then the frequency interpolation and phase compensation are
performed in the similar fashion to that in subsection 4.2.2.
The limitation of this decontamination method is the requirements for
the similarity of the clutter spectrum and high correlation of the phase
disturbance in adjacent resolution cells.
4.2.4. Decontamination Based on Phase Gradient
[27] As we point out in subsection 4.2.2, the segment processing may not successfully track the instantaneous frequency when the phase contamination appears nonlinear in short period. To overcome this problem, Parent and Bourdillon [1988] proposed to directly use the time derivative of the phase to estimate the instantaneous frequency of the filtered Bragg signal, i.e., f(t) = Δϕ/2πΔt, where Δϕ and Δt are the phase difference and time interval between adjacent time samples of the filtered Bragg signal, respectively.
[28] This method is simple and allows the
phase contamination with any period to be tracked. However, it was
found that when the amplitude of the filtered Bragg signal is low,
sudden and sharp variations may appear in the estimated instantaneous
frequency, and further degrade the phase decontamination [Parent and Bourdillon, 1988; Gauthier et al., 1990].
Thus it was proposed to employ the signal energy weighted average over
slightly different radar frequencies to improve the estimation
performance, i.e., 
, where ai
is the amplitude of the filtered Bragg signal for each radar frequency.
This method, although effective, presents a restriction on radar
transmitting signal. In practice, the signal energy weighted average may
be performed over adjacent range and azimuth cells on condition that
the phase disturbance is highly correlated and the signal amplitude
fluctuations are different in these resolution cells.
4.2.5. Decontamination Based on Time-Frequency Representation
[29] Time-frequency representation is an effective tool for nonstationary signal processing. Howland and Cooper [1993] published the results of applying the Wigner-Ville distribution (WVD) to phase decontamination. To compare the performance of linear and bilinear time-frequency representations, here we choose the Gabor transform (linear) and WVD (bilinear) to extract the instantaneous frequency of the filtered Bragg signal. The detailed introductions of these two time-frequency representations can be found in Yao et al. [1995] and Cohen [1989].
[30] When applying the time-frequency representation techniques to the phase decontamination, the first step is to calculate the time-frequency representation W(t, f) of the filtered Bragg signal, and then the instantaneous Bragg frequency can be derived from W(t, f) as the first-order moment of W(t, f) in frequency
where the obtained f(t) is truly instantaneous and no further frequency interpolation is needed.
4.2.6. Experimental Results
[31] Having discussed the phase decontamination techniques, we now apply them to a set of data to see how well the contaminated spectrum is recovered. The selected data is expected to be free from ionospheric distortion, whose spectrum is shown in Figure 5a. Then a phase modulation function with the form ejϕ(t), where ϕ(t) = 0.7e0.06t sin 0.4t is introduced and the contaminated spectrum is shown in Figure 5b. Figures 6 and 7display the decontaminated spectrums and the comparison of the estimated frequency modulation (dotted line) with the real one (solid line) for the maximum entropy spectral analysis, eigen-decomposition and phase gradient methods. In our processing, it is found that for the filtered Bragg signal, the maximum entropy spectral analysis method sometimes produces spurious peaks in the Doppler spectrum of segment data, which results in erroneous Bragg quasi-instantaneous frequency estimation and further leads the decontamination to fail. If not filtering the Bragg peak and directly applying the maximum entropy spectral analysis to the segment of original data, the better frequency estimation may be achieved. Such processing is employed in Figure 6a. 8-10 show the results when using time-frequency distribution techniques. It is clear that unlike the linear operator Gabor transform, the bilinear distribution WVD suffers from the cross-term interference. Even for the Bragg signal extracted only from the positive or negative Bragg peak, its WVD still suffers from the cross-terms due to the presence of other unwanted signal components such as noise and the simultaneously filtered second-order ocean echoes. As the result, many spikes appear in the estimated instantaneous frequency in Figure 10b. To suppress the cross-terms and offer more accurate instantaneous frequency estimation, the smoothed pseudo-WVD is adopted. In addition, from Figures 7 and 10, one can see that the edge effect exists in the estimated frequency modulation. Thus for better performance, the first and the last several seconds of the corrected data may be discarded.
(a) Original spectrum expected to be free from phase contamination. (b) Spectrum after introducing the phase contamination.
Decontaminated spectrum: (a) Maximum entropy spectral analysis, (b) Eigen-decomposition, and (c) phase gradient.
The comparison of the estimated frequency modulation (dotted line) with the real one (solid line): (a) Maximum entropy spectral analysis, (b) Eigen-decomposition, and (c) phase gradient.
Decontaminated spectrum: (a) Gabor Transform, (b) WVD, and (c) smoothed pseudo-WVD.
Time-frequency representation of the filtered Bragg signal: (a) Gabor Transform, (b) WVD, and (c) smoothed pseudo-WVD.
The comparison of the estimated frequency modulation (dotted line) with the real one (solid line): (a) Gabor Transform, (b) WVD, and (c) smoothed pseudo-WVD.
4.3. Decontamination of Multipath Propagation
[32] As Figure 2b and equation (2) illustrate, the multipath propagation produces the superposition of Doppler spectrum, which will enlarge the clutter region and influence the ship detection. Theoretically using two-dimensional antenna array can discriminate the multipath echoes in elevation. However, for HF band, the antenna will be of a large size and now high spatial resolution is only achieved in azimuth. Selecting a proper radar operating frequency may achieve the single-mode propagation, but this is not always applicable due to the limitations of ionospheric condition and the desired surveillance region.
[33] If multiple positive (or negative) Bragg peaks can be identified, according to the number, amplitudes and positions of the Bragg peaks, we can get the coarse estimation of the number of propagation modes P, the amplitude σi and the imposed Doppler shift fi0 of each mode. These parameters are subsequently used for multipath decontamination. Note that for simplicity, in this decontamination, the different resolution cell associated with the respective propagation mode is assumed to have the same Doppler spectrum F(f) [Anderson and Abramovich, 1998].
[34] When the phase disturbance and multipath propagation both exist, the contamination is generally severe. Here, it should firstly compensate the phase disturbance to sharpen the spectrum and then perform multipath decontamination. But in the presence of multipath, the performance of the phase decontamination will degrade. A simple derivation is as follows:
[35] Consider two simultaneous propagation modes with the same phase smearing, thus the filtered Bragg signal can be modeled by
where A is the amplitude ratio between the two multipath components, φ and φ1 are the initial phase, f10 is the ionosphere-imposed Doppler shift, and ϕ(t) is the phase smearing function. Thus
From equation (6), it is clear that when multipath propagation exists (i.e., A ≠ 0), the amplitude of the filtered Bragg signal will fluctuate more intensely than that when multipath propagation is absent (i.e., A = 0). And the amplitude fluctuation will significantly influence the instantaneous frequency estimation. The following is the performance evaluation of four mentioned phase decontamination techniques in the presence of multipath propagation. (a) For the phase gradient method, the processing results show that at the time sample where the amplitude of the filtered Bragg signal is low, large spikes may appear in the estimated instantaneous frequencies and further degrade the phase decontamination [Parent and Bourdillon, 1988; Gauthier et al., 1990]. When multipath propagation exists, the amplitude fluctuation of the filtered Bragg signal will be more intense. Thus to reduce the effects of signal amplitude fluctuation on instantaneous frequency estimation, the signal energy weighted average processing is required. (b) For the time-frequency representation techniques, large spikes may also appear during weak signal amplitude and the signal energy weighted average processing is also necessary. (c) For the eigen-decomposition method, its performance generally depends on the correlation of the phase disturbance and the similarity of clutter spectrum in the adjacent resolution cells. With respect to the signal amplitude fluctuation, if the signal amplitudes of adjacent resolution cells are low simultaneously, large spikes may also appear. The average processing over slightly different radar frequencies can be applied to improve the frequency estimation. (d) For the maximum entropy spectral analysis method, our processing results show that when multipath propagation exists, erroneous instantaneous frequency estimation often occurs. So in the authors' view, the maximum entropy spectral analysis method is unsuitable to be applied when multipath propagation exists.
[36] As a conclusion, in the presence of both phase contamination and multipath propagation, it is more necessary to employ the signal energy weighted average to reduce the effects of intense signal amplitude fluctuation on the instantaneous frequency estimation.
[37] After compensating the phase disturbance, the clutter spectrum will be sharpened. Then based on this narrow spectrum, the parameters (P, σi and fi0) of propagation modes can be estimated and the multipath contamination can be corrected.
[38] Here, we would point out that the above conclusions are based on the assumption that the phase disturbance is same for all paths. If this condition is not satisfied, i.e., different ionospheric reflection layer induces distinct phase disturbance, all mentioned phase desmearing methods cannot get satisfactory results. As a summary, Figure 11 presents the performance evaluations of the decontamination of phase disturbance and multipath propagation.
Performance evaluations of decontamination of ionospheric distortion.
5. Target Detection
[39] The targets observed by the OTHR mainly include the aircraft, ballistic missile, cruise missile and ship, etc. For the aircraft, since it is far away from the ocean/ground clutter in the Doppler spectrum, the main limitation for its detection is the external noise. Therefore, the OTHR operating frequency should be selected to achieve the maximal clutter-to-noise ratio. Besides, by employing the wide-aperture receiving antenna array and adaptive beam forming, the external noise can be effectively suppressed.
[40] In addition, the OTHR has potential performance in stealth aircraft detection. The stealth mechanisms generally include shape design and coating wave-absorption material. Since the wavelength of OTHR approximate to the physical size of most aircraft, these targets fall into the resonant scattering region. And their radar cross sections (RCS) are much more dependent on the gross dimensions than on shape detail. Concerning the stealth by coating wave-absorption material, to obtain a satisfactory stealth performance, the coat should be λ/4 thick (where λ is the radar wavelength). But it is impossible for HF band that OTHR operates in.
[41] For the ballistic missile, its physical size approximates to the wavelength of OTHR, which puts them into the resonant scattering region. Besides, the ballistic missile has very fast speed. So its detection is relatively easy. For cruise missile, it belongs to small-sized target and its speed is relatively low. Thus to enhance the detectability, it is better to choose high operating frequency to put them in or approximate to the resonant scattering region.
[42] For the ship target, it is close to the powerful ocean clutter in Doppler spectrum, thus increasing the signal-to-clutter ratio is very important. Additionally, some environmental factors such as the ionospheric phase contamination and multipath propagation will also affect the ship detection. Therefore, when detecting ship target, the following two aspects should be noted: one is selecting appropriate operating frequency with the lowest level of multipath contamination to guarantee the purity of Doppler spectrum; the other is enhancing the resolution in range, azimuth and Doppler to obtain high signal-to-clutter ratio. Currently the 2 km-length receiving antenna array can provide the necessary azimuth resolution. Considering the main limitation for ship detection is the ocean clutter and the effects of external noise are relatively small, the OTHR generally selects the broadband waveform to get high range resolution. For the choosing of CIT, when the discrete Fourier transform is used, the integration time of 25 s or longer is required to get sufficient Doppler resolution. However, such long CIT will produce two problems [Olkin et al., 1997; Root, 1998]: One is the long CIT will decrease the radar revisit frequency over each surveillance area, thereby interfering with the work of the tracker. The other is that as the CIT increase, the probability that the ionosphere's statistics change significantly during the CIT also increases. Hence, there is a requirement for realizing ship detection with short CIT. Presently, at lease two schemes have been proposed. One is directly employing the high-resolution spectral estimator instead of the discrete Fourier transform to obtain the Doppler information [Olkin et al., 1997]. The other is firstly canceling the strong ocean clutter, thus FFT can still extract the ship target from the short-time series [Root, 1998]. This clutter cancellation proceeds by iteration. At each iteration, the dominant ocean clutter peak is modeled as sinusoid in time domain, and then subtracted from data. This subtraction means that the frequency, amplitude and initial phase of the sinusoid should be accurately estimated.
[43] Next, we will illustrate some processing results of short-time data. To compare with the previous results, we extract the first 11s data (16-point time samples) from Figure 4a, and its Doppler spectrum is shown in Figure 12a, which is obtained by direct 16-point FFT. Due to the poor Doppler resolution, in this figure the ship peak is not visible. Figure 12b shows the processing result by high-resolution spectral estimation via modified covariance method. (For details about this method, see Marple [1987].) The ship peak reappears. Figure 12c shows the ship detection result using clutter cancellation by the second iteration, where zero-complemented 128-point FFT is employed to offer more accurate estimation of clutter parameter. In this figure, the ship peak is shown up but other clutter peaks also exist as well as the clutter residues due to imperfect clutter cancellation. These clutter residues can be cancelled at further iterations. However, to avoid mistakenly regarding the ship as the clutter residue and then canceling it, Doppler frequency is used as a criterion to demarcate the clutter and ship.
Ship detection with short CIT: (a) Doppler spectrum of the first 16-point data extracted from Figure 4a. (b) The spectrum obtained by high- resolution spectral estimation. (c) The spectrum after clutter cancellation by the second iteration.
[44] Compared with the aircraft detection, the ship detection is much more difficult, especially with short coherent integration time. For these two types of targets, the optimal radar parameters are different, including the operating frequency, emission bandwidth, pulse repetition period and coherent integration time, etc. Therefore, in a different scan period, we may employ different operating parameters to detect these two types of targets, respectively. However, if we need to detect them simultaneously, the radar operating parameters should be compromised.
6. Conclusion
[45] In this paper, the characteristics, architecture, and signal processing schemes of OTHR are reviewed. In particular, the techniques on interference suppression, ionospheric distortion decontamination, and target detection are discussed in detail. As a summary, the flow chart of OTHR signal processing is shown in Figure 13.
Flow chart of signal processing of OTHR.
[46] The OTHR has the potential for a significant contribution in both military and civilian applications. However, the operating environment of OTHR is very complicated, and so a lot of equipment is involved. It still needs to improve and introduce new technology; however, there is no doubt that it will play an important role in modern radar in the future.
References
No comments:
Post a Comment