Interference Cancellation for High-Frequency Surface ... - IEEE Xplore

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 46, NO. 7, JULY 2008

1879

Interference Cancellation for High-Frequency Surface Wave Radar Xin Guo, Hongbo Sun, and Tat Soon Yeo, Fellow, IEEE

Abstract—The performance of high-frequency surface wave radar (HFSWR) is known to suffer from external environmental interference and noise, such as cochannel radio-frequency interference from other radiating source, ionospheric clutter, lightning impulsive noise, etc. This paper experimentally evaluates the interference cancellation performance of various adaptive beamforming schemes with respect to the aforementioned three types of interferences in an attempt to find the most promising adaptive cancellation scheme in practical HFSWR environment. Index Terms—Adaptive beamforming, high-frequency surface wave radar (HFSWR), interference cancellation.

I. I NTRODUCTION

A

IDED by the good conductivity of ocean water with respect to the vertically polarized electromagnetic wave in the high-frequency (HF) band (3–30 MHz), the HF surface wave radar (HFSWR) is able to detect surface vessels and lowflying aircraft targets beyond the line-of-sight limitation and, consequently, has the potential to be an economical sensor for wide-area coastal surveillance [1], [2]. Although HFSWR possesses many advantages, the powerful external interference and noise in the HF band may cause significant degradation in radar performance. First of all, cochannel radio-frequency interference (RFI) is the main reason for the dramatically increased noise level at night and could mask the targets that have similar Doppler frequencies. Ionospheric clutter is also often observed, which can result in poor target detection and tracking. Moreover, in tropical areas where Singapore is situated, lightning impulsive noise is also of particular concern as tropical thunderstorms are extremely active. Therefore, in this paper, we pay particular attention to the cancellation of cochannel RFI, ionospheric clutter, and lightning impulsive noise. To minimize the impact of strong HF interferences, a number of ameliorative design on radar facilities and signal processing techniques have been developed for HFSWR. An intuitive solution to avoid cochannel RFI is real-time monitoring of the HF spectrum to find an unoccupied frequency channel for radar operation. However, in the user-congested HF band, an Manuscript received July 26, 2007. X. Guo was with the Department of Electrical and Computer Engineering, National University of Singapore, Singapore 119260. She is now with Temasek Laboratories, Nanyang Technological University, Singapore 637553. H. Sun is with Temasek Laboratories, Nanyang Technological University, Singapore 637553. T. S. Yeo is with the Radar and Signal Processing Laboratory, Department of Electrical and Computer Engineering, National University of Singapore, Singapore 119260 (e-mail: [email protected]). Digital Object Identifier 10.1109/TGRS.2008.916482

available channel with sufficient bandwidth (> 50 kHz) is at times extremely difficult to find, particularly during the night and/or at the lower end of the HF band. A good transmitting antenna and receiving antenna array with a deep broad null at high elevation can reduce the cochannel RFI and ionospheric clutter that reach the receiver through the ionospheric reflection [1], [3]. However, due to the very long wavelength of HFSWR, such a design generally requires the antenna to be of substantial height. Some time-domain excision techniques were also proposed in [4]–[6]. However, good suppression performance only occurs when used to cancel the transient interference such as lightning impulsive noise or meteor trail echo. As the cochannel RFI, ionospheric clutter, and lightning impulsive noise all possess some directional characteristics, they can be reduced by adaptive beamforming [3], [7]–[11]. Thus, in case where unoccupied frequency channel and ideal radar antenna facilities are unavailable, the spatial adaptive beamforming is a good candidate for HFSWR interference cancellation. For cochannel RFI cancellation, experimental evaluations on adaptive beamforming had been reported in [8] and [9]. All of them were focused on time-domain cancellation that may be associated with two problems: 1) in time-domain cancellation, the spatial covariance matrix of cochannel RFI must be trained by using the ocean/ground clutter-free time-domain samples. Under some radar parameters for short-range observation, all observation ranges may contain strong ocean/ground clutter. In such situations, time-domain cancellation requires a very effective ocean/ground clutter filtering process in order to obtain clutter-free time-domain samples for interference training; and 2) the performance of time-domain cancellation may be limited by the insufficient degree of freedom of the HFSWR system. When multiple cochannel RFIs appear simultaneously, the estimated spatial covariance matrix using time samples will contain all cochannel RFIs as they generally cannot be separated in the time domain. In this case, the effective cancellation of all cochannel RFIs requires a sufficient degree of freedom of the HFSWR system. Doppler-domain cancellation can cope with these two problems. In Doppler-domain cancellation, the Doppler processing is performed prior to adaptive beamforming, and the Doppler samples that are free of ocean/ground clutter could then be used for interference training. Moreover, in the Doppler domain, the multiple cochannel RFIs may be isolated, hence presenting a possibility of individual cancellation for each RFI, thereby reducing the requirement for the system’s degree of freedom. The effectiveness of Dopplerdomain cancellation for cochannel RFI suppression has been demonstrated in [7]. However, in [7], the antenna weight is

0196-2892/$25.00 © 2008 IEEE

1880


trained by using range averaging along each cochannel RFI range line, and a range guard zone was employed to prevent the target signal from being included in interference training and cancelled subsequently. In our experience, however, we found that the range guard zone alone is insufficient to protect the target signal, particularly in the case where the target signal is strong or is moving rapidly across multiple range cells. Thus, in this paper, an improved Doppler-domain cancellation scheme is proposed, which combines the range guard zone with the exclusion of strong target-like components in the interference training, thereby providing better target protection in adaptive cancellation. With respect to ionospheric clutter cancellation, some experimental investigations on adaptive beamforming have been addressed in [3] and [7]. As the ionospheric clutter may be associated with directional fluctuation due to the dynamic properties of the ionosphere, a time-domain instantaneous cancellation, where the antenna weight vector is calculated for each time instant, was recommended in [7]. However, such timedomain instantaneous cancellation may result in antenna weight vector fluctuating within coherent integration time (CIT) and further result in the severe loss of temporal coherence of the beamformed output. Moreover, unlike the cochannel RFI, the ionospheric clutter only exists in limited range cells, and its spatial properties are different among ranges that are far apart. Therefore, for ionospheric clutter cancellation, the adaptive weight must be trained on adjacent range cells, where the strong ocean/ground clutter may also be present. Thus, the successful application of time-domain cancellation requires a very effective ocean/ground clutter filtering to obtain ocean/ground clutter-free samples for ionospheric clutter training. To avoid the effect of ocean/ground clutter, Doppler-domain cancellation for ionospheric clutter was discussed in [3], but its performance is only compared with conventional beamforming. For lightning strike cancellation based on adaptive beamforming, some experimental results were reported in [11], which shows that lightning is likely to be confined to a narrow spatial sector. However, the need to obtain ocean-clutterfree samples for lightning interference training was again not discussed. Generally, both time-domain and Doppler-domain cancellations are applicable for lightning interference mitigation as long as the strong ocean/ground clutter is excluded in interference training. Noting that the lightning interference is present in the entire range–Doppler map with the direction of arrival (DOA) almost identical among different ranges and Doppler, we propose a new Doppler-domain cancellation scheme which uses the Doppler samples in a region outside the range of expected target-Doppler shifts for interference training. Therefore, it eliminates the need to locate and exclude the ocean/ground clutter and strong target-like components, thereby reducing the computational complexity. This paper is concerned with the unification and extension of previous researches in four areas: 1) propose an improved Doppler-domain cancellation scheme which combines range guard zone and exclusion of strong target-like components in the interference training to better protect the target signal; 2) propose a new Doppler-domain cancellation scheme for lightning interference suppression that eliminates the need to

locate the ocean/ground clutter and strong target-like components; 3) discuss the separate cancellation of multiple simultaneous interferences using Doppler-domain cancellation techniques, although it has been omitted in all previous publications; and 4) provide experimental performance evaluation of various adaptive beamforming schemes with respect to different types of interference, in an attempt to find the most promising adaptive cancellation scheme in practical HFSWR environment. II. I NTERFERENCE C HARACTERISTICS As mentioned in Section I, three sources of interferences, i.e., cochannel RFI, ionospheric clutter, and lightning impulsive noise, are of particular concern. In this section, the characteristics of the three types of interferences in the timedomain, range–Doppler domain, and spatial domain are briefly introduced, on which the subsequent adaptive beamforming schemes are based. Cochannel RFI from other HF radiating stations is the dominant source of interferences encountered for HFSWR. The important characteristics of the cochannel RFI are as follows: 1) it appears in all range cells; 2) it could occur in both high and low Doppler region, thereby having an effect on both aircraft and ship detections; 3) it is highly directional, and its directional characteristics do not change much among different range cells; and 4) the DOA of ionosphere-propagated RFI could fluctuate over time due to the dynamic properties of ionosphere. The ionospheric clutter is a self-generated interference arising when the radar transmitting signal is reflected back from the ionosphere. The important characteristics of the ionospheric clutter are as follows: 1) It could be localized or spread in the range–Doppler map and is likely to be present in the range between 100 and 400 km. However, depending on the pulse repetition frequency adopted in the practical HFSWR system, the far-range ionospheric clutter could be range-folded to the near range cells; 2) unlike the cochannel RFI, ionospheric clutter only exists in limited range cells and is mainly confined to the low Doppler region, thereby having more effect on ship detection than on aircraft detection; 3) it shows some directional characteristics, but its directional characteristics are different for ranges that are far apart. However, if the two ranges of the ionospheric clutter are quite close, their directional characteristics may be similar; and 4) its DOA could fluctuate over time due to the dynamic properties of the ionosphere. Regarding the lightning impulsive noise, the following are observed: 1) it is present in all range and Doppler cells, and 2) it is highly directional, and the directional characteristics are almost identical for different ranges and Doppler. Since all three types of interferences of interest possess some directional characteristics, they can be mitigated by adaptive beamforming. However, due to many other differences as summarized previously, the appropriate adaptive beamforming scheme for each interference mechanism also differs. We would highlight that the characteristics of ionospheric clutter are mostly different from others. First, the ionospheric clutter only exists in limited range cells. Thus, for ionospheric clutter cancellation, range is an important consideration factor as the

GUO et al.: INTERFERENCE CANCELLATION FOR HIGH-FREQUENCY SURFACE WAVE RADAR

adaptive weight must be trained on the range cells that contain the ionospheric clutter. Second, the directional characteristics of the ionospheric clutter are distinct for two ranges that are far apart, whereas the cochannel RFI and lightning have similar directivity among different ranges. Hence, as far as the cochannel RFI and lightning cancellations are concerned, it is feasible to train the antenna weight on a set of ranges and then apply to other ranges. However, for ionospheric clutter cancellation at a given range cell, the antenna weight must be trained on adjacent range cells. Third, unlike the cochannel RFI and lightning that could affect the detection of both ship and aircraft targets, the ionospheric clutter generally only masks the ship targets. Therefore, adaptive cancellation of ionospheric clutter is not necessary for aircraft detection mode. III. A DAPTIVE B EAMFORMING S CHEMES FOR I NTERFERENCE C ANCELLATION IN HFSWR Generally, the adaptive beamforming is based on the famous minimum variance distortionless response (MVDR) criterion [12]. The antenna weight vector is given by w=

R−1 n S H S R−1 n S

(1)

where Rn is the spatial covariance matrix of interference plus noise and S is the array steering vector. The theoretical analysis of the MVDR-based adaptive beamforming algorithm can be found in many literatures such as [12]. However, in practical applications, the methods of covariance matrix construction may differ in order to handle different types of interference. Thus, different adaptive beamforming schemes can be derived based on the same fundamental of MVDR. For the interference cancellation in HFSWR, the adaptive beamforming can be implemented in time or Doppler domain. In this section, three time-domain cancellation schemes and four Doppler-domain cancellation schemes are introduced. Some of them have been reported in other literatures, and others are newly proposed here. Their performance will be compared and evaluated by using experimental data in Section IV. A. Time-Domain Cancellation In time-domain cancellation, the interference training and beamforming are both performed on the range–time data recorded by antenna elements across the array. Then Doppler processing is carried out on the beamformed time series at each range cell. To avoid the effects of strong ocean/ground clutter, the interference training should be performed at far range cells where the ocean/ground clutter is absent. 1) Time-Domain Cancellation Scheme 1 (Applicable for Cochannel RFI and Lightning Cancellation): Use CIT averaging and fixed-range-window averaging in a set of ocean/ground clutter-free range cells to estimate the interference covariance matrix, and then, apply the computed antenna weight vector to all samples in the range–time map [as shown in Fig. 1(a)]. This scheme was first proposed in [8]. It is applicable to cochannel RFI and lightning strikes, as they are present in all

1881

range cells and the directional characteristics do not change much among different ranges. For ionospheric clutter, however, this scheme is not appropriate because its directional characteristics are distinct among range cells that are far apart. 2) Time-Domain Cancellation Scheme 2 (Applicable for Ionospheric Clutter Cancellation): Use CIT averaging and sliding-range-window averaging (with a range guard zone) in a set of ocean/ground clutter-free range cells to estimate the interference covariance matrix, and then, apply the computed antenna weight vector to all samples at the current range cell [as shown in Fig. 1(b)]. This scheme is proposed here and can be used for ionospheric clutter cancellation as its directional characteristics show similarities among adjacent range cells, provided that the ionospheric clutter is present in the range cells that do not contain strong ocean/ground clutter (or the ocean/ground clutter has been filtered). A range guard zone around a current range cell is employed to prevent the potential target signal from being included in the interference training and then mitigated. Note that in this scheme, the number of range cells used in range averaging should be limited due to the different directional characteristics of ionospheric clutter among different ranges. 3) Time-Domain Cancellation Scheme 3 (Considering Interference DOA Fluctuation): Divide the total CIT into several nonoverlapped time segments, and in each time segment, use time averaging and range averaging to estimate the quasiinstantaneous covariance matrix of the interference, then apply the computed antenna weight vector to all samples at the current range cell of current time segment [as shown in Fig. 1(c)]. Intuitively, the quasi-instantaneous cancellation can provide better cancellation performance for some DOA-fluctuated interference, such as ionosphere-propagated cochannel RFI and ionospheric clutter. However, such a time-varying antenna weight vector will result in the beamformed output to suffer from a severe temporal decoherence and will further result in Doppler spreading in the subsequent Doppler processing. To solve this problem, an extra data-driven constraint is added in [8] and [9] to reduce the temporal coherence loss caused by the time-varying antenna weight vector while still allowing the antenna weight vector to vary from one time segment to another. However, as will be demonstrated later, appropriate parameter selection in the data-driven constraint is crucial for the successful application of this scheme. B. Doppler-Domain Cancellation As mentioned in Section I, all time-domain cancellation schemes need to work on the ocean/ground clutter-free range cells and may suffer from the insufficient system degrees of freedom when multiple interferences appear simultaneously. The Doppler-domain cancellation can cope with these two problems. In Doppler-domain cancellation, the range processing and Doppler processing are first performed on each antenna element, and then, the interference training and beamforming are both carried out on the range–Doppler data recorded by different antenna elements across the array. Thus, the Doppler samples that are free of ocean/ground clutter could be selected for interference training. More importantly, the multiple

1882


Fig. 1. Diagrammatic illustrations of three time-domain cancellation schemes. (a) Time-domain cancellation scheme 1. (b) Time-domain cancellation scheme 2. (c) Time-domain cancellation scheme 3.

simultaneous interferences, particularly the cochannel RFIs and ionospheric clutter, may be isolated in the range–Doppler domain, which means that separate cancellation for each interference is possible, and the requirement for the system degrees of freedom can be significantly reduced. This is an important advantage of Doppler-domain cancellation, although it was omitted in all previous publications. 1) Doppler-Domain Cancellation Scheme 1 (Applicable for Cancellation of All Three Types of Interferences): Use Doppler averaging at each range cell to estimate the interference covariance matrix (with ocean/ground clutter and strong target-like components excluded), and then, apply the computed antenna weight vector to the Doppler samples at the current range cell [as shown in Fig. 2(a)]. In this scheme, as the antenna weight vector training and application are both performed at the current range cell, it could be used to cancel all the three types of interferences. Compared with the schemes where the antenna weight vector is trained on a set of range cells and then applied to other range cells, the interference training of this scheme is performed in the current

range cell and, thus, will provide better interference cancellation performance. However, a big flaw of such processing is that all potential target signals at the current range cell are also included in the interference training and, hence, may be cancelled. To protect the target signal, Abramovich et al. [10] proposed to first use the conventional processing to recognize the strong target-like components and, then, to exclude both the ocean/ground clutter and strong target-like components in the interference training. 2) Doppler-Domain Cancellation Scheme 2 (Applicable for Cancellation of All Three Types of Interferences): Use Doppler averaging and sliding-range-window averaging (with range guard zone) to estimate the interference covariance matrix (with ocean/ground clutter and strong target-like components excluded), and then, apply the computed antenna weight vector to the Doppler samples at the current range cell [as shown in Fig. 2(b)]. In the experimental processing of Doppler-domain cancellation scheme 1, we often found that the exclusion of strong target-like components in the interference training alone is


1883

Fig. 2. Diagrammatic illustrations of four Doppler-domain cancellation schemes. (a) Doppler-domain cancellation scheme 1. (b) Doppler-domain cancellation scheme 2. (c) Doppler-domain cancellation scheme 3. (d) Doppler-domain cancellation scheme 4.

insufficient to provide necessary target protection, particularly for the target that is masked by the interference and has moderate signal level. Such a target is likely to be unidentified by the target recognition procedure and, therefore, included in the interference training and impaired after the adaptive cancellation. The range guard zone is another method for target protection, which uses the samples from neighboring range cells for interference training. However, we also find that, although the range guard zone has perfect performance in weak target protection, its performance may drop in the case where the target signal is strong or with fast motion that occupies multiple range cells. Thus, to provide reliable protection for both strong and weak targets in adaptive cancellation, it will be better to combine the range guard zone with the exclusion of strong target-like components in interference training. Based on this idea, the aforementioned scheme is proposed here. Compared with the Doppler-domain cancellation scheme 1, this scheme has a slightly worse cancellation performance for ionospheric clutter due to the use of sliding-range-window averaging with a range guard zone. Fortunately, in most cases, this

degradation of cancellation performance is trivial. However, its target protection capability is enhanced significantly. In addition, as an important but often omitted advantage, the Doppler-domain processing is able to reduce the requirement for system’s degree of freedom. Specifically, in the Doppler cancellation schemes 1 and 2, if the number of interference directions is few and the system’s degree of freedom is sufficient, the interference training can be simply performed by using the averaging of the whole Doppler range of expected target-Doppler shifts. Otherwise, the interference training can be performed separately in the Doppler region around each interference, provided that these multiple simultaneous interferences (such as RFI and ionospheric clutter) are isolated in the Doppler domain. As a result, the limitation of the system’s degree of freedom is released, and the cancellation performance is enhanced. 3) Doppler-Domain Cancellation Scheme 3 (Applicable for Cochannel RFI Cancellation): For each cochannel-RFIoccupied Doppler cell, use the sliding-range-window averaging (with a range guard zone) on this Doppler cell to estimate the

1884


interference covariance matrix (with strong target-like components excluded), and then, apply the computed antenna weight vector to the current range–Doppler cell [as shown in Fig. 2(c)]. This scheme was originally proposed in [7] based on the fact that the cochannel RFI is present in all range cells like a range line and that its DOA does not change much among different ranges. Here, as an improvement, we still combine the range guard zone and exclusion of strong target-like components in the interference training to provide the better target protection. In this scheme, the cancellation is performed along each cochannel RFI range line. Thus, it effectively releases the limitation of system degrees of freedom in the case where multiple cochannel RFIs occur simultaneously. 4) Doppler-Domain Cancellation Scheme 4 (Applicable for Lightning Cancellation): Use Doppler averaging (or combined with range averaging) in a region outside the range of expected target-Doppler shifts to estimate the interference covariance matrix, and then, apply the computed antenna weight vector to all range–Doppler cells [as shown in Fig. 2(d)]. Although the Doppler-domain cancellation schemes 1 and 2 are both applicable to cancel lightning interference, it should be noted that both of them need to locate and exclude the ocean/ground clutter and strong target-like components in interference training, which may not be possible all the time as the lightning can significantly increase the noise floor. Considering that the lightning interference is present in the entire range–Doppler map with the DOA almost identical among different ranges and Doppler, the aforementioned scheme is proposed here, which uses the Doppler samples outside the range of expected target-Doppler shifts for interference training. Thus, it eliminates the need to locate and exclude the ocean/ground clutter and strong target-like components, thereby significantly reducing the computational complexity. IV. E XPERIMENTAL P ERFORMANCE E VALUATION In this section, the interference cancellation performances of various adaptive beamforming schemes are evaluated by using the raw data collected by an experimental HFSWR in Singapore area. This HFSWR system works in an onshore configuration and monostatic mode. The transmit configuration is a monopole antenna with a monopole reflector. The receiving system is based on a uniform linear array of 16 vertical polarization monopole antenna elements that occupy 90 m in length, where the outputs of four adjacent antenna elements are first combined via an analog beamformer and, then, fed into one of the four receiving channels to perform digital beamforming. Thus, this HFSWR system has three degrees of freedom and is capable of canceling the interferences from three directions. More detailed information about this experimental HFSWR system can be found in [13]. A. Cancellation of Cochannel RFI Fig. 3(a) shows a range–Doppler map of applying conventional beamforming to an experimental data segment. As can be observed, many ship targets, as well as several cochannel RFIs, were detected. From the target tracking results, we know

that a ship target should be present at the range cell 16 and Doppler cell 165. Unfortunately, its Doppler frequency happens to coincide with that of a cochannel RFI, causing the target to be submerged and undetectable. To cancel the cochannel RFIs, several adaptive beamforming schemes presented in Section III are employed. First, to illustrate the effects of strong ocean/ground clutter on adaptive cancellation, we use the time-domain samples at the range cell 18 to train the interference and then apply the calculated weight to all range cells. The results are shown in Fig. 3(b). It can be seen that the cancellation performance is very poor. Apparently, it is important to exclude the strong ocean/ground clutter in interference training. Fig. 3(c) shows the results of applying the time-domain cancellation scheme 1 to the data, where the time-domain samples at the range cell 150 are employed for interference training. Since the interference training is performed at the range where the cochannel RFIs are the dominant components rather than the ocean/ground clutter, it yields better cancellation performance than that in Fig. 3(b), and the ship target masked by the cochannel RFI has emerged (as indicated by the arrow). However, it can be seen that the residual cochannel RFIs still exist. This may be due to the fact that the degree of freedom of the experimental HFSWR (which is only three for our system) is insufficient to reject all RFIs. Fig. 3(d) shows the results of applying the Doppler-domain cancellation scheme 1 to the data, where the antenna weight vector is trained on the current range cell using all the Doppler samples in the expected target-Doppler region (with ocean/ground clutter and strong target-like components excluded). In comparison with Fig. 3(a), the interference level is reduced, and the ship target masked by the cochannel RFI has emerged. However, similar to the scheme in Fig. 3(c), the scheme in Fig. 3(d) suffers from the insufficient system degrees of freedom, because all the Doppler samples in the expected target-Doppler region are employed for interference training. In addition, we see that the magnitude of the revealed target (indicated by the arrow) is slightly weaker than that in Fig. 3(c). Such a result implies that excluding the strong target-like components alone in interference training may be insufficient for reliable target protection, particularly for the targets that are masked by the interference and have moderate signal level. This is because such targets are likely to be unidentified by the target recognition procedure and therefore included in the interference training. Fig. 3(e) shows the result of applying the Doppler-domain cancellation scheme 2 to the data, where the antenna weight vector is trained on a group of range cells using the slidingrange-window averaging with a range guard zone, and all the Doppler samples in the expected target-Doppler region (with ocean/ground clutter and strong target-like components excluded) are employed for interference training. In comparison with Fig. 3(d), as this scheme combines the range guard zone and the strong target-like components exclusion, it offers better target protection capability, and the revealed target is more prominent. However, as can be observed, its RFI cancellation performance is still limited by the insufficient degrees of freedom of the radar system.


1885

Fig. 3. Range–Doppler map of cochannel-RFI-affected experimental HFSWR data with (a) conventional beamforming, (b) time-domain cancellation scheme 1, where the antenna weight vector is trained on range cell 18, (c) time-domain cancellation scheme 1, where the antenna weight vector is trained on range cell 150, (d) Doppler-domain cancellation scheme 1, without separate cancellation, (e) Doppler-domain cancellation scheme 2, without separate cancellation, (f) Doppler-domain cancellation scheme 2, with separate cancellation.

Since the cochannel RFIs can be isolated in the Doppler domain, the separate cancellation is possible. Fig. 3(f) shows the result of applying the Doppler-domain cancellation scheme 2 to the data. However, different from that used in Fig. 3(e), the Doppler axis is now divided into several regions (Doppler cells 1–70, 71–180, 181–220, and 221–250). Then, the antenna weight vector training and application are performed separately for each Doppler region. Obviously, this separate cancellation effectively releases the limitation of system degrees of freedom and, therefore, provides better cancellation performance than that in Fig. 3(c)–(e).

Fig. 3(g) shows the result of applying the Doppler-domain cancellation scheme 3. In this scheme, the cancellation is performed along each cochannel RFI range line and, thus, effectively overcomes the limitation of system degrees of freedom. As observed, its cochannel RFI cancellation performance is even better than that in Fig. 3(f), and the target masked by the RFI is very prominent after cancellation. In the aforementioned adaptive cancellation schemes, we did not consider the DOA fluctuation of cochannel RFI with time. The ionosphere-propagated cochannel RFI may be associated with DOA fluctuation due to the dynamic properties of the

1886


Fig. 3. (Continued.) Range–Doppler map of cochannel-RFI-affected experimental HFSWR data with (g) Doppler-domain cancellation scheme 3, (h) time-domain cancellation scheme 3, with Pa = 16 and K = 1, and (i) time-domain cancellation scheme 3, with Pa = 16 and K = 3.

ionospheric layer. Since the aforementioned time-domain cancellation scheme 1 and Doppler-domain cancellation schemes employ the interference data within the entire CIT, either in time domain or in Doppler domain for interference training, the DOA fluctuation of interference is averaged over time. When the interference direction is time varying, the resulting antenna patterns of these schemes will yield a broad null, which covers the DOA variation of the interference. Thus, even when the interference direction is fluctuating, these schemes can still achieve effective interference cancellation. For comparison, the quasi-instantaneous interference cancellation, i.e., time-domain cancellation scheme 3, is applied. In this scheme, the entire CIT is divided into a series of time segments with each containing Pa time samples, in which the first K samples are used for data-driven constraint, and three range cells from 151 to 153 that are free of strong ocean/ground clutter are used for range averaging in interference training. Fig. 3(h) and (i) shows the results with Pa = 16, K = 1 and Pa = 16, K = 3, respectively. In Fig. 3(h), the interference level is reduced, and the ship target that is masked by the cochannel RFI has emerged. However, compared with Fig. 3(a), the beamformed ocean/ground clutter in Fig. 3(h) spreads in the Doppler domain, due to the temporal coherence loss caused by the time-varying antenna weight vector. In Fig. 3(i), the value of K is increased to three. Consequently, the data-driven constraint offers a better preservation of temporal coherence of the beamformed output. However, the large K results in significant performance degradation in interference cancellation.

Fig. 4.

SINR improvements gained in Fig. 3(c)–(i) as a function of range.

Therefore, the appropriate parameter selection of Pa and K is crucial to the successful application of time-domain cancellation scheme 3. Based on our experience in experimental data processing, the appropriate parameter selection of Pa and K is data dependent. In addition, this scheme also suffers from the insufficient system degrees of freedom. Fig. 4 shows the signal-to-interference-plus-noise ratio (SINR) improvements gained in Fig. 3(c)–(i) as a function of range. The findings are as follows. 1) Doppler-domain cancellation scheme 3, which works along each cochannel RFI line, yields the best cancellation performance with the mean improvement of about 9.89 dB. The Doppler-domain cancellation scheme 2,


where separate cancellation is applied, achieves the second best performance with the mean improvement of about 6.55 dB. Both schemes greatly benefit from the release of the limitation of insufficient system degrees of freedom. However, it should be noted that the Dopplerdomain cancellation scheme 3 is only applicable for individual cancellation of cochannel RFI, whereas the Doppler-domain cancellation scheme 2 can also be applied to separate cancellation of other interference mechanisms. 2) Time-domain cancellation scheme 1 and Doppler-domain cancellation schemes 1 and 2 without separate cancellation only give the mean improvements of 2.89, 3.09, and 3.02 dB, respectively, because all of them suffer from the insufficient system degrees of freedom. The Dopplerdomain cancellation scheme 1 provides slightly better performance than the time-domain cancellation scheme 1 and the Doppler-domain cancellation scheme 2. This is within our expectation as in Doppler-domain cancellation scheme 1, the antenna weight vector training and application are both performed at the current range cell, which can avoid the influence of interference directional variation across range. 3) As to quasi-instantaneous interference cancellation, the time-domain cancellation scheme 3 with Pa = 16 and K = 1 yields the mean improvement of 3.53 dB, which is slightly better than the Doppler-domain cancellation schemes 1 and 2 without separate cancellation. However, this result is obtained with the cost of significant temporal coherence loss and Doppler spreading in the near range cells. The performance of time-domain cancellation scheme 3 with Pa = 16 and K = 3 gets the worst SINR improvement of about 1.56 dB. Obviously, when multiple cochannel RFIs occur simultaneously and the system degree of freedom is limited, the separate cancellation scheme working on each cochannel RFI is more effective than that working on quasi-instantaneous interference direction. Furthermore, the former will also not cause the temporal coherence loss and the Doppler spreading. B. Cancellation of Ionospheric Clutter Fig. 5(a) shows the range–Doppler map of an experimental data set, which is corrupted by the ionospheric clutter, after applying the conventional beamforming. To evaluate the target detection performance after ionospheric clutter cancellation, a target signal is artificially injected to these data at the range cell 280 and Doppler cell 57. In Fig. 5(a), the injected target is masked by the ionospheric clutter. First, to illustrate the effect of directional characteristics of ionospheric clutter on interference cancellation, we use the Doppler samples at a fixed range cell 245 to train the interference, and then, we apply the computed antenna weight vector to all range cells corrupted by the ionospheric clutter. Fig. 5(b) shows the corresponding results. It is observed that the cancellation performance is satisfactory near the range cell 245. However, the performance decreases when the applied range is

1887

away from the antenna weight trained range. The injected target is still unable to be distinguished. Such a result actually proves that the directional characteristics of ionospheric clutter may be similar among the adjacent range cells but could have great difference among the range cells that are far apart. As the ionospheric clutter in this scenario is present at the range cells that do not contain the strong ocean/ground clutter, the time-domain cancellation scheme 2 could be applied, and the result is shown in Fig. 5(c). Compared with Fig. 5(a), the ionospheric clutter level is reduced. Since the interference training of this scheme is performed in the neighboring range cells, its cancellation performance is evidently better than that in Fig. 5(b), and the injected target masked by the ionospheric clutter has emerged (as indicated by the arrow). Fig. 5(d) and (e) shows the results of applying the Dopplerdomain cancellation schemes 1 and 2 to the data, where all the Doppler samples in the expected target-Doppler region are employed for interference training. It is observed that both schemes reduce the ionospheric clutter level and that the injected target submerged by the ionospheric clutter is revealed. Fig. 5(f) shows the result of applying the Doppler-domain cancellation scheme 2 to the data, where the Doppler axis is divided into several regions (Doppler cells 1–70, 71–156, and 157–193), and then, the separate cancellation is applied. Clearly, it provides the better ionospheric clutter cancellation performance than that in Fig. 5(c)–(e), and the revealed target is more prominent. At last, considering the DOA fluctuation of the ionospheric clutter with time, the quasi-instantaneous interference cancellation, i.e., time-domain cancellation scheme 3, is applied and evaluated. Fig. 5(g) and (h) shows the results with the parameters Pa = 16, K = 1 and Pa = 16, K = 3, respectively. Similar to the results of using this scheme to RFI cancellation, in Fig. 5(g), the ionospheric clutter level is reduced, and the injected target masked by the ionospheric clutter has emerged. However, compared with Fig. 5(a), the beamformed ground clutter in Fig. 5(g) has somewhat spreading in the Doppler domain due to the temporal coherence loss caused by the timevarying antenna weight vector, particularly at the range cells from 260 to 310. In Fig. 5(h), a larger value of K offers a better temporal coherence preservation, but the ionospheric clutter cancellation performance is significantly degraded. Such results also imply that the appropriate parameter selection of Pa and K is crucial to the successful application of quasi-instantaneous interference cancellation and may be data dependent. Fig. 6 shows the improvement of ionospheric clutter cancellation gained in Fig. 5(b)–(h) as a function of range. The following can be observed. 1) The interference cancellation, where the antenna weight vector is trained on a fixed range cell 245, offers a similar cancellation performance to that of Doppler-domain cancellation schemes 1 and 2 at the range cells close to the trained range. However, when the applied range cell is far away from the trained range, particularly between range cells 260 and 310, its cancellation performance is significantly reduced in comparison with the Dopplerdomain cancellation schemes 1 and 2. Such results can

1888


Fig. 5. Range–Doppler map of ionospheric-clutter-corrupted HFSWR data with (a) conventional beamforming, (b) Doppler-domain cancellation, where the antenna weight vector is trained on the fixed range cell 245, (c) time-domain cancellation scheme 2, (d) Doppler-domain cancellation scheme 1, without separate cancellation, (e) Doppler-domain cancellation scheme 2, without separate cancellation, (f) Doppler-domain cancellation scheme 2, with separate cancellation, (g) time-domain cancellation scheme 3, with Pa = 16 and K = 1, and (h) time-domain cancellation scheme 3, with Pa = 16 and K = 3.


Fig. 6.

SINR improvements gained in Fig. 5(b)–(h) as a function of range.

be expected because the directional characteristics of ionospheric clutter could have great difference among the range cells that are far apart. 2) The time-domain cancellation scheme 2 and the Dopplerdomain cancellation schemes 1 and 2 without separate cancellation yield the mean improvements of 4.71, 5.52, and 5.15 dB, respectively. Such results are reasonable because the time-domain cancellation 2 uses the timedomain sample for interference training, in which, although the ocean/ground clutter is not the dominant component, its presence can still cause the computed antenna weights not to be well matched to the ionospheric clutter. Therefore, its cancellation performance is slightly worse than that of Doppler-domain cancellation schemes 1 and 2. 3) As expected, the Doppler-domain cancellation scheme 2 with separate cancellation yields the best cancellation performance with the mean improvement of about 6.41 dB. 4) The time-domain cancellation scheme 3 with Pa = 16 and K = 1 yields the mean improvement of 5.03 dB, which is slightly better than that of the time-domain cancellation scheme 2 but worse than that of the Dopplerdomain cancellation schemes 1 and 2. The time-domain cancellation scheme 3 with Pa = 16 and K = 3 yields the worst cancellation performance throughout almost all of the corrupted range cells with the mean improvement of only 1.75 dB. C. Cancellation of Lightning Impulsive Noise Fig. 7(a) shows a conventionally beamformed range–Doppler map of an experimental data set, which is severely affected by lightning interference. The noise floor is significantly increased and masks almost all potential targets. As the lightning strike is the dominant component at the far range cells rather than the ocean/ground clutter, the timedomain cancellation scheme 1 could be applied. Fig. 7(b) shows the corresponding results, where the antenna weight vector is trained on the fixed range cell 152. It can be found that the lightning energy is significantly reduced and that the ship targets as well as the ocean/ground clutter are clearly visible. Fig. 7(c) and (d) shows the results of applying the Dopplerdomain cancellation schemes 1 and 2 to the data, in which all

1889

the Doppler samples in the expected target-Doppler region are employed for interference training (with ocean/ground clutter and strong target-like components excluded). It is shown that both schemes greatly decrease the lightning level. However, note that two ship targets that are well revealed in Fig. 7(b) (as indicated by the circle) are impaired and cannot be detected in Fig. 7(c). The likely reason for this is that these two targets have moderate signal level and are completely masked by the lightning. As a result, they are included in the interference training and impaired after adaptive interference cancellation. Different from the Doppler-domain cancellation scheme 1, the Doppler-domain cancellation scheme 2 combines the range guard zone and strong target-like components exclusion in the interference training, its target protection capability is therefore enhanced, and the aforementioned two targets are clearly visible in Fig. 7(d). Although the Doppler-domain cancellation schemes 1 and 2 are all applicable to cancel lightning interference, it should be noted that both schemes need to locate and exclude the ocean/ground clutter and strong target-like components in interference training, which may not be possible all the time as the lightning can significantly increase the noise floor. Considering that the lightning interference is present in the entire range–Doppler map with the DOA almost identical among different ranges and Doppler, it is feasible to use the Doppler samples in a region outside the range of expected targetDoppler shifts to train the interference. Fig. 7(e) shows the results of applying the Doppler cancellation scheme 4, where the antenna weight vector is trained by using the samples in the region of Doppler cells 1–15 and range cells 80–82. Obviously, this scheme eliminates the need to locate and exclude the ocean/ground clutter and strong target-like components, and very good cancellation performance is also achieved. Fig. 8 shows the performance improvements of lightning cancellation gained in Fig. 7(b)–(e) as a function of range. It is shown that the aforementioned four cancellation schemes provide the similar cancellation performance with mean improvements of 17.64, 17.81, 17.75, and 17.4 dB, respectively. V. D ISCUSSION From the aforementioned experimental results, it can be found that Doppler-domain cancellation is more preferred in practical HFSWR applications for the following reasons: 1) Doppler-domain cancellation can simply avoid the effects of ocean/ground clutter, and 2) multiple interferences may be isolated in the Doppler domain, hence presenting a possibility of separate cancellation for each interference, which greatly releases the limitation of system degree of freedom. With respect to different interference mechanisms, the most promising cancellation schemes in practical HFSWR applications are suggested as follows. In the HFSWR operating environment, multiple cochannel RFIs often occur simultaneously. Thus, the system degree of freedom is an important factor to be considered. If the system degree of freedom is sufficient, the Doppler-domain cancellation scheme 2 without separate cancellation is a good candidate for cochannel RFI cancellation due to its good cancellation

1890


Fig. 7. Range–Doppler map of lightning-impulsive-noise-corrupted HFSWR data with (a) conventional beamforming, (b) time-domain cancellation scheme 1, where the antenna weight vector is trained on the fixed range cell 152, (c) Doppler-domain cancellation scheme 1, without separate cancellation, (d) Dopplerdomain cancellation scheme 2, without separate cancellation, (e) Doppler-domain cancellation scheme 4.

Fig. 8. SINR improvements gained in Fig. 7(b)–(e) as a function of range.

performance, reliable target protection capability, and simple processing procedure (there is no need to locate the Doppler position of the cochannel RFI). If the degree of freedom of the HFSWR system is insufficient, the Doppler-domain cancellation scheme 3 is more suitable, as its cancellation is performed along each cochannel RFI line, thereby effectively releasing the limitation of system’s degree of freedom. Alternatively, the Doppler-domain cancellation scheme 2 with the separate cancellation is also applicable for multiple RFI suppression when the degree of freedom is limited. In particular, when other an interference mechanism (such as ionospheric clutter) occurs simultaneously with the cochannel RFI, Doppler cancellation 2 with separate cancellation is also applicable.


For the cancellation of ionospheric clutter, since its directional characteristics are different among range cells that are far apart, the Doppler-domain cancellation scheme 2 is preferred, which provides better target protection capability than the Doppler-domain cancellation scheme 1. However, some specular ionospheric clutter only occupies a very small number of range cells, thereby making the use of slidingrange-window averaging with a range guard zone unrealistic. In this case, the Doppler-domain cancellation scheme 1, where the weight training and application are both performed at the current range cell, is the appropriate scheme. For the cancellation of lightning impulsive noise, the Doppler-domain cancellation scheme 4 is the best solution, as it eliminates the need to locate the ocean/ground clutter and the strong target-like components, thereby reducing the computational complexity. VI. C ONCLUSION The cochannel RFI, ionospheric clutter, and lightning impulsive noise often significantly degrade the target detection performance of the HFSWR system. This paper summarizes various adaptive beamforming schemes (including two Doppler-domain cancellation schemes that are newly proposed in this paper) for the cancellation of the aforementioned three types of interferences. The cancellation performances of different schemes with respect to different types of interferences are evaluated by using experimental HFSWR data. The processing results show good agreements with the theoretical analysis and may serve as a guideline for the selection of appropriate adaptive cancellation schemes in practical HFSWR applications. ACKNOWLEDGMENT The authors would like to thank the Defense Science and Technology Agency of Singapore for providing the experimental HFSWR data.

1891

[8] G. A. Fabrizio, A. B. Gershman, and M. D. Turley, “Robust adaptive beamforming for HF surface wave over-the-horizon radar,” IEEE Trans. Aerosp. Electron. Syst., vol. 40, no. 2, pp. 510–525, Apr. 2004. [9] Y. I. Abramovich, N. K. Spencer, S. J. Anderson, A. Y. Gorokhov, “Stochastic-constraints method in nonstationary hot-clutter cancellation—Part I: Fundamentals and supervised training applications,” IEEE Trans. Aerosp. Electron. Syst. vol. 34, no.4, pp.1271–1292, Oct. 1998. [10] Y. Abramovich, P. Turcaj, N. K. Spencer, R. M. Ellard, and Y. Lyudviga, “Surface wave radar,” U.S. Patent 7 145 503. [11] J. D. R. Kramer, Jr. and R. T. Williams, “High frequency atmospheric noise mitigation,” in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process., 1994, pp. VI-101–VI-104. [12] R. A. Monzingo and T. W. Miller, Introduction to Adaptive Arrays. Hoboken, NJ: Wiley, 1980. [13] M. Lesturgie, G. Auffray, S. Saillant, H. L. Chan, X. S. Goh, and H. W. Seah, “Contribution of new technologies to HF surface wave radar for maritime surveillance,” in Proc. Int. Conf. Radar Syst., Toulouse, France, 2004.

Xin Guo received the B.Eng. and Ph.D. degrees in electronic engineering from the Nanjing University of Science and Technology, Nanjing, China, in 1999 and 2004, respectively. From 2004 to 2006, she was a Research Fellow with the Department of Electrical and Computer Engineering, National University of Singapore. Currently, she is a Research Scientist with Temasek Laboratories, Nanyang Technological University, Singapore. Her research interests include adaptive array signal processing, time–frequency distribution techniques, and high-resolution spectral analysis with application to radar.

Hongbo Sun received the B.Eng. and Ph.D. degrees in electronic engineering from the Nanjing University of Science and Technology, Nanjing, China, in 1997 and 2002, respectively. From 2002 to 2004, he was with the School of Electrical and Electronic Engineering, Nanyang Technological University (NTU), Singapore, as a Research Fellow. He is currently a Research Scientist with Temasek Laboratories, NTU. His current research areas include passive radar systems, highfrequency surface wave radar systems, multipleinput–multiple-output radar technology, and radar signal processing.

R EFERENCES [1] L. Sevgi, A. Ponsford, and H. C. Chan, “An integrated maritime surveillance system based on high-frequency surface-wave radars, Part 1: Theoretical background and numerical simulations,” IEEE Antennas Propag. Mag., vol. 43, no. 4, pp. 28–43, Aug. 2001. [2] A. Ponsford, L. Sevgi, and H. C. Chan, “An integrated maritime surveillance system based on high-frequency surface-wave radars, Part 2: Operational status and system performance,” IEEE Antennas Propag. Mag., vol. 43, no. 5, pp. 52–63, Oct. 2001. [3] Y. Abramovich, S. Anderson, Y. Lyudviga, N. Spencer, P. Turcaj, and B. Hibble, “Space-time adaptive techniques for ionospheric clutter mitigation in HF surface wave radar systems,” in Proc. Int. Conf. Radar Syst., Toulouse, France, 2004. [4] J. R. Barnum and E. E. Simpson, “Over-the-horizon radar sensitivity enhancement by impulse noise excision,” in Proc. IEEE Nat. Radar Conf., 1997, pp. 252–256. [5] T. F. Quan, J. W. Li, C. J. Yu, H. Wang, and Z. L. Ma, “An approach and experiment of suppressing burst interference in high-frequency radar,” Acta Electron. Sinica, vol. 27, no. 12, pp. 23–25, 1999. [6] X. Guo, H. Sun, and T. S. Yeo, “Transient interference excision in overthe-horizon radar using adaptive time-frequency analysis,” IEEE Trans. Geosci. Remote Sens., vol. 43, no. 4, pp. 722–735, Apr. 2005. [7] H. C. Chan and E. K. L. Hung, “An investigation in interference suppression for HF surface wave radar,” Def. Res. Establishment Ottawa, Ottawa, ON, Canada, DREO Tech. Rep. 2000-028, 1999.

Tat Soon Yeo (M’79–SM’93–F’03) received the B.Eng. degree (with first class honors) from the University of Singapore, Singapore, in 1979, the M.Eng. degree from the National University of Singapore (NUS), Singapore, in 1981, and the Ph.D. degree under a Colombo Plan Scholarship from the University of Canterbury, Christchurch, New Zealand, in 1985. He is currently a Professor with the Department of Electrical and Computer Engineering, NUS, where he is also the Vice Dean of the Faculty of Engineering. He is concurrently the Director of the Radar and Signal Processing Laboratory and Antennas and Propagation Laboratory, Department of Electrical and Computer Engineering, NUS. He is also the Director of the Temasek Defence Systems Institute, a teaching institute established jointly by NUS and the U.S. Naval Postgraduate School. His current research interests include scattering analysis, synthetic aperture radar, antenna and propagation study, numerical methods in electromagnetics, and electromagnetic compatibility. Dr. Yeo received the Singapore Ministry of Defence—NUS 1997 Joint R&D Award, the IEEE Millennium Medal in 2000, and the Singapore Standard Council’s Distinguish Award in 2002. He is the past-Chairman and Executive Committee Member of the MTT/AP and EMC Chapters, IEEE Singapore Section, and the Chairman of the Singapore EMC Technical Committee.