The principle of a global navigation satellite system is that satellites are launched into space to form a constellation around the surface of the Earth. The relative distance between satellites and receivers can be calculated by measuring the time delay between the transmission from multiple satellites and the reception at the receivers on the ground. Then, the three-dimensional coordinates of receivers can be solved [1]. In October 1957, the Soviet Union launched the first artificial satellite into space. In 1958, the United States Navy decided to research and develop the Navy Navigation Satellite System (NNSS) based on the Doppler frequency shift and launched the first satellite of the system in April 1960. The system was called Transit because all six satellites orbited about the poles of the Earth. The NNSS was the first successfully operating satellite navigation system in the world [2]. In 1973, the U.S. Department of Defense proposed a plan to develop a new generation of satellite navigation systems, which led to what is now known as the Global Positioning System (GPS). The system launched its first experimental satellite in February 1978 and entered complete operation in 1995 [3]. Subsequently, to eliminate dependence, some countries and organizations have established their own satellite navigation systems. Currently, the Global Navigation Satellite System (GNSS) is roughly divided into four systems, including GPS in the U.S., the Galileo Navigation Satellite System (Galileo) in the European Union [4], the Global Navigation Satellite System (GLONASS) in Russia [5] and the Beidou Navigation Satellite System (BDS) in China [6]. In addition, India and Japan have established their own regional navigation systems, the Indian Regional Navigational Satellite System (IRNSS) in India [7] and the Quasi-Zenith Satellite System (QZSS) in Japan [8]. Since their birth and development, navigation systems have played an essential role in both military and civil fields. In the military field, they provide precise guidance information for weapons. In the civil field, they provide positioning and navigation services for aircraft, fishing boats and vehicles and are widely used for the location information of mobile devices. Therefore, the strategic significance is remarkable, and the economic benefits cannot be underestimated.

A satellite navigation system is vulnerable to radio frequency interference because the signal transmitted from the satellite to the ground is feeble. Because of the weak signal strength, the sensitivity of a navigation receiver is important, and interference with the satellite navigation receiver is the main interference mode at present. In addition, the broadcasting frequency band and data format of satellite navigation signals are relatively fixed, so it is easy to receive intentional or unintended interference of similar signals, such as multipath interference or spoofing interference that mimics real signals [9].

When a satellite navigation receiver is interfered with, its performance will degrade, and the navigation and positioning function will fail. Many incidents of interference and deception against satellite navigation have occurred. In December 2011, Iran managed to take control of a U.S. RQ-170 Sentry drone and land it unharmed inside Iran by making it believe it was at a base in Afghanistan. The incident has been described as a successful example of spoofing GPS navigation systems. In June 2012, Todd Humphreys, an assistant professor at the University of Texas at Austin, and his students successfully captured a drone, demonstrating that spoofing against GPS systems could happen again [10]. In 2016, the United States conducted a large-scale GPS interference experiment. The tests were centered in Nevada’s 1.1-million-acre China Lake installation, but it affected ten different states and regions of GPS systems, including Oregon, Idaho, California, Nevada, Utah, Arizona, New Mexico, Colorado, Wyoming, and Sonora (Mexico) [11].

GNSS interference monitoring technologies are developed based on radio monitoring. Aviation authorities first raised the need for radio monitoring. The Federal Aviation Administration (FAA) established a civil aviation radio interference monitoring detection system (IMDS) covering the mainland United States. This system consists of dozens of fixed stations, movable stations, mobile stations, airborne interference monitoring systems, and a monitoring center data communication network. It monitors radio interference signals for civil aviation airports in the United States [12]. With the wide application of radio technology in communication, radar, and navigation, the electromagnetic environment has become increasingly complex. To address the influence of the worsening electromagnetic environment on radio systems, radio interference monitoring systems have developed rapidly, and extensive research has been done in the field of radio interference detection, direction finding, positioning, and so on. On the basis of radio interference monitoring, some achievements have also been made in the research of interference monitoring systems in the field of satellite navigation.

The research on these interference monitoring systems can be divided into three categories: spaceborne, airborne, and ground platforms. The ultimate purpose of these interference monitoring systems is to locate an interference source [13].

Ground platform monitoring equipment, such as fixed ground monitoring stations and handheld or vehicle-mounted portable monitoring equipment, can monitor the interference signals around it. A joint network with multiple stations can realize the accurate positioning of interference sources and electromagnetic signal monitoring for the whole satellite navigation system working environment. The interference monitoring equipment is greatly affected by the ground environment. Sometimes, handheld or vehicle-mounted portable monitoring devices cannot be close to interference sources for monitoring because of the impact of the ground environment. In 2020, Li H. et al. designed a ground interference monitoring platform composed of ground monitoring stations and monitoring vehicles. The system can effectively measure the direction of the interference signal and locate the interference source [14].

Airborne platform monitoring equipment, which is monitoring equipment carried on planes, can sometimes be carried on unmanned aerial vehicles (UAVs). Planes or UAVs can take the edge off the terrain effects. Airborne satellite monitoring equipment can approach an interference source more effectively for monitoring to a certain extent. Another advantage of UAV monitoring equipment is that after locating an interference source using a localization algorithm, it can directly approach the interference source, and an image of the interference source can be transmitted back by the camera. Wu G. researched interference source localization based on UAVs [15]. Then, he considered a case where multiple UAVs cooperate in locating interference sources [16]. Sun X. designed and implemented UAV-based direction finding and positioning of interference sources, whose direction finding accuracy can reach 3° [17].

Spaceborne platform monitoring equipment generally refers to electronic reconnaissance satellites. Monitoring equipment based on ground or airborne platforms covers a limited area, while spaceborne platform monitoring equipment can monitor interference over a wide range. Among them, the principle of triple-satellite positioning systems is mostly a single method based on the time difference of arrival (TDOA) or frequency difference of arrival (FDOA). The TDOA method is a relatively mature positioning technology among many positioning methods and generally has better positioning accuracy and performance than the FDOA [18]. The principle of dual-satellite positioning systems is mainly based on the joint positioning technology of the TDOA and FDOA. This technology combines the TDOA with the FDOA to reduce the number of positioning satellites needed. Compared with a triple-satellite positioning system, a dual-satellite positioning mode reduces the number of satellites, simplifies the system complexity and reduces the launch cost. The advantage of this method is that it can effectively save orbit resources. However, this method uses the FDOA, which requires high Doppler frequency shift measurement accuracy. For a satellite with a slight Doppler frequency shift, it will cause a significant positioning error [19].

The principle of a single satellite positioning system is to reduce the number of satellites by combining an ellipsoidal model of the Earth’s surface. In this way, only one measurement information is needed to locate an interference source. A common method is based on the angle of arrival (AOA) of the interference signal. Although this positioning method has no requirement for Doppler shift, its disadvantage is that it also needs high measurement accuracy, which brings difficulties to single-satellite positioning [20]. Generally, an interference monitoring system based on a spaceborne platform has a broader monitoring range, and the more satellites used, the better the monitoring system performs.

Many interference monitoring systems have been produced and applied based on these principles. LOCO GPSI, which began in 1997, is a development project directed by the Congress of the United States. LOCO GPSI is based on a short baseline interferometer to find the direction of the interference source and can determine the location of the interference source. The Space and Naval Warfare Systems Center (SSC) subsequently developed the LOCO GPSI Mini Lite, a miniaturized interference monitoring system. The LOCO GPSI interference monitoring system was first created on an airborne platform. Its technology can be adapted to ground platform equipment, such as vehicle-mounted or handheld equipment [21]. Interference monitoring systems based on a spaceborne platform include the SatID interference positioning system in Britain and the TLS Model 2000 interference positioning system in the U.S., both of which can achieve a positioning accuracy of 10–20 km. The first satellite-based commercial system that sought to provide comprehensive and near-real-time monitoring on a global scale was a constellation of small satellites called Hawkeye 360. The plan was for a group of three satellites to eventually lead to a worldwide radio monitoring system. On 3 Dec 2018, the first three satellites were launched. The initial plan was to launch 18 satellites, gradually increasing to 30. Although the three satellites were in a group, a TDOA/FDOA joint positioning method is used. As long as two of the three satellites are within the visual range, the interference source can be located [22].

Because a satellite navigation system is of great strategic and economic significance, easily interfered with, and the damage caused by the interference is great, interference monitoring technology in a satellite navigation system is crucial. Interference monitoring technologies can be divided into the following parts:

Interference detection, which aims to determine whether there is interference in a specific area;

Interference recognition, based on the time or frequency characteristics of interference, to identify the type of interference;

Interference source direction finding, to estimate the incoming direction of the interference signal;

Interference source location, to estimate the location of an interference source;

Interference source tracking, to estimate the speed and acceleration of an interference source, track it, and predict its trajectory.

Interference monitoring technologies aim to comprehensively perceive and continuously monitor interference in a particular area. Although there is no direct interference processing at the receiver, interference monitoring technologies actually provide considerable prior information for interference suppression. For example, if the direction of an interference source is known, the signal received by an array antenna can be processed using spatial filtering [23]. Interference monitoring technologies improve the pertinence of interference suppression algorithms, which can suppress interference more effectively and retain more useful signals [24]. In addition, knowing the location of an interference source can enable a receiver to avoid the interference source to ensure stable receiver operation.

This review is structured as follows. In Section 2, Section 3, and Section 4, interference monitoring technologies are introduced in the order of interference detection and recognition, interference source direction finding, interference source location and tracking. In Section 2, interference detection technology is the main technology, and interference recognition is introduced briefly. In Section 3, interference source direction finding technology is introduced completely. In Section 4, interference source location technology is introduced, and interference tracking is briefly introduced. In Section 5, the challenges faced by interference monitoring technologies are summarized and possible development directions for interference monitoring technologies in the future are proposed.

Due to the functionality of satellite navigation systems, different from other radio systems, it is necessary to consider intentional interference. The types of navigation signal interference mainly include blanket interference and spoofing interference [26]. Compound interference, a combination of blanket interference and spoofing interference, is also common. Generally, in the GNSS domain, jamming refers to intentional suppression interference, and spoofing refers to intentional deception interference. Blanket interference can be interpreted as radio frequency interference (RFI). Its purpose is to suppress the navigation signal from the time domain or frequency domain using a high-power interference signal to prevent a navigation receiver from working normally [27]. The RFI usually referred to may be unintentional interference in the electromagnetic environment. Spoofing refers to the use of signals similar to navigation signals as interference signals to affect satellite navigation receivers, thereby causing the navigation system to deviate from the correct positioning and navigation and provide incorrect navigation information. Since satellite navigation systems are also used for timing services, there is also deception in the timing services [28]. On the other hand, a multipath interference signal in a satellite navigation system is very similar to a real signal, but multipath interference is the unintended interference caused by different signal paths [29]. Spoofing interference is usually intentional.

Blanket interference is simple in principle, easy to implement, low in cost, and obvious in effect. Most existing interference is blanket interference. Generally, the higher the power of the interference signal is, the better the interference effects.

Blanket interference can be distinguished into two types in terms of the frequency domain: narrowband interference and broadband interference. Narrowband interference, whose frequency is concentrated in a small bandwidth, has strong pertinence. Partial band interference or targeting interference also expresses a similar meaning [30]. The most common narrowband interference is Gaussian narrowband interference, which is generated by passing white Gaussian noise through a filter. Another common narrowband interference is tone interference, that is, an interference signal only at a single frequency point, single-tone interference or an interference signal at several frequency points, and multitone interference [31]. For wideband interference, the bandwidth of the interference signal is wide. Blocking interference has a similar meaning [32]. The most common wideband interference is Gaussian wideband interference, which has a wide bandwidth. Another common broadband interference is frequency modulation interference (also known as sweeping interference) [33]. The frequency of the interference signal varies over a wide range, and is typically linear frequency modulation interference (LFM) [34]. Chirp interference means the same thing. In terms of generation mode, random binary codes can be modulated to the interference frequency band in BPSK mode to form narrowband BPSK interference or wideband BPSK interference. They are also called spread spectrum interference or matched spectrum interference [35]. The frequency spectra of common blanket interference types are shown in Figure 2.

In terms of the time domain, blanket interference can be divided into pulse interference and continuous wave interference (CWI). Pulse interference refers to the interference that focuses the interference energy on a certain period of time in a pulse cycle to transmit [36]. The opposite of pulse interference is continuous wave interference, which has a longer duration [37]. Because of the cost, pulse interference with a short duration is usually high-power broadband interference, while continuous wave interference is mostly narrowband interference. However, this is not absolute. For better interference effects, broadband interference can also be used in the form of continuous wave interference. The classification of blanket interference is shown in Figure 3.

A spoofing interference signal is similar to a real signal in form, and its purpose is not only to make a target receiver lose its original working performance but also to make it operate in a specified way [38]. In the field of satellite navigation, spoofing interference can make the target positioning be offset or be unable to complete the positioning and can make the target positioning to its specified position, which makes the harm of spoofing interference significant.

Spoofing interference in the field of satellite navigation can be generally divided into two categories, namely, generative spoofing interference and repeater spoofing interference. Generative spoofing interference equipment is mainly composed of a signal analog source, power amplifiers, and transmit antennas. Since the civil signal structure, signal characteristics, pseudocode characteristics and navigation message structure of existing navigation systems are all public, generative spoofing is easy to generate by imitating a real signal [39]. After receiving a real navigation signal, repeater spoofing will delay and amplify the power of the signal. Although the specific structure of the real navigation signal is not known, it can still have a certain impact on the positioning result [40]. Repeater spoofing is more commonly used in the military field. The military pseudocode of a navigation system is secret, so generating generative spoofing is almost impossible. Based on these two basic spoofing interference generation methods, spoofing interference continues to develop new methods, such as complex spoofing interference generated according to the estimated information after receiving the real signal or spoofing interference forwarded by using multiple antennas. Spoofing is sometimes divided into simple spoofing and complex spoofing according to their different functions.

One of the theoretical bases of interference detection technology is detection theory, which is used to determine whether a signal exists, that is, whether there is a signal or only noise in the case of noise [41]. Let s n denote a signal and w n denote noise; then, we can express it as the following two hypotheses: H 0 : x n = w n n = 0,1 , . . . , N − 1 H 1 : x n = s n + w n n = 0,1 , . . . , N − 1 (1)

In mathematics, if H 0 is true and we think H 1 is true, it is called a type I error. If H 1 is true and we think H 0 is true, it is called a type II error. In detection theory, the first type of error can be regarded as a misjudgment that there is a signal when there is no signal. Therefore, this event can also be called a false alarm (FA), and the probability of the occurrence of a false alarm is called the false alarm probability, which is usually denoted by P F A . If H 1 is true, and we think H 1 is true, we can consider that we have successfully detected a signal. This event is called detection, and the probability of successful detection is called the detection probability, which is usually denoted by P D . In statistics, the detection probability is also called the power of the test.

For each value of x , the ratio of the probability of H 1 to the probability of H 0 is called the likelihood ratio L x , where x = x 0 x 1 ⋯ x N − 1 . If we consider that the likelihood ratio is greater than a certain threshold, that is, the ratio of the probability of H 1 to the probability of H 0 is greater than a certain threshold γ , L x = p x ; H 1 p x ; H 0 > γ (2)

You can say that event H 1 happened. This test is called the likelihood ratio test (LRT). If p x ; H 1 ; p x ; H 1 are known and the false alarm probability P F A = α is set, the threshold γ can be obtained using the Neyman-Pearson theorem (NP theorem), p F A = ∫ x : L x > γ p x ; H 0 d x = α (3)

The false alarm probability is usually a very small value, also known as the significance level or detection scale in statistics, which indicates that the NP theorem seeks the decision with the maximum detection probability when limiting the false alarm probability to a small value. It aims to protect hypothesis H 0 and not accept hypothesis H 1 easily. In some practical applications, once a signal is detected, it is necessary to make a response, which requires paying a certain price. If the false alarm probability is too large, it will cause unacceptable losses. If no signal is detected, no action is needed. If the cost caused by not detecting the signal is large, the NP theorem is not necessarily applicable, and some adjustment is needed.

For example, in the case of deterministic signals, if signal s n is known and noise w n is white Gaussian noise with variance σ 2 , the specific probability density function can be substituted into Eq. 2 and simplified to obtain: T x = ∑ n = 0 N − 1 x n s n > γ ′ (4)

The right side of the inequality is denoted by a new threshold γ ′ . The left side of the inequality T x is called the test statistic, which is actually the cross-correlation between the known signal and the actual observed value. This is the NP detector, which consists of a test statistic T x and a threshold γ ′ . The threshold is also obtained by the NP theorem. The NP detector at this time is also called the matched detector.

If the signal is a random signal, assume that it is a Gaussian random process with variance σ s 2 ; then, T x = ∑ n = 0 N − 1 x 2 n > γ ″ (5)

Here, the test statistic T x is the energy of the observed data, and the NP detector at this time is also called the energy detector. A matched filter and energy detector are basic and common detection methods.

In a satellite navigation system, a common signal model is x i = s i + j i + n i (6)

This means satellite navigation signal, jamming and noise. The noise is not always white Gaussian noise, so it is directly denoted as n i . From energy detection, blanket interference detection technology mainly includes energy detection in the time domain and the frequency domain. Energy detection in the frequency domain is based on the fast Fourier transform (FFT). A time-frequency plane detection method based on a short-time Fourier transform (STFT) is also developing rapidly.

There have been some developments in blanket interference detection based on energy detection. The same detector has different detection performances for different interference types. Some detectors have good detection effects for specific interference. Based on energy detection, some algorithms improve the detection performance of some particular disturbances. Nunes F. D. et al. proposed blind interference detection based on fourth-order cumulants, which is more suitable for FM interference. Compared with the kurtosis algorithm, although the computational amount is increased, it achieves better performance [42]. Huo S. et al. improved the consecutive mean excision (CME) algorithm and proposed a block-flow method based on the backward CME (BCME) algorithm. This method can reduce the amount of storage and computation under the condition that the performance of monitoring pulse interference is similar [43].

Another rapidly developing direction is interference detection algorithms based on time-frequency analysis technology. Sun K. et al. proposed a new reassigned spectrogram method for interference detection for GNSS receivers. It has good detection performance for LFM interference [44]. They also provide another class of interference detection algorithms, aiming to transform the time scale using a wavelet transform, which can be used to detect weak RF signals. A new STFT method based on time-frequency domain analysis proposed by Wang P. improves the performance of narrowband and broadband detection in a signal with a low dry-to-noise ratio [45]. Lv Q. et al. also provided a time-frequency domain-based detection method using a goodness-of-fit (GoF) test [46]. Compared with the Hough transform of the Wigner-Ville distribution, the reassigned smoothed pseudo Wigner-Ville distribution (RSPWVD) has better performance in reducing cross-term interference and requires less computation [47].

Other interference detection algorithms are also feasible. Motella B. et al. applied detection theory to the field of satellite navigation and used a goodness-of-fit method to detect interference in satellite navigation systems [48]. Wu Q. proposed an interference detection algorithm for GNSS receivers based on adaptive subspace tracking technology and receiver autonomous integrity monitoring (RAIM) [49]. Zhai S. proposed an interference detection algorithm based on fuzzy logic fusion. In particular, for the problem of reduced detection performance caused by noise fluctuations in a complex electromagnetic environment, the proposed algorithm introduces the idea of fuzzy logic [50]. Silva F. B. et al. proposed a new precorrelation interference detection technology based on non-negative matrix factorization (NMF). The proposed technology uses NMF to extract the time and frequency properties of a received signal from its spectrogram. The estimated spectral shape is then compared with the time slices of the spectrogram using a similarity function to detect the presence of RFI [51].

Spoofing interference detection technology is quite complex, and the classification methods are different. This paper simply divides spoofing interference detection technology into signal categories and other categories.

Methods in signal categories are mainly related to the characteristics of the signal itself. The power of spoofing interference is usually larger than the real signal power, so it can be detected using the signal strength. The common detection method is absolute power monitoring [52]. A method comparing the power of the L1 channel and L2 channel is called relative power detection. It is also effective in detecting the carrier-to-noise ratio (CNR) of the signal [53]. Among these methods, a special method is auto gain control (AGC) detection. AGC detection is located at the RF front end of a receiver before the correlation operation. The principle is that the AGC at the RF front end of the receiver will produce large fluctuations when spoofing interference exists [54]. Because spoofing interference is forwarded, the arrival angle of the signal is different from that of the real signal. Therefore, spoofing interference can also be detected from the arrival angle of the signal [55]. Magiera J. combined array signal processing to detect spoofing interference according to the difference between the arrival angle of the spoofing signal and that of the real signal and used beamforming for spoofing interference mitigation [56]. The repeater spoofing interference has a delay, so it can also be detected by the arrival time [57]. Carrier Doppler can also be used for spoofing interference detection. This method compares the Doppler shift change of the real signal with the Doppler shift change calculated from the reported position [58]. Signal quality monitoring (SQM) is a very common detection method that mainly uses the output of the correlator. When there is spoofing interference, the related summit will be distorted [59]. Wang W. et al. proposed a detection method based on the S-curve bias (SCB) [60]. This algorithm works well and is expected to be combined with other algorithms. Multi-peak detection can also be classified into this category [61]. There are also methods such as residual signal detection and carrier phase detection, but their principles are mostly the same.

Different from the methods in signal categories, methods in other categories detect spoofing signals from aspects such as data layer and positioning results. At the data layer, navigation messages can be encrypted, which is useful for detecting spoofing interference. Lewis S. W. et al. presented a GNSS interferometric reflectometry (GNSS-IR) signature-based method [62]. Humphreys T. E. demonstrated that navigation message authentication (NMA) is effective for anti-spoofing [63]. Receiver autonomous integrity monitoring (RAIM) methods can detect repeater spoofing interference. For repeater spoofing, the repeater spoofing signal increases the satellite delay, so the calculated pseudorange value will mutate after the receiver receives the repeater spoofing signal, causing the consistency of the calculated pseudorange value of each satellite to worsen. Han S. et al. combined particle filtering and the RAIM algorithm, which can effectively deal with the spoofing interference of multiple satellites [64]. By comparing the positioning results of the satellite navigation system with the positioning results based on other methods, we can judge whether the satellite navigation system receives spoofing interference. Khanafseh S. et al. used the results of an inertial navigation system (INS) combined with RAIM to detect spoofing signals [65]. Jeong S. et al. proposed a method to detect spoofing interference, which uses the GNSS augmentation system as the correction data [66]. Spoofing interference detection can be combined with machine learning. Shafiee E. al. extracted the early-late phase, delta, and signal level as the three main features from the correlation output of the tracking loop. Using these features, spoofing detection can be performed by exploiting conventional machine learning algorithms such as K-nearest neighbor (KNN) and naive Bayesian classifiers [67]. Li J. et al. proposed a GNSS anti-spoofing method based on the idea of confrontation evolution of a general adverse network (GAN) [68].

The recognition of spoofing interference can be regarded as distinguishing the real signal and spoofing interference in the received signal, and then the effective part of the received signal can be used to continue the localization instead of using the spoofing interference signal for the subsequent localization solution. To a certain extent, the detection of spoofing interference basically completes the suppression of spoofing interference. The research in this direction mainly involves the suppression of spoofing interference, so we will not introduce it. The recognition of blanket interference is similar to the detection of blanket interference. The characteristics of the interference signal can be extracted from the time domain, the frequency domain and the time-frequency domain. By analyzing the characteristics of the blanket interference, the types of blanket interference can be identified effectively, and the corresponding filters can be designed to suppress the blanket interference.

Most interference recognition is based on pattern recognition [69] and machine learning [70] theory, although there are other ways. In the field of radar, the application of machine learning algorithms is more common. In the field of GNSS, research on interference recognition is limited. Kang C. et al. proposed a time-domain identification method based on an adaptive cascading IIR notch filter to identify single-tone, multitone, swept continuous wave interference (CWI) and band-limited white Gaussian noise (BLWN) [71]. Ferre R. M. proposed applying machine learning approaches to sort the interference signal into five classes [72]. Chen X. et al. proposed a deep convolutional neural network (DNN) named FPS-DNN on the basis of a convolutional neural network (CNN) for fingerprint recognition. This method can classify the interference types very well [73].

Interference detection technology is divided into blanket interference detection technology and spoofing interference detection technology. The development of blanket interference detection technology is relatively perfect. Blanket interference detection starts from energy detection and accumulates the sum in different transform domains, such as the time domain, frequency domain, time-frequency domain, and wavelet domain, to detect interference signals. According to the characteristics of interference signals, different detection algorithms have different effects. With the complexity and change of the interference signal, the targeted algorithm can often achieve better results. The detection technology of spoofing is quite complex, and the classification methods are different. In this paper, spoofing detection technology is simply divided into a signal class and other classes. Spoofing interference detection from the signal level involves signal processing technology, and the signal power and signal correlation peak are monitored. From the information level, we can verify whether the obtained information is wrong. We can use inertial navigation, data encryption and other methods to verify the positioning results and navigation messages. These two aspects involve different technologies, but both can play a role in spoofing detection, and their development is relatively independent.

Interference recognition technology classifies interference signals by extracting their features. The classification methods rely on pattern recognition techniques that emerged in the 1960s and 1970s, including support vector machines, clustering, principal component analysis and so on. In the 1980s, machine learning became an independent subject field and developed rapidly, and various machine learning techniques appeared. With the rise of machine learning, its functionality is almost perfect to replace the pattern recognition methods of the past. Interference recognition technology has developed rapidly on this basis.

Among amplitude direction finding methods, the most common method is the comparative amplitude direction finding method. By comparing the amplitude of the signals received by antennas at different positions, the angle information of the signals can be calculated. Typical direction finding systems include four-antenna array Adcock direction finding [77] and Watson-Watt direction finding [78]. An Adcock direction finding system is based on a four-antenna array [79], which calculates the direction of the signal based on the amplitude of the signal received by the north‒south and west‒east antennas [80]. A dual-channel Watson-Watt direction finding system requires two receiving channels with exactly the same amplitude and phase responses [81]. There are also single-channel Watson-Watt direction finding systems.

There are also some traditional amplitude methods, such as the maximum signal amplitude method, which uses a strong directional antenna to directly take the direction of the maximum signal amplitude as the direction of the incoming wave. Alternatively, the minimum signal method uses a figure-8 antenna to receive the signal, with the zero point of the antenna pattern being the direction of the incoming wave [82]. These methods are simple, but the direction finding error is large.

Phase direction finding methods are mainly based on interferometers. One is a phase interferometer. It estimates the angle information of the incoming wave through the phase difference between the signals received by the antennas at different positions [83]. A phase interferometer can realize the direction finding of a single pulse, so it is also called phase single pulse direction finding. The other type of interferometer, a correlative interferometer, estimates the direction of the signal by comparing the measured phase difference distribution with the one that has been stored beforehand. The direction corresponding to the maximum of the correlation coefficient is the direction of the signal [84]. The direction finding accuracy of interferometers often depends on the baseline, and multibaseline interferometers are mainly used in practical applications.

In addition, there are some other phase direction finding methods, such as Doppler direction finding based on the Doppler characteristics of the signal [85]. A Wullenweber direction finding system, which is one of the older methods, uses a circular array of antennas [86]. A system that uses the array antenna and the time difference of arrival to find the direction is also a phase direction finding method in essence.

Spatial spectral estimation direction finding has superresolution capability, also known as superresolution direction finding. Compared with traditional methods, a spatial spectrum estimation direction finding method is more accurate and widely used and has been developed rapidly in recent years. Spatial spectrum estimation direction finding is developed on the basis of array signal processing and spectrum estimation. It uses array antennas to receive signals, estimate the spatial spectrum of signals, and estimate the direction information of signals. Similar to the frequency spectrum, the spatial spectrum only takes the direction in space as the abscissa, reflecting the direction information of the signal. The diagram of an array antenna is shown in Figure 4.

Among spatial spectrum estimation methods, the most typical algorithm is multiple signal classification (MUSIC) [87]. MUSIC estimates the power spectrum of a signal by decomposing it into a signal subspace and a noise subspace. In array signal processing, MUSIC is used to estimate the spatial spectrum of a signal.

The observed data are regarded as the sum of the signal and noise, x t = s t + w t (7)

The wave surface of a signal needs to travel d cos ⁡ θ more distance to reach array element 1 than to reach array element 2, so the time to receive the signal at array element 1 will be later than that at array element 2. This time interval is τ 0 = d ⁡ cos ⁡ θ c = d ⁡ cos ⁡ θ λ f (8) where c represents the speed of light, λ represents the wavelength of the signal, and f represents the frequency of the signal. For the exponential signal, this delay can be expressed as the difference in phase, i.e., s t − τ 0 = s t e − j ω τ 0 = s t e − j ω d ⁡ cos ⁡ θ λ f = s t e − j 2 π λ d ⁡ cos ⁡ θ (9)

Assuming that the number of array elements is N and x i t is used to represent the observed signal of the i th array element, the received signal of the array can be expressed as: x 1 t x 2 t ⋮ x N t = 1 e − j 2 π λ d ⁡ cos ⁡ θ ⋮ e − j 2 π λ N − 1 d ⁡ cos ⁡ θ s t + w 1 t w 2 t ⋮ w N t (10)

For multiple signal sources, assume that the number of signal sources is D , let s i t denote the signal transmitted by the i th signal source, and let θ i denote the incident angle of the signal. In this case, Equation 10 is extended as follows: x 1 t x 2 t ⋮ x N t = 1 1 ⋯ 1 e − j 2 π λ d ⁡ cos ⁡ θ 1 e − j 2 π λ d ⁡ cos ⁡ θ 2 ⋯ e − j 2 π λ d ⁡ cos ⁡ θ D ⋮ ⋮ ⋱ ⋮ e − j 2 π λ N − 1 d ⁡ cos ⁡ θ 1 e − j 2 π λ N − 1 d ⁡ cos ⁡ θ 2 ⋯ e − j 2 π λ N − 1 d ⁡ cos ⁡ θ D s 1 t s 2 t ⋮ s D t + w 1 t w 2 t ⋮ w N t (11)

Define the vector x = x 1 t x 2 t ⋯ x N t T , which represents a data vector of observations and is a vector of N × 1 dimension;

vector S = s 1 t s 2 t ⋯ s D t T represents the signal source vector, which is a vector of D × 1 dimension;

vector w = w 1 t w 2 t ⋯ w N t T represents the noise vector, which is a vector of N × 1 dimension;

vector a θ i = 1 e − j 2 π λ d cos θ i ⋯ e − j 2 π λ N − 1 d cos θ i T , called the steering vector, is a vector of N × 1 dimension;

matrix A = a θ 1 a θ 2 ⋯ a θ D represents the direction matrix, which is a matrix of N × D dimensions;

s denotes the signal vector received by the array antenna, s = A ⋅ S . Therefore, Eq. 11 can be expressed as x = s + w = A ⋅ S + w (12)

Since noise and signal are not correlated, the autocorrelation R x of the observed data can be expressed as the sum of the autocorrelation of the signal received by the array antenna and the noise autocorrelation R x = R s + R w = A S A S H + σ w 2 I = A R S A H + σ w 2 I (13)

After eigenvalue decomposition, R x = R s + R w = ∑ i = 1 N λ s e s e s H + ∑ i = 1 N λ w e w e w H = U s Σ s U s H + U w Σ w U w H (14)

R S = S S H = d i a g P 1 P 2 ⋯ P D is a diagonal array, and the diagonal element P i is actually the signal power of the signal source. At this time, the autocorrelation R s of the signal just conforms to the definition of eigenvalue decomposition, which indicates that among the N eigenvalues of the autocorrelation R s of the signal, only D are non-zero values, which are the signal power of the signal source. R s = A R S A H = ∑ i = 1 D P i a θ i a H θ i (15)

Thus, the eigenvalues and eigenvectors of the autocorrelation matrix R x of the observed data are calculated. For D signal sources, λ 1 ≥ λ 2 ≥ ⋯ ≥ λ D ≥ λ D + 1 = ⋯ λ M = σ w 2 (16)

The eigenvectors corresponding to the first D eigenvalues, called principal eigenvectors, constitute the signal subspace. The eigenvectors corresponding to the last N − D eigenvalues constitute the noise subspace. Assuming that the signal-to-noise ratio (SNR) of the received signal is very high and contains almost no noise, we can approximate σ w 2 = 0 . Because eigenvectors with different eigenvalues are orthogonal to each other, the signal subspace is orthogonal to the noise subspace. a H θ U N = 0 (17)

The MUSIC spectrum is defined as P M U S I C = 1 a H θ U N U N H a θ (18)

If the signal does not contain noise, the denominator is zero, and the MUSIC spectrum is meaningless, so the MUSIC spectrum is also called the MUSIC pseudospectrum. However, in fact, noise is always there, and it is just approximately zero. By taking the inverse, we can obtain the peak, which is the estimate of the angle of the signal.

The number of interference sources is set as 3, and the arrival directions of the interference signals are − 3 0 ∘ , 0 ∘ , and 6 0 ∘ . Select the linear array element, the number of array elements is 8, and the number of fast beats is 500. The MUSIC spatial spectrum estimation is carried out in the cases of SNRs of 10 and 1. As seen from the simulation results, when the SNR is high, the MUSIC algorithm can simultaneously identify the interference signals in three directions and provide relatively accurate direction information, as shown in Figure 5A. When the SNR decreases, the performance of the MUSIC algorithm is affected when the simulation is carried out under the same three interference sources, as shown in Figure 5B.

In 1969, Capon proposed the minimum variance spectrum estimation method (MVM), and to some extent, spatial spectrum estimation was developed on this basis [88].

In 1979, Schmidt R. proposed the MUSIC algorithm, whose basic idea is to divide the received signal into an orthogonal signal subspace and a noise subspace. The performance of the method is close to that of the maximum likelihood method, but the computational amount is enormous. This is a landmark achievement of spatial spectrum estimation theory. In 1986, Roy R. et al. proposed the ESPRIT algorithm, the core of which is to rely on the rotational invariance of the signal subspace and use the least square method to obtain the angle of arrival. The search process of the spectral peak in the MUSIC algorithm is avoided, and the calculation amount is effectively reduced. This method not only has good resolution but also obtains high real-time performance [89].

In the late 1980s, angle estimation algorithms for subspace fitting emerged. This kind of method obtains the desired objective function by deducing and constructing a fitting relation, so the fitting model determines the accuracy of the algorithm. Typical algorithms include the maximum likelihood (ML) algorithm [90], the weighted subspace fitting (WSF) algorithm [91], and the multidimensional MUSIC (MD-MUSIC) algorithm [92]. The performance of this kind of algorithm is better, and the requirement of SNR is reduced to some extent, but the computational amount is large.

In 1995, Marcos S. et al. proposed the propagator method (PM) algorithm, which avoided the covariance matrix decomposition of the received signals of the array. Although compared with the MUSIC algorithm, the performance of partial resolution was lost, the computational amount was greatly reduced [93].

The development in recent years mainly has the following aspects. First, it is no longer satisfied with the use of second-order statistics (SOS) but with the use of high-order cumulants [94]. Common algorithms are usually based on fourth-order cumulants (FOCs) [95]. The second is compressed sensing (CS) [96], which breaks through the bandwidth limitation of traditional sampling theory and uses far less measured data than Nyquist sampling data, making superresolution signal processing possible. The theory of compressed sensing can greatly reduce the number of antenna elements and sampled data to reduce the amount of signal processing data. Therefore, the application of CS in direction estimation is increasingly widespread [97]. Deep learning is a development of machine learning. It has also been introduced into the research of direction finding technology. Deep neural networks have stronger representation and learning ability, so they can obtain better direction finding performance than traditional neural networks [98].

The principle of amplitude methods is simple and easy to implement, but the precision is not good. Since the interference to the satellite signals is often attached to the carrier frequency, the wavelength of the interference signals is short, and the phase information is obvious. Therefore, phase methods, such as interferometer direction finding, have higher measurement accuracy than the previous amplitude method. Based on the theory of array signal processing and spatial spectrum estimation, the direction finding of multiple interference sources can be realized simultaneously. This makes the array antennas and spatial spectrum estimation more widely used in practice. Spatial spectrum estimation methods have high measurement accuracy, breaking the Rayleigh limit, and have the ability of super resolution direction finding. However, the algorithm has a high SNR requirement and a large amount of calculation, so it is difficult to carry out real-time processing. Compared with the amplitude method, the phase method improves the sensitivity, but the number of interference sources that can be measured is limited. The spatial spectrum method further improves the sensitivity based on the phase method. It can measure multiple interference sources at the same time, but the number of interference sources cannot be more than the number of array elements.

From the received signal strength, a path loss model can be used to infer the distance between the monitoring point and the interference source [101]. Then, a system of equations can be established to find the location of the interference source through the observation data of multiple nodes. R S S i = P 0 − 10 n log 10 d i d 0 + σ i (19) where P 0 is the received power from the transmitter to the reference distance d 0 and P 0 and d 0 are known. n and σ i require actual measurements of the environment to be roughly determined. Small errors in RSS measurements can lead to large position errors. Even simple ground reflections can seriously affect accuracy. Even if the path model is optimized, its positioning accuracy is poor. One improvement of the RSS is the difference in the RSS (DRSS). D R S S i − j = R S S i − R S S j (20)

Thompson, R. J. R. et al. used an RSS method to locate RF interference sources in the GPS L1 band and considered the effect of ground refraction [102]. Wang, P. et al. proposed a simple and computationally efficient GNSS RFI localization technique based on DRSS measurements from crowdsourced devices. In this method, the weighted centroid of receiver position estimation is used as the RFI position estimation, which reduces the computational amount [103].

The TDOA evolved from the time of arrival (TOA). The principle of the TOA algorithm is similar to that of the GNSS. By measuring the arrival time of the signal transmission process from the unknown position point to the known position point, the distance from the unknown point to the known point is calculated. Then, enumerate the equation to solve the coordinates of the unknown point.

Let t i denote the observed time at the i th known point and t denote the observed time at the unknown point. τ is used to represent the time error caused by a series of processes such as modulation and demodulation. Sometimes, this term can also be expressed as white Gaussian noise. Then, the distance from the unknown point to the i known point is d i = c t i − t + c τ (21)

Let x i , y i , z i denote the coordinates of the known point and x , y , z denote the coordinates of the unknown point; then, the distance from the unknown point to the known point can also be expressed as d i = c t i − t + c τ = x i − x 2 + y i − y 2 + z i − z 2 (22)

The observation data of multiple nodes are used to establish the equation, and then the unknown position of the three-dimensional coordinates can be solved. It is worth noting that the TOA method used for interference source location is not applicable. The TOA algorithm requires highly accurate clock synchronization between the point to be measured and the known point. In the process of monitoring interference, whether intentional or unintentional interference, it is naturally impossible to expect the interference signal to cooperate, and there is no synchronization on the clock. The TOA algorithm is introduced as a passive positioning algorithm, which is generally used in indoor positioning [104].

Select a primary node x 0 , y 0 , z 0 . Its distance from the unknown point is d 0 . Then, the distance difference Δ d i between different nodes and the unknown point can be expressed as Δ d i = d 0 − d i = c t i − t 0 + c τ (23)

This is how the TDOA algorithm works. The TDOA algorithm calculates the distance difference between the unknown point and the known position point by measuring the difference between the arrival time of the signal transmitted by the unknown point and multiple known points. The TDOA algorithm still requires highly accurate clock synchronization between known measurement points, but unlike the TOA algorithm, the TDOA algorithm no longer requires the time information of unknown measurement points. This enables the TDOA algorithm to be used to locate interference sources. Common solutions of the TDOA algorithm are as follows: the Chan algorithm [105], Taylor series method [106], and weighted least squares method [107].

A common TDOA modification is the asynchronous time difference of arrival (A-TDOA). Díez-González J. et al. optimized the A-TDOA algorithm to allow higher refresh rates of localization signals and provide higher accuracy in the A-TDOA architecture [108]. Another improved method is the sparse time difference of arrival (S-TDOA). Uysal C. used sparse sensor data to estimate the TDOA, which reduced the computational amount. This method is more suitable for low SNR signals [109]. The particle swarm optimization (POS) algorithm is widely used to solve non-linear optimization problems, so it is often used to solve the TDOA equation [110].

The FDOA makes use of the Doppler effect. When a moving wave source approaches the receiver, the wave will be compressed, and the wavelength will become shorter, resulting in a higher frequency. Conversely, when the source moves away from the receiver, the wave is stretched, and its wavelength increases, causing the frequency to decrease. The motion between the wave source and the receiver refers to the relative motion, with the direction of its velocity pointing toward each other.

It is assumed that the wave source and the receiver are moving toward each other on the same horizontal line. Let f 0 denote the initial signal frequency emitted by the wave source and f ′ denote the signal frequency observed by the receiver. The moving speed of the wave is v , the moving speed of the wave source is v S and the moving speed of the receiver is v R . According to the Doppler effect, f ′ = v + v R v − v s f 0 (24)

Let v x i , v y i , v z i denote the velocity of the known position, and v x , v y , v z denote the velocity of the unknown position.

The unit direction vector can be expressed as e = 1 x i − x 2 + y i − y 2 + z i − z 2 x i − x , y i − y , z i − z (25) then v S i = v x , v y , v z ⋅ e (26) v R i = − v x i , v y i , v z i ⋅ e (27)

Then, the signal frequency received by the i th known position point can be expressed as f i = c + v R i c − v s i f 0 (28)

The velocity information of the unknown point can be solved by establishing the equation based on the observation data of multiple nodes. It is not enough to estimate only the velocity information of the interference source. The FDOA algorithm is often used together with the TDOA algorithm in interference source location.

Wang D. et al. proposed the use of a UAV with a known position to cooperate with the TDOA and FDOA algorithms for localization, which can achieve good results when locating multiple targets [111]. Li G. et al. proposed a virtualization approach consisting of the establishment of a virtual reference station and virtual frequency conversion to correct systematic errors in the system to address the problem of low positioning accuracy in low-orbit dual-satellite systems [112]. The FDOA curves and surfaces are very complicated in the near field, but in the far field, we showed that they simplify dramatically to easily understood curves and surfaces. Pine K. C. et al. analyzed the far-field situation and paid more attention to the study of the FDOA [113].

The AOA algorithm enumerates the equations by estimating the arrival direction of the signal and then determines the coordinates of unknown points. Using interference direction finding technology, the angle information of an interference source can be obtained. For a target with some known information, such as an interference source on the ground, only one angle of information is needed to solve the location of the interference source. This is because the surface of the Earth can be used to create equations. For the general case, as shown in Figure 6, the observed data of at least two nodes are needed to establish the equation.

Let θ i denote the horizontal angle at which the i th known position point receives the signal to be monitored, and φ i denote the pitch angle at which the i th known position point receives the signal to be monitored. Therefore, Δ x = d i ⁡ cos ⁡ φ i ⁡ sin ⁡ θ i Δ y = d i ⁡ cos ⁡ φ i ⁡ cos ⁡ θ i Δ z = d i ⁡ sin ⁡ φ i (29)

By solving the system of equations, the three-dimensional coordinates can be obtained. This is called the AOA method. The common way to solve the problem is to use geometric relations. The distance between two known position points is calculated and the sine law is used to determine d i and the three-dimensional coordinates.

The accuracy of the TDOA is proportional to the bandwidth of the received signal, and the accuracy of the AOA is not affected by it. In the case of a large signal bandwidth, TDOA measurements usually have higher acquisition accuracy than AOA measurements. Compared with the TDOA, the AOA localization performance is generally poor, so it is often combined with other algorithms. Sun Y. et al. proposed a new method called the constrained eigenspace (CES) to obtain AOA positioning solutions in modified polar representation (MPR) [114]. Yin J. et al. combined the AOA method and TDOA method and proposed a simple closed-form solution method, which only needed to use two observation stations [115]. Costa M. S. et al. combined the AOA method and RSS method and used the second-order cone programming (SOCP) relaxation technique to transform the non-convex estimator into a convex estimator. It has good performance and reduces computation [116]. Zuo P. et al. performed similar work based on the AOA method and RSS method [117].

From theory to simulation, interference source location technology often focuses on static interference sources. The purpose of interference source tracking is to track the interference source in motion. Based on the theory of interference source location, a Kalman filter algorithm is often involved. Biswas S. K. et al. combined the AOA and TDOA algorithms to compare the performance of various Kalman filtering algorithms for interference source tracking [118]. Biswas S. K. also proposed an interference source tracking algorithm based on the particle filter and compared it with the Kalman filter [119]. Qin N. et al. proposed a two-iteration interval extension method to determine the motion source using TDOA and FDOA information from multiple receivers [120]. Although interference tracking technology is only an extension of interference location technology, it still has room for development [121].

The use of measurements from multiple nodes for positioning dates back to the 1990s, partly because of satellite navigation technology. Among RSS methods, i.e., the TDOA method, FDOA method and AOA method, the TDOA method has the most mature technology, and the time information required by the TDOA method is also the easiest to obtain. In the field of satellite navigation, high-precision time alignment can be obtained through satellite signals, which provides support for positioning by the TDOA method. On the other hand, compared with the TDOA, the measurement accuracy of the FDOA has a greater impact on the positioning accuracy. This means that to ensure the accuracy of positioning, the measurement requirements for frequency measurement are more stringent than those for time measurement.

Interference tracking technology depends on Kalman filtering technology. The Kalman filter was developed in the 1960s. Using a Kalman filter, the estimation of the object motion trajectory is more accurate, and it is widely used in tracking moving objects. Compared with the classical algorithm of the 1940s, the Wiener filter and Kalman filter have been extended and achieved better performance, but the amount of computation is increased.

Interference detection technology is divided into blanket interference detection technology and spoofing interference detection technology. The development of blanket interference detection technology is relatively complete. On the one hand, in many fields, such as communication and radar, there is similar RF interference, and the field of research is relatively rich. However, blanket interference forms are relatively simple, and the interference signal can be detected by energy detection in the time or frequency domain, which is easy to realize. The research on spoofing interference is different. Because of the characteristics of satellite navigation systems and their signals, spoofing interference in satellite navigation systems is specific. Some traditional methods have poor performance in detecting spoofing interference in the navigation field. In detection technology, there are several development directions. The first is the detection of weaker signals. Although the interference signals are often strong when subjected to human interference, to monitor the changes in the electromagnetic environment and achieve the function of early warning, it is necessary to have a certain monitoring ability for weak interference signals. The second is the interference of multiple interference sources, which is also common in actual scenarios. Therefore, the monitoring system must be able to detect multiple interference sources. Third, the monitoring system is faced with time-varying interference, which requires low computational complexity, fast computing speed, real-time performance and adaptive ability of interference detection and recognition equipment. Fourth, blanket interference and spoofing interference can exist at the same time, which requires the equipment to have the ability to recognize compound interference. It is common for both blanket interference and spoofing interference to exist in a real environment, while current theoretical research and simulation generally only consider one of them.

Interference recognition technology is mainly blind recognition. The frequency, amplitude and phase of the interference signal are not known before the received signal is processed. Therefore, feature parameters with a high separation degree should be extracted before interference signal identification. In the field of satellite navigation, there is little research on interference recognition technology. In the past, interference recognition technology was rarely analyzed alone. When an interference signal is detected, the relevant characteristics of the interference signal can be roughly obtained. With the increasing complexity of the electromagnetic environment and the increase in artificial interference means, it becomes important to distinguish the types of interference. With the development of machine learning, pattern recognition, deep learning and other computer fields, interference recognition technology has also been developed to a certain extent.

The principle of amplitude methods is simple and easy to implement, but the precision is not good. Since the interference signals to the satellite signals are often attached to the carrier frequency, the wavelength of the interference signals is short, and the phase information is obvious. Therefore, phase methods, such as interferometer direction finding, have higher measurement accuracy than the previous amplitude method. Based on the theory of array signal processing and spatial spectrum estimation, the direction finding of multiple interference sources can be realized simultaneously. This makes the array antennas and spatial spectrum estimation more widely used in practice. Spatial spectrum estimation methods have high measurement accuracy, breaking the Rayleigh limit, and have the ability of super resolution direction finding. However, the algorithm has a high SNR requirement and a large amount of calculation, so it is difficult to carry out real-time processing. The main development of direction finding technology is the spatial spectrum estimation algorithm. First, spatial spectrum estimation algorithms usually require a large amount of computation, so GPU and other technologies can be used to improve the computational speed of spatial spectrum estimation algorithms. The second is compressed sensing, which can greatly reduce the number of antenna elements and sampled data. The third is to combine with cutting-edge fields such as machine learning.

The first development direction of interference source location technology is the joint algorithm, which can improve the accuracy of positioning and reduce the difficulty and cost of measurement, such as TDOA and FDOA common joint algorithms. Other forms of joint algorithms are generally feasible, such as the TDOA and AOA. The second is higher real-time performance. The interference source location algorithm also pursues higher computation speed. The third is the location of multiple interference sources. It is very common to face multiple interference sources in practice, and it is difficult to determine their locations.

Interference tracking technology is one of the possible development directions in the future. It is worth trying to improve the real-time performance of the interference location algorithm or improve the performance of a Kalman filter.