Analysis of Interference Magnetic Field Characteristics of Underwater Gliders

Abstract

Underwater gliders are a new type of unmanned underwater vehicle, characterized by high energy efficiency, long endurance, and low operational costs. They hold broad application prospects in fields such as ocean exploration, resource surveying, maritime surveillance, and military defense. This paper takes underwater gliders as the research subject, analyzing the characteristics of magnetic interference signals under different operational conditions. The study found that during full operational states, the motor’s operation generates interference signals at 17 Hz; during attitude adjustment, the movement of the moving block generates significant interference magnetic fields, especially during the forward and backward motion of the block, where interference signals at 20 Hz are particularly pronounced. To meet the objective of equipping underwater gliders with magnetic field sensors for underwater target detection, this paper proposes an adaptive filtering method based on the Recursive Least Squares (RLS) algorithm. The experimental results indicate that after filtering with the RLS algorithm, the amplitude of the noise signal has been reduced by over 60%, and it can effectively eliminate the noise components at 17 Hz and 20 Hz caused by the glider’s motor. This algorithm achieves an average increase in the signal-to-noise ratio (SNR) of 12 dB, which is equivalent to an approximately 80% improvement in accuracy. It significantly enhances the stability and signal-to-noise ratio of the magnetic field signals of underwater targets. This provides a feasible solution for equipping underwater gliders with magnetic field sensors for underwater target detection, holding important practical engineering significance.

1. Introduction

Underwater gliders are a new type of autonomous underwater vehicle that achieves underwater navigation by modulating its buoyancy and generating lift through its wings. This mode of operation endows underwater gliders with several advantages, including high energy efficiency, extended endurance, and low operational costs [1]. Moreover, hybrid-powered underwater gliders can combine their buoyancy-driven propulsion with a propeller system to achieve versatile and variable motion trajectories [2]. As a result, underwater gliders are widely used in oceanographic surveys, meteorological data collection, and CTD (Conductivity, Temperature, and Depth) measurements in typhoon-affected areas [3]. In terms of design, a biomimetic approach was proposed in [4], which optimized the hydrodynamic performance of the glider based on the leatherback turtle. This design successfully increased the lift coefficient and reduced the drag coefficient at low angles of attack, thereby enhancing the performance of the underwater glider. For positioning and tracking, underwater gliders typically rely on complex underwater acoustic systems or surface support vessels. While this reliance somewhat limits their application scope, it also endows them with a high degree of stealth during mission execution [5]. Its potential applications are vast, including ocean exploration, classification of large marine mammals, maritime surveillance, reconnaissance missions, and maritime defense [6]. With significant military and civilian value, it is increasingly playing a crucial role in underwater mining, hydrological data collection, and other fields. Moreover, it has become an essential tool for major military nations to enhance their underwater combat capabilities and gain strategic advantages in maritime warfare.

Currently, nations worldwide are placing significant emphasis on the development of marine resources, with increasing interest in their exploration and surveying. Marine magnetic exploration has emerged as a crucial component of marine science and technology [7]. Building on the principles of marine magnetic surveying and acoustic positioning, reference [8] introduces a novel approach that integrates acoustic technology with marine magnetic measurement. This combined method has enhanced the effectiveness of marine magnetic prospecting by approximately 33% compared to previous methods, proving particularly practical for the exploration of submarine pipelines. In the field of geophysical exploration, the detection of marine magnetic anomalies using diverse methodologies is increasingly becoming the mainstream approach [9], and the use of mobile equipment for extensive data acquisition has been demonstrated to be an effective method [10]. In the realm of ocean magnetic field measurement, underwater gliders have proven to be an effective detection platform, capable of conducting long-term, large-scale measurements of oceanic physical and geophysical parameters. Owing to the compact structure of underwater gliders, the sensors they carry must be lightweight. Given the continuous motion of these gliders, the acceleration response resulting from structural vibrations and changes in velocity presents a significant source of noise.

For the measurement of weak magnetic signals underwater, flux gate sensors are typically employed. Dalian University of Technology has developed a new type of fluxgate sensor by integrating magnetization remanence time and neural networks, which has reduced the linear error by more than 40 times [11]. In reference [12], based on SIP multidimensional integration technology, a U-shaped micro three-component fluxgate sensor is realized. A closed-loop feedback structure of a double-core self-excited fluxgate zero-flux current sensor is proposed, and its working principle is analyzed [13]. Meanwhile, reference [14] studies the manufacturing process of micro fluxgate sensors and introduces MEMS and semiconductor processes to improve performance. From the above analysis, it can be seen that equipping fluxgate sensors on underwater mobile platforms will become an important plan for marine magnetic exploration in the future, while currently there is no research on the assembly of electromagnetic sensors on wave gliders. For unstable underwater platforms such as gliders, studying their electromagnetic field characteristics holds practical engineering significance for advancing underwater electromagnetic field detection.

This manuscript focuses on underwater gliders as the primary research subject, with underwater magnetic field detection serving as the research context. By measuring the magnetic interference signals generated by underwater gliders under various operational conditions and analyzing the characteristics of these interference signals, adaptive filtering is applied to the glider’s interference signals in conjunction with the properties of underwater target magnetic signals. This approach facilitates the integration of magnetic field sensors into underwater gliders for the detection of underwater magnetic signals.

2. Structure and Motion Mechanism of Underwater Glider

The underwater glider comprises a streamlined shell and an internal cabin, which is partitioned into three distinct sections: the front, middle, and rear sections. The front end features a diffuser, while the rear section of the cabin is equipped with wings and tail fins. The wings are typically constructed from non-metallic materials and are designed to be foldable, whereas the tail fins are fabricated from the same material as the main body. The tail fairing is positioned ahead of the thruster, and the tail communication antenna facilitates wireless communication when the glider surfaces. The overall structure is illustrated in Figure 1.

2.1. Motion Mechanism of Underwater Glider

During the movement of an underwater glider, internal moving blocks are actuated by a motor to shift the glider’s center of mass, thereby generating attitude angles such as pitch and roll. Additionally, the tail section of the glider’s cabin houses a variable-volume oil bladder, which functions as a buoyancy control mechanism. By pumping oil in and out of the bladder, its volume can be altered to regulate the buoyancy of the glider, thereby changing its buoyancy in water. The interplay of buoyancy, attitude angles, and lift results in a sawtooth-like motion trajectory, as depicted in Figure 2.

Gliders typically employ two primary turning mechanisms: (1) generating a roll angle to alter the direction of lift, thereby creating a lateral force that induces rotation of the body; and (2) rotating by manipulating the tail rudder to generate lateral force. The first method of turning is influenced by various hydrodynamic factors, such as lift, resulting in a larger turning radius; the second method offers a smaller turning radius but requires a well-sealed servo to function effectively.

Underwater gliders have effectively surmounted the energy supply limitations inherent in small motion platforms, offering several notable advantages: extended endurance, vast sailing distances, reduced maintenance costs, adaptability to severe maritime conditions, straightforward deployment and retrieval processes, independence from chemical fuels, and a minimal radar cross-section. In recent years, they have garnered considerable research and application value across both military and civilian sectors. For civilian applications, these gliders serve as innovative mobile observation platforms, capable of integrating a diverse array of sensors to facilitate comprehensive studies of the marine environment, resource surveys, seabed mineral exploration, and marine biology. In the realm of military defense, underwater gliders are employed for critical missions such as communication relays and target reconnaissance.

2.2. Analysis of Sources of Magnetic Field Interference Noise

Although gliders lack propellers, the mechanical structures that adjust buoyancy and center of gravity still generate some self-noise. The self-noise of the platform primarily originates from three adjustment motors: the buoyancy adjustment motor, the center of gravity adjustment motor, and the tail rudder adjustment motor. The buoyancy and center of gravity adjustment mechanism comprises an oil bladder and two slider mechanisms. When the platform dives, the slider moves forward, shifting the center of mass forward, and oil from the oil bladder flows into the storage tank, reducing buoyancy. After reaching the specified depth, the sliding block moves backward, shifting the center of mass backward, and oil flows from the storage tank to the oil bladder, increasing buoyancy, causing the platform to ascend. According to the gliding mechanism, gliders need to adjust these three major motors to maintain their posture during movement. The motors required for adjusting different postures are located in different compartments, resulting in varying levels of noise.

To date, a variety of detection sensors have been successfully integrated into gliders, including temperature and salinity depth sensors, turbidity meters, ocean current meters, hydrophones, dissolved oxygen sensors, chlorophyll fluorometers, and optical backscatter meters [15,16]. However, research on the self-noise of gliders within the academic community remains relatively limited, with a primary focus on the acoustic field noise of underwater gliders equipped with sound sensors. In 2015, the Ocean University of China utilized an autonomous underwater glider for turbulence observations. The experimental data revealed that the vibration of the Autonomous Underwater Glider (AUG) platform primarily originated from the oil pump, airbag pump, ballast pump, and moving battery pack [17]. In 2017, the University of the Chinese Academy of Sciences, through analysis of data from tests conducted in the South China Sea, concluded that mechanical noise was the predominant source of self-noise, and that motor operation also generated significant noise [18]. In 2018, Cheng Jiang et al. [19] conducted two sea trials using the “Haiyan II” gliders, and their findings indicated that the self-noise of the glider was primarily attributed to two motors. Data from these two sea trials showed that the proportion of time without motor operation exceeded 50%, suggesting that the self-noise of the glider can be considered negligible during periods of motor inactivity.

Underwater gliders possess robust and versatile sensor mounting capabilities. For the detection of underwater electromagnetic targets, their structural and motion characteristics indicate that magnetic field sensors should be installed within the glider’s internal cabin. Considering the inherent characteristics of magnetic field sensors, they are susceptible to three types of magnetic field interference: firstly, the glider’s own static magnetic field; secondly, the internal motors of the glider. During the glider’s movement, continuous attitude adjustments are necessary, and the operation of the three internal motors generates magnetic field interference; thirdly, as moving bodies, gliders cut through magnetic field lines and induce magnetic fields during their motion. The oil bladder motor and solenoid valve operate only at the water surface or seabed and remain inactive during the intermediate process, thus having minimal impact; however, the slider is in constant motion, and attention must be paid to the magnetic fields generated when the slider moves forward, backward, left, and right. The movement of the slider is also motor-driven.

Based on actual sea trial data, this manuscript examines the time-frequency characteristics of various motors under different operational states. By selecting targeted time-frequency analysis results for distinct positions, the self-noise characteristics of gliders are identified, which include pulse noise interference, line spectrum noise interference, and continuous spectrum noise interference. This analysis provides a foundation for the subsequent research on noise interference suppression methods presented in this study.

3. Analysis of Magnetic Field Noise Characteristics

Given the intricate variations in the attitude angle of the glider’s buoy, establishing an accurate correlation model between the induced electric field and the attitude angle is difficult. Therefore, this study analyzes the magnetic field characteristics of the glider using measured data. In the experimental setup, the “Haiyan” glider was utilized as the subject, and measurements were conducted in a low magnetic field environment. The magnetic field sensor was strategically placed beneath the glider (as shown in Figure 3), and the electromagnetic noise was assessed under different electrode conditions by varying the glider’s motion state.

Analysis of Magnetic Field Characteristics of Gliders Under Different Operating Conditions

(a) The magnetic field sensor is positioned directly beneath the glider’s midpoint.

Position the magnetic field sensor directly below the glider’s center, maintaining a horizontal distance of 10 cm from the glider, and proceed to measure the magnetic field interference as follows:

Analysis from Figure 4, Figure 5, Figure 6, Figure 7, Figure 8, Figure 9 and Figure 10 reveals that the glider induces significant variations in the magnetic field under various motion states. (1) The operation of the oil bag mechanism results in alterations to the static magnetic field; (2) during attitude adjustments, the movement of the sliding block also generates substantial magnetic field changes. The left and right movements of the sliding block can produce a gradual magnetic field change in up to 200 nT, with distinct trends in magnetic field variations when the block moves left versus right, yet the power spectrum change patterns are largely consistent. (3) The magnetic field generated by the forward and backward movements of the sliding block is relatively large, exhibiting a sine interference with frequency peaks at 0.23 Hz and 20 Hz, which is markedly different from the left and right movements of the block.

(b) The magnetic field sensor is placed directly below the head of the glider

Place the magnetic field sensor directly beneath the glider’s head, maintaining a horizontal distance of 10 cm from the glider, and proceed to measure the magnetic field interference as follows, the measurement results are shown in Figure 11, Figure 12, Figure 13, Figure 14, Figure 15, Figure 16 and Figure 17:

From the analysis above, it can be observed that when the sensor is placed at the glider’s head, the magnetic field variation patterns under different motion states are largely consistent, exhibiting sinusoidal interference with a subtle disturbance signal at 17 Hz. However, when the moving block performs forward and backward actions, it generates a more significant magnetic field with frequency peaks at 17 Hz and 20 Hz, which is markedly different from the block’s left and right movements.

Overall, the following conclusions have been drawn.

(a)

While the glider is in operation, it induces significant alterations in the magnetic field, with the maximum amplitude reaching approximately 200 nT;

(b)

During the operation of the oil bag mechanism, a pronounced interference signal at 5 Hz is generated, yet the amplitude of this interference signal exhibits relatively minimal variation;

(c)

When the glider’s electrode is functioning, it produces a subtle interference signal at 17 Hz;

(d)

For the motion block utilized to adjust the glider’s attitude, the amplitude of the interference signal resulting from the block’s lateral movements (left and right) is comparatively low, predominantly confined within the 5 Hz frequency range. Conversely, when the motion block moves longitudinally (forward and backward), it generates an interference signal at 20 Hz.

4. Magnetic Field Noise Suppression Method Based on Target Features

The ship’s magnetic field serves as a crucial signal source for underwater weaponry and is also a primary target for ship degaussing and magnetic protection. According to the principles of the ship’s physical field, the intensity of the ship’s magnetic field diminishes with increasing distance following an approximate dipole attenuation law (which can be roughly considered as 1/r³). Based on this, it can be inferred that for medium-sized vessels with a beam width of 12–15 m, operating at a water depth of 50–60 m, the magnetic field intensity is approximately 20–300 nT.

Based on the principles of physical fields, a ship’s magnetic field is a composite of multiple magnetic field sources, encompassing the ship’s inherent magnetic field, the propeller shaft frequency magnetic field, the magnetic field from mechanical equipment vibrations, and the magnetic field radiated by electrical equipment. These magnetic field signals from various sources can overlap and interfere with each other in frequency, resulting in a more complex frequency distribution for the ship’s magnetic field. The magnetic field of a ship falls within an extremely low frequency range, typically spanning from 0 to 1 Hz. For large- and medium-sized ships sailing below their economic speed, the upper frequency limit is generally around 0.1 Hz. In contrast, when small speedboats travel at high speeds, their upper frequency limit usually does not exceed 1 Hz. The peak frequency of the magnetic field signal for large- and medium-sized ships typically occurs around 0.025 Hz. Considering the magnetic field generated by the internal mechanisms of a glider, applying a 1 Hz low-pass filter to the interference signal can effectively filter out the interference; the most prominent low-frequency component in a ship’s magnetic field is usually the shaft frequency signal produced by the propeller’s rotation. The frequency of this shaft frequency signal is directly proportional to the propeller’s speed, typically ranging from 0.5 Hz to 30 Hz.

Scholar Sun Qiang et al. [20] proposed an improved detection method based on adaptive threshold line spectrum energy summation to address the issue of electric field signal detection on buoy platforms. This method has been validated through both simulation and real-measured experiments, demonstrating its effectiveness in detecting target electric field signals while significantly reducing the false alarm rate while maintaining the probability of target detection. In addition, Scholar Yu Peng et al. [21] developed an adaptive filtering algorithm for sway-related electric field noise suppression on underwater floating platforms, using attitude angles as reference variables. The effectiveness of this algorithm was verified using real-measured environmental electric field data, and the results showed that it can effectively suppress sway-related electric field noise. These studies provide valuable references for electric field signal detection and noise suppression in underwater environments. During the operation of underwater gliders, interference signals from the internal structure can impact the detection of underwater magnetic field signals. By leveraging the characteristics of the ship’s magnetic field signals, adaptive filtering techniques can be effectively employed to suppress the associated magnetic field noise.

During the complex operation of underwater gliders, the interference of magnetic field signals is intricate. As can be seen from the real-time data analysis in Section 3, the glider exhibits unique and complex characteristics of interference signals under different operating conditions:

(a)

The interference signals generated by various parts of the glider are not fixed; different directional movements of the moving blocks produce interference signals of different frequencies, with varying amplitudes. The RLS (Recursive Least Squares) filtering can continuously adjust the filter coefficients based on real-time changes in the signal, adaptively tracking the dynamic variations in the signal. In contrast, some fixed-coefficient filtering methods (such as simple low-pass and high-pass filters) struggle to cope with these time-varying characteristics.

(b)

When the glider operates, the frequency distribution of interference signals is broad, such as the oil bag generates a 5 Hz interference signal, the electrodes produce a 17 Hz interference signal, and the forward and backward movements of the moving block generates a 20 Hz interference signal, etc. RLS filtering can effectively suppress multiple interference signals of different frequencies through adaptive adjustment.

(c)

The amplitude of the magnetic interference signals also has different characteristics; for instance, the amplitude of the oil bag interference signal changes little, while the amplitude of the interference signal from the moving block’s movement also varies to varying degrees. RLS filtering can adjust the filtering gain according to changes in signal amplitude, maintaining good filtering effects even when the amplitude of the interference signal changes. In summary, due to the time-varying, multi-frequency, and amplitude-varying characteristics of the interference signals generated during the glider’s operation, along with the presence of weak interference signals, RLS filtering has a distinct advantage over other filtering methods in dealing with these interference signals and is therefore more suitable for filtering the relevant signals of the glider.

4.1. Basic Principles of Adaptive Filtering

Recursive Least Squares (RLS) is a classic adaptive filtering algorithm. Its core idea is to recursively update the weights of the filter in order to minimize prediction errors.

When applying the Recursive Least Squares (RLS) algorithm to address the magnetic field signal interference generated by underwater gliders, it is assumed that the motion of the internal mechanisms of the underwater glider can be modeled as a discrete-time linear time-invariant (LTI) system. The state-space model of this system can be represented as follows:

$$\begin{array}{l}x(k+1)=Ax(k)+Bu(k)+w(k)\\ y(k)=Cx(k)+v(k)\end{array}$$

(1)

$x(k)$ represents the state vector at time k, $u(k)$ denotes the control input, $y(k)$ signifies the observable output, and $w(k)$ and $v(k)$ represent the process and observation noises, respectively, which are typically assumed to be Gaussian white noise. A, B, and C are the state transition, control input, and observation matrices.

The purpose of the RLS algorithm is to find the optimal weight vector w that minimizes the sum of the squares of the prediction errors. For the polynomial model, the output $d(k)$ can be represented as a linear combination of the input $x(k)$ and a set of unknown coefficients w.

$$d(k)=wTx(k)+v(k)$$

(2)

$d(k)$ represents the desired signal, w is the filter weight vector, $x(k)$ is the input signal vector, and $v(k)$ is the observation noise, which is typically assumed to be zero-mean Gaussian white noise. The goal of the RLS algorithm is to find the weight vector w that minimizes the cost function $J(k)$:

$$J(k)={\sum }_{i=1}^k{\mathsf{\lambda }}^{k-i}{\parallel e(i)\parallel }^2$$

(3)

In this context, $e(i)$ represents the error signal, and $\mathsf{\lambda }$ is the forgetting factor (0 < $\mathsf{\lambda }$ ≤ 1), which is used to adjust the degree of memory of past data.

The RLS algorithm updates the weight vector $w(k)$ by minimizing the weighted sum of the squared prediction errors. The recursive equations of the algorithm are as follows:

$$w(k)=w(k-1)+K(k)e\backslash *(k)$$

(4)

In the equation, $w(k)$ is the gain vector, and $e\backslash *(k)$ is the conjugate of the error signal. The computation of the gain vector relies on the inverse of the autocorrelation matrix of the input signal, which determines the extent to which new data contribute to the weight update. The gain vector $w(k)$ is calculated using the following formula:

$$K(k)={\displaystyle \frac{{\mathsf{\lambda }}^{-1}P(k-1)x(k)}{1+{\mathsf{\lambda }}^{-1}xH(k)P(k-1)x(k)}}$$

(5)

$P(k)$ represents the recursive error covariance matrix, which is updated at each iteration as follows:

$$P(k)={\mathsf{\lambda }}^{-1}[P(k-1)-1{\displaystyle \frac{K(k)x^H(k)P(k-1)}{+{\mathsf{\lambda }}^{-1}{x}^H(k)P(k-1)x(k)}}]$$

(6)

By updating the recursive error covariance matrix, the RLS algorithm can effectively estimate new weight vectors at each iteration. This process ensures that the algorithm can quickly respond to changes in the input signal and adapt to a constantly varying environment.

The core goal of the RLS algorithm is to find an optimal weight vector W that minimizes the sum of squared prediction errors. The prediction error $e(m)$ is defined as the difference between the actual output $d(m)$ and the filter output $y(m)$, i.e., $e(m)=d(m)-y(m)$. By continuously adjusting the weight vector, the RLS algorithm can adapt to changes in the signal in real time, achieving effective noise suppression and precise signal extraction.

RLS is mainly achieved through the following process:

(1)

Prediction Value Calculation:

$$y(m)={\mathbf{W}}^T{\mathbf{R}}_m$$

(7)

Among them, ${\mathbf{R}}_m$ is the reference input vector, which contains signal samples from the current time and several previous times; $\mathbf{W}$ is a weight vector that determines the contribution of each reference signal sample in the prediction.

(2)

Error Calculation:

$$e(m)=d(m)-y(m)$$

(8)

The error $e(m)$ reflects the deviation between the predicted value and the actual value at the current time, which serves as the basis for the RLS algorithm to optimize the weight vector.

(3)

Gain Vector Calculation:

$$\mathbf{k}(m)={\displaystyle \frac{\mathbf{P}(m-1){\mathbf{R}}_m}{\mathsf{\lambda }+{\mathbf{R}}_m^T\mathbf{P}(m-1){\mathbf{R}}_m}}$$

(9)

The gain vector $\mathbf{k}(m)$ determines the magnitude and direction of the weight vector update. Here, $\mathbf{P}(m-1)$ is the error covariance matrix from the previous time step, which is used to measure the uncertainty of the prediction error; λ is the forgetting factor, which controls the algorithm’s emphasis on new data.

(4)

Weight Update:

$$\mathbf{W}(m)=\mathbf{W}(m+1)+\mathbf{k}(m)e(m)$$

(10)

The weight vector $\mathbf{W}(m)$ is updated based on the gain vector and the error to reduce the prediction error and improve the filtering accuracy.

(5)

Error Covariance Matrix Update:

$$\mathbf{P}(m)={\displaystyle \frac{\mathbf{P}(m-1)-\mathbf{k}(m){\mathbf{R}}_m^T\mathbf{P}(m-1)}{\mathsf{\lambda }}}$$

(11)

The update of the error covariance matrix $\mathbf{P}(m)$ reflects the algorithm’s re-evaluation of the uncertainty of the prediction error, providing a basis for the next weight update.

The weight vector $\mathbf{W}(0)$ is typically initialized as a zero vector, indicating that there is no prior knowledge of the signal prediction at the initial moment. The error covariance matrix $\mathbf{P}(0)$ is initialized as $\mathsf{\delta }\mathbf{I}$, where $\mathsf{\delta }$ is a large positive number and $\mathbf{I}$ is the identity matrix, indicating a high degree of uncertainty in the prediction error at the initial moment. For each time step $m$, the predicted value $y(m)$ is first calculated based on the current weight vector and the reference input vector.

Then, the prediction error $e(m)$ is calculated, and the weight vector $\mathbf{W}(m)$ is updated based on the error and the gain vector. Finally, the error covariance matrix $\mathbf{P}(m)$ is updated, preparing for the prediction and weight update at the next moment.

4.2. Empirical Data Validation

During the filtering process, the forgetting factor λ controls the algorithm’s emphasis on new data. A larger value of λ makes the algorithm more stable, but may slow down the convergence speed. Considering the small amount of signal data, λ is set to 0.98. The initial diagonal element value δ is used to initialize the error covariance matrix $\mathbf{P}(0)$, affecting the initial state of the algorithm. A larger $\mathsf{\delta }$ value indicates a greater uncertainty in the initial prediction error, allowing the algorithm to more flexibly adjust the weights in the initial stage; thus, $\mathsf{\delta }$ is set to 100. The filter order L determines the complexity and memory length of the filter. Since the signal is relatively stable, a lower value of L can be used.

Based on the analysis of magnetic field noise in various operational states of underwater gliders, it has been observed that significant noise with large amplitude and considerable power is generated when the moving block undergoes forward and backward motion. Consequently, data from this particular operational condition have been selected for validation purposes.

The time-domain and power spectral density diagrams after adaptive filtering are shown in Figure 18 and Figure 19. As can be seen from Figure 18a and Figure 19a, the magnetic field components have been significantly improved in the time-domain diagram after adaptive filtering. However, the filtered signal still exhibits considerable fluctuations in the first 5 s (more pronounced when the moving block moves backward). This is considered to be due to the lack of prior knowledge for signal prediction at the initial moment, leading to a high degree of uncertainty in the prediction error at the beginning.

As shown in Figure 18b and Figure 19b, after adaptive filtering, the three components of the magnetic field exhibit reduced peak noise in the frequency domain, resulting in a more stable measured magnetic field with a significant decrease in power spectral density. The RLS algorithm is capable of filtering out the 17 Hz and 20 Hz noise components caused by the underwater glider’s motor.

Figure 20 and Figure 21 depict the signal diagrams after adaptive filtering when the magnetic field sensor is placed at the head of the glider. In Figure 20a and Figure 21a, the three components of the magnetic field show significant improvement in the time-domain diagram after adaptive filtering, with a substantial reduction in signal fluctuation. Unlike when the sensor is placed in the middle, the filtered signal exhibits minimal fluctuation in the first 5 s. This is attributed to the reference signal for filtering being By, which has minimal fluctuation, thereby also reducing the related signals in the filtering process.

From Figure 21b, it is evident that the power spectral density has been significantly reduced after filtering. In the frequency-domain diagram, a decrease in peak noise is observable, resulting in a more stable measured magnetic field. The filtering process has effectively removed some of the 17 Hz and 20 Hz components caused by the electrodes.

5. Conclusions

When the underwater glider is in operation, magnetic field interference mainly originates from two aspects. (1) Interference characteristics of the internal structure and motor: The internal structure of the vehicle generates magnetic field interference, and the motor produces a 17 Hz interference signal when operating at full capacity. (2) Interference during attitude adjustment: During the attitude adjustment process, the movement of the moving block generates a larger amplitude of interference magnetic field, especially when the block moves back and forth; the amplitude of the interference magnetic field is larger, and the 20 Hz interference signal is particularly evident.

To meet the objective of equipping underwater gliders with magnetic field sensors for underwater target detection, adaptive filtering is used to suppress the glider’s noise. The results show that (1) the RLS algorithm can significantly reduce the noise of the underwater glider under any working conditions and that (2) the RLS algorithm can effectively suppress the interference noise at the significant spectral components of 17 Hz and 20 Hz. By effectively suppressing the interference signals, the feasibility of installing magnetic field sensors on underwater gliders is demonstrated, providing a new idea for underwater target detection and underwater confrontation, and holding significant engineering practicality for the detection of underwater target magnetic field signals.