Unmanned aerial vehicle real-time positioning and direction finding tracking method for complex electromagnetic environment

Abstract

The present application belongs to the field of radio positioning and navigation technology, and is used for solving the problems that the prior art lacks the discrimination mechanism for whether the data of each channel is originated from the same physical radiation event before the data frame is encapsulated, and also lacks the flexible connection strategy for continuously quantitatively evaluating the frame quality and adjusting the tracking filter accordingly, and specifically is a real-time positioning and direction finding tracking method for unmanned aerial vehicle in complex electromagnetic environment, comprising: acquiring a multi-channel digital baseband signal sampling sequence; reading the health degree coefficient output by the last frame, and dynamically determining the masking window and trigger sensitivity of the current frame according to the coefficient; detecting the first sampling time point which simultaneously satisfies the trigger sensitivity condition and the third characteristic trend condition in the remaining sequence as an effective arrival fingerprint; through the joint extraction and step-by-step constraint of the first characteristic, the second characteristic and the third characteristic, the present application can distinguish the rising edge of the monotonous aggregation of signal energy from the recovery period of the energy oscillation after the interference overload in time.

Description

Technical Field

[0001] This invention belongs to the field of radio positioning and navigation technology, specifically a real-time positioning and direction finding tracking method for unmanned aerial vehicles (UAVs) in complex electromagnetic environments. Background Technology

[0002] Multi-channel array direction finding and positioning technology is one of the core means for UAVs to achieve autonomous navigation and target tracking in complex electromagnetic environments. This type of method usually uses multiple antenna array elements distributed in different locations in space to simultaneously receive the target's radiated signals. By comparing the amplitude differences, phase differences, or arrival time differences between the signals of each channel, the azimuth and pitch angles of the target relative to the UAV are inverted, and then combined with the platform's motion information to complete continuous tracking. To ensure direction finding accuracy, existing systems usually perform factory consistency calibration on the receiving links of each channel and perform preprocessing operations such as time synchronization and gain normalization before data processing.

[0003] In actual operations, the electromagnetic environment faced by UAVs often contains multiple non-cooperative radiation sources, including high-power radar pulses, communication harmonic interference, and multipath reflection signals. When such strong pulse signals enter the receiver front end, the low-noise amplifier and automatic gain control loop may enter a saturation or overload state, followed by a continuous recovery process. During this recovery period, residual oscillation components with decreasing amplitude and non-stationary characteristics are superimposed on the channel output signal. Due to individual differences, aging levels, and temperature characteristics of the front-end devices in each channel, the recovery period length and oscillation mode induced by the same interference pulse are not consistent in each channel. Existing technologies, when processing continuously sampled data, typically extract multi-channel data into frames using a fixed-length time window and align the frames according to a globally unified timestamp or synchronization pulse, assuming that the data in each channel within the frame comes from the same external radiation event. However, when strong interference and target signals are temporally adjacent or interleaved, the data buffers of different channels may be affected by different physical triggering conditions. The data acquisition process is as follows: some channels actually contain residual components from the interference recovery period, while other channels have recovered and mainly carry the target signal components. The backend data processing flow cannot detect this inconsistency in the sources of different channels within a frame, and still treats all channel data as observations from the same source and sends them to the direction finding solution stage. Under these conditions, direction finding algorithms based on phase difference or time difference of arrival may converge to a false azimuth angle that compromises between the real interference direction and the target direction. This false azimuth angle may remain stable for multiple consecutive frames, thus continuously misleading the tracking filter. Even if the system can identify abnormal data quality and discard it, if the interference lasts for a long time, continuously discarding observation frames will cause the tracking filter's state update to rely on extrapolation prediction for a long time. As time accumulates, the prediction covariance matrix expands rapidly. When the real observations recover after the interference ends, the filter may not be able to converge in time due to gain misalignment, which manifests as drift or oscillation in the positioning trajectory.

[0004] It is evident that existing technologies lack a mechanism to determine whether data from each channel originates from the same physical radiation event before data frame encapsulation, and also lack a flexible connection strategy for continuously quantifying and evaluating frame quality and adjusting the tracking filter accordingly. This results in significant degradation of positioning and tracking performance in scenarios with strong interference. Areas for improvement include: how to establish a method for measuring the homogeneity of intra-frame data by utilizing the inherent response characteristics of each channel signal at the transient initial stage before the data frame enters the direction finding solution link, and how to progressively integrate the measurement result into the observation weighting and state recursion stages to smoothly handle the state switching between interference contamination and observation recovery. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments, which solves the problems of existing technologies that lack a mechanism to determine whether the data of each channel originates from the same physical radiation event before data frame encapsulation, and also lack a flexible connection strategy for continuously quantifying and evaluating frame quality and adjusting the tracking filter accordingly.

[0006] The technical problem to be solved by this invention is: how to provide a real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments that can determine whether the data of each channel originates from the same physical radiation event before data frame encapsulation.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] Real-time positioning and orientation tracking methods for UAVs in complex electromagnetic environments include:

[0009] Acquire multi-channel digital baseband signal sampling sequences;

[0010] The signal amplitude features are extracted based on the sampling sequence of each channel, and a first feature, a second feature and a third feature are generated accordingly. The first feature is used to characterize the degree of dispersion of signal energy within the local window, the second feature is used to characterize the instantaneous change in amplitude, and the third feature is used to characterize the changing trend of the first feature.

[0011] Read the health coefficient output from the previous frame, and dynamically determine the masking window and trigger sensitivity for the current frame based on this coefficient;

[0012] Skip the masking window based on the amplitude characteristics of each channel, and detect the first sampling moment in the remaining sequence that simultaneously satisfies the trigger sensitivity condition and the third feature trend condition, as the valid arrival fingerprint;

[0013] A dispersion index is generated based on the dispersion between each valid fingerprint, a consistency index is generated based on the third feature symbol at each fingerprint time, and the health coefficient of the current frame is generated by mapping the intermediate state of the two indices.

[0014] The observation noise basis is dynamically constructed based on the health coefficient and injected into the array covariance matrix to perform direction finding; when the health coefficient is lower than the threshold, the current direction finding result is discarded and the state covariance matrix is ​​dynamically expanded.

[0015] The present invention has the following beneficial effects:

[0016] 1. By jointly extracting and progressively constraining the first, second, and third features, it is possible to distinguish in time the rising edge of monotonically accumulating signal energy from the recovery period of energy oscillation after interference overload. When the value of the third feature is non-negative, it indicates that the energy change at that moment belongs to the monotonically accumulating process, which is consistent with the physical law of the arrival of the real signal wavefront; when its value is negative, it indicates that the moment is in the downward interval of energy oscillation, corresponding to the interference recovery period. The judgment condition composed of the three features makes the detection of the effective arrival fingerprint physically unique and specific, reducing the probability of misjudging the interference residue as the edge of the effective signal at the source.

[0017] 2. By dynamically adjusting the masking window width and trigger sensitivity based on the health coefficient of the previous frame, the detection parameters for effectively reaching fingerprints can adaptively track changes in the pollution level of the electromagnetic environment. When the health coefficient of the previous frame decreases, the masking window width automatically widens to skip the longer potential recovery period, and the trigger sensitivity automatically increases to avoid false triggers caused by residual jitter. When the health coefficient of the previous frame rises again, both parameters automatically shrink to maintain the sensitivity to capture weak target signal edges. This feedback mechanism enables the system to follow the time-varying characteristics of the electromagnetic environment without relying on external interference recognition equipment.

[0018] 3. By constructing positive interactive proof constraints through the temporal discreteness index of effective fingerprint arrival and the consistency index of third feature symbol, the frame quality assessment is extended from binary decision to a health coefficient with continuous values. By combining the diagonal regularization injection of observation noise basis multiplier with the saturation expansion processing of state covariance matrix, flexible degradation of direction finding and tracking links in polluted environments is realized at both the data domain and state domain levels. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of the method according to Embodiment 1 of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Multi-channel array direction finding and positioning technology is one of the core means for UAVs to achieve autonomous navigation and target tracking in complex electromagnetic environments. The basic physical principle of this method is that when antenna array elements distributed in different locations in space receive electromagnetic waves radiated by the same target, due to the difference in propagation path length, the phase or arrival time of the signal obtained by each array element has a definite difference relationship with the direction of the target. By solving the phase difference, arrival time difference or amplitude ratio between multi-channel signals, the spatial azimuth and pitch angles of the target relative to the UAV platform can be retrieved, and then continuous tracking can be completed by combining the platform attitude and position information. The premise for this physical relationship is that the signal samples entering the processing links of each channel at the same time do indeed come from the same radiation source and the same wavefront emitted at the same time. Existing array direction finding systems are designed with this premise as a default, and usually eliminate the amplitude and phase errors of each channel through factory consistency calibration, and perform global time synchronization at the data processing front end, aligning the data of each channel according to a unified frame structure before sending it into the direction finding solution link.

[0023] In actual operation, the electromagnetic environment faced by UAVs is far from ideal; non-cooperative radiation sources such as high-power radar pulses, communication harmonic interference, and multipath reflection signals are common. When such strong pulse signals enter the receiver front end, the low-noise amplifier may enter the saturation region, and the automatic gain control loop undergoes significant adjustments before gradually recovering. During the recovery period, residual oscillation components with decreasing amplitude and irregular shape are superimposed on the channel output signal. Due to individual differences, aging levels, and temperature characteristics of the front-end devices in each channel, the recovery period length and oscillation pattern induced by the same interference pulse are not consistent in different channels. More complex situations occur with strong interference. When the target signal is temporally adjacent to or overlaps with the target signal: some channels may be triggered by strong interference pulses to initiate the front-end response, and the target signal wavefront arrives before the recovery period is over. Other channels only begin to respond when the target signal wavefront is triggered. In this case, the physical triggering source of each channel in the multi-channel data captured by the system with a fixed frame structure is no longer singular. The sampling sequence of some channels carries residual information from the interference recovery period, while the other channels carry information about the leading edge of the target signal. The existing data processing flow cannot detect the inconsistent sources of each channel within this frame and still treats all channel data as observations from the same source and sends them to the direction finding algorithm.

[0024] Under these conditions, direction-finding algorithms based on phase difference or time difference of arrival may calculate a false azimuth angle that compromises between the true interference direction and the target direction. This false azimuth may remain stable over multiple consecutive frames—because the order in which each channel is triggered and the source mixing method during each frame capture are determined by the relative temporal relationship between the interference and target signals. If this temporal relationship is stable, the direction-finding output will also stably converge to the false azimuth. After receiving this series of stable false direction-finding results, the tracking filter will assign them a high confidence level and update the target state estimate accordingly, causing the positioning trajectory to gradually deviate from the true position. Even if the system can identify data quality anomalies and handle them by discarding frames, when the interference lasts for a long time, the discarding of multiple consecutive data frames means that the filter will not receive effective observation information for a long time. The state prediction covariance matrix will continue to expand over time. When the true observations recover after the interference ends, the filter gain may have been excessively amplified, resulting in a large jump in the state estimate or an excessively long reconvergence time.

[0025] The root cause of the aforementioned failure modes lies in the fact that existing technologies lack a step to determine and quantify whether the data from each channel within a frame originates from the same physical radiation event before the data frame enters the direction finding solution link. They also lack a flexible connection strategy that integrates the quantification assessment results into observation weighting and state recursion in a progressive manner. If a homology measurement mechanism for intra-frame data cannot be established in scenarios where interference and target signals coexist, the correctness of the direction finding solution cannot be guaranteed, thereby affecting the reliability of the entire positioning and tracking link.

[0026] Example 1: As Figure 1 As shown, a real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments includes:

[0027] Step S1: Obtain the multi-channel digital baseband signal sampling sequence; extract the signal amplitude features according to the sampling sequence of each channel, and generate a first feature, a second feature and a third feature accordingly. The first feature is used to characterize the degree of signal energy distribution within the local window, the second feature is used to characterize the instantaneous change in amplitude, and the third feature is used to characterize the changing trend of the first feature.

[0028] Step S1 specifically includes the following data processing steps executed sequentially.

[0029] The system synchronously samples the digital baseband signal output by the multi-channel receiver at a fixed sampling rate, with each channel obtaining a complex sampling sequence; assuming the total number of array channels is... , No. The channel is in The complex sampled data at each sampling time point are denoted as ,in For in-phase components, These are orthogonal components. The system is continuous. Each sampling point constitutes a data frame, and all channels are within the same time window. Point sampling constitutes the input data for the current frame.

[0030] The complex sampled data from each channel is first sent to the amplitude extraction module; this module performs amplitude extraction on each channel at each sampling time. and Perform coordinate rotation numerical calculations, approximating the modulus of the complex number through several angular rotation iterations, with the number of iterations being a preset constant; after this calculation, the... The channel is in Complex data at each sampling time point are converted into real-valued amplitude features. ,satisfy Amplitude characteristic sequence The time resolution is equal to the system sampling period, and its dimension is consistent with the engineering unit of the input signal amplitude. This sequence serves as the basic input for all subsequent feature calculations.

[0031] Based on the amplitude characteristic sequences of each channel, the system sequentially generates the first, second, and third features in the time domain using a sliding window approach; for the first... Channel, based on the current sampling time As the endpoint, take a length of The sliding window contains amplitude sampling points. ,in The preset window length constant is determined by the receiver front-end bandwidth and the system sampling rate to ensure that the sliding window contains at least two sampling periods of signal rising edges.

[0032] First characteristic Defined as the arithmetic mean of the squared deviations of each amplitude characteristic within the current window relative to the mean amplitude within the window; specifically, calculating the mean amplitude within the window. ,calculate This feature characterizes the degree of signal energy distribution within the current local time window, with the dimension being the square of the signal amplitude. The value is low when the signal is stable and rises rapidly when the signal edge is reached or interference occurs.

[0033] Second feature Defined as the absolute value of the difference between the amplitude characteristic at the current time and the amplitude characteristic at the previous time, i.e. This feature characterizes the instantaneous change in signal amplitude at adjacent sampling times, with dimensions consistent with the signal amplitude, and is used to detect the steepness of signal edges.

[0034] Third characteristic Defined as the first feature The second-order difference in the time domain, i.e. This feature characterizes the second-order trend direction of the first feature value changing over time; when the rising edge of the signal arrives, the energy accumulates monotonically and rapidly. It shows a one-sided upward trend, at this time It is positive or close to zero; when the channel is in the recovery period after interference overload, Oscillate up and down, at this time It exhibits an alternating pattern of positive and negative values; effective extraction of this feature depends on The system needs to cache the values ​​of the sequence at three consecutive time points in the time domain during the sliding window computation. and Historical values.

[0035] Thus, step S1 completes the transformation from the original multi-channel complex sampling sequence to three latent feature sequences; for the current data frame, each channel of data acquired by the system is... , , , The four sets of sequences are used in subsequent steps, among which... , , These correspond to the first feature, the second feature, and the third feature, respectively. All of the above features are updated in real time at each sampling moment and are cached in registers on a frame-by-frame basis for use when reading the data from the previous frame in step S2.

[0036] Step S2: Read the health coefficient output from the previous frame, and dynamically determine the masking window and trigger sensitivity of the current frame based on the coefficient;

[0037] Step S2 is executed based on step S1. The data base on which it depends is the value of the first feature sequence within the preset window length at the end of the previous frame that has been cached at the start time of the current frame in step S1, and the smoothed health coefficient of the previous frame stored in the history register.

[0038] The system first reads the smoothed health coefficient of the previous frame from the loop history register. This coefficient is a dimensionless quantity, and its range of values ​​has been clamped. For the first frame after system startup, there is no valid frame smoothness health coefficient available for reading in the history register. Its value is determined according to the cold start rule defined in step S4 - that is, when the history register is empty, it is initialized with 0.5 to participate in the calculation of the masking window and trigger sensitivity of the current frame.

[0039] The masking window width is determined as follows; the system has a preset minimum masking width parameter. and maximum mask width parameter Both are measured in units of the number of sampling points; This corresponds to the minimum number of sampling points required for the receiver to recover from a strong signal overload under clean, interference-free operating conditions. The maximum number of recovery sampling points under strong interference saturation conditions is obtained during both the factory calibration phase and the calibration phase. At this time, the receiver input power is increased to at least 3dB above the 1dB compression point of its front-end low-noise amplifier, causing the limiter to enter deep limiting mode. Under this condition, the statistical maximum value of the recovery period duration is measured, multiplied by a safety margin factor of 1.5, and rounded down to obtain the result. This safety margin covers extreme disturbance scenarios beyond the rated operating conditions, ensuring... It is always greater than the actual recovery period.

[0040] System read Then, calculate the width of the physical transient masking window for the current frame. :

[0041] ;

[0042] in This is a rounding function that guarantees... The number of sampling points is an integer. The physical meaning of this formula is: the lower the health coefficient of the previous frame after smoothing, the worse the data quality of the previous frame and the more serious the interference pollution. The more starting sampling points need to be skipped in the current frame to avoid possible channel residual recovery period jitter. Conversely, when the health coefficient of the previous frame is high, the system trusts that the channel is in a normal state and only skips the shortest necessary protection interval. exist Approaching Time approaches ,exist Approaching Time approaches .

[0043] Determining the trigger sensitivity involves obtaining the current frame's noise floor estimate; the system reads the last value of the previous frame. The first feature sequence value of each sampling point is... Calculate their arithmetic mean as the noise floor basis estimate for the current frame. For the processing of the first frame after system startup, there is no end of the previous frame. Data, at this time The value is taken before the start time of the current frame. sampling points Value; since the sliding window statistics at the start of the first frame are not yet fully ready, the system needs to wait for sufficient data collection. Calculations can only be performed after sampling points. Before this, step S2 does not output valid parameters, and the frame processing pipeline is inserted. The waiting delay for each sampling period; here The preset window length is the same as that used when generating the first feature in step S1 to ensure the consistency of the data source. The reason for using the data at the end of the previous frame instead of the data at the beginning of the current frame is that the sliding window statistics at the beginning of the current frame are not yet fully ready, while the data at the end of the previous frame has been fully calculated and cached in the register at the frame boundary.

[0044] The system also has two pre-set proportional coefficients. and Both are dimensionless and positive, and are obtained through finite condition calibration before leaving the factory; among them These are the basic mapping coefficients from the noise floor to the trigger sensitivity. The gain modulation coefficient is used for health feedback; the system based on , , and Calculate the virtual trigger sensitivity of the current frame : ;

[0045] This sensitivity is measured in terms of signal amplitude; where... This item reflects the unhealthiness of the previous frame: the more unhealthy the previous frame, the closer this item is to the previous frame. The closer the entire bracketed item is to The higher the sensitivity, the healthier the previous frame, and the closer this property is to the target. The closer the terms in parentheses are to each other The sensitivity drops back to the baseline level. The physical meaning of sensitivity is: when the quality of the previous frame is poor, the system expects that the current frame may still be affected by residual interference, so it actively raises the trigger threshold to avoid misjudging the residual jitter during the recovery period as a valid signal edge; when the quality of the previous frame is good, the system lowers the threshold to maintain the ability to capture weak target signal edges. The lower limit is The upper limit is .

[0046] The above and The calibration logic is as follows: Under pure, interference-free calibration conditions, Should approach ,at this time The amplitude must be lower than the normal test pulse edge value to ensure triggering; under strong interference saturation calibration conditions... Should approach ,at this time The amplitude needs to be higher than the maximum amplitude of residual jitter during the recovery period to suppress false triggering; by using two sets of boundary condition constraints, the solution can be obtained that simultaneously satisfies the above conditions. and The value range is selected, and the midpoint of the range is written as the preset value into the system parameter memory.

[0047] After the calculation is completed, and It is written to the current frame control register for use in step S3 when performing masking skipping and valid fingerprint detection on the amplitude feature sequence of each channel; simultaneously The value remains valid only during the processing of the current frame and will be updated at the start of the next frame. .

[0048] Step S3: Skip the masking window on the amplitude features of each channel, and detect the first sampling moment in the remaining sequence that simultaneously satisfies the trigger sensitivity condition and the third feature trend condition, as a valid arrival fingerprint;

[0049] Step S3 performs a fingerprint search on the amplitude feature sequences of each channel, with the input data coming from the amplitude feature sequences of each channel of the current frame cached in step S1. Second characteristic sequence Third characteristic sequence And the masking window width written to the current frame control register in step S2. and virtual trigger sensitivity .

[0050] For the The system locates the first channel in the amplitude feature sequence, starting from the initial sampling point of the current frame. The first sampling point is used as the starting position for the search, i.e., skipping the previous sampling point. If there are sampling points, If the number of sampling points exceeds the total number of sampling points in the current frame, then after skipping all sampling points, there are no remaining sequences to search, and this channel is directly marked as not finding a valid fingerprint in the current frame. Within the remaining sampling point range from this starting position to the end of the current frame, the system checks two conditions for each sampling point; the first condition is the second feature. The value is greater than the virtual trigger sensitivity of the current frame. ,Right now The second condition is the third feature. The value is greater than the preset small positive threshold. ,Right now The tiny positive threshold Used to exclude signals with stable amplitude. False triggering caused by values ​​of zero or slight fluctuations; when Equal to zero or a slightly positive value but less than At any given time, these are not considered valid monotonic energy accumulation features, and the system does not classify them as valid rising edges. The system sequentially determines rising edges based on the increasing sampling point numbers, starting from the search start position, and records the first sampling time that simultaneously satisfies both conditions as the [number missing]. Valid arrival fingerprint of the channel .

[0051] The detection process can be executed independently and in parallel on each channel within a frame. For any channel, if no sampling moment that simultaneously satisfies both conditions is found after traversing all sampling points from the search start position to the end of the frame, then that channel is marked as having no valid fingerprint found in the current frame. The marked channels do not participate in subsequent fingerprint aggregation and dispersion statistics. The number of channels that actually find valid fingerprints in the current frame is denoted as... The subsequent step S4 requires the use of valid arrival fingerprints from each valid channel. ,in Iterate through all channel indices that yield valid fingerprints, along with the third feature values ​​of these channels at their respective fingerprint moments. The symbol information of the third feature value is used to generate dispersion index and consistency index.

[0052] For example, suppose that within a certain frame, a certain channel starts searching from the end position of the masking window, and at the... Detected at each sampling time (Dimensions consistent with signal amplitude) Greater than ,at the same time (The dimension is the second-order difference of the first feature, which is non-negative). If the sampling time is recorded as a valid arrival fingerprint and the search stops, the subsequent sampling points of this channel will no longer participate in the fingerprint detection of the current frame. If it is not triggered until the end of the frame, the channel is marked as an invalid channel. During the search process, all sampling times are counted with the global intra-frame offset and the sampling point number is used as the unit. There is no need to perform absolute time conversion, which ensures that the temporal discreteness calculation of fingerprints of each channel in the same frame has a unified time reference.

[0053] Step S4: Generate a dispersion index based on the dispersion between each valid fingerprint, generate a consistency index based on the third feature symbol at each fingerprint time, and generate the current frame health coefficient through the intermediate state mapping of the two indices.

[0054] Step S4 receives the valid arrival fingerprints of each valid channel output in step S3. and the value of the third feature at each fingerprint time. subscript The system iterates through all channels in the current frame that have found valid fingerprints; it first determines the number of valid channels. Whether it is less than a preset effective channel number threshold, which is set to 2 in this embodiment; if If the value is less than 2, then the health coefficient of the current frame is directly set. The lower limit value is set to 0, skipping subsequent mapping calculations, and the frame directly enters the processing branch corresponding to low health in step S5; after this pre-judgment is completed, all subsequent downstream processing of the current frame only passes through... Driven by the value (set to 0), step S5 no longer reads or judges repeatedly. The value of .

[0055] like If the value is not less than 2, the system will generate two independent intermediate indicators; the first indicator is the dispersion index. The system calculates the fingerprint times of all valid channels. arithmetic mean Then, calculate the sum of squares of the deviations of each fingerprint time point from the average value, and then take the square root of the average. This index, measured in units of sampling point number, characterizes the degree of aggregation of effective arrival fingerprints across channels on the time axis; when the fingerprints of all channels almost overlap, A value close to 0 indicates that all channels are triggered by the same signal wavefront, demonstrating good time consistency; however, when the fingerprint dispersion between channels is large... An increase in the value indicates that the data in each channel within the frame may come from triggering events from different sources.

[0056] The second indicator is the sign consistency index. ; Check the third feature value of each valid channel at its fingerprint time. Statistical values ​​greater than Number of channels Calculate this number and the total number of valid channels. The ratio, i.e. This index is a dimensionless ratio, ranging from 0 to 1, representing whether the second-order trend of energy change in each channel at its respective fingerprint time converges; when all channels... When all are non-negative, An equal value of 1 indicates that all channels are triggered at the rising edge of a monotonically accumulating energy, consistent with the physical causal consistency of arrival at the same wavefront; when some channels... When it is negative, The corresponding decrease indicates that the fingerprints of some channels fell into the downward range of energy oscillation.

[0057] After obtaining the above two indicators, the system generates the health coefficient of the current frame through intermediate state mapping. The mapping consists of two steps: the first step is to determine the dispersion index. Construct the first intermediate variable, which is in the form of ,in The preset scaling factor, with the dimension being the reciprocal of the sampling point index, is used to... Mapped to an appropriate sensitive interval; this intermediate quantity follows Increasing and monotonically decreasing, When it approaches 0, it approaches 1; the second step is to use the sign consistency index. Constructing a second intermediate quantity This amount is to from Interval linear mapping to Interval; when hour This means that when the sign is evenly split between positive and negative, the contribution of the intermediate quantity is zero; the system then multiplies the two intermediate quantities to obtain the original health coefficient. ;because It may take a negative value. It may be a negative number; the system will perform a lower limit clamping operation on this, i.e. Ensure that the output value is within Within the interval; the meaning of this clamping operation is: when the fingerprint dispersion is too large, or when the third feature symbol consistency index... Below 0.5 (i.e.) When this occurs, it indicates that more than half of the valid channels within the frame are in the downlink range of energy oscillation at their fingerprint time, the causal consistency of the intra-frame data has been lost, the system directly sets the health coefficient to zero, and subsequent steps will be handled as an extreme contamination case; only when Furthermore, when the dispersion is within an acceptable range, the health coefficient takes a positive value.

[0058] For example, assuming the number of effective channels in a given frame is 8, the dispersion of fingerprint time points... Each sampling point, with a preset scale factor. Then the first intermediate quantity If 7 out of 8 channels are... If non-negative, then Second intermediate quantity Original coefficients After clamping If the value remains at 0.469, this frame is considered a medium confidence observation; if another frame... , ,but , , If the value is 0 after clamping, the frame is considered untrusted.

[0059] In generating the current frame health coefficient The system then stores it at a depth of The circular history register, in which In this embodiment, the value is 5; this register stores the health coefficient sequence of the most recent 5 frames, including the current frame. The system performs an arithmetic mean on the sequence to obtain a smoothed health coefficient. This smoothing value will be written to the frame health history register for use in the next frame step S2. For the startup phase where the frame number is less than 5, frames not stored in the history register are used for averaging with an initial value of 1.

[0060] During the cold start phase, missing frames in the history register are calculated using the average health coefficient of valid frames preceding the current frame; if there are no valid frames preceding the current frame, the smoothed health coefficient of the current frame is used. Set it directly to 0.5; this 0.5 serves as the global initial value during system cold start, and is also used in step S2 to read the smoothness health coefficient of the previous frame - when the history register is empty, the value read in step S2 is 0.5.

[0061] Step S5: Dynamically construct the observation noise basis according to the health coefficient and inject it into the array covariance matrix to perform direction finding; when the health coefficient is lower than the threshold, discard the current direction finding result and dynamically expand the state covariance matrix.

[0062] Step S5 receives the current frame health coefficient output in step S4. and the smoothed health coefficient Two parallel processing links are initiated: one for the direction-finding data domain of the current frame, and the other for the state domain of the tracking filter.

[0063] Data domain processing link according to Construct the observation noise basis multiplier; the system has a preset minimum noise power basis parameter. and maximum noise power basis parameters Both have the same dimensions as the elements of the array covariance matrix, which are power dimensions. The value represents the thermal noise power level of the receiver under clean operating conditions. The value is taken as 0.1 times the maximum eigenvalue of the covariance matrix under strong disturbance saturation conditions, and both are written into the parameter memory during the factory calibration stage; the system is based on exist and Linear interpolation is performed between them to obtain the observation noise basis multiplier for the current frame. ;when Approaching 1 Approaching Direction finding calculations are performed with the highest precision; when Approaching 0 Approaching The direction finding solution is performed with the lowest confidence level.

[0064] The value of 0.1 is determined empirically. In a standard array direction finding system, this ratio is equivalent to reducing the observation signal-to-noise ratio to 1 / 0.1 = 10 times the original value, which corresponds to a decrease in direction finding accuracy of about one order of magnitude. This can effectively suppress the weight contribution of interference channels in contaminated frames without making the covariance matrix excessively ill-conditioned.

[0065] The system then reads the complete data of each channel of the current frame from the data buffer. Point complex sampling sequence ,in , for Dimensional column vector; Note that the original data in this buffer is not truncated by the masking skip operation in step S3. The skip operation in step S3 only applies to the fingerprint search of the amplitude feature sequence and does not affect the integrity of the complex data here; The system calculates the array covariance matrix. superscript This represents the conjugate transpose; based on this, the product of the observation noise basis multiplier and the unit diagonal matrix is ​​added to the covariance matrix to obtain the corrected direction-finding input matrix. ,in for An identity diagonal matrix of order 1; this operation is mathematically equivalent to applying a regularization penalty to the main diagonal of the covariance matrix, reducing the weight contribution of contaminated channels in subsequent eigenvalue decomposition or matrix inversion, while ensuring that the matrix is ​​always full-rank and invertible; the modified The azimuth and elevation angles are calculated by multiple signal classification or minimum variance distortionless response direction finding algorithms, and the direction finding loop is not interrupted due to data quality issues.

[0066] The state domain processing link is executed after the direction-finding solution is completed; the system will smooth the health coefficient of the current frame. Compared with preset extreme thresholds Comparison, In this embodiment, the value is taken as 0.15; if Not less than The direction finding results are marked as reliable and directly fed into the tracking filter for routine observation updates; if Below The system determines that the current frame is in a heterogeneous collapse state, marks the direction finding result as unreliable and discards it, and does not inject it into the observation innovation filter; at this time, the system maintains a continuous collapse frame counter. Its update rule is: when the current frame is in a crash state. When the current frame is not in a crash state ; Calculate the state covariance inflation coefficient based on the counter value. ,in A preset time constant, measured in frames, is used to control the rate at which uncertainty increases. The preset maximum expansion factor is set to 10 in this embodiment; this coefficient monotonically increases with the number of consecutive crash frames and approaches a certain value. It will not diverge indefinitely; the system will use the state prediction covariance matrix of the Kalman filter. Multiplying by the inflation factor yields the inflated covariance matrix. During the time update phase, the dilated matrix replaces the original matrix for recursion; simultaneously, the state estimation result of the current frame is set to an extrapolated prediction value that relies solely on the inertial kinematics model, without introducing any radio direction finding information from the current frame; once a subsequent frame... rebounded to above, Reset to zero Reverting to 1, the filter resumes its normal observation update mode.

[0067] Given that the feedback control in step S2 has a one-frame lag, in the current frame Before the calculation is completed, both step S3 (valid fingerprint detection) and step S5 (direction finding) are based on the previous frame. This means that sudden interference may cause false triggering within the current frame; to mitigate this problem, step S5 stipulates that regardless of whether the direction-finding result of the current frame is abnormal due to false triggering, in Calculation completed and below At that time, the direction finding result is discarded, and its impact on the new information is completely blocked.

[0068] When multiple consecutive frames are in a state of insufficient effective channels, the smoothed health coefficient per frame The rate gradually recovers, and the system state domain covariance inflation coefficient... The system gradually releases data to 1 at the same rate; the introduction of this smoothing mechanism results in the system recovering from the disturbance state to normal tracking within a time limit not exceeding 1. Frame lag.

[0069] The above steps constitute a bottom-up intra-frame data homology measurement and adaptive observation quality control mechanism; the operational logic of this mechanism can be traced back to its physical meaning along the data processing chain: Step S1 extracts three latent features from the original sampling sequence, of which the third feature... The role of this step is particularly crucial, enabling the system to distinguish between the "rising edge of a signal with monotonically concentrated energy" and the "interference recovery period of energy oscillation" in time—this is the foundation for all subsequent discriminations. Step S2 uses the evaluation results of the previous frame to adjust the acquisition parameters of the current frame in reverse, allowing the system to actively tighten the detection threshold in polluted environments to avoid false triggers, and actively relax the threshold in clean environments to capture weak signals, thus achieving active adaptation of the feature extraction level to the environmental state. Steps S3 and S4 use the temporal dispersion of the effective fingerprint and the consistency of the third feature symbol to cross-validate the intra-frame data, improving the frame quality assessment from two dimensions. The "good / bad" of the element is upgraded to a continuous "health coefficient", which allows the degradation process of data quality to be quantified and tracked rather than triggered abruptly. Step S5 applies regularization to the direction finding covariance matrix and saturation expansion to the state covariance matrix based on this coefficient, so that the observation noise basis and filter uncertainty change synchronously and smoothly with the data quality. This embodiment can effectively reduce the confidence weight of false direction finding results when interference and target coexist, maintain the boundedness of the filter state when the data is completely unavailable, and realize the rapid recovery of the direction finding and tracking link after the interference dissipates, thereby enabling the positioning and direction finding tracking system to have dynamic adaptive capability to complex electromagnetic environments.

[0070] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0071] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0072] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A real-time positioning and direction finding tracking method for unmanned aerial vehicles (UAVs) in complex electromagnetic environments, characterized in that, include: Acquire multi-channel digital baseband signal sampling sequences; The signal amplitude features are extracted based on the sampling sequence of each channel, and a first feature, a second feature and a third feature are generated accordingly. The first feature is used to characterize the degree of dispersion of signal energy within the local window, the second feature is used to characterize the instantaneous change in amplitude, and the third feature is used to characterize the changing trend of the first feature. Read the health coefficient output from the previous frame, and dynamically determine the masking window and trigger sensitivity for the current frame based on this coefficient; Skip the masking window on the amplitude features of each channel, and detect the first sampling moment in the remaining sequence that simultaneously satisfies the trigger sensitivity condition and the third feature trend condition, as a valid arrival fingerprint; A dispersion index is generated based on the dispersion between each valid fingerprint, a consistency index is generated based on the third feature symbol at each fingerprint time, and the current frame health coefficient is generated by mapping the intermediate state of the two indices. Based on the health coefficient, an observation noise basis is dynamically constructed and injected into the array covariance matrix to perform direction finding; When the health coefficient is below the threshold, the current direction finding result is discarded and the state covariance matrix is ​​dynamically expanded.

2. The real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments according to claim 1, characterized in that, The signal amplitude characteristics are obtained by performing coordinate rotation digital calculation on the in-phase and quadrature components in the complex data of each channel at each sampling time; in the first characteristic, the current time value is the arithmetic mean of the squared deviations of each amplitude characteristic relative to the mean amplitude within a preset window length with the current time as the endpoint; in the second characteristic, the current time value is the absolute value of the difference between the amplitude characteristic at the current time and the amplitude characteristic at the previous time; in the third characteristic, the current time value is the difference between the first characteristic value at the current time and the previous time minus the difference between the first characteristic value at the previous time and the time before that.

3. The real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments according to claim 1, characterized in that, The dynamic determination of the masking window and trigger sensitivity of the current frame based on the health coefficient output from the previous frame specifically includes: the width of the masking window is negatively correlated with the health coefficient of the previous frame; when the coefficient approaches the upper limit, the width approaches the preset minimum masking width; when the coefficient approaches the lower limit, the width approaches the preset maximum masking width; the trigger sensitivity is proportional to the product of the square root of the current frame's noise floor estimate and the dynamic gain factor; the dynamic gain factor is negatively correlated with the health coefficient of the previous frame.

4. The real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments according to claim 1, characterized in that, The dispersion index is the square root of the arithmetic mean of the squared deviations of each valid fingerprint from its arithmetic mean; the consistency index is the ratio of the number of channels with non-negative values ​​of the third feature at each valid fingerprint time to the total number of valid channels; the intermediate state mapping includes constructing a first intermediate quantity negatively correlated with the dispersion index, constructing a second intermediate quantity obtained by performing a linear mapping on the consistency index, and taking the larger of the product of the two intermediate quantities and a preset lower limit value as the current frame health coefficient.

5. The real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments according to claim 1, characterized in that, After generating the health coefficient of the current frame, a smoothing process is also included: the coefficient is stored in a circular history register with a preset depth, the health coefficients of multiple consecutive frames in the register are arithmetically averaged, and the average value is used as the smoothed health coefficient. The reading of the health coefficient output from the previous frame specifically refers to reading the smoothed health coefficient corresponding to the previous frame; the noise floor basis estimate is the arithmetic mean of the first feature values ​​within a preset window length at the end of the previous frame.

6. The real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments according to claim 1, characterized in that, The specific steps of constructing the observation noise basis and injecting it into the array covariance matrix include: obtaining the smoothed health coefficients of the current frame and the preset minimum and maximum noise power basis; mapping the smoothed health coefficients to the minimum and maximum noise power basis to obtain the observation noise basis multiplier; reading the complete complex sampling sequence of each channel of the current frame and calculating the array covariance matrix; and superimposing the product of the observation noise basis multiplier and the unit diagonal matrix into the array covariance matrix.

7. The real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments according to claim 1, characterized in that, The process of discarding the current direction finding results and dynamically expanding the state covariance matrix includes: determining the current frame as a heterogeneous collapse state and marking the direction finding results as unreliable when the health coefficient is lower than a preset extreme threshold; maintaining a continuous collapse frame counter, which increments in the collapse state and resets to zero in the non-collapse state; constructing an expansion coefficient based on the counter value, a preset time constant, and a preset maximum expansion factor, which increases with the counter value and approaches a preset upper limit; and using the expansion coefficient to multiply the state prediction covariance matrix for state recursion, with the current frame state estimation using only the inertial navigation extrapolation prediction value.

8. The real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments according to claim 1, characterized in that, It also includes handling when the number of effective channels is insufficient: count the number of channels in the current frame that detect the effective fingerprint, and when the number is less than the preset effective channel number threshold, directly set the current frame health coefficient to the lower limit value, perform the operation of the injection array covariance matrix with the maximum noise power basis, and at the same time determine that the current frame is in a heterogeneous collapse state and apply the dynamic expansion processing to the state prediction covariance matrix.

9. The real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments according to claim 1, characterized in that, The dispersion index and the consistency index constitute a positive interaction constraint: the dispersion index characterizes the degree of aggregation of each valid fingerprint on the time axis to reflect the temporal consistency of wavefront arrival; the consistency index characterizes whether the value of the third feature at each valid fingerprint moment is greater than a preset small positive threshold to reflect whether each channel is triggered in the effective rising edge interval of monotonically accelerating energy aggregation; the health coefficient only approaches its upper limit value when the dispersion index is lower than the preset dispersion threshold and the consistency index is higher than the preset consistency threshold; when the consistency index is lower than 0.5, regardless of the value of the dispersion index, the health coefficient is clamped to zero.

10. The real-time positioning and direction finding tracking method for UAVs in complex electromagnetic environments according to claim 5, characterized in that, The masking window width and the trigger sensitivity are dynamically generated according to their respective negative correlation with the health coefficient after smoothing of the previous frame. The proportional coefficient in the negative correlation is obtained through limited working condition calibration before leaving the factory. The noise floor estimate is the arithmetic mean of the first feature values ​​within the preset window length at the end of the previous frame. The preset window length is equal to the preset window length used when generating the first feature.

Images