Millimeter wave radar multi-dimensional performance self-sensing method and intelligent monitoring platform

Abstract

The application discloses a kind of millimeter wave radar multidimensional performance self-sensing method and intelligent monitoring platform, belong to millimeter wave radar system technical field, first endow radar radio frequency front-end online self-sensing ability, through built-in coupling loop and time division multiplexing calibration sequence, real-time extraction transmission and reception link key performance parameters under not interrupting task, no external instrument. Three layers of ten index health measurement quantitative evaluation system is constructed, environmental coupling compensation is introduced, environmental fluctuation and self degradation are distinguished, false alarm is avoided. Based on the time series analysis and sliding window regression of degradation index, the remaining life is predicted in advance, and the upgrade from passive maintenance to predictive maintenance is realized. For cluster deployment, abnormality is automatically screened and hardware failure and environmental problems are distinguished by reference group division and Z-score statistics. Temperature adaptive reference window updating mechanism adapts to long-term aging drift, reduces the false alarm rate in the middle and late stages, cross-index correlation analysis assists in root cause positioning, and improves operation and troubleshooting efficiency.

Description

Technical Field

[0001] This invention belongs to the field of millimeter-wave radar system technology, specifically relating to a multi-dimensional performance self-sensing method and intelligent monitoring platform for millimeter-wave radar. Background Technology

[0002] Millimeter-wave radar, due to its all-weather operation and excellent range and velocity resolution, has been widely used in fields such as automotive driver assistance, intelligent transportation, perimeter security, and industrial measurement. With the continuous expansion of deployment scale and the increasing complexity of application scenarios, the long-term reliable operation of millimeter-wave radar equipment has become a critical issue in engineering practice. However, current industry research and development and applications of millimeter-wave radar technology almost entirely focus on how to better detect external targets, lacking systematic attention to the fundamental issue of whether the radar itself is in optimal operating condition.

[0003] Specifically, existing technologies have shortcomings in the following three aspects: First, there is a lack of online monitoring capabilities for RF front-end performance. The RF front-end of millimeter-wave radar includes core components such as power amplifiers, quadrature modulators, mixers, and low-noise amplifiers. Over long-term operation, these components degrade due to aging, temperature drift, and power supply fluctuations. Typical degradation forms include enhanced power amplifier nonlinearity (AM-AM and AM-PM distortion), increased amplitude and phase imbalance in the I / Q branches, increased local oscillator phase noise, receiver link gain drift, and increased noise figure. While these degradations may not cause complete radar failure in the initial stages, they gradually decrease the target detection signal-to-noise ratio and increase the false alarm rate, eventually leading to a large number of missed and false detections after a certain critical point. Currently, measurements of these RF front-end parameters typically rely on external vector network analyzers, spectrum analyzers, or signal generators, requiring the radar to be offline, the antenna disconnected, and the test equipment connected. This makes continuous acquisition during normal equipment operation impossible. The calibration operations described in existing patent literature are all calibrations of the external environmental background or registration of multi-sensor coordinate systems, not online real-time extraction of the radar's own RF link performance parameters.

[0004] Second, there is a lack of assessment methods to integrate multi-dimensional heterogeneous performance indicators into a unified health metric. The operating status of a millimeter-wave radar involves multiple dimensions, including radio frequency link indicators such as transmit linearity, IQ balance, phase noise, receive gain, and image rejection capability; environmental condition indicators such as background noise level, clutter intensity, and multipath interference; and actual detection performance indicators such as effective detection rate, ranging consistency, and false alarm density. These indicators have different dimensions, value ranges, and physical meanings, and there are complex coupling relationships between them. For example, in heavy rain or dense fog, detection performance indicators will naturally decline, but this is not due to equipment failure but rather environmental factors. Existing technologies do not provide effective solutions for distinguishing between environmentally caused performance fluctuations and equipment performance degradation in assessments. Furthermore, there is no existing method to integrate the above-mentioned multi-dimensional indicators into a single standardized health score with hierarchical weights.

[0005] Third, there is a lack of intelligent operation and maintenance platforms for multi-node cluster deployments. In actual engineering projects, dozens or even hundreds of millimeter-wave radars of the same model are often deployed in the same park or road section. Currently, the operation and maintenance management of these devices is still mainly based on post-fault repair, that is, intervention is only carried out after the equipment reports obvious fault codes or the user discovers missed or false diagnoses. This approach cannot provide early warnings during the performance degradation stage, nor can it automatically identify abnormal nodes through horizontal comparison. Especially under the same environmental conditions, if the health indicators of a radar deviate significantly from the statistical distribution of other nodes in the same group, it often means that the node has an individual hardware failure; while if multiple radars in the same group show the same direction of deviation, it may point to systemic environmental interference or firmware version defects. Existing patent literature does not have a cross-node performance comparison anomaly detection method or a unified operation and maintenance platform solution for face-to-face millimeter-wave radar clusters.

[0006] In summary, the industry urgently needs a method to acquire RF front-end performance status parameters in real time while the radar is operating normally, without relying on external instruments or interrupting the detection mission, as well as a platform solution that can uniformly quantify multi-dimensional indicators into health scores, predict degradation trends, and automatically screen for anomalies in multi-node clusters. Summary of the Invention

[0007] To address the shortcomings of the existing technologies, this application provides a multi-dimensional performance self-sensing method and intelligent monitoring platform for millimeter-wave radar.

[0008] In its first aspect, this application proposes a multi-dimensional performance self-sensing method for millimeter-wave radar, comprising the following steps: Step S1: Set up a directional coupler between the power amplifier output and the transmitting antenna of the millimeter-wave radar transmitting link. Couple the sample signal from the transmitted signal through the directional coupler and feed it into the auxiliary receiving channel, which shares the local oscillator source with the main receiving channel. Process the sampled digital baseband signal obtained after down-conversion and analog-to-digital conversion of the auxiliary receiving channel to extract the transmitting link performance parameters. These transmitting link performance parameters include the power amplifier nonlinearity characteristics, I / Q imbalance, and phase noise. Step S2: Insert a known calibration sequence into the gap of the radar transmitted waveform in a time-division multiplexing manner. Feed the calibration sequence into the front end of the receiving link through the internal radio frequency path. Perform matched filtering and spectrum analysis on the calibration sequence acquired after passing through the receiving link to extract the receiving link performance parameters, including gain drift, image rejection ratio and noise figure. Step S3: Extract environmental condition parameters from the range-Doppler spectrum generated during normal radar detection. These environmental condition parameters include the mean background noise energy, clutter spread rate, multipath reflection salience, and target signal-to-noise ratio attenuation. The environmental complexity index is obtained by weighted fusion of the various environmental condition parameters. Step S4: Using a fixed reference scatterer with known location and stable reflection characteristics in the deployment scenario, the detection quality of the radar is evaluated online, and detection quality parameters are extracted. These detection quality parameters include effective detection rate, ranging consistency index, false alarm density, and detection stability index. Step S5: Convert each indicator in the transmit link performance parameters, receive link performance parameters and detection quality parameters into a percentage-based individual score through a piecewise linear mapping rule. Merge each individual score according to a hierarchical weight allocation. Apply decoupling compensation based on the environmental complexity index to the detection dimension score to obtain a standardized health score. Step S6: Select key degradation indicators from various indicators, perform preprocessing on the historical time series data of key degradation indicators, and then perform sliding window linear regression to extract the degradation rate. Based on the degradation rate and the preset failure threshold, estimate the remaining normal operating time of the radar. Step S7: Divide multiple radars into reference groups according to preset dimensions, calculate the statistical deviation of each radar in the reference group, determine abnormal nodes based on the statistical deviation, and classify the anomalies into isolated anomalies or group anomalies according to the number of abnormal nodes and the direction of deviation.

[0009] In some embodiments, the process of extracting the power amplifier nonlinear characteristic index in step S1 includes: The acquired digital baseband signal and the original baseband transmitted signal are cross-correlated to achieve time-domain alignment. An AM-AM characteristic curve is constructed based on the amplitude relationship between the aligned acquired signal and the original transmitted signal. The linear gain deviation and third-order nonlinear intensity are extracted from the AM-AM characteristic curve. An AM-PM characteristic curve is constructed based on the phase relationship between the aligned acquired signal and the original transmitted signal, and a phase deviation index is extracted from the AM-PM characteristic curve. The intensity index of memory effect is extracted based on the amplitude-phase two-dimensional distribution of the acquired signal.

[0010] In some embodiments, the process of extracting the I / Q imbalance index and the phase noise index in step S1 includes: The energy of the acquired digital baseband signal on the in-phase branch and the quadrature branch are calculated separately, and the I / Q amplitude imbalance and I / Q phase imbalance are obtained by cross-correlation calculation of the two branches. The phase error sequence of the acquired digital baseband signal is estimated by power spectrum estimation. The phase noise power value corresponding to each frequency offset interval is obtained by integration in different frequency offset intervals.

[0011] In some embodiments, the insertion frequency of the calibration sequence in step S2 is adjusted in conjunction with the rate of change of the transmit link performance parameters extracted in step S1: when the rate of change of any indicator in the transmit link performance parameters exceeds a preset threshold, the insertion frequency of the calibration sequence is automatically increased; when the rate of change falls back below the preset threshold, it is gradually restored to the baseline insertion frequency.

[0012] In some embodiments, the target signal-to-noise ratio attenuation in step S3 is: using a fixed reference scatterer with a known location and known radar cross-section in the scene, the attenuation of the signal-to-noise ratio of the fixed reference scatterer in the current frame relative to the historical signal-to-noise ratio reference value of the fixed reference scatterer under clear weather conditions is calculated.

[0013] In some embodiments, the hierarchical weight allocation in step S5 includes: The individual scores are categorized into three dimension groups: radio frequency (RF), environment, and detection. First, a weighted average is calculated within each dimension group according to the weights within the group to obtain the overall dimension score. Then, the overall scores of the RF dimension, environment dimension, and detection dimension are weighted and summed according to the weights between dimensions. The weights between dimensions of the RF dimension are greater than those of the detection dimension, and the weights between dimensions of the detection dimension are greater than those of the environment dimension. The decoupling compensation is as follows: when the environmental complexity index exceeds a preset threshold, positive compensation is applied to the comprehensive score of the detection dimension, and the compensation magnitude increases with the increase of the environmental complexity index, and the compensated score does not exceed the full score.

[0014] In some embodiments, the preprocessing in step S6 includes: Outliers are removed from the historical time-series data of key degradation indicators, and data points that deviate from the median of the sliding window by more than a preset multiple of the median absolute deviation are replaced with interpolated values ​​within the window. Smooth the data after outlier removal using a moving average; The smoothed data is periodically normalized to eliminate regular fluctuations caused by periodic environmental factors; the window length of the sliding window linear regression is a preset number of days.

[0015] In some embodiments, the criteria for classifying anomalies as isolated anomalies or clustered anomalies in step S7 include: If an abnormal event involves only a single node within the reference group within a preset time window, and no other indicators of that node show a chain of abnormalities, it is determined to be an isolated abnormality. If an abnormal event involves at least three nodes within a reference group exhibiting the same directional shift in the same indicator, it is determined to be a cluster-type abnormality.

[0016] Secondly, this application proposes a millimeter-wave radar intelligent monitoring platform, including: The data aggregation module is used to receive performance monitoring data frames reported by multiple millimeter-wave radars and write the performance monitoring data frames in parallel into the time-series database according to the node identifier. The performance monitoring data frame includes transmit link performance parameters, receive link performance parameters, environmental condition parameters, detection quality parameters, and health score. The trend analysis module is used to read historical data of key degradation indicators of each radar node from the time series database, perform sliding window linear regression after preprocessing the historical data to extract the degradation rate, and estimate the remaining normal working time of each radar node based on the degradation rate and preset failure threshold. The anomaly detection module is used to automatically divide multiple radars into reference groups according to preset dimensions, calculate the statistical deviation of each radar node in the reference group, determine the abnormal nodes based on the statistical deviation, and classify the anomalies into isolated anomalies or group anomalies according to the number of abnormal nodes and the direction of deviation. The visualization and alarm module is used to display the real-time health status of each radar node on the interactive interface in a color-coded manner, draw the health score trend curve, and generate alarm notifications when an anomaly is detected or the remaining normal working time is lower than a preset threshold.

[0017] In some embodiments, the performance monitoring data frame includes: The node identifier field group contains the radar's unique serial number and model code; the timestamp field group contains the timestamp of the data acquisition time. The RF specification field group includes power amplifier nonlinearity characteristics, I / Q imbalance, phase noise, gain drift, image rejection ratio, and noise figure. The environmental metrics field group includes the environmental complexity index and its sub-parameters; The detection index field group includes effective detection rate, ranging consistency index, false alarm density, and detection stability index; The health score field group includes the comprehensive score of each dimension, the decoupling compensation amount, the final health score, and the health level identifier; After dividing the reference group, the anomaly detection module first verifies the homogeneity of the distribution of each indicator within the reference group. When the variance within the group exceeds the preset variance threshold, it performs secondary clustering subdivision on the reference group before performing anomaly detection.

[0018] Thirdly, this application proposes an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0019] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above.

[0020] The beneficial effects of this invention are: While the millimeter-wave radar is performing its detection mission normally, it extracts the performance status parameters of the transmission and reception links in real time without relying on external testing instruments or interrupting the target detection process. It integrates multiple heterogeneous performance indicators from the radio frequency, environmental, and detection dimensions into a standardized health score that can be uniformly measured and horizontally compared, based on their physical importance and mutual coupling relationship. The score also eliminates pseudo-degradation effects caused by environmental factors. Based on the historical time-series data of the health score, it identifies the trend characteristics of performance degradation, estimates the remaining effective working time of the radar, and in multi-node cluster deployment scenarios, it automatically distinguishes between individual hardware failures and systemic environmental problems through cross-node statistical comparison, realizing an upgrade from passive maintenance to predictive maintenance. Attached Figure Description

[0021] Figure 1 This is the overall flowchart of the present invention.

[0022] Figure 2 This is a diagram of the intelligent monitoring platform architecture of the present invention. Detailed Implementation

[0023] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein; rather, these embodiments are provided so that a more thorough understanding of the invention can be achieved and that the full scope of the invention can be conveyed to those skilled in the art.

[0024] Firstly, this application proposes a multi-dimensional performance self-sensing method for millimeter-wave radar, such as... Figure 1 As shown, it includes the following steps: Step S1: Set up a directional coupler between the power amplifier output and the transmitting antenna of the millimeter-wave radar transmitting link. Couple the sample signal from the transmitted signal through the directional coupler and feed it into the auxiliary receiving channel, which shares the local oscillator source with the main receiving channel. Process the sampled digital baseband signal obtained after down-conversion and analog-to-digital conversion of the auxiliary receiving channel to extract the transmitting link performance parameters. These transmitting link performance parameters include the power amplifier nonlinearity characteristics, I / Q imbalance, and phase noise. In some embodiments, the process of extracting the power amplifier nonlinear characteristic index in step S1 includes: The acquired digital baseband signal and the original baseband transmitted signal are cross-correlated to achieve time-domain alignment. An AM-AM characteristic curve is constructed based on the amplitude relationship between the aligned acquired signal and the original transmitted signal. The linear gain deviation and third-order nonlinear intensity are extracted from the AM-AM characteristic curve. An AM-PM characteristic curve is constructed based on the phase relationship between the aligned acquired signal and the original transmitted signal, and a phase deviation index is extracted from the AM-PM characteristic curve. The intensity index of memory effect is extracted based on the amplitude-phase two-dimensional distribution of the acquired signal.

[0025] In some embodiments, the process of extracting the I / Q imbalance index and the phase noise index in step S1 includes: The energy of the acquired digital baseband signal on the in-phase branch and the quadrature branch are calculated separately, and the I / Q amplitude imbalance and I / Q phase imbalance are obtained by cross-correlation calculation of the two branches. The phase error sequence of the acquired digital baseband signal is estimated by power spectrum estimation. The phase noise power value corresponding to each frequency offset interval is obtained by integration in different frequency offset intervals.

[0026] Transmit link coupling loopback monitoring is the first core technological innovation of this solution, specifically including: ① Hardware architecture: A directional coupler is added between the power amplifier output and the transmitting antenna of the radar transmit link. The coupling port of the coupler is connected to a dedicated auxiliary receiving channel. This auxiliary receiving channel includes an adjustable attenuator, a down-conversion mixer, an intermediate frequency amplifier, and an analog-to-digital converter sampler connected in sequence, and finally sends the sampled data to the digital baseband processing unit.

[0027] Key design considerations: The auxiliary receiving channel shares the local oscillator source with the main receiving channel to ensure phase coherence between the transmitted signal sample and the ideal reference signal. The coupling degree of the directional coupler needs to strike a balance between "not affecting the output power of the main transmitting link" and "ensuring that the auxiliary channel obtains a sampled signal with a sufficient signal-to-noise ratio." A coupling degree of 20dB is recommended.

[0028] ② Signal Acquisition and Time Domain Alignment: While the radar transmits a frequency-modulated continuous wave detection signal, the auxiliary receiving channel performs radio frequency sampling on the transmitted signal at a sampling rate no less than twice the signal bandwidth. After down-conversion and digitization, the actual transmitted signal sampling sequence is obtained. Simultaneously, the waveform generator in the digital baseband processing unit retains the ideal baseband reference sequence for the corresponding time period.

[0029] Due to group delay in the transmission link, there is a fixed offset between the acquired signal and the ideal reference signal on the time axis. The system calculates the cross-correlation function between the two sequences, uses the time delay corresponding to the peak value of the cross-correlation function as the alignment offset, and precisely aligns the two sequences in the time domain, extracting the effective data segment after alignment for subsequent analysis.

[0030] ③ Extraction of Nonlinear Distortion Feature Parameters: The actual transmitted sequence and the ideal reference sequence, after time-domain alignment, are treated as complex baseband signals, and their instantaneous amplitude variation trajectories are extracted. Using the instantaneous amplitude of the ideal reference signal as the reference axis and the instantaneous amplitude of the actual transmitted signal as the observation axis, the amplitude-amplitude characteristic relationship of the transmission link is constructed. Under ideal linear conditions, this characteristic relationship is a straight line with a unit slope; in reality, due to the nonlinear effect of the power amplifier, this characteristic relationship exhibits a curved characteristic deviating from a straight line, and the degree of curvature intuitively reflects the strength of nonlinear distortion.

[0031] Similarly, by extracting the instantaneous phase difference between the two sequences, and using the instantaneous amplitude of the ideal reference signal as the reference axis and the phase difference as the observation axis, an amplitude-phase characteristic relationship is constructed. This characteristic relationship reflects the additional phase shift introduced by the power amplifier at different input power levels.

[0032] A polynomial model is used to model the amplitude-amplitude characteristic relationship, considering both the impact of the current input on the output and the residual impact of the input from previous time steps on the current output. The model coefficients are solved using least-squares fitting, and the following nonlinear distortion characteristic parameters are extracted from the coefficients: Linear gain deviation: reflects the degree to which the actual gain of the power amplifier deviates from the nominal design value, and is determined by the difference between the first-order term coefficient in the fitting coefficient and the theoretical value; Odd-order nonlinearity intensity: The relative proportions of the coefficients of the third-order and fifth-order nonlinear terms to the coefficients of the first-order term are extracted, reflecting the severity of third-order and fifth-order intermodulation distortion. Odd-order nonlinearity is the most destructive component of power amplifier nonlinearity because the intermodulation products it generates fall into the signal band and cannot be removed by filters; Memory effect strength: The sum of the energy of all coefficients with memory terms is compared with the sum of the energy of coefficients without memory terms to reflect the overall strength of the power amplifier's memory effect.

[0033] ④ IQ Imbalance Parameter Extraction: The sampled signal is decomposed into in-phase and quadrature branch components. Ideally, the two signals should have equal amplitude and a strict 90-degree phase difference. However, in actual IQ modulators, due to mixer gain mismatch and phase shifter phase error, amplitude and phase imbalances will occur between the two signals.

[0034] The system characterizes IQ imbalance from two dimensions: Amplitude imbalance: This is measured by comparing the average energy difference between in-phase and quadrature branches over the entire observation period. Ideally, the two branches should have equal energy; the greater the difference, the more severe the amplitude imbalance. Phase imbalance: Measured by analyzing the correlation between in-phase and quadrature branches. When two paths are perfectly orthogonal, their correlation is zero. The presence of correlation indicates that the orthogonal phase relationship has been disrupted, and the degree of deviation can be mapped to the phase error angle.

[0035] ⑤ Phase noise index extraction: Calculate the instantaneous phase error sequence between the sampled signal and the ideal reference signal—that is, take the phase angle of the conjugate product of the two signals to obtain the phase deviation as a function of time. Perform power spectral density estimation on this phase deviation sequence, using a method of segmented windowing, overlapping of adjacent segments, and multi-segment averaging to obtain statistically stable estimation results.

[0036] Phase noise values ​​at several key frequency offset positions are extracted from the estimated phase noise power spectrum. These frequency offset positions correspond to different noise sources: the phase noise at the near-end frequency offset is mainly determined by the noise of the reference crystal and phase detector within the PLL loop bandwidth; the phase noise at the mid-range frequency offset reflects the combined noise contribution of the PLL and the voltage-controlled oscillator; and the phase noise at the far-end frequency offset is mainly determined by the broadband noise floor of the voltage-controlled oscillator.

[0037] By comparing the current phase noise values ​​at each frequency offset with the factory calibration reference values, it can be determined which link in the local oscillator link has experienced performance degradation.

[0038] ⑥ Monitoring parameter output: All the parameters extracted above are packaged into a transmit link performance monitoring data set, including: linear gain deviation, third-order nonlinear intensity, fifth-order nonlinear intensity, memory effect intensity, IQ amplitude imbalance, IQ phase imbalance, and phase noise values ​​at four key frequency offsets.

[0039] Each parameter in this data set is configured with a corresponding normal range threshold and a warning threshold, which are determined during factory calibration based on the statistical distribution of multiple devices in the same batch. When any parameter exceeds the warning threshold, the radar triggers the corresponding alarm flag.

[0040] Step S2: Insert a known calibration sequence into the gap of the radar transmitted waveform in a time-division multiplexing manner. Feed the calibration sequence into the front end of the receiving link through the internal radio frequency path. Perform matched filtering and spectrum analysis on the calibration sequence acquired after passing through the receiving link to extract the receiving link performance parameters, including gain drift, image rejection ratio and noise figure. In some embodiments, the insertion frequency of the calibration sequence in step S2 is adjusted in conjunction with the rate of change of the transmit link performance parameters extracted in step S1: when the rate of change of any indicator in the transmit link performance parameters exceeds a preset threshold, the insertion frequency of the calibration sequence is automatically increased; when the rate of change falls back below the preset threshold, it is gradually restored to the baseline insertion frequency.

[0041] Specifically, this includes: ① Calibration strategy design: Performance monitoring of the receiver link requires the response of a known reference signal after passing through the receiver link as a basis for judgment. This scheme adopts a "time-division multiplexing frame insertion calibration" strategy: dedicated calibration frames are periodically inserted between the detection frames during normal radar operation, utilizing the complete path from the radar's own transmit link to spatial coupling and then to the receiver link to achieve online calibration of the receiver link.

[0042] The insertion frequency of calibration frames is linked to the rate of change of the transmit link monitoring data. The logic is as follows: when the parameters in the transmit link monitoring data set remain stable, calibration frames are inserted according to the default period; when a rapid change in transmit link performance parameters is detected, the calibration interval is automatically shortened to increase the frequency of receive link monitoring. This linkage mechanism avoids over-monitoring of the receive link and concentrates limited calibration resources on periods of performance fluctuation.

[0043] ② Calibration Sequence Design: The calibration frame transmits a specially designed linear frequency modulation (FM) sequence. This sequence design must meet three conditions: First, the FM bandwidth must cover the entire operating frequency band of the radar to ensure the calibration results are representative across the entire band; second, the chirp duration is designed to be an integer multiple of the operating chirp duration to maintain frame synchronization continuity when inserting the operating frame sequence; third, a constant amplitude envelope is maintained throughout the entire chirp duration to avoid envelope variations introducing additional amplitude nonlinearities that could interfere with the evaluation of the receiver link linearity.

[0044] ③ Receiver Link Gain Drift Detection: The receiver performs matched filtering on the received signal of the calibration sequence. Using the original calibration sequence as a reference template, the peak amplitude after matched filtering is extracted as a measure of the current total gain of the receiver link. This current gain measure is compared with the reference gain measure measured under the same calibration sequence conditions during factory calibration, and the relative deviation between the two is calculated. The positive or negative direction of this relative deviation indicates the direction of gain drift (increase or decrease), and its absolute value reflects the severity of the drift. When the relative deviation exceeds the preset allowable range, the system determines that the receiver link gain has abnormally drifted and automatically applies a reverse gain compensation coefficient in the digital domain to restore the equivalent total gain to the nominal level.

[0045] ④ Receiver link image rejection ratio detection: Perform spectrum analysis on the received calibration sequence signal to simultaneously identify the main signal frequency component and its corresponding image frequency component in the spectrum. The image frequency component originates from the quadrature imbalance of the receiver link IQ mixer—when the gains of the in-phase and quadrature branches are not completely equal or their phases are not completely quadrature, the signal will generate a parasitic component at its image frequency position.

[0046] The system measures the power of the main signal component and the power of the image frequency component, and calculates their ratio as a quantitative indicator of the image rejection ratio. The higher the ratio, the stronger the receiver link's ability to suppress image interference. When this ratio is lower than a preset threshold, it indicates that the receiver link's IQ imbalance has become severe enough to potentially affect normal target detection, and the system triggers a receiver link IQ compensation procedure or a maintenance alarm.

[0047] ⑤ Receiver Link Noise Figure Estimation: Using the spectral data of the no-signal intervals in the received signal of the calibration frame (i.e., frequency bands outside the calibration chirp band), the noise floor power spectral density of the receiver link itself is measured. Combined with the known injected power of the calibration signal and the measured receiver link gain, the equivalent noise figure of the receiver link is calculated. A continuous increase in the noise figure is a key early warning signal of performance degradation of the low-noise amplifier at the receiver link front end; its degradation usually precedes a significant change in gain.

[0048] ⑥ Monitoring parameter output: The receiver link performance monitoring data set includes three parameters: gain drift, image rejection ratio, and noise figure.

[0049] Step S3: Extract environmental condition parameters from the range-Doppler spectrum generated during normal radar detection. These environmental condition parameters include the mean background noise energy, clutter spread rate, multipath reflection salience, and target signal-to-noise ratio attenuation. The environmental complexity index is obtained by weighted fusion of the various environmental condition parameters. In some embodiments, the target signal-to-noise ratio attenuation in step S3 is: using a fixed reference scatterer with a known location and known radar cross-section in the scene, the attenuation of the signal-to-noise ratio of the fixed reference scatterer in the current frame relative to the historical signal-to-noise ratio reference value of the fixed reference scatterer under clear weather conditions is calculated.

[0050] The specific implementation of step S3 is as follows: Extract the following four environmental feature parameters from the distance-Doppler spectrum of the current frame: ① Average background noise energy: In the range-Doppler spectrum, the average energy level of the clean noise region determined to be targetless after constant false alarm rate detection is calculated. This indicator reflects the overall background noise level of the electromagnetic environment, and its variation may originate from the enhancement of external electromagnetic interference sources or the increase in noise of the receiving link itself.

[0051] ② Clutter Spread Rate: The proportion of range cells in the statistical range-Doppler spectrum whose energy exceeds the adaptive noise threshold out of the total number of range cells. This indicator reflects the spatial coverage of static and slow-moving clutter in the environment. In open outdoor scenes, this proportion is usually low (below 5%); in complex environments such as industrial plants with dense metal structures, this proportion may exceed 30% or even higher.

[0052] ③ Multipath Reflection Significance: In the range-Doppler spectrum, for each detected target, a search is performed along the range dimension to both sides to check for the existence of secondary peaks whose distances are integer multiples of the main peak (this is a typical characteristic of multipath reflection: coherent superposition occurs when the reflection path length is an integer multiple of the direct path). The proportion of the sum of the energies of all such secondary peaks to the sum of the energies of all main peaks is calculated; the larger this proportion, the more severe the multipath reflection phenomenon.

[0053] ④ Target Signal-to-Noise Ratio (SNR) Attenuation: The average SNR of the detected targets in the current frame is compared with the baseline SNR value measured against targets at the same distance under standard test conditions during factory calibration, and the attenuation is calculated. This indicator comprehensively reflects the degree of SNR degradation caused by both environmental factors and the radar's own performance.

[0054] The four parameters mentioned above are combined into a single environmental complexity index through a weighted summation. The weights of the four parameters are configurable, with equal weights by default to ensure that each parameter contributes equally to the environmental complexity. The environmental complexity index value is normalized to an easily understandable and comparable range (e.g., between 0 and 1), with a higher value indicating a less favorable environment for radar detection.

[0055] Step S4: Using a fixed reference scatterer with known location and stable reflection characteristics in the deployment scenario, the detection quality of the radar is evaluated online, and detection quality parameters are extracted. These detection quality parameters include effective detection rate, ranging consistency index, false alarm density, and detection stability index. Specifically, step S4 evaluates the following indicators: ① Effective Detection Rate: Several reference scatterers with fixed locations are pre-set within the radar monitoring area (such as building corners, streetlights, specially installed corner reflectors, etc., whose precise locations are calibrated and recorded during the deployment phase). The proportion of these reference scatterers successfully detected in the current frame is calculated. Since the reference scatterers are fixed, a continuous decrease in the detection rate eliminates the uncertainty of "whether the target is present," and can directly reflect the attenuation of the radar's detection capability.

[0056] ② Ranging Consistency: For a successfully detected reference scatterer, calculate the statistical dispersion of the deviation between its current frame ranging value and the deployment calibration value. An increase in this dispersion indicates a deterioration in ranging accuracy, which may be related to degradation of the linearity of the transmitted signal, a decrease in the signal-to-noise ratio, or drift in signal processing parameters.

[0057] ③ False Alarm Density: The number of all threshold-crossing points in the constant false alarm rate (CFRD) detection output of the current frame's range-Doppler spectrum is counted. The number of valid points that can be correlated with the real target or reference scatterer is subtracted to obtain the suspected false alarm points. Dividing the suspected false alarm points by the total number of detection units yields the false alarm density index. An increased false alarm density indicates that noise or interference levels have caused the detection threshold to be triggered frequently, leading to a decrease in the reliability of the detection results.

[0058] ④ Detection Stability: Tracking the continuity of target detection across multiple consecutive frames. Statistically calculate the proportion of frames in the most recent observation window where the target was consistently and stably detected. A decrease in this proportion indicates intermittent or flickering target detection, typically related to signal quality fluctuations or the target being at the detection sensitivity threshold. The detection quality data set includes the above four indicators.

[0059] Step S5: Convert each indicator in the transmit link performance parameters, receive link performance parameters and detection quality parameters into a percentage-based individual score through a piecewise linear mapping rule. Merge each individual score according to a hierarchical weight allocation. Apply decoupling compensation based on the environmental complexity index to the detection dimension score to obtain a standardized health score. In some embodiments, the hierarchical weight allocation in step S5 includes: The individual scores are categorized into three dimension groups: radio frequency (RF), environment, and detection. First, a weighted average is calculated within each dimension group according to the weights within the group to obtain the overall dimension score. Then, the overall scores of the RF dimension, environment dimension, and detection dimension are weighted and summed according to the weights between dimensions. The weights between dimensions of the RF dimension are greater than those of the detection dimension, and the weights between dimensions of the detection dimension are greater than those of the environment dimension. The decoupling compensation is as follows: when the environmental complexity index exceeds a preset threshold, positive compensation is applied to the comprehensive score of the detection dimension, and the compensation magnitude increases with the increase of the environmental complexity index, and the compensated score does not exceed the full score.

[0060] This is the second core technological innovation of this solution, which specifically includes: ① Constructing an evaluation index system: The constructed radar health evaluation index system covers three dimensions and ten primary indicators: Radio frequency dimension: reflects the working status of the radar hardware itself; Transmit link linearity score: derived from the gain deviation and nonlinearity intensity parameters monitored by the transmit link, reflecting the linearity status of the power amplifier; IQ balance score: derived from IQ amplitude imbalance and phase imbalance parameters, reflecting the state of the IQ modulator; Phase noise score: The phase noise measurement values ​​are derived from the mid-range and far-end frequency offsets, reflecting the spectral purity of the local oscillator signal; Receive link gain score: derived from the receive link gain drift, reflecting the stability of the receive link gain; Image rejection score: derived from the image rejection ratio of the receiver link, reflecting the rejection capability of the receiver link IQ mixer.

[0061] Environmental dimension: Reflects the current external conditions of the radar; Noise environment score: derived from the average background noise energy, reflecting the background noise level of the electromagnetic environment; Clutter environment score: derived from clutter spread rate, reflecting the breadth of clutter coverage in the environment.

[0062] Detection dimensions: Reflecting the actual effectiveness of radar under current conditions: Effective detection score: derived from the detection rate of the reference scatterer; Ranging accuracy score: derived from the ranging consistency index; False alarm control score: derived from the false alarm density index.

[0063] ② Individual indicator scoring function: Each primary indicator defines a mapping rule from its original measurement value to a standardized score (e.g., 0 to 100 points). The design of the mapping rule follows the following logical principles: First, the factory calibration value is used as the benchmark for full marks: when the current measured value of a certain indicator equals the factory calibration value, that indicator receives full marks. Second, a deviation tolerance range is set: when the measured value deviates from the calibration value but the deviation is within the normal fluctuation range, the score decreases slowly, meaning that each unit of deviation corresponds to a smaller deduction. Third, an accelerated deduction zone is set: when the measured value exceeds the normal range and enters the warning zone, the score decreases more rapidly, meaning that the same unit of deviation corresponds to a larger deduction, reflecting the accelerated accumulation of performance degradation risk. Finally, a failure threshold is set: when the measured value approaches or reaches the preset performance failure threshold, the score drops below the passing line (e.g., below 60 points), clearly indicating that the indicator no longer meets the minimum usage requirements.

[0064] Taking the transmit link linearity rating as an example: At the factory, there is a baseline value for the third-order nonlinear strength. The system-set warning threshold corresponds to a score dropping to 80 points, and the failure threshold corresponds to a score dropping to 60 points. When the current measured value moves from the baseline value towards the warning threshold, the score drops from 100 points to 80 points at a relatively slow rate; when the measured value exceeds the warning threshold and continues to move towards the failure threshold, the score drops from 80 points to 60 points at a steeper rate. This piecewise linear mapping rule is simple, interpretable, and can effectively distinguish between "normal fluctuations" and "trend degradation."

[0065] ③ Multi-dimensional weighted fusion: Since different dimensions and indicators have different abilities to answer the ultimate question of "whether the radar is healthy", they need to be assigned differentiated weights. This invention adopts a hierarchical weight allocation method: first, the relative importance weights of the three dimensions of radio frequency, environment, and detection are determined at the dimensional level, and then the relative importance weights of each primary indicator are determined within each dimension.

[0066] The weighting follows these principles: the radio frequency (RF) dimension has a higher weight than the environmental dimension because the state of the radar's hardware itself reflects the true health of the equipment better than the external environment: harsh environments are temporary and recoverable, while hardware degradation is continuous and irreversible; within the RF dimension, the transmit link indicators have a slightly higher weight than the receive link indicators because the quality of the transmitted signal directly determines the upper limit of the detection effect, and the power devices in the transmit link usually age faster than the devices in the receive link; within the detection dimension, the effective detection score has the highest weight because it most directly answers the core question of "whether the radar can still complete basic detection tasks".

[0067] The recommended weighting scheme is as follows: the radio frequency dimension accounts for half of the overall health score, the environmental dimension accounts for one-fifth, and the detection dimension accounts for three-tenths. The sub-indicators within each dimension are further weighted according to the above principles.

[0068] ④ Overall Health Score Calculation: The overall health score of the machine is a weighted sum of the scores of each dimension and indicator: First, within each dimension, the scores of each indicator of that dimension are weighted and summed according to their respective weights to obtain the dimension score; then, the dimension scores of the three dimensions are weighted and summed according to their dimension weights to obtain the final overall health score of the machine.

[0069] This comprehensive score is mapped to four health levels: Health (score not lower than 85): All radar performance indicators are normal, no need to pay attention; Sub-health (score between 70 and 85): Some indicators deviate slightly from the nominal value. It is recommended to include them in the observation list, but this will not affect normal use. Degradation (score between 60 and 70): The radar performance has significantly deteriorated, and a maintenance and inspection plan is recommended. Abnormal (score below 60): Radar detection capability is severely reduced, requiring immediate shutdown and maintenance.

[0070] ⑤ Decoupling of Health Score from the Environment: Since the detection quality index is affected by both the radar's own performance and the external environment, if environmental factors are not isolated, situations may occur where "severe weather conditions cause a sharp drop in the health score, leading to misjudgment as equipment failure." Therefore, this invention introduces a decoupling mechanism, the core logic of which is: When the environmental complexity index exceeds a preset high environmental pressure threshold, a positive compensation is applied to the scores of each indicator in the detection dimension. The amount of compensation is proportional to the extent to which the environmental complexity index exceeds the threshold. The compensation coefficient is obtained based on regression analysis of a large amount of measured data, and its physical meaning is: "the expected decrease in the score of the detection indicator due to pure environmental factors for each unit increase in environmental complexity." Through this compensation, the scores "suppressed by the environment" are restored to their true hardware level under harsh environmental conditions, ensuring that the health score reflects the radar's own health status, rather than external environmental conditions. Step S6: Select key degradation indicators from various indicators, perform preprocessing on the historical time series data of key degradation indicators, and then perform sliding window linear regression to extract the degradation rate. Based on the degradation rate and the preset failure threshold, estimate the remaining normal operating time of the radar. In some embodiments, the preprocessing in step S6 includes: Outliers are removed from the historical time-series data of key degradation indicators, and data points that deviate from the median of the sliding window by more than a preset multiple of the median absolute deviation are replaced with interpolated values ​​within the window. Smooth the data after outlier removal using a moving average; The smoothed data is periodically normalized to eliminate regular fluctuations caused by periodic environmental factors; the window length of the sliding window linear regression is a preset number of days.

[0071] In some embodiments, the criteria for classifying anomalies as isolated anomalies or clustered anomalies in step S7 include: If an abnormal event involves only a single node within the reference group within a preset time window, and no other indicators of that node show a chain of abnormalities, it is determined to be an isolated abnormality. If an abnormal event involves at least three nodes within a reference group exhibiting the same directional shift in the same indicator, it is determined to be a cluster-type abnormality.

[0072] This is the third core technological innovation of this solution, specifically including: ① Selection of key degradation indicators: From all monitoring indicators, the indicators most sensitive to radar hardware performance degradation were selected as key degradation indicators for trend analysis. The selection criteria were based on the typical change pattern of the indicator throughout the radar's entire life cycle, specifically meeting two conditions: first, there must be a clear causal relationship between the change of the indicator and the physical degradation process of the device; second, the indicator must exhibit an observable monotonic change trend within the normal service life of the radar.

[0073] Based on the above principles, three key degradation indicators were selected: Third-order nonlinearity: During long-term use, the linearity of a power amplifier will gradually deteriorate due to the increase in transistor junction temperature and the degradation of the gate oxide layer, and the third-order intermodulation distortion will increase monotonically. Mid-frequency offset phase noise: During long-term operation, the phase noise of the reference crystal and voltage-controlled oscillator in the local oscillator phase-locked loop will slowly deteriorate due to device aging and decreased frequency stability. IQ amplitude imbalance: The mixer gain matching in the IQ modulator gradually deteriorates with the accumulation of temperature cycling and device aging, and the degree of amplitude imbalance tends to increase.

[0074] ② Time-series data processing: For each key degradation indicator, its complete historical time-series data is maintained in the monitoring platform. To avoid random fluctuations from a single measurement interfering with trend judgment, the following preprocessing is performed on the raw time-series data: First, outlier removal: isolated data points that deviate significantly from the mean of their immediate and neighboring regions are marked and removed, and are not included in trend analysis; Second, moving average smoothing: For the sequence after removing outliers, take a sliding window of appropriate width, replace the original value with the arithmetic mean of the data in the window, filter out short-term random noise components, and make the potential long-term trend more obvious. Third, periodic normalization: Considering that radar performance parameters have daily periodic fluctuations caused by diurnal temperature variations, data from the same time period on different dates are grouped and normalized to eliminate the interference of periodic components on long-term trend judgment.

[0075] ③ Degradation rate estimation: The degradation rate is estimated using a sliding window linear regression method. Specifically, based on the preprocessed time-series data, a sufficiently long time window is selected, typically 30 days, and a best-fit line is used to describe the relationship between the index value and time within that time window.

[0076] The slope of the fitted line is the degradation rate, representing the average daily change in the indicator value. Simultaneously, the goodness of fit is calculated to measure whether the indicator value sequence truly exhibits a linear trend within the observation window, or merely fluctuates randomly around a certain mean.

[0077] The determination is based on the degree of goodness of fit: If the goodness of fit is low (below 0.5), it indicates that the index does not show a significant linear degradation trend within the observation window and is currently in a stable state. If the goodness of fit is high and the degradation rate is positive, it is determined that the index does indeed have a statistically significant monotonic degradation trend. If the goodness of fit is high but the degradation rate is negative or close to zero, the index is considered to be in a stable or even improved state.

[0078] ④ Remaining Normal Operating Time Estimation: When a key degradation indicator is determined to have a "significant degradation trend," the remaining normal operating time of that indicator is estimated based on the degradation rate and failure threshold. The estimation logic is as follows: subtract the current value from the preset failure threshold for that indicator, and then divide by the degradation rate. The quotient is the number of days expected to reach the failure threshold linearly extrapolated from the current degradation rate. This number of days is the estimated remaining normal operating time from the perspective of that indicator.

[0079] The monitoring platform automatically generates a predictive maintenance alert when any of the following conditions are met: If the estimated remaining normal working time for any key degradation indicator is shorter than the preset maintenance lead time, it is generally recommended to use 7 days. The current value of any key degradation indicator has exceeded 80 percent of the failure threshold.

[0080] The alarm information includes: the node identifier and installation location that triggered the alarm, the reason for the alarm, the current value and degradation rate of the indicator, the estimated remaining normal working time based on the current trend, and the recommended latest maintenance time window.

[0081] Step S7: Divide multiple radars into reference groups according to preset dimensions, calculate the statistical deviation of each radar in the reference group, determine abnormal nodes based on the statistical deviation, and classify the anomalies into isolated anomalies or group anomalies according to the number of abnormal nodes and the direction of deviation.

[0082] In multi-node deployment scenarios, the platform leverages the premise that radar nodes from the same batch, of the same model, and operating conditions should have similar performance characteristics to achieve cross-node anomaly comparison and detection. Specifically, this includes: ① Reference group construction: The platform automatically groups radar nodes based on their model, firmware version, deployment date, and deployment environment type. Radar nodes within the same group constitute a reference group, and theoretically, nodes within the group should have similar performance baselines. For example, radars of the same model manufactured in the same batch and deployed at the same time on the same side of the same perimeter wall in the same industrial park will naturally form a reference group.

[0083] ② Anomaly Detection Algorithm: For each monitoring indicator of each node within the reference group, calculate the within-group statistical mean and within-group statistical standard deviation. For each node within the group, calculate the degree to which each indicator value deviates from the within-group mean, i.e., how many standard deviations the deviation is equivalent to. When the deviation of a node's indicator from the mean exceeds a preset deviation multiple threshold (3 standard deviations is recommended), the node is determined to have a statistically significant anomaly in that indicator.

[0084] ③ Anomaly type identification: Cross-node anomalies are divided into two categories, which are distinguished by the combination of anomaly indicators: Isolated anomaly: Only a single node exhibits an abnormal indicator, while the same indicator on all other nodes in the same group remains within the normal range. This strongly suggests an individual hardware failure or installation problem on that node (such as partial damage to the node's power amplifier, water ingress into the antenna connector, or installation angle deviation). A targeted alarm, "Single-point inspection of this node is recommended," is generated.

[0085] Cluster-type anomaly: Multiple nodes in the same group exhibit synchronized deviations in the same metric (even if the deviation of individual nodes does not reach the strict threshold of 3 times the standard deviation, the group as a whole shows a consistent deviation trend). This situation tends to point to external environmental factors or firmware version defects. For example, a new strong electromagnetic interference source has recently appeared in the area, or a parameter configuration in the firmware is not suitable for the current environment. A targeted alarm is generated suggesting "investigating external environmental factors or checking the firmware version" to avoid maintenance personnel performing ineffective, one-by-one repairs on each node.

[0086] The performance monitoring data frame format (communication protocol) is standardized, and data transmission between the radar and the monitoring platform uses this format. The frame structure includes the following fields: Table 1: Fields and their descriptions

[0087] The length of the entire frame is controlled to be on the order of 100 bytes to reduce the occupation of communication bandwidth and enable it to be transmitted efficiently on low-bandwidth IoT links.

[0088] In summary, this invention proposes a three-layer progressive architecture: a perception layer, an evaluation layer, and a platform layer. The perception layer, deployed at each radar node, is responsible for extracting RF performance parameters of the transmit and receive links online, while simultaneously collecting environmental condition parameters and detection quality parameters, without interrupting normal detection tasks. The evaluation layer, deployed at the radar or edge computing nodes, is responsible for normalizing the multi-dimensional indicators acquired by the perception layer and, through hierarchical weight allocation and decoupling compensation, fusing and outputting a standardized percentage-based health score. The platform layer, deployed in the cloud or on edge servers, is responsible for aggregating health data from multiple nodes, estimating the degradation rate and remaining normal operating time of each node based on time-series trend analysis, and achieving automatic cross-node anomaly screening by constructing reference groups and statistical deviation detection. The three layers communicate with each other through a standardized performance monitoring data frame protocol.

[0089] The following section provides a complete and clear explanation of this technical solution, taking into account specific application scenarios and engineering implementation details.

[0090] Example 1: Complete self-sensing and health assessment process for a single radar node.

[0091] This embodiment takes a frequency modulated continuous wave millimeter-wave radar operating in the 77 GHz band and using a three-transmitter, four-receiver antenna configuration as an example to fully illustrate the entire process of a single radar node from factory calibration, power-on initialization, continuous monitoring and operation to abnormal response and handling.

[0092] Factory calibration phase: Before leaving the factory, the radar is calibrated using standard instruments such as a vector network analyzer and a spectrum analyzer under a standard laboratory ambient temperature of 25 degrees Celsius, completing the benchmark value calibration of all ten indicators. For the transmit link, the radar transmits a frequency-modulated continuous wave signal at its rated power, which is attenuated by a standard load before being input to the instruments. The benchmark shape parameters of the AM-AM and AM-PM curves are recorded, and the benchmark values ​​for linear gain, third-order nonlinear intensity, fifth-order nonlinear intensity, and memory effect intensity are extracted. Simultaneously, the benchmark values ​​for I / Q amplitude imbalance and I / Q phase imbalance, as well as the benchmark values ​​for phase noise power in the near-end, mid-range, and far-end frequency offset intervals, are recorded. For the receive link, a single-tone signal and a modulated signal of known power are injected through a signal source, and the benchmark values ​​for receive link gain, image rejection ratio, and noise figure are recorded. Regarding environmental conditions, the benchmark values ​​for the average background noise energy and clutter spread rate are recorded in a microwave anechoic chamber. Regarding detection quality, a standard corner reflector was installed at a fixed position of the radar in the anechoic chamber to record the effective detection rate benchmark, ranging consistency benchmark, false alarm density benchmark, and detection stability benchmark. All benchmark values, along with environmental conditions such as temperature, humidity, and air pressure during calibration, were written into the radar's built-in non-volatile memory. Additionally, a miniature temperature sensor was installed near the power amplifier chip package to record the power amplifier temperature benchmark value during calibration, which was used for subsequent temperature compensation.

[0093] Power-on initialization phase: After each power-on, before entering normal detection tasks, the radar automatically performs a full-parameter self-check. The radar first transmits a short test waveform with rated parameters, simultaneously activating the auxiliary receiving channel to collect the transmitted link feedback signal, extracting the current transmitted link index value according to the same processing procedure as the factory calibration phase; then, a preset calibration sequence is injected into the receiving link front end through an internal RF switch, extracting the current receiving link index value; next, ambient background noise is collected in silent mode; finally, a detection waveform is transmitted to detect a fixed reference scatterer. The self-check values ​​are compared with the factory baseline values: if all index values ​​are within the allowable range of the baseline values, the radar is considered normal, the self-check results are recorded, and it enters normal operating mode; if one or more index values ​​exceed the allowable range but do not reach the accelerated deduction range, it is recorded as a sub-healthy state and reported to the platform, simultaneously entering normal operating mode but automatically increasing the frequency of subsequent monitoring data collection and reporting; if any index value enters the accelerated deduction range or failure range, it pauses entering normal operating mode, reports an alarm message to the platform, and awaits confirmation from maintenance personnel or a remote reset command.

[0094] Continuous monitoring operation phase: After the radar enters normal operating mode, the following parallel operations are performed synchronously during the transmission of each frame of frequency-modulated continuous wave detection waveform: Operation 1: Transmit link coupling loopback monitoring.

[0095] A 30-decibel directional coupler at the power amplifier output couples out a sample signal with approximately one-thousandth of the power from the transmitted signal. Considering that the gain and linearity characteristics of the power amplifier chip will drift under different temperatures, this embodiment introduces a temperature compensation mechanism: while the auxiliary receiving channel is acquiring the retrieval signal, it reads the temperature sensor value near the power amplifier, and fine-tunes the coupling coefficient according to the difference between the current temperature and the calibrated reference temperature: for every ten degrees Celsius the temperature deviates from the reference temperature, the nominal value of the coupling coefficient is compensated for a deviation of 0.1 decibels according to a preset temperature-coupling coefficient lookup table. The temperature-compensated retrieval signal is normalized and cross-correlated with the original baseband transmit signal to achieve sub-sampling point level time-domain alignment. Then, the slope of the linear region fitting of the AM-AM curve is extracted as the current linear gain value, the gain compression of the AM-AM curve in the high input power region is extracted as the third-order nonlinear intensity, the residual nonlinearity in the deeper compression region is extracted as the fifth-order nonlinear intensity, and the hysteresis loop area is extracted from the joint two-dimensional distribution of AM-AM and AM-PM as the memory effect intensity. The root mean square amplitude of the I branch and the cross-correlation coefficient of the two branches are calculated respectively to obtain the I / Q amplitude imbalance and phase imbalance. The phase error sequence is extracted and the power spectral density is calculated. The phase noise power values ​​of the near end, middle end and far end are obtained by integration in three frequency bands: 1 to 10 kHz, 10 to 100 kHz and 100 kHz to 1 MHz.

[0096] Operation 2: Receive link calibration sequence injection monitoring.

[0097] The baseline insertion frequency of the calibration sequence is once every sixty seconds. Simultaneously, the background continuously monitors the rate of change of transmit link monitoring indicators: if the rate of change of the third-order nonlinear intensity or mid-range phase noise power value exceeds 5% per hour within the last five minutes, the calibration sequence insertion frequency is automatically increased to once every ten seconds; once the rate of change falls below the threshold, the insertion frequency gradually returns to the baseline value. This linked adaptive scheduling mechanism ensures that the receive link status is also closely tracked when the transmit link shows signs of rapid degradation, avoiding undetected degradation due to synchronous degradation of the receive link. During each calibration, the calibration sequence acquired by the receive link is matched and filtered with a locally stored copy of the original sequence, and the gain drift is obtained from the ratio of the filtered peak amplitude to the reference value; a fast Fourier transform is performed on the acquired sequence, and the power integral value in the image frequency interval is divided by the power integral value in the signal interval to obtain the image rejection ratio; within the silent window between two adjacent calibrations, the power spectrum of the receive channel output is estimated, and the noise figure is calculated based on the background noise power when there is no signal input.

[0098] Operation 3: Environmental condition perception.

[0099] Before performing two-dimensional constant false alarm rate (CFAR) detection on the range-Doppler spectrum of the current frame, the raw spectral data is analyzed: the average power of all range and velocity cells excluding the cells containing detected targets is calculated as the mean background noise energy; using the mean noise plus six dB as the clutter decision threshold, the proportion of range cells exceeding this threshold is calculated as the clutter spread rate; paired mirror-symmetric peaks are searched in the range-Doppler spectrum, and the ratio of the mirror peak power to the main peak power is calculated as the multipath reflection saliency; a pre-calibrated fixed reference scatterer in the scene (such as a concrete utility pole 35 meters from the radar) is located, its current signal-to-noise ratio (SNR) is read, and compared with the SNR benchmark value of the scatterer recorded under historical clear weather conditions to obtain the target SNR attenuation. The four parameters are weighted and summed according to preset weights, and the output is an environmental complexity index ranging from zero to one.

[0100] Operation 4: Self-assessment of detection quality.

[0101] Using the same fixed reference scatterer as in Operation 3, the proportion of frames in the most recent 300 frames where the scatterer was successfully detected is counted as the effective detection rate; the deviation between the ranging value and the calibrated distance of 35 meters at each detection in these 300 frames is calculated, and the root mean square value is obtained as the ranging consistency index; the number of false detection points in each frame other than the detection point corresponding to the scatterer is counted, and the average value of the 300 frames is taken as the false alarm density; the variance of the detection signal-to-noise ratio of the scatterer in the 300 frames is calculated as the detection stability index.

[0102] The above four operations are executed in parallel after each frame of detection is completed. The output of each operation, along with the current timestamp, is encapsulated into a monitoring data frame. The radar reports the data frame to the evaluation layer via an onboard Ethernet interface or wireless communication module using a message queue telemetry transmission protocol.

[0103] Health assessment execution: The assessment layer can be an embedded processor local to the radar or an edge computing node deployed near the radar. After receiving the monitoring data frames reported by the perception layer, the assessment layer performs health score calculation according to the following steps: Step 1, Individual index normalization. A piecewise linear mapping rule is applied to each of the ten indices. Taking the transmit link linearity as an example, let the factory-calibrated linear gain baseline value be G0, and the current gain value be G. When the deviation between G and G0 does not exceed 0.5 dB, the score is 100 points; when the deviation is between 0.5 and 1.5 dB, the score linearly decreases from 100 points to 70 points; when the deviation is between 1.5 and 3 dB, the score accelerates from 70 points to 40 points; when the deviation exceeds 3 dB, the score is fixed at 30 points and an alarm is triggered. The other nine indices each have an independent mapping parameter table. The mapping parameters are predetermined during factory calibration based on the hardware characteristics and design tolerances of the radar model and stored in non-volatile memory. Step 2, Intra-dimensional fusion. Five metrics in the radio frequency (RF) dimension are weighted and averaged according to preset weights: transmit link linearity (0.35), IQ balance (0.25), phase noise (0.20), receive link gain (0.10), and image suppression score (0.10) to obtain the RF dimension comprehensive score. The noise environment score and clutter environment score in the environment dimension are weighted equally (0.50 each) to obtain the environment dimension comprehensive score. The effective detection score, ranging accuracy score, and false alarm control score in the detection dimension are weighted (0.40), (0.35), and (0.25) to obtain the detection dimension comprehensive score. The third step is decoupling compensation. The environmental complexity index of the current frame is compared with three preset thresholds: 0.3, 0.5, and 0.7. When the environmental complexity index is below 0.3, no compensation is triggered; when it is between 0.3 and 0.5, the detection dimension comprehensive score receives 5 points of positive compensation; when it is between 0.5 and 0.7, it receives 10 points of positive compensation; and when it exceeds 0.7, it receives 15 points of positive compensation. The compensated score does not exceed 100 points. The fourth step is inter-dimensional fusion. The comprehensive score of the radio frequency dimension, the compensated comprehensive score of the detection dimension, and the comprehensive score of the environmental dimension are weighted and summed according to weights of 0.5, 0.3, and 0.2 respectively to obtain the final health score out of 100. The health level is determined based on the score: 85 points and above is healthy, 70 to 85 points is sub-healthy, 60 to 70 points is deteriorating, and below 60 points is abnormal. The evaluation results are encapsulated into an extended data frame along with the original indicator data and reported to the platform layer.

[0104] Anomaly Response and Alarm Escalation: When the health score drops from a healthy level to a sub-healthy level, or when the score of any single indicator drops from 100 to below 85, the radar automatically encrypts the frequency of data collection and reporting from the baseline value (e.g., one frame every ten seconds) to one frame every two seconds, and simultaneously sets a high-priority alarm flag in the data frame. When the health score enters a deterioration or anomaly level, the radar, while encrypting the report, pushes an alarm notification to the maintenance personnel's terminal through the platform layer. The alarm information includes the current score, the three most significantly deteriorated indicators and their deviation from the baseline value, and the estimated remaining normal working time based on the degradation rate. After confirming the alarm, maintenance personnel can choose to remotely reset, remotely switch to a backup working parameter set, or dispatch on-site maintenance.

[0105] Example 2: Multi-node cluster deployment and cross-node anomaly detection for perimeter security in the park.

[0106] This embodiment uses a perimeter security system of an industrial park as an example to illustrate the operation of the platform layer in a multi-node cluster deployment scenario. The park's perimeter is approximately four kilometers long, and thirty-six identical millimeter-wave radars are evenly deployed along the perimeter wall at intervals of approximately 110 meters, covering the east, west, south, and north sections of the wall. All radars are connected to an edge server deployed in the park's central computer room via a fiber optic ring network, on which the platform layer software runs.

[0107] Data aggregation and storage: Each of the thirty-six radars operates according to the process described in Example 1, reporting one extended monitoring data frame to the edge server every ten seconds. After receiving the data frames from all nodes, the message broker module of the edge server writes the data in parallel to the time-series database according to the node identifier. The time-series database is indexed by node identifier and timestamp, supporting flexible queries by node, time period, and indicator type. The data retention strategy is set as follows: original frame-by-frame data is retained for the most recent ninety days, and daily aggregated statistical values ​​are retained for three years.

[0108] Automatic Reference Group Segmentation and Homogeneity Verification: Upon initial startup, the platform layer automatically performs reference group segmentation based on the metadata of each node. The segmentation dimensions are as follows: radar model code, firmware major version number, deployment date grouped by quarter, and deployment environment type divided by perimeter segment (East, West, South, and North). Since the thirty-six radars are identical in model and firmware version, they are ultimately divided into four reference groups of nine radars each, based on perimeter segment. After the reference group segmentation, the platform layer performs homogeneity verification on each reference group: a normality test is performed on the distribution of various health indicators within the group, and the within-group variance is calculated. If the within-group variance of a certain indicator exceeds a preset variance threshold: for example, if the within-group standard deviation of the comprehensive score of the radio frequency dimension of each node exceeds eight points, it indicates that the internal differences within the reference group are too large, possibly indicating the inclusion of nodes from different batches or under different operating conditions. In this case, the platform layer automatically performs secondary clustering on the group: using deployment date, average daily temperature exposure, etc., as auxiliary features, the K-means clustering algorithm is used to split the original group into two subgroups, and the within-group variance is reassessed within each subgroup. After splitting, if the variance of each subgroup falls below the threshold, subsequent anomaly detection is performed within the subgroup; if the variance still exceeds the threshold, the group is marked as a non-homogeneous group, and only trend analysis is performed on the nodes within the group without performing Z-score anomaly detection, thus avoiding misjudgment due to an unreasonable reference group.

[0109] Cross-node anomaly comparison detection: After receiving a new frame of monitoring data, the platform layer performs incremental anomaly detection on the reference group to which the data belongs. For each key indicator—third-order nonlinear intensity, mid-segment phase noise power, I / Q amplitude imbalance, gain drift, and image rejection ratio—the mean and standard deviation within the group are calculated, and then the Z-score of the current node is calculated. When the absolute value of the Z-score for any indicator exceeds three, the node is marked as a statistical anomaly, and the name of the anomaly indicator, the Z-score value, and the trigger time are recorded. The platform layer maintains a sliding time window with a length of thirty minutes, and performs aggregate analysis on anomaly events within the window. If there are only anomaly events from the same node within the window, and other indicators of that node do not show cascading anomalies, it is determined to be an isolated anomaly, generating a suspected individual fault alarm. It is recommended that maintenance personnel prioritize checking the heat dissipation conditions, power supply quality, and physical installation status of that node. If the same index from three or more nodes within the same reference group is detected to have the same directional shift (i.e., the Z-scores have the same sign and the amplitude exceeds two), it is determined to be a cluster anomaly, generating a suspected systemic problem alarm. It is recommended that maintenance personnel prioritize investigating regional electromagnetic interference sources, firmware configuration inconsistencies, or recent environmental changes.

[0110] Visualization and Operations Decision Support: The platform layer provides a browser-based dashboard interface. The main dashboard displays the real-time health status of 36 radars using a map base overlaid with node location markers. Each node is represented by a circular icon, with icon colors coded according to health level: green for healthy, yellow for sub-healthy, orange for deteriorating, and red for abnormal. Clicking on any node icon expands the detailed panel for that node, displaying the current health score trend graph, bar charts of comprehensive scores across various dimensions, and historical curves for ten individual indicators. The platform layer also provides a dedicated page for deterioration trends, displaying the estimated remaining normal working time for all nodes in a list format, sorted from shortest to longest remaining time, and using colors to indicate urgency. Operations personnel can use this page to quickly identify the nodes requiring the most attention and plan upcoming maintenance schedules.

[0111] Example 3: Verification of the effect and technical expansion of the decoupling mechanism.

[0112] This embodiment illustrates the actual effect of the decoupling compensation mechanism through comparative experiments and introduces an extended temperature adaptive benchmark update method.

[0113] Comparison of Decoupling Effects: A radar with proven performance after factory calibration was selected and deployed in a fixed location in an open outdoor environment. During a two-week observation period, at least 48 hours of continuous monitoring data were collected under four typical weather conditions: sunny, cloudy, moderate rain, and dense fog. Without decoupling compensation, the radar's health score remained stable between 92 and 95 points under sunny and cloudy conditions, meeting expectations. However, under moderate rain conditions, due to the attenuation and scattering of millimeter-wave signals by rainwater, the target signal-to-noise ratio decreased significantly, and the effective detection rate dropped from over 99% under sunny conditions to approximately 88%. Consequently, the overall detection dimension score decreased from 95 points to approximately 72 points, ultimately lowering the health score to approximately 75 points, entering the sub-healthy range and generating false alarms. The situation was even more severe under dense fog conditions, with the final health score dropping to approximately 68 points, entering the degradation range. After enabling decoupling compensation, under the same conditions of moderate rain and dense fog, the environmental complexity index was calculated to be 0.48 and 0.63 respectively, triggering the corresponding positive compensation of 10 points. The comprehensive score of the detection dimension was reasonably corrected, and the final health scores were 84 and 82 respectively, which correctly reflected the real situation of normal equipment status but harsh environmental conditions, and effectively eliminated false alarms caused by environmental factors.

[0114] Temperature-Adaptive Baseline Update Mechanism: The characteristics of radar RF devices undergo slow but irreversible aging drift over time. For example, after thousands of hours of cumulative operation, the linear gain of a power amplifier chip may permanently decrease by a few tenths of a decibel. This is a normal physical aging phenomenon, not a fault. If the absolute baseline value at the factory is always used as a reference, the health score will remain consistently low in the middle and late stages of aging, generating a large number of false alarms. To solve this problem, this embodiment introduces a temperature-adaptive baseline sliding window update mechanism in the evaluation layer. Specifically, the evaluation layer maintains a baseline history window for each temperature range, with the temperature range divided in five-degree Celsius increments. When the system determines that the radar is in a healthy state (i.e., the current health score is 85 or higher and the environmental complexity index is below 0.3), the ten index values ​​of that frame are assigned to the baseline history window of the corresponding temperature range according to the current power amplifier temperature. Each temperature range window retains index data collected within the last ninety days. When the amount of data within a temperature range window exceeds the preset minimum sample size, the median of the data within that window is used as the dynamic baseline value for that temperature range, replacing the factory baseline value for subsequent piecewise linear mapping. If a certain indicator deviates from the dynamic baseline value by more than three times the standard deviation within the temperature range window during a certain data acquisition, that frame of data will not be included in the window update to prevent abnormal data from contaminating the baseline. Furthermore, the update of the dynamic baseline value is also subject to an aging rate limit constraint: if the cumulative deviation between the dynamic baseline value and the factory baseline value exceeds the allowable drift amount given in the device's datasheet for its entire lifecycle, updates will cease, and a device aging over-limit alarm will be generated, prompting maintenance personnel to consider replacing the device or the entire device. This mechanism allows the health score to automatically adapt to normal device aging drift during long-term operation, focusing the score's sensitivity on abnormal accelerated degradation events, significantly reducing the false alarm rate in the later stages of maintenance.

[0115] Example 4: Fault root cause localization through cross-index correlation analysis.

[0116] This embodiment describes how the platform layer can use cross-index correlation analysis to further locate the root cause of the fault after detecting an anomaly, providing more accurate guidance for maintenance personnel to make troubleshooting decisions.

[0117] When a node's health score triggers an anomaly alarm, the platform automatically performs pairwise correlation analysis on the time series of ten individual indicators for that node over the past 72 hours. Specifically, for any two indicators, the Pearson correlation coefficient is calculated, and their temporal relationship is examined to determine whether the degradation of indicator A precedes the degradation of indicator B. For example, if the absolute value of the correlation coefficient between third-order nonlinear strength and the effective detection rate exceeds 0.7, and the degradation of third-order nonlinear strength precedes the degradation of the effective detection rate by approximately 20 to 40 minutes, then the deterioration of third-order nonlinear strength is determined as the cause, and the decrease in the effective detection rate as the effect. The platform then compiles the correlation analysis results into a causal relationship diagram, marking the root cause indicator at the forefront of the causal chain. The alarm notification includes the name of this root cause indicator, its degradation start time, and degradation magnitude, helping operations personnel quickly focus on the troubleshooting direction. Furthermore, the platform cross-validates the results of cross-indicator correlation analysis with the results of cross-node anomaly detection. For example, if the third-order nonlinear intensity of a node is determined to be a root cause indicator and the node is identified as an isolated anomaly, it strongly suggests that there is an individual problem with the power amplifier device or heat dissipation conditions of that node. If the mid-segment phase noise of multiple nodes within the same reference group is simultaneously determined to be a root cause indicator and is identified as a cluster anomaly, it strongly suggests the presence of a new high-power radio frequency interference source in the region where the reference group is located. This cross-validation mechanism significantly improves the accuracy of fault location and the efficiency of operation and maintenance response.

[0118] Secondly, this application proposes a millimeter-wave radar intelligent monitoring platform, such as... Figure 2 As shown, it includes: The data aggregation module is used to receive performance monitoring data frames reported by multiple millimeter-wave radars and write the performance monitoring data frames in parallel into the time-series database according to the node identifier. The performance monitoring data frame includes transmit link performance parameters, receive link performance parameters, environmental condition parameters, detection quality parameters, and health score. The trend analysis module is used to read historical data of key degradation indicators of each radar node from the time series database, perform sliding window linear regression after preprocessing the historical data to extract the degradation rate, and estimate the remaining normal working time of each radar node based on the degradation rate and preset failure threshold. The anomaly detection module is used to automatically divide multiple radars into reference groups according to preset dimensions, calculate the statistical deviation of each radar node in the reference group, determine the abnormal nodes based on the statistical deviation, and classify the anomalies into isolated anomalies or group anomalies according to the number of abnormal nodes and the direction of deviation. The visualization and alarm module is used to display the real-time health status of each radar node on the interactive interface in a color-coded manner, draw the health score trend curve, and generate alarm notifications when an anomaly is detected or the remaining normal working time is lower than a preset threshold.

[0119] In some embodiments, the performance monitoring data frame includes: The node identifier field group contains the radar's unique serial number and model code; the timestamp field group contains the timestamp of the data acquisition time. The RF specification field group includes power amplifier nonlinearity characteristics, I / Q imbalance, phase noise, gain drift, image rejection ratio, and noise figure. The environmental metrics field group includes the environmental complexity index and its sub-parameters; The detection index field group includes effective detection rate, ranging consistency index, false alarm density, and detection stability index; The health score field group includes the comprehensive score of each dimension, the decoupling compensation amount, the final health score, and the health level identifier; After dividing the reference group, the anomaly detection module first verifies the homogeneity of the distribution of each indicator within the reference group. When the variance within the group exceeds the preset variance threshold, it performs secondary clustering subdivision on the reference group before performing anomaly detection.

[0120] Thirdly, this application proposes an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0121] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described above.

[0122] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0123] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0124] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0125] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus / computer devices and methods can be implemented in other ways. For example, the apparatus / computer device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. Multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0126] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0127] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0128] If an integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program may include computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. A computer-readable medium may include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in a computer-readable medium may be appropriately added to or subtracted according to the requirements of legislation and patent practice in a jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media may not include electrical carrier signals and telecommunication signals.

[0129] The above are merely preferred embodiments of the present invention. It should be noted that any modifications and improvements made by those skilled in the art without departing from the present technical solution should also be considered to fall within the scope of protection claimed by the present solution.

Claims

1. A method for self-sensing the multi-dimensional performance of millimeter-wave radar, characterized in that, Includes the following steps: Step S1: Set up a directional coupler between the power amplifier output and the transmitting antenna of the millimeter-wave radar transmitting link. Couple the sample signal from the transmitted signal through the directional coupler and feed it into the auxiliary receiving channel, which shares the local oscillator source with the main receiving channel. Process the sampled digital baseband signal obtained after down-conversion and analog-to-digital conversion of the auxiliary receiving channel to extract the transmitting link performance parameters. These transmitting link performance parameters include the power amplifier nonlinearity characteristics, I / Q imbalance, and phase noise. Step S2: Insert a known calibration sequence into the gap of the radar transmitted waveform in a time-division multiplexing manner. Feed the calibration sequence into the front end of the receiving link through the internal radio frequency path. Perform matched filtering and spectrum analysis on the calibration sequence acquired after passing through the receiving link to extract the receiving link performance parameters, including gain drift, image rejection ratio and noise figure. Step S3: Extract environmental condition parameters from the range-Doppler spectrum generated during normal radar detection. These environmental condition parameters include the mean background noise energy, clutter spread rate, multipath reflection salience, and target signal-to-noise ratio attenuation. The environmental complexity index is obtained by weighted fusion of the various environmental condition parameters. Step S4: Using a fixed reference scatterer with known location and stable reflection characteristics in the deployment scenario, the detection quality of the radar is evaluated online, and detection quality parameters are extracted. These detection quality parameters include effective detection rate, ranging consistency index, false alarm density, and detection stability index. Step S5: Convert each indicator in the transmit link performance parameters, receive link performance parameters and detection quality parameters into a percentage-based individual score through a piecewise linear mapping rule. Merge each individual score according to a hierarchical weight allocation. Apply decoupling compensation based on the environmental complexity index to the detection dimension score to obtain a standardized health score. Step S6: Select key degradation indicators from various indicators, perform preprocessing on the historical time series data of key degradation indicators, and then perform sliding window linear regression to extract the degradation rate. Based on the degradation rate and the preset failure threshold, estimate the remaining normal operating time of the radar. Step S7: Divide multiple radars into reference groups according to preset dimensions, calculate the statistical deviation of each radar in the reference group, determine abnormal nodes based on the statistical deviation, and classify the anomalies into isolated anomalies or group anomalies according to the number of abnormal nodes and the direction of deviation.

2. The method according to claim 1, characterized in that, The process of extracting the nonlinear characteristic index of the power amplifier in step S1 includes: The acquired digital baseband signal and the original baseband transmitted signal are cross-correlated to achieve time-domain alignment. An AM-AM characteristic curve is constructed based on the amplitude relationship between the aligned acquired signal and the original transmitted signal. The linear gain deviation and third-order nonlinear intensity are extracted from the AM-AM characteristic curve. An AM-PM characteristic curve is constructed based on the phase relationship between the aligned acquired signal and the original transmitted signal, and a phase deviation index is extracted from the AM-PM characteristic curve. The intensity index of memory effect is extracted based on the amplitude-phase two-dimensional distribution of the acquired signal.

3. The method according to claim 2, characterized in that, The process of extracting the I / Q imbalance index and phase noise index in step S1 includes: The energy of the acquired digital baseband signal on the in-phase branch and the quadrature branch are calculated separately, and the I / Q amplitude imbalance and I / Q phase imbalance are obtained by cross-correlation calculation of the two branches. The phase error sequence of the acquired digital baseband signal is estimated by power spectrum estimation. The phase noise power value corresponding to each frequency offset interval is obtained by integration in different frequency offset intervals.

4. The method according to claim 3, characterized in that, In step S2, the insertion frequency of the calibration sequence is adjusted in conjunction with the rate of change of the transmit link performance parameters extracted in step S1: when the rate of change of any indicator in the transmit link performance parameters exceeds a preset threshold, the insertion frequency of the calibration sequence is automatically increased; when the rate of change falls back below the preset threshold, it is gradually restored to the baseline insertion frequency.

5. The method according to claim 4, characterized in that, The target signal-to-noise ratio attenuation in step S3 is as follows: using a fixed reference scatterer with a known location and known radar cross-section in the scene, the attenuation of the signal-to-noise ratio of the fixed reference scatterer in the current frame is calculated relative to the historical signal-to-noise ratio reference value of the fixed reference scatterer under clear weather conditions.

6. The method according to claim 5, characterized in that, The hierarchical weight allocation in step S5 includes: The individual scores are categorized into three dimension groups: radio frequency (RF), environment, and detection. First, a weighted average is calculated within each dimension group according to the weights within the group to obtain the overall dimension score. Then, the overall scores of the RF dimension, environment dimension, and detection dimension are weighted and summed according to the weights between dimensions. The weights between dimensions of the RF dimension are greater than those of the detection dimension, and the weights between dimensions of the detection dimension are greater than those of the environment dimension. The decoupling compensation is as follows: when the environmental complexity index exceeds a preset threshold, positive compensation is applied to the comprehensive score of the detection dimension, and the compensation magnitude increases with the increase of the environmental complexity index, and the compensated score does not exceed the full score.

7. The method according to claim 6, characterized in that, The preprocessing in step S6 includes: Outliers are removed from the historical time-series data of key degradation indicators, and data points that deviate from the median of the sliding window by more than a preset multiple of the median absolute deviation are replaced with interpolated values ​​within the window. Smooth the data after outlier removal using a moving average; The smoothed data is periodically normalized to eliminate regular fluctuations caused by periodic environmental factors; the window length of the sliding window linear regression is a preset number of days.

8. The method according to claim 7, characterized in that, The criteria for classifying anomalies into isolated anomalies or clustered anomalies in step S7 include: If an abnormal event involves only a single node within the reference group within a preset time window, and no other indicators of that node show a chain of abnormalities, it is determined to be an isolated abnormality. If an abnormal event involves at least three nodes within a reference group exhibiting the same directional shift in the same indicator, it is determined to be a cluster-type abnormality.

9. A millimeter-wave radar intelligent monitoring platform, characterized in that, include: The data aggregation module is used to receive performance monitoring data frames reported by multiple millimeter-wave radars and write the performance monitoring data frames into the time-series database in parallel according to the node identifier. This performance monitoring data frame includes transmit link performance parameters, receive link performance parameters, environmental operating condition parameters, detection quality parameters, and health score; The trend analysis module is used to read historical data of key degradation indicators of each radar node from the time series database, perform sliding window linear regression after preprocessing the historical data to extract the degradation rate, and estimate the remaining normal working time of each radar node based on the degradation rate and preset failure threshold. The anomaly detection module is used to automatically divide multiple radars into reference groups according to preset dimensions, calculate the statistical deviation of each radar node in the reference group, determine the abnormal nodes based on the statistical deviation, and classify the anomalies into isolated anomalies or group anomalies according to the number of abnormal nodes and the direction of deviation. The visualization and alarm module is used to display the real-time health status of each radar node on the interactive interface in a color-coded manner, draw the health score trend curve, and generate alarm notifications when an anomaly is detected or the remaining normal working time is lower than a preset threshold.

10. The intelligent monitoring platform according to claim 9, characterized in that, The performance monitoring data frame includes: The node identifier field group contains the radar's unique serial number and model code; the timestamp field group contains the timestamp of the data acquisition time. The RF specification field group includes power amplifier nonlinearity characteristics, I / Q imbalance, phase noise, gain drift, image rejection ratio, and noise figure. The environmental metrics field group includes the environmental complexity index and its sub-parameters; The detection index field group includes effective detection rate, ranging consistency index, false alarm density, and detection stability index; The health score field group includes the comprehensive score of each dimension, the decoupling compensation amount, the final health score, and the health level identifier; After dividing the reference group, the anomaly detection module first verifies the homogeneity of the distribution of each indicator within the reference group. When the variance within the group exceeds the preset variance threshold, it performs secondary clustering subdivision on the reference group before performing anomaly detection.

Images