Microwave radio link multi-index comprehensive measurement method

Abstract

The application discloses a microwave radio frequency link multi-index comprehensive measurement method and relates to the technical field of radio frequency communication testing, and comprises the following steps: S001, a peak power prediction model is established, a peak power probability curve is generated based on statistical distribution characteristics of multiple carrier signals, and a peak early warning vector containing an estimated time point and an amplitude value is formed; S002, synchronous sampling scheduling is performed according to the peak early warning vector, the sampling trigger threshold and the time window parameter are adjusted, the sampling window is aligned with the peak power occurrence time, and the time interval covered by the amplitude limiting protection is covered. Through peak prediction, accurate sampling, multi-dimensional difference and distortion identification, the method realizes efficient capture and dynamic regulation of the amplitude limiting abnormality, constructs a prediction-identification-regulation-optimization closed loop process, significantly improves the monitoring integrity and adaptive ability of the microwave radio frequency link to the burst distortion, and guarantees stable operation of the system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of radio frequency communication test, in particular to a microwave radio frequency link multi-index comprehensive measurement method. BACKGROUND

[0002] The microwave radio frequency link multi-index comprehensive measurement refers to a measurement mode that simultaneously detects and analyzes multiple key performance indicators of a link system carrying microwave and radio frequency signal transmission in a single test process. This method usually covers parameters such as gain, noise figure, bandwidth, phase delay, group delay, frequency response, intermodulation distortion, return loss, etc. of the link within the working frequency band. Through multi-channel signal acquisition, wideband vector analysis, synchronous triggering and algorithm solving, multiple tests that need to be completed in steps and by different devices are integrated into a unified measurement platform for execution, so that more comprehensive, real-time and highly correlated performance data can be obtained without changing the working state of the link. This comprehensive measurement not only improves the test efficiency and data consistency, but also reveals the coupling relationship between different indicators, providing a reliable basis for the design optimization, fault diagnosis and quality evaluation of microwave radio frequency links.

[0003] The prior art has the following disadvantages:

[0004] Under the condition of wideband multi-carrier excitation, the nonlinear devices (such as power amplifiers) existing in the measured microwave radio frequency link may trigger the internal overload protection mechanism and enter the limiting amplitude working state when they bear the instantaneous peak power impact. This limiting amplitude action will have a sudden impact on the transmission characteristics of the signal in a very short time, causing the link gain to drop instantaneously, the phase noise level to rise sharply, and the degree of spectral distortion to increase significantly. However, the triggering time of the limiting amplitude protection is affected by the randomness of the appearance of the multi-carrier superposition peak, and it is difficult to predict or accurately locate before measurement. When the test system performs multi-index synchronous measurement in a discrete sampling manner, if the limiting amplitude anomaly occurs only in part of the sampling period, it is very likely that it will not be detected because the sampling time window does not cover the transient event. Such uncaught abnormal state will cause significant deviation between the test results and the real link performance, forming "false normal" data in the test report, which will introduce potential risks in link performance evaluation, tolerance design and field application, and may cause unexpected communication interruption or function failure of the system during high-power wideband service operation.

[0005] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present application is to provide a microwave radio frequency link multi-index comprehensive measurement method to solve the problems in the background.

[0007] In order to achieve the above object, the present application provides the following technical scheme: a microwave radio link multi-index comprehensive measurement method, comprising the following steps:

[0008] S001, a peak power prediction model is established, a peak power probability curve is generated based on the statistical distribution characteristics of a plurality of carrier signals, and a peak early warning vector containing an estimated time point and an amplitude value is formed;

[0009] S002, according to the peak early warning vector, synchronous sampling scheduling is performed, by adjusting the sampling trigger threshold and the time window parameter, the sampling window is aligned with the peak power occurrence time, and the time interval covered by the amplitude limiting protection;

[0010] S003, multi-dimensional differential operation is performed on the link response in the sampling window, the difference characteristics of gain change, phase offset and spectrum distribution are extracted, and a distortion fingerprint data set is generated;

[0011] S004, the distortion fingerprint data set is used to train an abnormality recognition neural network, and an amplitude limiting protection event is identified, a risk label containing trigger time, amplitude limiting strength and spectrum change is generated;

[0012] S005, based on the risk label, the amplitude of the excitation signal and the bias of the nonlinear device are adjusted, and the link parameters are monitored in real time, the un-suppressed transient distortion data is extracted, and residual distortion characteristics are generated;

[0013] S006, the residual distortion characteristics and the link operation data are fed back to the peak power prediction model, the probability curve and the early warning vector are optimized, and a closed-loop control process of prediction, detection, regulation and optimization is constructed.

[0014] Preferably, step S001 comprises:

[0015] The center frequency, bandwidth, amplitude and initial phase value of each carrier in the excitation signal are obtained, and a multi-carrier superposition time domain waveform data under high time resolution is constructed;

[0016] The waveform data is statistically processed point by point, the amplitude histogram is calculated, and the cumulative distribution function and the complementary cumulative distribution function are generated based on the histogram;

[0017] According to the complementary cumulative distribution function, the peak power probability curve is drawn, the time segment with a probability higher than the risk threshold is identified, and the corresponding start and end time, maximum amplitude value and power amplitude ratio are extracted;

[0018] The plurality of time segments are arranged into a two-dimensional peak early warning vector, and the time and power information are smoothed and combined and weighted, which are used as the input basis of the synchronous sampling scheduling.

[0019] Preferably, step S002 comprises:

[0020] The peak early warning vector is read row by row, a sampling time window is set with each prediction time point as the center, and the window length is dynamically adjusted according to the power change rate;

[0021] A trigger threshold is set according to the peak power value of the center point, and the trigger condition is judged in combination with the continuous power change trend;

[0022] When the trigger condition is established, sampling is started, amplitude, phase and spectrum data are collected at a fixed sampling rate, and a timestamp is marked for output;

[0023] By analyzing the interval between consecutive time points, the overlapping sampling windows are merged to generate a complete and non-repeating sampling scheduling sequence.

[0024] Preferably, step S003 comprises:

[0025] Based on the amplitude value, phase value and spectrum distribution of each sampling point, gain sequence, phase sequence and spectrum sequence are constructed;

[0026] First-order difference operation is performed on the obtained sequences respectively to generate difference gain sequence, difference phase sequence and difference spectrum sequence;

[0027] Set the difference threshold of gain, phase and spectrum, and mark the points in the difference sequence that meet the characteristic change;

[0028] Define the points that meet at least two characteristic change conditions in time as distortion trigger points, extract the observation data before and after them, generate standardized distortion fingerprint data and collect them into a distortion fingerprint data set.

[0029] Preferably, the specific steps of generating standardized distortion fingerprint data are as follows:

[0030] Based on each distortion trigger point, the gain change sequence, phase change sequence and spectrum change matrix of 50 nanoseconds before and after are intercepted, the maximum gain change value, maximum phase offset value, spectrum maximum power change frequency and its power change value in this time period are extracted, and the signal distortion duration is recorded. Together with the discrete data sequence of the original change curve, it constitutes a structured distortion fingerprint data entry.

[0031] Preferably, step S004 comprises:

[0032] Use the distortion fingerprint data set containing seven fields as training samples, standardize each sample and label the amplitude limiting event level;

[0033] Construct an identification model composed of an input mapping layer, a feature aggregation layer and an output judgment layer, and perform training based on the error minimization mechanism;

[0034] The real-time generated distortion fingerprint data is input into the trained identification model, and whether the clipping event occurs, the clipping intensity level, and the spectral anomaly features are output;

[0035] The identification result is integrated with the corresponding timestamp to form a risk marker, which is written into the link monitoring data set for subsequent control.

[0036] Preferably, step S005 comprises:

[0037] The clipping trigger time, peak power, distortion level, spectral anomaly frequency point, and affected bandwidth are analyzed to determine the adjustment trigger condition;

[0038] The excitation signal amplitude and the nonlinear device bias voltage or bias current are adjusted according to the power priority principle, and an adjustment log is recorded;

[0039] The link gain, phase, and 6GHz-12GHz spectrum data are monitored at a rate of 1GS / s to determine whether there is residual distortion after adjustment;

[0040] The residual distortion time, gain residual, phase fluctuation peak, spectral anomaly frequency point, and duration are extracted to generate residual distortion feature data.

[0041] Preferably, step S006 comprises:

[0042] The residual distortion feature data and the link operation data are fused to construct a unified structured feedback sample and perform standardization processing;

[0043] The feedback sample is input into the peak power prediction model to correct the power probability distribution function and the statistical decision parameter;

[0044] A peak early warning vector is generated based on the optimized distribution curve, including the prediction time, peak power value, frequency combination index, and risk level;

[0045] The newly generated early warning vector is written into the data set synchronously with the feedback sample and the model parameter, and is used for updating the next cycle sampling control process.

[0046] In the above technical solution, the present application provides technical effects and advantages:

[0047] The application realizes early warning of instantaneous power mutation by introducing a peak power prediction model, accurately guides sampling timing adjustment with a warning vector, ensures that the sampling window covers the amplitude limiting protection trigger interval, and avoids missing detection; the distortion characteristics of the key physical quantities of the link are extracted through multi-dimensional difference processing, and an abnormality recognition mechanism is constructed using distortion fingerprint data to realize automatic recognition and quantification of abnormal states; in combination with risk marking, the excitation signal and device working state are dynamically adjusted in a closed loop, and the residual distortion after adjustment is continuously monitored, the prediction model parameters are optimized in reverse, and a closed-loop control process of measurement-recognition-regulation-optimization is realized. This method not only significantly improves the monitoring integrity and measurement accuracy of sudden distortion behavior of the link, but also enhances the adaptive ability of the system to nonlinear dynamic risks, thereby providing key support for the design evaluation, fault warning and high-reliability operation of the microwave radio frequency system. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0049] Figure 1 The method flowchart of the microwave radio frequency link multi-index comprehensive measurement method of the present application. DETAILED DESCRIPTION

[0050] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the inventive concept of the example implementations to those skilled in the art.

[0051] The present application provides a microwave radio frequency link multi-index comprehensive measurement method as shown in Figure 1 The method flowchart of the microwave radio frequency link multi-index comprehensive measurement method of the present application.

[0052] S001, a peak power prediction model is established, the statistical distribution characteristics of a plurality of carrier signals in the excitation signal are analyzed, a probability curve for describing the occurrence probability of instantaneous peak power is generated, and a peak warning vector containing estimated time points and corresponding amplitude values is formed based on the probability curve;

[0053] In order to realize early identification and intervention of peak power impact, a peak power prediction is proposed for the amplitude limiting protection trigger problem that may occur in a microwave radio frequency link under wideband multi-carrier excitation. By analyzing the composition characteristics and time domain power behavior of the excitation signal, a prediction mechanism based on actual data statistics is established, which is used as the basis for decision-making for subsequent synchronous sampling scheduling. The specific steps are as follows:

[0054] The basic parameters of the multi-carrier signal of the excitation source output are obtained, including the center frequency, bandwidth, amplitude and initial phase value of all carriers. Based on this, time series signal data is constructed, and the instantaneous voltage value of each carrier at each time point is calculated at a granularity of time step less than or equal to 1 nanosecond. By linearly superimposing the voltage values of all carriers at each time point according to the amplitude and phase relationship, the complete broadband excitation signal time domain waveform data is formed. The waveform data should cover the entire measurement time period (for example, 10 milliseconds, with one sample per nanosecond to form 10 million sample points), to ensure sufficient time resolution for capturing high-frequency peak mutation behavior. This stage does not apply the signal to the microwave radio link, and only generates the waveform data in the simulation platform or digital modeling environment for subsequent power statistical analysis.

[0055] The generated time domain waveform data is analyzed point by point in amplitude, specifically: the signal amplitude value corresponding to each sampling point is calculated, and the frequency distribution of different amplitude values appearing in the entire time period is counted to form an amplitude histogram. The cumulative distribution function is calculated on the basis of the histogram, which is used to represent the probability trend of the signal amplitude exceeding a certain threshold at different times, and the complementary cumulative distribution function is further obtained, which is used to represent the probability of the signal amplitude being greater than a certain specific value (such as the system amplitude triggering threshold) at any given time point. According to the complementary cumulative distribution function, a peak power probability curve is drawn, with the horizontal axis as time and the vertical axis as the probability of power exceeding the threshold. By observing the wave crest, slope change and local peak interval of the curve, the time period with power impact risk in the entire excitation period can be clearly identified.

[0056] Cumulative Distribution Function (CDF) is the probability of a variable (in this case, the signal amplitude) being less than or equal to a certain value. In a multi-carrier excitation signal, CDF describes the overall trend of the signal amplitude distribution in the entire time sequence. For example, if the CDF is 0.95 at a certain amplitude value, it means that 95% of the time the signal amplitude is less than or equal to that value. The Complementary Cumulative Distribution Function (CCDF) is the complement of CDF, which represents the probability of a certain amplitude value or above, i.e. the likelihood of the signal being greater than that amplitude value. The formula is: CCDF(x) = 1 - CDF(x). In this step, the Complementary Cumulative Distribution Function is particularly important because the triggering of the amplitude limiting protection often depends on whether the instantaneous amplitude exceeds a certain threshold. Through CCDF, the probability distribution of a certain power threshold being exceeded can be quantified, thus predicting which time intervals are more likely to trigger the amplitude limiting mechanism. Therefore, CDF reflects the overall safety interval distribution, while CCDF directly indicates the potential risk area.

[0057] In the prediction mechanism proposed in the present application, the reason for calculating the cumulative distribution function and the complementary cumulative distribution function is that the amplitude limiting protection is usually triggered by the peak power exceeding the threshold that the device can withstand. Due to the irregular burst of peak values in a multi-carrier superimposed signal, it is difficult to accurately determine the risk area by observing the original time domain waveform, so it is necessary to statistically analyze the amplitude values of all sampling points to quantify the probability intensity and distribution range of exceeding the critical threshold. The specific calculation steps are as follows:

[0058] Extract the amplitude value of each sampling point of the entire time domain waveform to form an array containing all instantaneous amplitudes.

[0059] Sort the array to arrange the amplitude values from small to large.

[0060] Traverse the sorted amplitude sequence, and sequentially count the number of sampling points below each amplitude value, and divide by the total number of sampling points to obtain the cumulative distribution probability of that amplitude, i.e. CDF.

[0061] Calculate the complementary probability value, i.e. CCDF, by using "1-CDF" method, which reflects the probability of the signal appearing above that amplitude value. By plotting the CCDF curve and superimposing the preset amplitude limiting power threshold line, all amplitude points and corresponding time intervals with CCDF higher than the set safety probability (such as 0.01) can be found, which serve as an important basis for predicting the triggering of amplitude limiting. This method does not rely on physical measurement and is not affected by the response time delay of the device, and has strong predictive ability and system adaptability.

[0062] Extract all time periods in the peak power probability curve where the probability is higher than a set risk threshold (for example, when the probability is greater than 0.01, it is considered to have a clipping risk), and record three indicators in each high-risk time period: one is the start and end time points of the time period, the second is the maximum amplitude value corresponding to the time period, and the third is the increase ratio of the maximum amplitude value compared to the average power. These three indicators together constitute a risk time slice. Arrange all time slices that meet the conditions in chronological order to form a set of continuous power risk segments, each of which has a definite time position and intensity attribute. Structurally combine these risk segments to form a peak warning vector, which is a two-dimensional array in specific form, with the first column being the warning time point (in nanoseconds), the second column being the corresponding power peak (in milliwatts), and the third column being the power mutation gradient (in milliwatts per nanosecond). The peak warning vector is used as a key input to set the trigger time window and sampling threshold in subsequent sampling scheduling.

[0063] Perform sequence processing on the formed peak warning vector to improve its practicality and stability. The specific method includes weighted smoothing of the time coordinates, and merging adjacent warning points with a continuous occurrence time of less than 100 nanoseconds into a single risk segment to avoid repeatedly counting the same power fluctuation into multiple warning events. At the same time, time-weighted average processing is performed on the power value to eliminate the influence of occasional spikes on the overall warning judgment. The processed peak warning vector retains the most representative and most meaningful power impact data, with fixed data structure and clear content, which can be directly input into the sampling control process to dynamically set the sampling window position and sampling trigger threshold, thereby accurately covering the key time period with clipping trigger risk.

[0064] The step is to provide the pre-warning identification ability in the time domain and the amplitude domain for the high-precision measurement process of the subsequent microwave radio frequency link, and to establish the judgment basis for the peak power burst behavior in advance by extracting the statistical characteristics and analyzing the power fluctuation of the superposition behavior of multiple carriers in the excitation signal. In the multi-carrier synthesized signal, due to the differences in frequency, amplitude and phase of each carrier, the interference superposition at some time points may produce a transient peak much higher than the average power, thereby triggering the amplitude limiting protection mechanism of the nonlinear device in the microwave link, causing problems such as link gain drop, phase shift, spectrum distortion, etc. The traditional sampling method is prone to miss these transient events due to the fixed sampling window, resulting in distorted test results. Therefore, the step constructs a peak power probability curve to quantify the risk degree of the signal exceeding the set power threshold in each time period, and further forms a peak warning vector to explicitly express the high-risk time points and the corresponding amplitude values, thereby providing accurate time domain reference and power trigger basis for the synchronous sampling scheduling process. Through the introduction of this step, the measurement system can dynamically focus on the key time interval with actual power risk, effectively improving the capture rate and identification accuracy of abnormal events, laying a data foundation and time coordinate basis for subsequent distortion feature extraction, event identification and power control. It is the key pre-step for the invention to realize predictive measurement and closed-loop control.

[0065] S002, according to the peak warning vector, a synchronous sampling scheduling process is performed, the sampling trigger threshold and the sampling time window parameters are adjusted, the sampling time window is aligned with the peak power occurrence time, so that the sampling process completely covers the time interval triggering the amplitude limiting protection;

[0066] In order to ensure that the process of triggering the amplitude limiting protection event of the microwave radio frequency link under the condition of multi-carrier excitation due to transient peak power impact is accurately and completely captured, a synchronous sampling scheduling based on the peak warning vector is proposed. By accurately matching the predicted high-power risk time points with the measurement sampling time window, and dynamically adjusting the sampling trigger threshold and the sampling time window length, each actual amplitude limiting behavior can fall into the sampling time period, so that complete amplitude limiting event data is obtained. The specific steps are as follows:

[0067] The peak early warning vector constructed in the previous step is input into the sampling control flow. The early warning vector is a three-column structured data, where the first column is the prediction time point, expressed in nanoseconds; the second column is the corresponding instantaneous peak power value, expressed in milliwatts; and the third column is the power change rate, i.e., the change amount of the power value between adjacent sampling points, expressed in milliwatts per nanosecond. First, the vector is read row by row to define a symmetric sampling time window with each time point as the sampling center. The window length is a fixed value of 200 nanoseconds, with a 100 nanosecond extension before the center point and a 100 nanosecond extension after the center point. For time points with a large power gradient (e.g., a change rate exceeding 1 milliwatt / nanosecond), the window width is automatically expanded to 300 nanoseconds. In this way, each high-risk point is completely surrounded in the sampling interval, and the entire process of power rise, peak maintenance, and power decline is captured without omission.

[0068] For each set sampling time window, the trigger threshold value is dynamically set. The peak power value corresponding to the center point of the time period is read, and the value is multiplied by an empirical coefficient (e.g., 0.9) to obtain the sampling trigger threshold. When the actual input signal power reaches or exceeds this threshold value before the window is started, the high-speed sampling in the window is immediately triggered. For example, if the center point peak is 200 milliwatts, the trigger threshold is set to 180 milliwatts. To enhance the accuracy and stability of the trigger, the signal power continuous change slope is also monitored while setting the trigger threshold. If the power value shows a monotonic increase in the last 3 sampling points and the average increase is greater than 10 milliwatts, it is confirmed that the power burst trend is established, and the trigger condition is met. The trigger mechanism is implemented through an external control circuit or embedded control logic, without introducing complex operation flow, thereby ensuring the response speed.

[0069] When the trigger condition is met, the high-speed sampler is immediately started to continuously sample the link output signal. The sampling rate of the sampler is fixed at 20 GS / s (i.e., 20 billion data points per second), which can fully cover the signal bandwidth range. The sampling content includes the amplitude value, phase value, and corresponding frequency component data of the link output signal, which are obtained through amplitude detection circuit, phase tracking circuit, and real-time spectrum analysis circuit, respectively. During the entire sampling window, the three types of data are recorded synchronously at each sampling point and output as three groups of arrays in timestamp sequence, named amplitude data group, phase data group, and spectrum data group, respectively. These data groups are one-to-one corresponding through time labels and stored in the local cache area for subsequent feature extraction and amplitude limiting discrimination operations.

[0070] To avoid repetition or omission of the sampling window between multiple peak warning time points, the sampling scheduling process introduces a window overlap analysis mechanism. Compare the time difference between the two consecutive warning time points, if less than the current window length, merge into a longer sampling window, continue the previous window sampling and extend the sampling end time; if the interval is greater than the window length, enter the idle state after the current sampling ends, waiting for the trigger signal of the next time point. The sampling window merging logic is completed once through the preprocessing process, and is executed continuously throughout the measurement period without manual intervention. All actual executed sampling time windows will be marked in the scheduling table and correspond one by one with the sampling data, ensuring that the collected data accurately match the actual risk time points and avoiding data redundancy or loss.

[0071] The role of this step is to convert the prediction results of peak power into sampling control behavior with time specificity and amplitude sensitivity, in order to achieve high-precision and full-period capture of the clipping protection trigger process in the microwave radio link. Under the excitation of multiple carriers, the power peaks in the excitation signal are bursty and instantaneous, and their occurrence time is difficult to accurately determine in advance, and the duration is extremely short, often in the nanosecond level. If the traditional fixed-period sampling or static time window sampling method is still used, it is easy to cause the sampling of key events to be missing, resulting in the real state of the link being unable to be recorded completely. Therefore, this step uses the peak warning vector constructed in the previous step as the basis for active control, extracts the warning time points and power characteristics one by one, and dynamically adjusts the position, width and start condition of the sampling time window in these time segments with actual risks. By adaptively setting the trigger threshold, the sampling behavior is only started when the power surge reaches the critical point, which not only reduces the invalid sampling overhead, but also accurately captures the start, peak and recovery of the clipping behavior. This mechanism ensures that the sampling window accurately aligns with each time point with high clipping risk, avoids missing and false sampling, and at the same time ensures that the original measurement data has time continuity and event relevance, which is a prerequisite for realizing link anomaly identification, subsequent distortion modeling and closed-loop control. It can be said that this step converts power prediction from "static analysis" to "dynamic control", and is the key bridge from "discrimination" to "intervention" in the entire measurement process, significantly improving the response ability and identification success rate of the entire test system to nonlinear distortion events.

[0072] S003, after completing the sampling, multi-dimensional difference operation is performed on the microwave radio link response in the sampling time window, the difference features between the link gain change data, the link phase offset data and the link spectrum distribution data are extracted, and a distortion fingerprint data set for representing the transient distortion is generated;

[0073] In order to accurately characterize the transient distortion caused by amplitude protection events of microwave radio link under high power excitation conditions, a multi-dimensional differential processing based on sampling data is proposed. Based on the original response data of the link within a specific time window, the link gain response, link phase response and link spectrum response are sequentially calculated and analyzed. A feature data set is constructed to represent the instantaneous distortion for subsequent identification and control. The specific steps are as follows:

[0074] The original data obtained in the sampling time window is structured. The original data comes from the sampling operation of the microwave radio link in the previous step, and the data is indexed by time stamp. Each sampling point contains three physical quantities: the first is the amplitude value of the link output signal, with the unit of dBm, which represents the output power of the excitation signal after passing through the link at this time point; the second is the phase value of the link output signal, with the unit of angle, which is represented by the standardized angle value in the interval of -180° to +180°; the third is the spectral amplitude distribution of the link output signal in the frequency band of 6GHz to 12GHz, with the frequency step of 0.1GHz, with the unit of dBm / Hz. The above three physical quantities are respectively constituted into gain sequence, phase sequence and spectrum sequence, each sequence is sorted by time axis to form a data set with unified time reference.

[0075] First-order difference operation is performed on the above three sequences to mine the local change trend of the link output signal between adjacent time points. For the link gain sequence, the power difference between the adjacent two sampling points is calculated point by point, for example, if the power changes from -3dBm to -8dBm in 5ns interval, the difference is -5dB, which represents power compression; for the link phase sequence, the phase difference between adjacent sampling points is calculated, for example, from +30° to -150°, the difference is -180°, which indicates that there may be phase jump or phase locking abnormality; for the spectrum sequence, the power value difference between the current sampling point and the last sampling point is calculated at each fixed frequency point, for example, from -60dBm / Hz to -45dBm / Hz at 9.3GHz frequency point, the difference is +15dB, which reflects the sudden rise of spectral energy. After this step, three difference sequences or matrices are obtained, which record the amplitude, phase and frequency change behavior of the link response in a short time.

[0076] Based on the difference result, the abnormal change point with significant characteristics is extracted. Define a specific threshold: when the gain difference is greater than ± 3dB, the phase difference is greater than ± 20°, and the spectral power change is greater than ± 10dB, it is considered that the point has distortion characteristics. According to this threshold standard, the three difference sequences are judged point by point, and all the sampling points that meet the conditions are marked. Then, the feature points in different sequences are compared on the time axis. If a sampling time point meets the above feature criteria in at least two sequences at the same time, the time point is confirmed as a "distortion trigger point". On the basis of each distortion trigger point, the complete sampling data within 50 nanoseconds before and after the trigger point is intercepted as an observation segment, including the gain change curve, the phase change curve and the spectrum change curve in this time period. The above three curve data are aligned under the same time axis to form a complete observation sample.

[0077] Each observation sample is structured into a standardized distortion fingerprint data entry. Each data contains the following fields: trigger time (unit: nanosecond), maximum amplitude of gain change (unit: dB), maximum amplitude of phase shift (unit: degree), maximum gain frequency point of spectrum (unit: GHz), power change value of the frequency point (unit: dB), signal distortion duration (unit: nanosecond) and discrete data point sequence of the three original change curves. All distortion fingerprint data is recorded in a data table in a unified format, named distortion fingerprint data set, which can be stored in the local non-volatile memory of the collection device or uploaded to the remote server for processing. The data set is the direct input data for subsequent anomaly recognition, high-risk time period backtracking, link distortion modeling and dynamic control, with high timeliness, high correlation and high engineering practicability.

[0078] The role of this step is to extract and express the dynamic behavior of the microwave radio link during the amplitude protection trigger in a structured and quantitative manner, providing interpretable and operable input basis for subsequent anomaly identification and risk response. When the link enters the amplitude limiting state under the peak power impact of the excitation signal, its performance parameters will produce instantaneous and complex nonlinear disturbances, manifested as sudden drop in link gain, phase jump in output, and rapid diffusion of spectral energy, etc. Such changes last for a very short time and are unevenly distributed in the frequency domain. If not extracted, they are easily misjudged by the test system as normal state, or only part of the changes are recorded, leading to distortion that cannot be accurately identified and classified. Therefore, by performing gain difference, phase difference and spectral difference analysis on the original link response data obtained within the sampling window, not only can the local mutation degree of the signal between consecutive sampling points be quantified, but also the time synchronization between multiple physical parameters can be established, accurately locking the time point and distortion characteristics of the real distortion. Further, by converting these confirmed mutation behaviors into standardized "distortion fingerprint" data entries, not only does it provide high-quality sample data for the training of subsequent anomaly identification neural networks, but also lays the foundation for the construction of a link anomaly feature database for the measurement system, thereby realizing long-term tracking, rapid identification and dynamic response of the microwave link state. This step essentially plays a bridge role in "de-redundancy, feature extraction, and semantic formation" of the original data, and is a key link in the transition from passive measurement to active identification and from continuous waveform to intelligent tagging, with very important engineering practical significance and technical necessity.

[0079] S004, inputting the distortion fingerprint data set into an anomaly identification neural network for training, and using the trained neural network model to identify the amplitude protection trigger event in the microwave radio link, and outputting a risk label containing the amplitude trigger time, amplitude strength and spectral change characteristics;

[0080] In order to realize the accurate identification and quantitative output of the amplitude protection behavior triggered by the microwave radio link under peak power impact, an event determination based on distortion fingerprint data identification model is proposed. Relying on the extracted distortion fingerprint data set, an identification model with multi-dimensional input capability is trained, and through the model, the occurrence time, severity and influence on spectral characteristics of the amplitude event during the link operation are identified in real time, forming a complete risk label output. Specifically, the following steps are included:

[0081] The distortion fingerprint dataset generated by the aforementioned sampling process and differential extraction link is used as the training dataset of the recognition model. Each distortion fingerprint data entry is composed of fixed fields, including the sampling start time point (in nanoseconds), the gain compression amplitude (in decibels, representing the influence of amplitude limiting on signal amplification capability), the phase shift amplitude (in degrees, representing the degree of phase disturbance of the link when distorted), the spectral mutation frequency point (in GHz, representing the center frequency of the energy appearing sharply rising in the spectral response), the mutation frequency point energy change value (in decibels), the spectral distortion bandwidth (in GHz), and the duration (in nanoseconds). All fields are standardized to unify the input dimension and unit, forming a structured and sufficient multi-dimensional input sample set. Each data entry is manually labeled as whether it is an amplitude limiting event and its amplitude limiting strength level, represented by an integer from 0 to 3, where 0 represents no event, 1 represents mild amplitude limiting, 2 represents moderate amplitude limiting, and 3 represents severe amplitude limiting.

[0082] The structure of the recognition model is constructed and the training process is performed. In this embodiment, the recognition model is composed of three layers, namely the input mapping layer, the feature aggregation layer and the output judgment layer. The input mapping layer receives the above-mentioned seven-dimensional data fields and extracts the linear correlation between them; the feature aggregation layer is used to calculate the joint change trend between the input fields, especially the synchronization between gain change and phase shift, or the coupling degree between spectral distortion bandwidth and energy change amplitude; the output judgment layer classifies and analyzes the combined features, and outputs the judgment result and corresponding level of amplitude limiting event. The training process uses the error minimization iteration mechanism, which compares the error between the model output and the actual labeled value in each training round, gradually optimizes the model parameters, and improves the judgment accuracy. The training samples account for 80% of the total data, and the remaining 20% of the samples are used to evaluate the recognition effect of the trained model, to ensure stability and generalization ability in actual application.

[0083] The trained recognition model is used to recognize the distortion fingerprint data collected in real time. During the operation of the microwave radio link, once the sampling and differential extraction are completed, the newly generated distortion fingerprint data is immediately input into the recognition model as input, and the model outputs the following three types of structured results: (1) whether an amplitude limiting event occurs, output as "yes" or "no"; (2) the amplitude limiting strength level, output as an integer value from 0 to 3, corresponding to four levels from no event to severe event; (3) spectral feature parameters, including the frequency point position where the mutation occurs, the power mutation amplitude of the frequency point, and the affected frequency bandwidth. These output results are bound with the time stamp of the original input data for time positioning and intervention basis for subsequent actions.

[0084] The identification output result is integrated into a "risk label" data record and written into a link monitoring data set in time sequence. Each risk label record contains at least the following fields: clipping trigger time (in nanoseconds), clipping intensity level (discrete value of 0-3), spectrum anomaly frequency point position (in GHz), power mutation amplitude of the frequency point (in dB), and affected bandwidth (in GHz). The risk label data set is used as a direct basis for link state evaluation, power adjustment control, and signal processing strategy switching, and can realize rapid identification and response adjustment of clipping events without relying on manual intervention. The calculation time of the entire identification process is controlled within 1 millisecond, which can meet the response timeliness requirements of the measurement system in real-time communication or high-speed radar working conditions of the wideband microwave system.

[0085] The step is to realize the automatic identification, accurate classification and structured output of the clipping protection event triggered by the instantaneous peak power impact in the microwave radio frequency link, so as to provide real-time, reliable and executable risk information support for subsequent dynamic power control and link safety adjustment. The clipping protection behavior usually shows multi-dimensional abnormal phenomena such as gain compression, phase mutation, spectrum expansion in a very short time, which has high transient, nonlinear and coupling characteristics, and is difficult to accurately identify by traditional single parameter threshold comparison method, which is easy to cause missed detection or misjudgment, and seriously affects the link state evaluation and system control strategy. The step builds an identification model based on the distortion fingerprint data set, takes the gain change amplitude, phase shift amplitude, spectrum mutation characteristics, distortion duration and other physical quantities reflecting the abnormal state of the link as input, and trains the model to have the ability to identify whether the clipping event occurs, quantify its intensity level, and identify the spectrum distortion characteristics. The identification process is changed from relying on artificial experience or static rules to a data-driven active sensing mechanism. By judging each abnormal behavior in the sampling window, the structured risk label is output, including the specific time point of clipping trigger, the affected frequency band, the power change degree and the distortion intensity level, so that the link measurement system has the intelligent ability of "understanding signal behavior". The step not only realizes the real-time quantitative identification of link abnormal behavior, but also converts the originally complex and dispersed multi-dimensional observation information into a set of simple, efficient and feedback control decision basis. It is the core bridge of converting the previous measurement data into dynamic control information, plays a key role in the whole closed-loop system, has high engineering practical value and technical advancement.

[0086] S005, input the risk mark into the dynamic power regulation process, based on the power characteristics and spectrum characteristics in the risk mark, real-time adjust the amplitude parameter of the excitation signal and the bias operating point of the nonlinear device in the microwave radio frequency link, at the same time, continuously monitor the gain, phase and spectrum changes of the adjusted link, extract the transient distortion data that is not completely suppressed, and generate residual distortion feature data;

[0087] To achieve timely suppression and dynamic response adjustment of the microwave radio frequency link amplitude limiting protection event, a power regulation based on risk mark information driving is proposed. The power characteristics and spectrum characteristics of the identified amplitude limiting event are used to jointly adjust the output amplitude of the excitation signal and the bias voltage or bias current of the nonlinear device in the link, while the link response is monitored in real time to identify residual abnormalities and build a residual distortion feature data set for subsequent optimization. The specific steps are as follows:

[0088] The risk mark data generated in the amplitude limiting event identification process is received and parsed. The risk mark includes the following fields: amplitude limiting trigger time (in nanoseconds), peak power of the excitation signal at that time point (in milliwatts), distortion intensity level (integer value from 0 to 3), spectrum mutation center frequency (in GHz), power surge of the mutation frequency point (in dBm), and spectrum affected bandwidth (in GHz). The adjustment control process reads the risk mark one by one and processes them in time sequence. For records with distortion level marked as 2 (moderate) or 3 (severe), the amplitude voltage control and bias adjustment process is immediately started; for records with distortion level 1 (mild), small parameter adjustment is made according to the preset strategy to suppress the initial distortion development.

[0089] According to the peak power and spectrum distortion range recorded in the risk mark, the amplitude output of the excitation signal and the bias parameters of the nonlinear device in the link are adjusted. The amplitude of the excitation signal is reduced by the digital signal controller through the modulation control interface, for example, the excitation output power is reduced from 200 milliwatts to 170 milliwatts (i.e. reduced by about 1.7dB). The adjustment of the bias parameters of the nonlinear device is completed by the digital power supply controller, for example, the gate bias voltage of the power amplifier is adjusted from 3.3 volts to 3.0 volts to reduce the probability of working in the gain compression region of the device, so as to maintain it in a more linear amplification interval. The joint adjustment of amplitude and bias is performed according to the control logic of "power priority, bias following", that is, the excitation amplitude is adjusted first, and if the distortion is not alleviated, the bias value is adjusted. All adjustment actions are controlled by nanosecond-level time, driven by field programmable logic circuit, and the adjustment records are written into the adjustment log synchronously, including adjustment time, adjustment type, target parameter, adjustment amplitude and confirmation flag.

[0090] The "nonlinear device bias parameter" refers to a static electrical parameter used to control and set the working state of a nonlinear electronic device (such as a power amplifier, a low-noise amplifier, a mixer, etc.) in a radio frequency link. Its essential function is to set the working point of the device in a specific region through an external DC voltage, current or other control signal, thereby affecting the linearity, gain, stability and distortion characteristics of the device in the high-frequency signal processing process. In a radio frequency circuit, a nonlinear device is prone to enter a nonlinear compression region under high power and large dynamic input conditions, thereby causing problems such as gain reduction, harmonic enhancement and intermodulation distortion. By adjusting the bias parameter, the working point of the device can be shifted from the nonlinear region to the linear working region, thereby suppressing the frequent triggering of the amplitude limiting protection and improving the overall dynamic range and stability of the link. Common bias parameters include: gate bias voltage (for field effect transistor devices), base current or voltage (for bipolar devices), collector or drain static current (affecting output power capability), and digital potential setting value for current source control. Reasonable setting of the bias parameter is a key engineering means to achieve linear amplification, thermal stability and low distortion output of the device, and plays an irreplaceable important role in high-frequency communication, radar and radio frequency measurement systems.

[0091] After the power adjustment is completed, the running state of the microwave radio frequency link is continuously monitored, focusing on three physical indicators: the link gain variation curve, the link phase fluctuation trajectory, and the real-time spectrum distribution of the output signal. The link gain variation is obtained through a real-time power detector with a sampling rate of 1 GS / s (1 billion times per second); the phase variation is collected through a high-speed phase comparison circuit, and the phase difference between adjacent points is calculated at each sampling point; the spectrum data is obtained through an integrated spectrum analyzer in the range of 6 GHz to 12 GHz, with a frequency point measurement every 0.1 GHz. The above three types of measurement data are stored in the link response buffer area and compared with the pre-adjustment data. If the gain difference still exceeds 3 dB, or the phase offset exceeds 20°, or the spectral energy is continuously higher than the reference 5 dBm at some frequency points, it is considered that the adjustment has not completely suppressed distortion.

[0092] The abnormal response data of the non-suppressed part is extracted and structured to generate residual distortion feature data. The feature data includes: the time point (in nanoseconds) of abnormal response, the gain change residual after adjustment (in dB), the phase fluctuation peak value (in degrees) after adjustment, the spectral mutation frequency position (in GHz), the residual energy value (in dBm) of the mutation frequency point, and the distortion duration (in nanoseconds). All fields are combined into a record in a unified format and added to the residual distortion data set. The data set not only retains the nonlinear behavior of the link in the adjusted state, but also can be used for subsequent updating of the peak prediction model or evaluating the effectiveness of the current adjustment strategy. The residual distortion data is uploaded to the main controller in batches every 5 seconds for centralized storage and further modeling analysis, providing key support for building continuous improvement type regulation and control capability for the system.

[0093] The role of this step is to realize the active intervention and fine regulation of the microwave radio link after the amplitude limiting protection event occurs. The output power of the excitation source and the bias working point of the nonlinear device in the link are adjusted in real time based on the risk marked information, so as to quickly pull the device from the nonlinear working area to the linear area, suppress distortion expansion, and ensure stable recovery of the link performance. The amplitude limiting protection mechanism is usually triggered when the link is subjected to instantaneous high power impact, causing the device to enter the gain compression area, the phase instability area or the spectrum distortion area, and passive measurement cannot effectively cope with it. Therefore, relying on the amplitude limiting risk label output by the previous identification process, this step uses the power peak, spectral mutation frequency point and distortion level as the basis for adjustment decision, and reduces the amplitude of the excitation signal within milliseconds, for example, reducing the output power from 200 milliwatts to 170 milliwatts, while adjusting the bias voltage of the power amplifier or low noise amplifier from the high gain state to the low gain, low nonlinear region, for example, adjusting the gate voltage from 3.3 volts to 3.0 volts to reduce the degree of device nonlinearity. After adjustment, the gain change, phase stability and spectral energy distribution of the link are monitored in real time at a high sampling rate, and the distortion characteristics that are not completely eliminated after adjustment are extracted, such as residual gain mutation, phase jump or local surge of spectral energy, and finally these abnormal phenomena are structured into residual distortion feature data for subsequent model optimization and strategy iteration. This step not only timely corrects the abnormal state at the physical layer, but also establishes a data closed loop mechanism for the system, making risk identification, adjustment execution and performance verification logically connected, providing dynamic protection for the stable operation of the microwave radio link in a complex excitation environment, and is the core bridge for transforming "finding problems" into "solving problems".

[0094] S006, the residual distortion feature data and the link operation data after dynamic power adjustment are jointly input into a peak power prediction model for feedback optimization, the probability curve and the peak warning vector are updated, the peak prediction accuracy and the sampling scheduling accuracy are continuously improved during system operation, and a closed-loop control process of peak prediction, amplitude limiting detection, risk control and model optimization is formed;

[0095] To realize continuous adaptive control and closed-loop performance optimization of microwave radio link in high dynamic working environment, a peak power prediction model self-update based on real-time data feedback is proposed. The residual distortion feature data extracted in the dynamic power adjustment process and the current link operation state parameters are input, the statistical structure in the prediction model is periodically optimized, the accuracy of peak power prediction is continuously improved with the actual operation of the link, and a "prediction-detection-control-optimization" closed-loop technical process is constructed. The specific steps are as follows:

[0096] The residual distortion feature data and the link operation data collected and structured in the dynamic power adjustment process are fused and normalized. The residual distortion feature data includes: (1) the gain compression value observed after dynamic power adjustment, unit: dB, precision: 0.1 dB; (2) phase offset residual, unit: angle, range: ±180°, used to measure the amplitude of signal phase disturbance; (3) frequency mutation frequency point, unit: GHz, precision: 0.01 GHz, representing the frequency offset position of distortion; (4) the power residual value corresponding to the frequency point, unit: dBm, used to quantify the energy level of frequency disturbance; (5) abnormal state duration, unit: nanosecond. The link operation data includes the total power value of the current excitation signal (unit: milliwatt), the number of multi-carrier signals, the frequency distribution interval (unit: MHz), the signal modulation type (such as 16QAM, BPSK), the gate bias voltage and drain current (unit: volt and milliamperes) of the nonlinear device, and the device operating temperature (unit: Celsius). All data are processed by standardization mapping before input, and are unified as dimensionless floating point numbers to form structured feedback input samples.

[0097] The feedback data is input as a training increment into the peak power prediction model to optimize the original power probability distribution function and statistical parameter structure in the prediction model. The power distribution function in the original prediction model is usually constructed based on the theoretical statistics of the peak power of the multi-carrier composite waveform, but the model does not consider the dynamic response characteristics of the link device and the influence of external environmental variables. Therefore, in this step, the original power distribution function is modified in form using the newly input residual distortion data. For example, if an amplitude limiting event is frequently triggered in a certain combination frequency interval in the historical data, but the prediction probability is less than 20%, the curvature of the probability density function in this interval needs to be improved to enhance the perception ability of the high-risk section. In addition, by introducing a "residual offset factor", the error amount and the actual deviation value are mapped into a weight compensation parameter, which is used to update the judgment boundary of different amplitude power values in the model, to ensure that the risk assessment results output by the model are closer to the actual response behavior of the link.

[0098] The peak warning vector is regenerated based on the optimized probability distribution curve. The vector includes a series of estimated peak power event records arranged in time sequence, each record containing the predicted time point (in nanoseconds), the predicted peak power value (in milliwatts), the frequency combination mode index (representing the carrier combination mode), and the estimated distortion risk level (ranging from 0 to 3). For example, if the combined carrier power overlap probability exceeds the set threshold (such as 95%) in a certain period in the prediction model, the time point is marked as a "high-priority sampling period", and the peak power prediction value is corrected from the original 190 milliwatts to 205 milliwatts. At the same time, the record is attached with an identifier to prompt the sampling scheduling process to increase the sampling frequency and open the sampling window in advance to cover the potential amplitude limiting risk period. The warning vector structure is stable and the parameters are complete, which can be directly embedded into the sampling control unit as control instructions to support millisecond-level sampling scheduling optimization.

[0099] The updated peak warning vector is sent to the sampling control process of the next cycle, and the current feedback data, the optimized model parameters and the generated warning vector are stored in the system operation data set. This data set not only serves as input for the next round of model optimization, but also provides historical basis and traceability for the overall system operation state. The system is set to trigger model feedback update automatically every 1 minute of operation, ensuring that the peak power prediction model can evolve dynamically with the excitation environment, device state and distortion behavior, and always maintain high-precision prediction performance. Through this mechanism, the prediction model, sampling scheduling mechanism and power regulation mechanism form a linked closed loop: predict anomalies → perform sampling → identify risks → adjust responses → extract residuals → feedback optimization → re-predict, thereby building a complete, efficient and self-learning intelligent closed loop system in the field of microwave radio link measurement and control.

[0100] The step is to build a core closed-loop self-optimization mechanism in the microwave radio frequency link multi-index comprehensive measurement system. By introducing "residual distortion feature data" and "link operation data after dynamic power regulation" as feedback input, the peak power prediction model is updated and corrected in real time, so that the system has the ability of continuous learning and self-correction, so as to maintain high accuracy of instantaneous peak prediction and high precision of sampling scheduling efficiency under the conditions of dynamic excitation, complex carrier combination and frequent nonlinear response. In the actual microwave radio frequency link measurement process, due to the strong non-stationarity and statistical randomness of the instantaneous peak power change of carrier superposition, the prediction curve constructed by traditional static signal model often cannot reflect the response characteristics of the device in the real working state, especially under the conditions of high power, large bandwidth and high temperature drift, the theoretical prediction and actual behavior are easy to deviate. By inputting the "residual distortion features" detected in the dynamic measurement process, such as residual gain compression, phase mutation peak, spectrum mutation frequency point power offset, and the excitation power, carrier structure, bias voltage, current and temperature parameters of the link at that time into the prediction model, the model is not only based on ideal statistics, but also can capture the coupling characteristics between the physical response of the device and the actual operation of the system, so as to build a dynamic risk probability curve facing the real environment. The updated probability curve generates a new peak warning vector, which provides more accurate time positioning and amplitude prediction for the sampling control strategy of the next measurement period, improves the success rate of capturing amplitude anomalies, and reduces the occurrence of "false normal" and "missed judgment". At the same time, through closed-loop iterative optimization, the prediction model has the adaptability in long-term operation, supports the system to continuously maintain high sensitivity and high stability in measurement and control ability in different scenes, and is the key step to realize the evolution of intelligent measurement system from responsive control to predictive control.

[0101] By the above multi-index comprehensive measurement method of the microwave radio frequency link, the key problems such as transient distortion being difficult to capture, abnormal state being easy to be missed, and test results existing "false normal" when facing the nonlinear amplitude limiting event in the broadband multicarrier excitation environment can be effectively solved. The present application realizes early warning of instantaneous power mutation by introducing a peak power prediction model, accurately guides the sampling time sequence adjustment with the warning vector, ensures that the sampling window covers the amplitude limiting protection trigger interval, and avoids missing detection. By multi-dimensional differential processing to extract the distortion characteristics of the key physical quantities of the link, and using the distortion fingerprint data to construct an abnormal recognition mechanism, automatic recognition and quantification of abnormal state are realized. In combination with risk marking, the excitation signal and the device working state are dynamically adjusted in a closed loop, and the residual distortion after adjustment is continuously monitored, the prediction model parameters are optimized in the reverse direction, and the closed loop control process of measurement-identification-regulation-optimization is realized. The method not only significantly improves the monitoring integrity and measurement accuracy of the link sudden distortion behavior, but also enhances the adaptive ability of the system to nonlinear dynamic risks, thereby providing key support for the design evaluation, fault warning and high reliable operation of the microwave radio frequency system.

[0102] The above has described certain exemplary embodiments of the present application by way of illustration only, and it is needless to say that the described embodiments can be modified in various ways without departing from the spirit and scope of the present application for those skilled in the art. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present application.

Claims

1. A microwave radio link multi-index comprehensive measurement method, characterized in that, The method comprises the following steps: S001, a peak power prediction model is established, a peak power probability curve is generated based on statistical distribution characteristics of a plurality of carrier signals, and a peak early warning vector containing an estimated time point and an amplitude value is formed; S002, a synchronous sampling schedule is performed according to the peak early warning vector, the sampling window is aligned with the peak power occurrence time by adjusting the sampling trigger threshold and the time window parameter, and the time interval covered by the amplitude limiting protection is covered; S003, multi-dimensional difference operation is performed on the link response in the sampling window, the difference characteristics of the gain change, the phase offset and the spectrum distribution are extracted, and a distortion fingerprint data set is generated; S004, the distortion fingerprint data set is used for training an abnormality recognition neural network, and an amplitude limiting protection event is recognized, and a risk label containing a trigger time, an amplitude limiting strength and a spectrum change is generated; S005, the amplitude of the excitation signal and the bias of the nonlinear device are adjusted based on the risk label, and the link parameters are monitored in real time, the un-suppressed transient distortion data is extracted, and residual distortion characteristics are generated; S006, the residual distortion characteristics and the link operation data are fed back to the peak power prediction model, the probability curve and the early warning vector are optimized, and a closed-loop control process of prediction, detection, regulation and optimization is constructed.

2. The microwave radio link multi-metric integrated measurement method according to claim 1, characterized in that, Step S001 comprises: The center frequency, bandwidth, amplitude and initial phase value of each carrier in the excitation signal are obtained, and a multi-carrier superposition time domain waveform data under high time resolution is constructed; The waveform data is subjected to point-by-point amplitude statistics, the amplitude histogram is calculated, and the cumulative distribution function and the complementary cumulative distribution function are generated based on the histogram; The peak power probability curve is drawn according to the complementary cumulative distribution function, the time segments with probability higher than the risk threshold are identified, and the corresponding start and end time, maximum amplitude value and power amplitude ratio are extracted; The plurality of time segments are arranged into a two-dimensional peak early warning vector, and the time and power information are smoothed and combined and weighted, serving as the input basis of the synchronous sampling schedule.

3. The microwave radio link multi-metric integrated measurement method according to claim 1, characterized in that, Step S002 comprises: The peak early warning vector is read row by row, and a sampling time window is set around each prediction time point, and the window length is dynamically adjusted according to the power change rate; The trigger threshold is set according to the peak power value of the center point, and the trigger condition is judged in combination with the continuous power change trend; When the trigger condition is established, the sampling is started, the amplitude, phase and spectrum data are collected at a fixed sampling rate, and the time stamp is labeled and output; By analyzing the interval between consecutive time points, the overlapping sampling windows are combined, and a complete and non-repeating sampling schedule sequence is generated.

4. The microwave radio link multi-metric integrated measurement method according to claim 1, characterized in that, Step S003 comprises: Based on the amplitude value, phase value and spectrum distribution of each sampling point, gain sequence, phase sequence and spectrum sequence are constructed; First-order difference operation is performed on the obtained sequences respectively, and difference gain sequence, difference phase sequence and difference spectrum sequence are generated; The difference threshold of gain, phase and spectrum is set, and the points in the difference sequence that meet the characteristic change are marked; The points that meet at least two characteristic change conditions in time are defined as distortion trigger points, the observation data before and after the points are extracted, standardized distortion fingerprint data is generated, and the distortion fingerprint data set is collected.

5. The microwave radio link multi-metric integrated measurement method according to claim 4, c h a r a c t e r i z e d b y, The specific steps of generating standardized distortion fingerprint data are as follows: On the basis of each distortion trigger point, the gain change sequence, phase change sequence and spectrum change matrix of the previous and subsequent 50 nanoseconds are intercepted, the maximum gain change value, maximum phase offset value, frequency point of maximum spectrum power change and power change value in the time period are extracted, and the signal distortion duration is recorded. The structured distortion fingerprint data entry is composed of the discrete data sequence of the original change curve.

6. The microwave radio link multi-metric integrated measurement method according to claim 1, characterized in that, Step S004 includes: Using a distortion fingerprint data set containing seven fields as training samples, each sample is standardized and labeled with clipping event level; Build a recognition model composed of input mapping layer, feature aggregation layer and output judgment layer, and perform training based on error minimization mechanism; Input the real-time generated distortion fingerprint data into the trained recognition model, and output whether the clipping event occurs, the clipping intensity level and the spectrum anomaly characteristics; Integrate the recognition result with the corresponding timestamp to form a risk marker and write it into the link monitoring data set for subsequent control.

7. The microwave radio link multi-metric integrated measurement method according to claim 1, characterized in that, Step S005 includes: Analyze the clipping trigger time, peak power, distortion level, spectrum anomaly frequency point and affected bandwidth to determine the adjustment trigger condition; Adjust the excitation signal amplitude and nonlinear device bias voltage or bias current according to the power priority principle, and record the adjustment log; Monitor the link gain, phase and 6GHz to 12GHz spectrum data at a rate of 1GS / s to determine whether there is residual distortion after adjustment; Extract residual distortion time, gain residual, phase fluctuation peak, spectrum anomaly frequency point and duration to generate residual distortion feature data.

8. The microwave radio link multi-metric integrated measurement method according to claim 1, characterized in that, Step S006 includes: Fuse residual distortion feature data and link operation data to build a unified structured feedback sample and perform standardization processing; Input the feedback sample into the peak power prediction model to correct the power probability distribution function and statistical judgment parameters; Generate a peak early warning vector based on the optimized distribution curve, including prediction time, peak power value, frequency combination index and risk level; Write the newly generated early warning vector into the data set synchronously with the feedback sample and model parameters, and use it for the next cycle sampling control process update.

Images