A method and apparatus for suppressing phase noise of an oscillator

Abstract

This invention discloses a method and apparatus for oscillator phase noise suppression, belonging to the fields of signal processing and noise control. The method includes: identifying phase noise sources by traversing the oscillator; dynamically tracking multiple phase noise standard signals; predicting noise in the oscillator based on noise change data; filtering the multiple phase noise standard signals based on the phase noise prediction results; establishing a phase-locked loop based on the noise filtering data; performing multi-level noise suppression on the multiple phase noise standard signals using a multi-stage filtering structure combined with noise phase difference; performing suppression detection based on the noise suppression results; providing feedback on the suppression effect; and formulating a suppression strategy for adaptive suppression of the oscillator's phase noise. This application solves the technical problem in the prior art where oscillators have poor phase noise suppression performance in complex noise source environments, leading to signal instability and spectral broadening.

Description

Technical Field

[0001] This invention relates to the fields of signal processing and noise control, and more specifically to a method and apparatus for suppressing phase noise of an oscillator. Background Technology

[0002] In modern electronic communication and precision measurement systems, signal stability and accuracy are crucial, especially in wireless communication, radar, satellite navigation, and high-frequency devices. As the core component generating clock and carrier signals, the phase noise of the oscillator directly affects system performance. The presence of phase noise can cause spectral broadening, signal distortion, and even affect the demodulation and decoding processes of the entire system, reducing data transmission rates and signal quality. Therefore, effectively suppressing oscillator phase noise has become an important research direction for improving the reliability of high-precision and high-frequency electronic systems.

[0003] Currently, techniques for suppressing oscillator phase noise mainly include traditional passive and active filtering techniques, as well as phase-locked loop (PLL) based techniques. However, these methods have certain limitations when facing complex noise sources and dynamically changing noise conditions, such as incomplete noise filtering, large system delays, and insufficient dynamic response. With the increasing demands for high-frequency and low-noise operation in electronic systems, existing technologies are insufficient to meet the phase noise suppression requirements in complex application scenarios. Therefore, there is an urgent need for a method and apparatus for oscillator phase noise suppression to address the technical problem of poor phase noise suppression performance in complex noise source environments, leading to signal instability and spectral broadening. Summary of the Invention

[0004] This application provides a method and apparatus for suppressing phase noise in oscillators, aiming to solve the technical problem in the prior art where the phase noise suppression effect of oscillators is poor in complex noise source environments, resulting in signal instability and spectrum broadening.

[0005] In view of the above problems, this application provides a method and apparatus for suppressing oscillator phase noise.

[0006] The first aspect disclosed in this application provides a method for suppressing phase noise in an oscillator. This method includes: identifying phase noise sources in the oscillator to generate multiple phase noise signals; standardizing the multiple phase noise signals to determine multiple phase noise standard signals; dynamically tracking the multiple phase noise standard signals to construct noise change state data; predicting noise in the oscillator based on the noise change state data to obtain phase noise prediction results; filtering the multiple phase noise standard signals based on the phase noise prediction results to determine noise filtering data; performing a phase-locked loop based on the noise filtering data to generate a noise phase difference; constructing a multi-level filtering structure; using the multi-level filtering structure in conjunction with the noise phase difference to perform multi-layer noise suppression on the multiple phase noise standard signals to obtain a noise suppression result; and performing suppression detection based on the noise suppression result, providing feedback on the suppression effect, and formulating a suppression strategy for adaptive suppression of the oscillator's phase noise.

[0007] Another aspect of this application discloses an apparatus for oscillator phase noise suppression. The apparatus includes a phase noise standard signal determination module, used to traverse the phase noise sources of the oscillator for noise identification, generate multiple phase noise signals, and standardize the multiple phase noise signals to determine multiple phase noise standard signals; a noise change state data construction module, used to dynamically track the multiple phase noise standard signals and construct noise change state data; a noise filtering data determination module, used to predict noise in the oscillator based on the noise change state data, obtain phase noise prediction results, and filter the multiple phase noise standard signals based on the phase noise prediction results to determine noise filtering data; a noise suppression result acquisition module, used to perform phase-locked loops based on the noise filtering data, generate noise phase differences, construct a multi-level filtering structure, and perform multi-layer noise suppression on the multiple phase noise standard signals using the multi-level filtering structure combined with the noise phase differences to obtain a noise suppression result; and an adaptive suppression module, used to perform suppression detection based on the noise suppression result, provide feedback on the suppression effect, and formulate a suppression strategy to adaptively suppress the phase noise of the oscillator.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0009] By employing a technique that traverses the oscillator's phase noise sources for noise identification, generates multiple phase noise signals, and performs standardization processing, this approach solves the technical problem in existing technologies where oscillators exhibit poor phase noise suppression in complex noise source environments, leading to signal instability and spectral broadening. Through accurate noise source identification and extraction of effective features, it enables effective separation and standardization of different noise signals, significantly improving phase noise suppression, ensuring signal stability, reducing spectral broadening, and ultimately optimizing oscillator performance.

[0010] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0011] Figure 1 This application provides a flowchart illustrating a method for suppressing oscillator phase noise.

[0012] Figure 2 This application provides a schematic diagram of an oscillator phase noise suppression device.

[0013] Figure labeling: Phase noise standard signal determination module 11, noise change state data construction module 12, noise filtering data determination module 13, noise suppression result acquisition module 14, adaptive suppression module 15. Detailed Implementation

[0014] The overall concept of the technical solution provided in this application is as follows:

[0015] This application provides a method and apparatus for suppressing oscillator phase noise. By traversing the phase noise sources of the oscillator, noise is identified and multiple phase noise signals are generated. First, the noise sources of the oscillator are identified and initial noise signals are acquired. Then, noise features are extracted through spectral analysis, and these features are used for cluster analysis. Finally, the noise signals are separated by category, generating multiple phase noise signals. This process provides a precise data foundation for subsequent noise processing and suppression.

[0016] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0017] Example 1, as Figure 1 As shown in the figure, this application provides a method for suppressing oscillator phase noise, the method comprising:

[0018] Step S100: Traverse the phase noise sources of the oscillator to identify noise, generate multiple phase noise signals, perform standardization processing on the multiple phase noise signals, and determine multiple phase noise standard signals.

[0019] Specifically, an oscillator is an electronic component used to generate periodically changing electrical signals, typically used for generating clock or carrier signals. The stability of the oscillator directly affects the accuracy and reliability of the signal. Phase noise refers to the phase disturbance caused by a frequency signal deviating from its ideal frequency in the frequency domain. This noise leads to signal spectrum broadening and signal distortion. Phase noise usually originates from various noise sources, such as thermal noise, device-inherent noise, and external interference. A phase noise signal refers to the specific noise signal identified and extracted from the noise source. Each phase noise signal represents a noise characteristic, such as noise components of different frequency bands or intensities. Normalization refers to adjusting different noise signals to the same standard form by standardizing parameters such as signal amplitude, frequency, and time for subsequent analysis. A standard phase noise signal refers to a normalized phase noise signal with a uniform frequency distribution, amplitude range, and time synchronization.

[0020] First, the various phase noise sources of the oscillator are traversed, with each noise source treated as a separate signal input for noise identification. During this process, a spectrum analyzer is used to measure the spectrum of each noise source, thereby generating multiple distinct phase noise signals.

[0021] Then, each generated phase noise signal undergoes normalization, a process involving three main steps: amplitude normalization, frequency normalization, and time alignment normalization. For example, amplitude normalization uses a normalization algorithm to adjust the amplitude of all noise signals to the same range (e.g., 0 to 1); frequency normalization maps the frequency distribution of each signal to a specified range, giving it uniform frequency characteristics; and time alignment normalization adjusts the time axis of the signals to ensure that all noise signals are displayed synchronously within the same time window. These normalization processes can typically be performed using signal processing tools, such as MATLAB or Python's signal processing libraries. After these processes, multiple standard phase noise signals with uniform characteristics are obtained.

[0022] This step identifies multiple noise sources in the oscillator and generates corresponding noise signals. Standardization ensures these noise signals have consistent characteristics in subsequent analysis. This allows the suppression process to more accurately identify and remove noise of different frequency bands and types, improving the effectiveness of phase noise suppression.

[0023] Step S200: Dynamically track the multiple phase noise standard signals to construct noise change state data.

[0024] Specifically, dynamic tracking refers to real-time monitoring of changes in a phase noise standard signal to identify and record its fluctuation characteristics. Dynamic tracking typically utilizes data update and change capture algorithms to identify signal change patterns. Noise change state data refers to the state information of the signal over time recorded through dynamic tracking of the phase noise standard signal. This state data includes noise amplitude changes, frequency drift, and temporal fluctuations, reflecting the dynamic characteristics of the noise.

[0025] First, multiple phase noise standard signals are dynamically tracked, i.e., their changes over time are monitored to obtain detailed noise change data. This is usually achieved using time series analysis tools and dynamic tracking algorithms. For example, the pandas library in Python can be used to record real-time signal fluctuations, combined with the frequency tracking method of the scipy.signal library, to continuously acquire real-time changes of noise signals at specific frequencies and amplitudes.

[0026] The primary task of dynamic tracking is to construct a data stream that receives and processes changes in the phase noise signal in real time. Assuming the amplitude of a phase noise signal fluctuates between 0.1 and 1, while its frequency drifts between 100Hz and 200Hz, dynamic tracking can capture this frequency fluctuation range and record the oscillation amplitude over time. Whenever the noise signal exceeds a specific threshold or deviates from its initial state, the tracking system records this change, incorporating it into the noise state to form time-series data of the noise.

[0027] In constructing noise variation state data, a sliding window algorithm is used to improve data accuracy. For example, a 10-second sliding window is used to continuously collect noise signal data, enabling analysis of noise variation patterns within each 10-second interval. This time-series data is then input into time-series models in machine learning libraries such as statsmodels or scikit-learn to generate state data containing noise features, thereby reflecting the fluctuation trends and characteristic changes of the noise.

[0028] By constructing dynamic tracking and noise change state data, the real-time state changes of the phase noise signal can be obtained, thus providing a key reference for subsequent noise prediction, filtering, and suppression.

[0029] Step S300: Based on the noise change state data, perform noise prediction on the oscillator to obtain phase noise prediction results, and filter the multiple phase noise standard signals based on the phase noise prediction results to determine noise filtering data.

[0030] Specifically, phase noise prediction results refer to the future noise characteristics calculated using prediction algorithms based on noise change state data, helping to determine the noise characteristics that the oscillator will exhibit at future times. Noise filtering data refers to the data generated after filtering based on the predicted phase noise information, used to actually suppress and reduce the phase noise of the oscillator.

[0031] First, based on the collected noise change data, the noise trend of the oscillator is predicted to obtain a future change estimate of the phase noise (i.e., the phase noise prediction result). Temporal prediction algorithms in machine learning, such as the Autoregressive Ensemble Moving Average (ARIMA) model or Long Short-Term Memory (LSTM) network, are typically used to process the time-series noise data.

[0032] Once the phase noise prediction results are obtained, the next step is to filter multiple phase noise standard signals based on these results. Filtering is the process of suppressing noise fluctuations in specific frequency bands or amplitudes, typically using Kalman filtering or low-pass filters. Kalman filtering is a dynamic filtering method, particularly suitable for smoothing time-series prediction data.

[0033] Using the low-pass filter from Python's scipy.signal library, a cutoff frequency is set (e.g., 2.5kHz), and then a phase noise standard signal is filtered in real time. This filtering significantly reduces noise amplitudes above 2.5kHz, resulting in a cleaner signal. The filtered data, called noise-filtered data, contains information about the noise-suppressed signal.

[0034] By employing noise prediction and filtering processes based on noise change state data, fluctuation trends in phase noise can be identified and eliminated in advance. This not only improves the stability of the oscillator signal but also enhances the system's anti-interference capability, making phase noise suppression more accurate and adaptive.

[0035] Step S400: Perform a phase-locked loop based on the noise filtering data to generate a noise phase difference value, construct a multi-level filtering structure, and use the multi-level filtering structure in conjunction with the noise phase difference value to perform multi-layer noise suppression on the multiple phase noise standard signals to obtain a noise suppression result.

[0036] Specifically, a phase-locked loop (PLL) is a feedback control system that synchronizes the phase of the output signal with the phase of the input signal by comparing and locking the input signal's phase. PLLs are used to detect and adjust the phase difference of noise signals to achieve phase consistency. The noise phase difference refers to the difference between the phase of the input signal and the phase of the reference signal during the phase-locking process. By calculating this phase difference, fluctuation trends in the noise signal can be detected. A multi-stage filtering structure is composed of multiple filters connected sequentially or in parallel. Each stage of the filter suppresses noise components of different frequencies or phases in the noise signal layer by layer, ultimately resulting in a smoother and more stable output signal. Multi-layer noise suppression refers to using multiple filtering stages to suppress noise in layers. Each layer suppresses noise with different characteristics, thus achieving a more refined noise reduction effect.

[0037] First, a phase-locked loop (PLL) is applied to the noise-filtered data generated by the oscillator. The PLL ensures phase consistency of the signal by comparing the phase of the noise-filtered data with a reference signal to obtain the noise phase difference. For example, when detecting communication signals, the phase difference can reveal the real-time trend of noise changes, helping to identify the direction of signal deviation. The core of the PLL includes a phase detector and a low-pass filter; these components work together to capture the phase difference in the signal and adjust it in time to maintain lock.

[0038] Next, based on the noise phase difference, a multi-stage filtering structure is constructed to achieve hierarchical suppression of phase noise signals. Specifically, the multi-stage filtering structure can be divided into three filtering stages: low-frequency, mid-frequency, and high-frequency. Each filter attenuates noise components of different frequencies. For example, the first low-frequency filter can use a Butterworth low-pass filter to remove low-frequency disturbances, while subsequent mid-frequency and high-frequency filters suppress noise in other frequency bands. Multiple filters are connected layer by layer or in parallel to ensure that noise components of different frequencies are effectively suppressed. The scipy.signal library in Python provides tools for designing multi-stage filters, such as the `butter` function for designing Butterworth filters, and the `filtfilt` function for bidirectional filtering to achieve a phase-shift-free filtering effect. Based on the multi-stage filtering, the results of each filtering stage are corrected by calculating the noise phase difference, thereby enhancing the filtering effect.

[0039] By combining a phase-locked loop and a multi-stage filtering structure, multi-level suppression of oscillator noise signals is achieved, effectively improving signal stability and accuracy. The phase-locked loop controls the signal phase, making the noise signal more consistent, while the multi-stage filtering structure ensures effective suppression of noise across different frequency bands.

[0040] Step S500: Based on the noise suppression results, perform suppression detection, provide feedback on the suppression effect, and formulate a suppression strategy to adaptively suppress the phase noise of the oscillator.

[0041] Specifically, suppression detection refers to analyzing the suppressed signal to verify the noise suppression effect. The effectiveness of noise suppression is confirmed by detecting the difference between the suppressed signal and the original signal. Feedback suppression effect refers to transmitting the suppression detection results (i.e., the effect of noise suppression) back to the control system or filter as a basis for adjustment and optimization. Through the feedback mechanism, the system can be further adjusted and optimized based on the actual effect to improve noise suppression performance. Suppression strategy refers to formulating corresponding control measures based on the detection results and feedback signals to further optimize the noise suppression process. This includes adjusting filter parameters, changing filtering methods, or enhancing noise suppression in specific frequency bands. Adaptive suppression refers to the system automatically adjusting the suppression strategy based on real-time feedback signals and environmental changes to achieve the best noise suppression effect.

[0042] First, the signal after multi-stage filtering and noise suppression is tested to evaluate the noise suppression effect. This can be done by calculating the signal-to-noise ratio (SNR), performing spectral analysis, or measuring the difference between the suppressed noise amplitude and the original signal. If the test results indicate poor noise suppression or excessive noise, the system triggers a feedback mechanism. The detected suppression effect is returned as a feedback signal to the control system or filter. This feedback signal can be used to guide filter adjustments.

[0043] Based on feedback information, new suppression strategies are formulated. Specifically, if the noise suppression effect in a certain frequency band is poor, a new strategy will be adopted, such as increasing the filtering intensity of that frequency band, changing the filtering algorithm, or optimizing the filter parameters. Furthermore, the suppression strategy can be adjusted based on the system's real-time detection data to better adapt to the current noise characteristics. By combining feedback effects and detection results, the system can adaptively adjust the noise suppression strategy, achieving dynamic optimization.

[0044] Through this adaptive suppression process, the system can dynamically adjust its noise suppression strategy based on real-time noise conditions. Compared to static noise suppression schemes, it can more effectively cope with various complex noise environments, ensuring that the system continuously maintains high-quality signal output.

[0045] Furthermore, noise identification is performed by traversing the phase noise sources of the oscillator to generate multiple phase noise signals. The method includes: identifying the phase noise sources of the oscillator; acquiring noise according to the phase noise sources of the oscillator to obtain multiple initial noise signals; performing spectral analysis based on the multiple initial noise signals; extracting features from the multiple initial noise signals according to the noise spectral information to generate multiple noise features; performing cluster analysis on the multiple initial noise signals according to the multiple noise features to determine multiple phase noise classes; and performing separation processing on the multiple initial noise signals according to the multiple phase noise classes to generate the multiple phase noise signals.

[0046] Specifically, the initial noise signal refers to the raw signal collected from different phase noise sources, typically containing unprocessed noise data. These signals form the basis for further analysis and processing. Spectrum analysis involves analyzing the signal's characteristics in the frequency domain to identify the frequency distribution of noise. Spectrum analysis can reveal the main frequency bands and intensity variations of the noise signal, usually performed using a spectrum analyzer or FFT technique. Noise features refer to the characteristic parameters extracted from each initial noise signal through spectrum analysis, including frequency range, amplitude, and phase characteristics; these features describe the attributes of the noise signal. Cluster analysis is a data analysis method used to group noise signals with similar characteristics into the same category. Initial noise signals can be divided into different phase noise classes for more targeted processing. A phase noise class refers to the noise type group determined after cluster analysis, containing specific types of phase noise signals, such as high-frequency noise, mid-to-low-frequency noise, etc. Each phase noise class represents a specific noise characteristic.

[0047] First, the various phase noise sources of the oscillator are identified. This process involves using noise acquisition equipment (such as a spectrum analyzer or digital oscilloscope) to acquire signals from each noise source, thereby obtaining multiple initial noise signals. For example, thermal noise around the oscillator, electromagnetic interference, and the device's own radio frequency noise are all acquired as different noise sources, and the signals from each noise source will exhibit different frequency characteristics.

[0048] Next, spectral analysis is performed on these initial noise signals. By converting the signals from the time domain to the frequency domain, the frequency characteristics of different noise signals are extracted. For example, using FFT technology, the amplitude distribution of each noise signal at different frequencies can be obtained. This allows the extraction of a series of noise characteristics, such as higher frequency noise having larger amplitudes and lower frequency noise having smaller amplitudes. The frequency and amplitude characteristics of each noise signal will serve as the main parameters of its noise characteristics.

[0049] Next, by performing cluster analysis on these noise features, noise signals with similar characteristics are classified into different phase noise classes. For example, the K-means clustering algorithm can be used to group noise signals with similar frequency distributions into the same class, thereby generating multiple phase noise classes. Different types of noise, such as high-frequency noise, mid-frequency noise, and low-frequency noise, can then be classified into the corresponding noise classes based on their frequency characteristics.

[0050] Finally, the original noise signal is separated according to these phase noise categories, extracting the signal from each category to generate multiple independent phase noise signals. For example, high-frequency noise signals are separated from the total noise signal to generate an independent high-frequency phase noise signal. The purpose of this is to enable targeted optimization and suppression of different types of noise signals in subsequent processing.

[0051] This process enables precise identification and classification of oscillator phase noise sources, decomposing complex noise signals into multiple phase noise classes with similar characteristics. Based on this, each noise signal can be processed individually, facilitating more efficient filtering and suppression of noise in different frequency bands and with different characteristics, thereby significantly improving the stability and clarity of the oscillator output signal.

[0052] Furthermore, the multiple phase noise signals are standardized to determine multiple phase noise standard signals. The method includes: standardizing the amplitude of the multiple phase noise signals according to the multiple noise characteristics to generate a noise amplitude range; standardizing the frequency of the multiple phase noise signals according to the noise spectrum information to generate a noise frequency distribution range; standardizing the time alignment of the multiple phase noise signals according to the signal acquisition timing to generate a synchronization time window; and performing consistency verification based on the noise amplitude range, the noise frequency distribution range, and the synchronization time window. When the verification passes, the multiple phase noise standard signals are determined.

[0053] Specifically, amplitude normalization refers to adjusting the amplitude of phase noise signals to a uniform range for consistent comparison and analysis. Amplitude normalization is generally achieved through normalization or scaling transformation. The noise amplitude range refers to the amplitude range of the normalized noise signal, such as adjusting all signal amplitudes to the range of 0 to 1. Frequency normalization refers to uniformly adjusting the frequency components of the phase noise signal to a specific frequency range. Frequency normalization commonly uses frequency normalization or other scaling transformation methods to unify signals from different frequency bands into the same frequency range. The noise frequency distribution range refers to the frequency distribution range of the signal after frequency normalization. Time alignment normalization refers to synchronizing the signals on the time axis to ensure that different noise signals are aligned within the same time window. Time alignment can be performed through signal acquisition timing to ensure that the sampling points of each noise signal are consistent. The synchronization time window refers to the unified time range of multiple normalized noise signals, ensuring that different signals are analyzed and compared on the same time reference. Consistency verification refers to verifying whether the normalization process of each phase noise signal is successful, ensuring matching and consistency in amplitude, frequency, and time.

[0054] First, amplitude normalization is performed on multiple phase noise signals. For example, suppose some noise signals have amplitudes ranging from 0.1 to 5, while others range from 0 to 2. Using amplitude normalization tools (such as the sklearn.preprocessing library in Python), these amplitudes can be adjusted to the range of 0 to 1, forming a uniform noise amplitude range. This amplitude normalization not only ensures the consistency of noise signal amplitudes but also facilitates subsequent filtering and suppression processing.

[0055] Next, frequency normalization is performed on the frequency distribution characteristics of the noise signal. This step typically involves mapping the signal's frequency range to a uniform distribution interval. For example, using frequency transformation functions in the scipy.signal library, noise signals in different frequency bands (such as signals in the 1MHz to 10MHz range and signals in the 10Hz to 100Hz range) can be adjusted to the same 0 to 1kHz frequency interval, thus forming a consistent noise frequency distribution interval. Based on this, the signal's frequency characteristics are uniformly processed, allowing subsequent analysis to more accurately capture noise variations.

[0056] Next, time alignment and normalization are performed on the noise signals. By standardizing the timing of signal acquisition, different noise signals are adjusted to the same synchronization time window. For example, assuming that the acquisition time periods of different noise signals are not exactly the same, this step ensures that each noise signal is displayed in the same time period by aligning the timestamps. Time alignment and normalization can be performed using interpolation tools in signal processing libraries (such as NumPy) to resample the signals to the same time reference.

[0057] After completing the standardization process described above, a consistency verification is performed, which checks the degree of matching in amplitude, frequency, and time for each phase noise signal. This consistency verification ensures that all signals behave consistently under a unified amplitude, frequency, and time standard. Once verified, these signals can be recognized as standard phase noise signals, providing reliable input data for subsequent noise suppression.

[0058] This standardized processing procedure adjusts multiple phase noise signals with different characteristics to a unified amplitude, frequency, and time standard, making the characteristics of the noise signals clearer and more consistent. This not only improves the comparability of the noise signals but also provides a high-quality data foundation for subsequent filtering and noise suppression, thereby enhancing the accuracy and effectiveness of phase noise suppression.

[0059] Furthermore, the method for predicting oscillator noise based on the noise change state data to obtain phase noise prediction results includes: segmenting data based on the noise change state data to generate a noise data training set and a noise data validation set; learning from the noise data training set to generate a phase change trend map for unsupervised training to construct a phase noise prediction model; extracting a real-time noise dataset based on the multiple phase noise standard signals, synchronizing the real-time noise dataset to the phase noise prediction model for alternating prediction to obtain an initial noise matrix; validating the initial noise matrix using the noise data validation set to obtain prediction error fluctuation values; and dynamically correcting the initial noise matrix based on the prediction error fluctuation values ​​to obtain the phase noise prediction results.

[0060] Specifically, the noise data training set refers to a dataset generated from noise-varying state data, used during the model training phase to help the model learn the characteristics and trends of noise. The noise data validation set refers to a subset of data segmented from the noise-varying state data, used to validate the model's performance and ensure the model's accuracy in predicting unseen data. The initial noise matrix refers to the noise data matrix obtained from the real-time noise dataset through preliminary calculations by the prediction model, representing preliminary prediction information of phase noise. In this matrix, rows represent time series, and columns represent state variables such as phase and frequency shifts. The prediction error fluctuation value refers to the difference between the predicted noise value and the actual noise value, used to correct the prediction model to improve prediction accuracy.

[0061] First, the noisy, changing state data is split into a training set and a validation set. This allows the model to learn the trends and characteristics of the noise during training, while the validation set is used to evaluate the model's performance. Data splitting can be done using the `train_test_split` method from the scikit-learn library in Python, randomly dividing the original data into an 80% training set and a 20% validation set. The training set is used for model learning, while the validation set is used for subsequent performance validation.

[0062] Next, unsupervised learning is performed on the training set to generate a phase change trend map and construct a phase noise prediction model. Unsupervised algorithms such as K-means clustering or principal component analysis (PCA) can be used to capture latent patterns in the noisy data, thereby extracting key features. K-means clustering groups the training data so that similar noise features are concentrated in one cluster; while PCA is used for dimensionality reduction, transforming multidimensional noisy data into lower-dimensional data that is easier to predict.

[0063] After establishing the phase noise prediction model, real-time noise datasets extracted from multiple phase noise standard signals are input into the prediction model for alternating prediction to obtain an initial noise matrix. For example, assuming that each time point in the real-time noise dataset contains the amplitude and frequency characteristics of the phase noise, by inputting these characteristics into the prediction model, the expected value of the phase noise at that time point can be obtained, thereby constructing an initial noise matrix containing multiple time points.

[0064] Next, the initial noise matrix is ​​validated using the validation set to obtain the prediction error fluctuation value. This step assesses the model's accuracy by comparing the deviation between the predicted and actual values, for example, using mean squared error (MSE) or mean absolute error (MAE) as evaluation metrics. A higher error fluctuation value indicates that the model needs correction.

[0065] Finally, the initial noise matrix is ​​dynamically corrected based on the fluctuation value of the prediction error to ensure that the phase noise prediction results output by the model are more accurate. Correction methods typically include techniques such as Kalman filtering or exponential weighted averaging, which smooth the prediction results and suppress short-term fluctuations.

[0066] This step enables accurate prediction of oscillator phase noise. The real-time extracted noise dataset and dynamic correction process allow the prediction model to better handle the randomness and temporal characteristics of noise, thereby improving the model's adaptability and prediction accuracy.

[0067] Furthermore, based on the multiple phase noise standard signals, a real-time noise dataset is extracted, and the real-time noise dataset is synchronized to the phase noise prediction model for alternating prediction to obtain an initial noise matrix. The method includes: labeling the multiple phase noise standard signals based on the multiple noise features to determine multiple noise reference benchmarks; comparing and matching the multiple phase noise standard signals according to the multiple noise reference benchmarks to obtain multiple similarity scores; extracting a real-time noise dataset based on the multiple similarity scores; synchronizing the real-time noise dataset to the phase noise prediction model for offset prediction to obtain offset prediction results, the offset prediction results including phase offset prediction results and frequency offset prediction results; feeding back the phase offset prediction results and the frequency offset prediction results to the phase noise prediction model for alternating calculation and update to obtain state variables of phase offset and frequency offset; arranging the state variables of phase offset and frequency offset according to the signal acquisition time sequence to construct the initial noise matrix.

[0068] Specifically, the noise reference benchmark refers to a set of benchmark values ​​determined based on noise characteristics, used to compare with real-time data to determine the characteristics and similarity of real-time noise signals. The similarity score refers to the score obtained by comparing real-time noise characteristics with the noise reference benchmark, used to quantify signal similarity and facilitate the extraction of highly correlated real-time noise data. The offset prediction result includes information on phase and frequency offsets; the prediction model calculates the noise variation trend based on these offsets. State variables refer to variables representing the system state in the prediction model, describing the phase and frequency states of the noise signal over time, used for further noise prediction and suppression.

[0069] First, feature annotation is performed based on noise characteristics from multiple phase noise standard signals, and multiple noise reference benchmarks are determined. For example, statistical measures such as mean and variance can be used as benchmarks for subsequent similarity matching. Through annotation, multiple benchmarks can be obtained, such as the "typical offset value" or "frequency fluctuation range" of phase noise.

[0070] Next, based on these benchmarks, each phase noise signal is compared and matched with a reference benchmark to generate a similarity score, thereby extracting the real-time noise dataset. For example, if the reference benchmark shows that the noise signal is more significant in a specific frequency range, the real-time data closest to the benchmark is assigned a higher similarity score and selected into the real-time noise dataset. Algorithms such as Euclidean distance or cosine similarity can be used to measure similarity; for example, the `cosine_similarity` function in Python can help calculate the similarity score.

[0071] Then, the extracted real-time noise data is synchronized to the phase noise prediction model for offset prediction, and the offset prediction results are obtained. In this prediction process, the model calculates the offset trends of phase and frequency.

[0072] Based on the obtained phase and frequency shifts, these results are fed back into the phase noise prediction model for alternating calculations and updates, thus forming new state variables. This feedback process, through multiple alternating updates, gradually approximates the true noise trend. Specifically, techniques such as Kalman filters or recurrent neural networks (RNNs) can be used for the prediction and updating of time-series data to improve the accuracy of noise prediction.

[0073] Finally, all state variables are arranged according to the time sequence of signal acquisition to generate an initial noise matrix. This matrix, composed of state variables arranged in chronological order, reflects the overall picture of noise changes over time, laying the foundation for subsequent filtering and suppression. The initial noise matrix not only contains the state information of existing noise but also includes predicted trends, providing data support for subsequent suppression operations.

[0074] This step enables more accurate data extraction and trend capture in oscillator phase noise prediction. The application of noise reference benchmarks and similarity scores makes the extracted real-time noise dataset more representative and relevant, while the alternating updates of offset prediction results and state variables improve adaptability to phase and frequency offsets and prediction accuracy.

[0075] Furthermore, a phase-locked loop is constructed based on the noise filtering data to generate a noise phase difference value, thereby building a multi-level filtering structure. The method includes: setting a target signal to be locked; comparing the noise filtering signal with the target signal according to the noise filtering data to obtain a noise phase difference value; adjusting the noise phase difference value using an oscillator through the feedback loop of the phase-locked loop to generate a phase difference value signal; performing dynamic filtering adjustment based on the phase difference value signal combined with Kalman filtering to construct a first-level filter; dynamically adjusting the weights of the first-level filter according to the noise phase difference value to construct a second-level filter; performing long-term and short-term time series analysis based on the second-level filter, and constructing a third-level filter based on the analysis results; and integrating the first-level filter, the second-level filter, and the third-level filter to construct the multi-level filtering structure.

[0076] Specifically, the target signal is the reference signal used in the phase-locked loop (PLL) for comparison with the noise signal. The target signal is typically a stable signal with a known frequency and phase, and is the desired output in the PLL. The phase of the noise signal is compared to this target signal and adjusted accordingly. The noise phase difference is the difference between the phases of the target signal and the noise signal, representing the phase deviation between them. By calculating the phase difference, the signal can be effectively adjusted to eliminate phase errors. Kalman filtering is a dynamic filtering method based on a linear system model, which denoises signals by estimating the system's state. Kalman filters use historical system data to infer the current state, thus smoothing and predicting dynamic systems, and are widely used in signal processing and navigation. First-stage, second-stage, and third-stage filters refer to multiple filtering stages, each processing noise signals with different characteristics. The first-stage filter is used for preliminary signal processing, the second-stage filter further adjusts the first-stage filtered signal, and the third-stage filter is used for more complex time-series signal analysis. Long Short-Term Time-Series Analysis (LSTM) is a neural network-based time-series data prediction method that can learn long-term dependencies in data. LSTM is often used for modeling time series data, as it can capture both long-term and short-term information, helping to make accurate noise predictions.

[0077] First, a target signal needs to be established, which is a known stable signal and serves as the reference for feedback adjustment. During signal processing, the noise-filtered signal is compared with the target signal in phase, and the noise phase difference value is obtained by calculating the phase difference. For example, in a communication system, the target signal is a stable carrier signal, while the noise signal is a disturbance signal with phase drift. Calculating the phase difference value can reveal the deviation caused by the noise to the signal.

[0078] Next, a phase-locked loop is used to feed the noise phase difference into the system via a feedback control mechanism to adjust the oscillator, thereby reducing phase error and generating a phase difference signal. This process is achieved through a phase detector and a loop filter, which can adjust the oscillator's phase in real time to keep it consistent with the target signal.

[0079] Next, a Kalman filter is used to dynamically filter and adjust the generated phase difference signal, constructing a first-stage filter. Based on historical data and the current noise phase difference, the Kalman filter predicts the future state, thus smoothing out random fluctuations in the noise signal. In this process, the Kalman filter provides an effective noise removal method, reducing errors caused by transient noise.

[0080] Then, the output signal of the first-stage filter enters the second-stage filter, where the signal weights are dynamically adjusted. This stage mainly relies on changes in the noise phase difference, combined with the filter's adaptive characteristics, to dynamically adjust the filter parameters, making the signal smoother. The working principle of the second-stage filter is to adjust the weights based on the real-time feedback signal, enabling it to better handle noise of different frequencies and amplitudes.

[0081] In further processing, a third-level filter is constructed based on the output of the second-level filter and combined with Long Short-Term Time-Tracking Method (LSTM). LSTM can capture long-term dependencies in time-series data, making it particularly suitable for processing time-dependent noise data. By analyzing the temporal changes in noise, the third-level filter can predict future noise trends and perform corresponding filtering to achieve more accurate noise suppression.

[0082] Finally, the first-stage, second-stage, and third-stage filters will be integrated to construct a multi-stage filtering structure. Through the coordinated work of filters at each level, noise signals are suppressed in multiple dimensions and at multiple levels. This multi-stage filtering structure can efficiently filter out noise in various frequency bands, especially noise components that change over time.

[0083] By implementing this multi-stage filtering structure, phase noise of the oscillator can be effectively processed and suppressed within different frequency bands and time windows. Dynamic adjustment using Kalman filtering smooths the signal and reduces errors caused by transient interference, while the use of LSTM captures long-term and short-term dependencies in the time series, effectively predicting and processing complex noise patterns. Ultimately, through the comprehensive processing of multiple filters, the phase noise of the oscillator output is significantly suppressed, thereby improving the stability and accuracy of the system.

[0084] Furthermore, by combining the multi-stage filtering structure with the noise phase difference value to perform multi-layer noise suppression on the multiple phase noise standard signals to obtain a noise suppression result, the method includes: inputting the multiple phase noise standard signals to the first-stage filter according to the noise phase difference value for high-frequency noise processing to generate a first suppression result, the first suppression result including a first suppression effect; synchronizing the first suppression result to the second-stage filter for mid-to-low frequency noise processing, updating the first suppression result to obtain a second suppression result, the second suppression result including a second suppression effect; synchronizing the second suppression result to the third-stage filter for long-sequence noise processing, updating the second suppression result to obtain a third suppression result, the third suppression result including a third suppression effect; integrating the first suppression result, the second suppression result, and the third suppression result according to the first suppression effect, the second suppression effect, and the second suppression effect to obtain the noise suppression result.

[0085] Specifically, high-frequency noise processing refers to suppressing noise in the high-frequency components of a signal. High-frequency noise typically originates from rapidly changing interference in the system. Processing this noise can improve signal stability, especially in high-frequency applications (such as radio frequency signals). Mid-to-low frequency noise processing refers to suppressing mid-frequency and low-frequency noise in a signal. This noise originates from sources such as power supply fluctuations or vibrations of the equipment itself; its low frequency typically has a long-term impact on system stability. Long-sequence noise processing refers to suppressing slowly varying noise in a signal. This noise has a long time period and is caused by environmental changes or long-term signal drift. This type of noise is usually difficult to eliminate quickly and requires long-term tracking and adjustment for smoothing. The first suppression result, second suppression result, and third suppression result are the signal results after processing by different filters (first-stage, second-stage, and third-stage filters). Each stage of the filter suppresses noise in different frequency bands of the signal, resulting in multiple suppression effects.

[0086] First, multiple phase noise standard signals are input to a first-stage filter using the noise phase difference value for high-frequency noise processing. The function of the first-stage filter is to eliminate high-frequency noise components in the signal, which are usually caused by external disturbances or equipment noise.

[0087] Next, the first suppression result is synchronously input into the second-level filter for mid-to-low frequency noise processing. The second-level filter is used to eliminate mid-to-low frequency noise, which typically originates from factors such as power fluctuations and temperature changes caused by prolonged operation of equipment. Based on the characteristics of the input signal, the second-level filter dynamically adjusts the filtering process, updates the signal, and generates a second suppression result. This result includes the signal after the mid-to-low frequency noise has been removed, as well as the second suppression effect after noise removal.

[0088] Based on this, the second suppression result is simultaneously input into the third-stage filter for long-term noise processing. The third-stage filter suppresses slowly changing noise in the signal by analyzing its long-term variations.

[0089] Finally, the first, second, and third suppression results are integrated based on their respective suppression effects (first, second, and third suppression effects) to obtain the final noise suppression result. This process, through weighted integration of the suppression effects of each filter stage, ensures that noise from high frequency to low frequency and from short time to long time is effectively suppressed, ultimately outputting a signal with minimized noise.

[0090] This multi-stage filtering structure effectively suppresses noise with different frequency bands and timing characteristics, improving the overall system performance. By processing noise layer by layer according to different frequency bands and timing characteristics, the system can accurately eliminate various noise components, providing a high-quality, noise-minimized output signal.

[0091] Furthermore, after performing multi-layer noise suppression on the multiple phase noise standard signals by combining the multi-level filtering structure with the noise phase difference, the method includes: feeding back the first suppression effect, the second suppression effect, and the third suppression effect to the first-level filter, the second-level filter, and the third-level filter layer by layer through the feedback loop to obtain a first-level optimized filter, a second-level optimized filter, and a third-level optimized filter; updating and optimizing the multi-level filtering structure based on the first-level optimized filter, the second-level optimized filter, and the third-level optimized filter to obtain a multi-level optimized filtering structure.

[0092] Specifically, a feedback loop refers to a portion of the system output signal being returned to the input, typically used to adjust system behavior or improve system performance. In noise suppression methods, feedback loops are used to provide feedback on the processing results of each filter stage, allowing the filters to be adjusted based on actual performance and further optimize noise suppression capabilities. A multi-stage optimized filter structure refers to a filter system composed of multiple optimized filters, including first-, second-, and third-stage filters. By integrating these optimized filters into a single system, different types of noise can be handled more comprehensively, achieving highly efficient noise suppression.

[0093] After multi-layer noise suppression—that is, after the initial noise suppression by first-, second-, and third-stage filters—the system feeds back the suppression effects of the first, second, and third-stage filters through feedback loops. These feedback signals reflect the processing effect of each filter layer and provide a basis for improvement. The first-, second-, and third-stage filters adjust their operating states based on the feedback signals. For example, if the first-stage filter is ineffective in handling high-frequency noise, the feedback signal will prompt it to strengthen its high-frequency noise suppression; similarly, the second- and third-stage filters will optimize their ability to handle low-frequency and long-sequence noise based on the feedback signals. After feedback optimization, optimized filters are obtained, including first-stage optimized filters, second-stage optimized filters, and third-stage optimized filters. These optimized filters are more accurate in handling noise in different frequency bands and can adapt to environmental changes.

[0094] Finally, the system updates and optimizes the original multi-stage filtering structure based on the optimized filter, forming a multi-stage optimized filtering structure. This optimized structure can more efficiently suppress different types of noise, ensuring higher quality of the final output signal.

[0095] By optimizing the filter layer by layer through a feedback loop, the accuracy and effectiveness of noise suppression can be significantly improved. The optimized filter can accurately suppress different types of noise, making it particularly suitable for complex signal environments and applications requiring high precision.

[0096] In summary, the oscillator phase noise suppression method provided in this application has the following technical effects:

[0097] 1. By traversing the phase noise sources of the oscillator to identify noise and generate multiple phase noise signals, the characteristics of different noise sources can be accurately captured and extracted. This method improves the identification accuracy of noise sources, provides a reliable data foundation for subsequent noise processing and suppression, and thus optimizes the phase noise suppression effect of the oscillator.

[0098] 2. By dynamically tracking and predicting noise change data, the future phase noise trend of the oscillator can be accurately predicted. This method utilizes real-time noise datasets for alternating prediction and error correction, effectively improving the accuracy of phase noise prediction and providing a more accurate basis for the noise suppression process.

[0099] 3. By constructing a multi-stage filtering structure and combining it with noise phase difference for noise suppression, the oscillator's phase noise suppression capability is effectively improved. This method employs a phase-locked loop and dynamic adjustment of the Kalman filter to ensure efficient suppression of noise in different frequency bands, thereby significantly improving the system's signal quality.

[0100] Example 2, based on the same inventive concept as the oscillator phase noise suppression method in the foregoing examples, such as... Figure 2 As shown, this application embodiment provides an oscillator phase noise suppression device, the device comprising:

[0101] The module 11 for determining phase noise standard signals is used to traverse the phase noise sources of the oscillator for noise identification, generate multiple phase noise signals, standardize the multiple phase noise signals, and determine multiple phase noise standard signals. The module 12 for constructing noise change state data is used to dynamically track the multiple phase noise standard signals and construct noise change state data. The module 13 for determining noise filtering data is used to predict noise in the oscillator based on the noise change state data, obtain phase noise prediction results, filter the multiple phase noise standard signals based on the phase noise prediction results, and determine noise filtering data. The module 14 for obtaining noise suppression results is used to perform phase-locked loop based on the noise filtering data, generate noise phase difference, construct a multi-level filtering structure, and perform multi-layer noise suppression on the multiple phase noise standard signals through the multi-level filtering structure combined with the noise phase difference to obtain noise suppression results. The module 15 for adaptive suppression is used to perform suppression detection based on the noise suppression results, provide feedback on the suppression effect, and formulate suppression strategies to adaptively suppress the phase noise of the oscillator.

[0102] Furthermore, the phase noise standard signal determination module 11 is also used to perform the following steps:

[0103] The phase noise source of the oscillator is identified, and noise is acquired according to the phase noise source of the oscillator to obtain multiple initial noise signals; spectral analysis is performed on the multiple initial noise signals, and feature extraction is performed on the multiple initial noise signals according to the noise spectrum information to generate multiple noise features; cluster analysis is performed on the multiple initial noise signals according to the multiple noise features to determine multiple phase noise classes; the multiple initial noise signals are separated according to the multiple phase noise classes to generate the multiple phase noise signals.

[0104] Furthermore, the phase noise standard signal determination module 11 is also used to perform the following steps:

[0105] The amplitude of the multiple phase noise signals is standardized according to the multiple noise characteristics to generate a noise amplitude range; the frequency of the multiple phase noise signals is standardized according to the noise spectrum information to generate a noise frequency distribution range; the time alignment of the multiple phase noise signals is standardized according to the signal acquisition timing to generate a synchronization time window; consistency verification is performed based on the noise amplitude range, the noise frequency distribution range, and the synchronization time window, and when the verification is successful, the multiple phase noise standard signals are determined.

[0106] Furthermore, the noise filtering data determination module 13 is also used to perform the following steps:

[0107] Data segmentation is performed based on the noise change state data to generate a noise data training set and a noise data validation set. Learning is then performed on the noise data training set to generate a phase change trend map for unsupervised training, constructing a phase noise prediction model. A real-time noise dataset is extracted based on the multiple phase noise standard signals, and this dataset is synchronized to the phase noise prediction model for alternating prediction to obtain an initial noise matrix. The initial noise matrix is ​​then validated using the noise data validation set to obtain prediction error fluctuation values. Finally, the initial noise matrix is ​​dynamically corrected based on these prediction error fluctuation values ​​to obtain the phase noise prediction result.

[0108] Furthermore, the noise filtering data determination module 13 is also used to perform the following steps:

[0109] Based on the multiple noise features, feature annotations are performed on the multiple phase noise standard signals to determine multiple noise reference benchmarks; the multiple phase noise standard signals are compared and matched according to the multiple noise reference benchmarks to obtain multiple similarity scores, and a real-time noise dataset is extracted based on the multiple similarity scores; the real-time noise dataset is synchronized to the phase noise prediction model for offset prediction to obtain offset prediction results, which include phase offset prediction results and frequency offset prediction results; the phase offset prediction results and the frequency offset prediction results are fed back to the phase noise prediction model for alternating calculation and update to obtain the state variables of phase offset and frequency offset; the state variables of phase offset and frequency offset are arranged according to the signal acquisition time sequence to construct the initial noise matrix.

[0110] Furthermore, the noise suppression result obtaining module 14 is also used to perform the following steps:

[0111] A target signal is set, and the noise-filtered signal is compared with the target signal according to the noise filtering data to obtain the noise phase difference. The noise phase difference is adjusted by an oscillator through the feedback loop of the phase-locked loop to generate a phase difference signal. Based on the phase difference signal, dynamic filtering adjustment is performed using Kalman filtering to construct a first-level filter. The weights of the first-level filter are dynamically adjusted according to the noise phase difference to construct a second-level filter. Long-term and short-term time series analysis is performed based on the second-level filter, and a third-level filter is constructed based on the analysis results. The first-level filter, the second-level filter, and the third-level filter are integrated to construct the multi-level filtering structure.

[0112] Furthermore, the noise suppression result obtaining module 14 is also used to perform the following steps:

[0113] The multiple phase noise standard signals are input to the first-stage filter according to the noise phase difference value for high-frequency noise processing to generate a first suppression result, which includes a first suppression effect. The first suppression result is synchronized to the second-stage filter for mid-to-low frequency noise processing, and the first suppression result is updated to obtain a second suppression result, which includes a second suppression effect. The second suppression result is synchronized to the third-stage filter for long-sequence noise processing, and the second suppression result is updated to obtain a third suppression result, which includes a third suppression effect. The first suppression result, the second suppression result, and the third suppression result are integrated according to the first suppression effect, the second suppression effect, and the second suppression effect to obtain the noise suppression result.

[0114] Furthermore, the system is also used to perform the following steps:

[0115] The first suppression effect, the second suppression effect, and the third suppression effect are fed back to the first-stage filter, the second-stage filter, and the third-stage filter through the feedback loop to obtain the first-stage optimized filter, the second-stage optimized filter, and the third-stage optimized filter; the multi-stage filter structure is updated and optimized based on the first-stage optimized filter, the second-stage optimized filter, and the third-stage optimized filter to obtain the multi-stage optimized filter structure.

[0116] In summary, any step of the method described above can be stored as a computer instruction or program in an unrestricted computer memory, and can be called and identified by an unrestricted computer processor to implement any method in the embodiments of this application, without any additional restrictions.

[0117] Furthermore, the "first" or "second" mentioned above not only represents a sequential relationship but also a specific concept and / or refers to the possibility of selecting individual or all of multiple elements. Clearly, those skilled in the art can make various modifications and variations to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for suppressing phase noise in an oscillator, characterized in that, The method includes: The phase noise sources of the oscillator are traversed to identify noise, generating multiple phase noise signals. The multiple phase noise signals are then standardized to determine multiple phase noise standard signals. Dynamic tracking is performed based on the multiple phase noise standard signals, that is, monitoring the changes of the phase noise standard signals over time to construct noise change state data; Based on the noise change state data, noise prediction is performed on the oscillator to obtain phase noise prediction results. Based on the phase noise prediction results, the multiple phase noise standard signals are filtered to determine noise filtering data. A phase-locked loop is constructed based on the noise filtering data to generate a noise phase difference value. A multi-level filtering structure is then built. The multi-level filtering structure, combined with the noise phase difference value, is used to perform multi-layer noise suppression on the multiple phase noise standard signals to obtain a noise suppression result. Based on the noise suppression results, suppression detection is performed, the suppression effect is fed back, and a suppression strategy is formulated to adaptively suppress the phase noise of the oscillator. Feedback on the suppression effect refers to transmitting the suppression detection results back to the control system or filter as a basis for adjustment and optimization. The method involves identifying phase noise sources in the oscillator and generating multiple phase noise signals, including: Identify the phase noise source of the oscillator, collect noise according to the phase noise source of the oscillator, and obtain multiple initial noise signals; Spectral analysis is performed on the multiple initial noise signals, and features are extracted from the multiple initial noise signals based on the noise spectral information to generate multiple noise features; Based on the multiple noise characteristics, the multiple initial noise signals are clustered to determine multiple phase noise classes; The multiple initial noise signals are separated according to the multiple phase noise categories to generate the multiple phase noise signals; The method for predicting noise in the oscillator based on the noise change state data to obtain phase noise prediction results includes: Based on the noise change state data, data segmentation is performed to generate a noise data training set and a noise data validation set. Based on the noise data training set, a phase change trend map is generated for unsupervised training to construct a phase noise prediction model. Based on the multiple phase noise standard signals, a real-time noise dataset is extracted, and the real-time noise dataset is synchronized to the phase noise prediction model for alternating prediction to obtain an initial noise matrix. The initial noise matrix is ​​validated using the noise data validation set to obtain the prediction error fluctuation value. The initial noise matrix is ​​dynamically corrected based on the prediction error fluctuation value to obtain the phase noise prediction result.

2. The method for suppressing oscillator phase noise as described in claim 1, characterized in that, The method for standardizing the multiple phase noise signals to determine multiple phase noise standard signals includes: Based on the multiple noise characteristics, the amplitude of the multiple phase noise signals is standardized to generate a noise amplitude range; Based on the noise spectrum information, the multiple phase noise signals are frequency-normalized to generate a noise frequency distribution range; Based on the signal acquisition timing sequence, the multiple phase noise signals are time-aligned and standardized to generate a synchronization time window; Consistency verification is performed based on the noise amplitude range, the noise frequency distribution range, and the synchronization time window. When the verification is successful, the multiple phase noise standard signals are determined.

3. The method for suppressing oscillator phase noise as described in claim 1, characterized in that, Based on the multiple phase noise standard signals, a real-time noise dataset is extracted. This real-time noise dataset is then synchronized to the phase noise prediction model for alternating prediction to obtain an initial noise matrix. The method includes: Based on the multiple noise characteristics, the multiple phase noise standard signals are characterized and multiple noise reference benchmarks are determined. The multiple phase noise standard signals are compared and matched according to the multiple noise reference benchmarks to obtain multiple similarity scores, and a real-time noise dataset is extracted based on the multiple similarity scores; The real-time noise dataset is synchronized to the phase noise prediction model for offset prediction to obtain offset prediction results, which include phase offset prediction results and frequency offset prediction results. The phase offset prediction result and the frequency offset prediction result are fed back to the phase noise prediction model for alternating calculation and updating to obtain the state variables of phase offset and frequency offset; The state variables of the phase offset and the state variables of the frequency offset are arranged according to the signal acquisition timing to construct the initial noise matrix.

4. The method for suppressing oscillator phase noise as described in claim 1, characterized in that, Based on the noise filtering data, a phase-locked loop is constructed to generate a noise phase difference value, and a multi-level filtering structure is built. The method includes: Set the target lock signal, and compare the phase of the noise filter signal with the target lock signal according to the noise filter data to obtain the noise phase difference value; The noise phase difference is adjusted by an oscillator through the feedback loop of the phase-locked loop to generate a phase difference signal; Based on the phase difference signal, dynamic filtering adjustment is performed using Kalman filtering to construct a first-stage filter; A second-level filter is constructed by dynamically adjusting the weights of the first-level filter according to the noise phase difference value. Long-term and short-term time series analysis is performed based on the second-level filter, and a third-level filter is constructed based on the analysis results. The first-level filter, the second-level filter, and the third-level filter are linked and integrated to construct the multi-level filtering structure.

5. The method for suppressing oscillator phase noise as described in claim 4, characterized in that, The method involves performing multi-level noise suppression on multiple phase noise standard signals using the multi-stage filtering structure and the noise phase difference to obtain noise suppression results. The method includes: The plurality of phase noise standard signals are input to the first-level filter according to the noise phase difference value for high-frequency noise processing to generate a first suppression result, wherein the first suppression result includes a first suppression effect. The first suppression result is synchronized to the secondary filter for low- and mid-frequency noise processing, and the first suppression result is updated to obtain a second suppression result, which includes a second suppression effect. The second suppression result is synchronized to the third-level filter for long-term noise processing, and the second suppression result is updated to obtain a third suppression result, which includes a third suppression effect. The first suppression result, the second suppression result, and the third suppression result are integrated according to the first suppression effect, the second suppression effect, and the second suppression effect to obtain the noise suppression result.

6. The method for suppressing oscillator phase noise as described in claim 5, characterized in that, After performing multi-layer noise suppression on the multiple phase noise standard signals by combining the multi-level filtering structure with the noise phase difference, the method includes: The first suppression effect, the second suppression effect, and the third suppression effect are fed back to the first-stage filter, the second-stage filter, and the third-stage filter through the feedback loop to obtain the first-stage optimized filter, the second-stage optimized filter, and the third-stage optimized filter; The multi-stage filtering structure is updated and optimized based on the first-stage optimized filter, the second-stage optimized filter, and the third-stage optimized filter to obtain a multi-stage optimized filtering structure.

7. A device for suppressing phase noise of an oscillator, characterized in that, For performing a method for suppressing oscillator phase noise according to any one of claims 1 to 6, the apparatus comprises: The phase noise standard signal determination module is used to traverse the phase noise sources of the oscillator to identify noise, generate multiple phase noise signals, perform standardization processing on the multiple phase noise signals, and determine multiple phase noise standard signals. The noise change state data construction module is used to dynamically track the multiple phase noise standard signals, that is, to monitor the changes of the phase noise standard signals over time and construct noise change state data. The noise filtering data determination module is used to perform noise prediction on the oscillator based on the noise change state data, obtain phase noise prediction results, filter the multiple phase noise standard signals based on the phase noise prediction results, and determine noise filtering data. The noise suppression result acquisition module is used to perform a phase-locked loop based on the noise filtering data, generate a noise phase difference, construct a multi-level filtering structure, and perform multi-layer noise suppression on the multiple phase noise standard signals by combining the multi-level filtering structure with the noise phase difference to obtain a noise suppression result. The adaptive suppression module is used to perform suppression detection based on the noise suppression results, provide feedback on the suppression effect, and formulate a suppression strategy to adaptively suppress the phase noise of the oscillator. The feedback on the suppression effect refers to transmitting the suppression detection results back to the control system or filter as a basis for adjustment and optimization.

Images