Method for removing phase noise of millimeter wave interferometric radar based on first path signal

Abstract

The application discloses a millimeter wave interferometric radar phase noise elimination method based on a first path signal, which comprises the following steps: based on the original echo data of the millimeter wave interferometric radar, the first path signal coupled by the radar transmitting and receiving antenna is locked in the time delay domain through the power time delay spectrum, and the original first path phase sequence is extracted; the original first path phase sequence is subjected to the purification treatment of the fusion Hampel robust pre-filtering and the Gaussian mixture model to obtain a pure phase noise sequence representing the radar hardware local oscillator phase fluctuation; a space-time double difference interferometric model is constructed, the pure phase noise sequence is taken as a reference benchmark, the target observation phase sequence is subjected to the space difference and time difference processing, the common mode phase noise and static phase constant are eliminated, and a target real dynamic micro-deformation sequence is reconstructed. The application directly utilizes the first path signal generated by the inherent coupling between the transmitting and receiving antennas of the millimeter wave interferometric radar as the phase noise reference, and realizes the effective elimination of the hardware phase noise.

Description

Technical Field

[0001] This invention relates to the field of radar noise reduction methods, and in particular to a phase noise reduction method for millimeter-wave interferometric radar based on the first-path signal. Background Technology

[0002] High-precision micro-deformation monitoring of large infrastructure such as bridges and super high-rise buildings is crucial for accurately assessing structural operational status, identifying abnormal dynamic responses, and ensuring public safety. Millimeter-wave interferometric radar, with its advantages of high precision, high dynamics, and non-contact operation, has become an important technical means for monitoring micro-deformation in infrastructure. Its sensing process is based on the principle of microwave interferometry, transmitting microwave signals and receiving target echoes, using the phase changes of the echoes to calculate the target's minute displacement. However, in practical applications, the measurement accuracy of millimeter-wave interferometric radar is severely limited by hardware performance. Since the transmitter and receiver of a radar system typically share a local oscillator for frequency synthesis and conversion, the frequency instability of the local oscillator during operation introduces random phase fluctuations, forming nonlinear multiplicative phase noise. This phase noise is further converted into displacement measurement errors in the interferometric calculation. Considering that the dynamic response of infrastructure is typically on the millimeter or even sub-millimeter scale, the phase noise introduced by the local oscillator severely interferes with the extraction of real micro-deformation signals, disrupts the continuity and stability of long-term phase sequences, and thus reduces the reliability of the monitoring data.

[0003] Currently, handling phase errors in dynamic measurements of radar systems still faces multiple challenges. On the one hand, common spatial domain or transform domain filtering methods often struggle to balance denoising and signal fidelity when dealing with hardware phase noise exhibiting non-stationary characteristics. While conventional linear low-pass filtering can reduce the overall variance, it tends to forcibly suppress effective deformation signals containing real high-frequency deflection, leading to the loss of details in the structural dynamic response. On the other hand, the phase noise generated by the local oscillator contains low-frequency components with long-term memory characteristics, and the ranging errors it causes exhibit random walk and divergence as the observation time increases. Such non-steady-state random processes cannot achieve statistical convergence through simple time-domain averaging or increasing the radar coherence accumulation time. Furthermore, in complex urban open spaces, the echoes received by the radar not only contain phase noise from the hardware itself but also are mixed with environmental electromagnetic pulse interference and multipath clutter, making it even more difficult to isolate the underlying real hardware errors.

[0004] Because the radar transmit and receive links share the same local oscillator, all types of echoes involved in the system's transmission and reception are affected by phase noise from the same source, exhibiting significant common-mode characteristics. Therefore, accurately finding and extracting phase noise reference signals unaffected by external target deformation modulation from complex environments without altering the existing radar hardware architecture, and then precisely eliminating common-mode phase errors while simultaneously reconstructing the target's true dynamic micro-deformation sequence with high fidelity, has become a critical technical challenge for achieving high-precision monitoring with millimeter-wave interferometric radar. Therefore, a phase noise elimination method for millimeter-wave interferometric radar based on the first-path signal is needed. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention provides a method for phase noise elimination of millimeter-wave interferometric radar based on the first-path signal.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention includes the following steps: S1: Based on the raw echo data of millimeter-wave interferometric radar, the initial path signal of the radar transmitting and receiving antenna coupling is locked in the time delay domain by the power delay spectrum, and the raw initial path phase sequence is extracted. S2: The original first-path phase sequence is purified by fusing Hampel robust pre-filtering and Gaussian mixture model to remove transient pulse interference and environmental clutter, and a clean phase noise sequence characterizing the local oscillator phase fluctuation of the radar hardware is obtained. S3: Construct a spatiotemporal double-difference interferometry model, using the pure phase noise sequence as a reference, perform spatial and temporal difference processing on the target observation phase sequence, remove common-mode phase noise and static phase constant, and reconstruct the target's true dynamic micro-deformation sequence.

[0007] Furthermore, the step of locking the initial path signal of the radar transceiver antenna coupling in the time delay domain through the power delay spectrum and extracting the original initial path phase sequence specifically includes: S11: Acquire multi-carrier channel frequency response data within the continuous radar observation period, and convert the complex frequency domain sequence into a time-delay domain channel impulse response sequence through inverse fast Fourier transform; S12: Calculate the squared magnitude of the channel impulse response and generate a power delay spectrum to characterize the signal energy distribution of different physical paths; S13: Set the near-field delay observation window, and locate the delay position of the first path signal by jointly searching for the shortest delay and the maximum energy within the window; S14: Extract the phase angle of the channel impulse response at the corresponding time delay of the first path signal to obtain the original first path phase sequence.

[0008] Furthermore, the purification process of fusing Hampel robust pre-filtering and Gaussian mixture model on the original first-path phase sequence specifically includes: S21: Perform one-dimensional phase unwrapping on the original first-path phase sequence to restore the continuous physical phase change trend and obtain the unwrapped phase sequence; S22: The unwrapped phase sequence is processed by Hampel robust pre-filtering to remove abnormal outliers of transient pulse jumps and obtain the initial filtered phase sequence; S23: Construct a Gaussian mixture model to model the statistical characteristics of the initial filtered phase sequence, estimate the model parameters iteratively through the expectation-maximization algorithm, separate and extract the dominant principal components that characterize the local oscillator phase noise, and obtain the pure phase noise sequence.

[0009] Furthermore, the Hampel robust pre-filtering process is as follows: Set a sliding window and calculate the median and absolute median difference of the phase samples within the window; Anomaly detection thresholds are determined based on the absolute median difference and the scaling factor. Phase samples that deviate from the median by more than the threshold are identified as impulse interference, and the median is used to replace the abnormal sample.

[0010] Furthermore, the iterative estimation of model parameters using the expectation-maximization algorithm specifically includes: iteratively estimating the weights, mean, and variance of each Gaussian component until the likelihood function converges; The extraction of the dominant principal component characterizing the local oscillator phase noise specifically involves: extracting the dominant principal component from the separated components based on the variance aggregation characteristics of the hardware phase noise.

[0011] Furthermore, the spatial and temporal difference processing of the target observation phase sequence is used to reconstruct the target's true dynamic micro-deformation sequence, specifically including... S31: Extract the original observation phase sequence of the far-field echo of the monitored target, perform sampling point-by-sampling spatial difference with the clean phase noise sequence, cancel common-mode hardware phase noise, and obtain the spatial difference phase sequence; S32: Using the spatial difference phase at the start of the observation as a reference, perform time difference on the spatial difference phase sequence to eliminate the initial static path phase constant and obtain a pure dynamic deformation phase sequence; S33: Based on the principle of microwave interferometry, the pure dynamic deformation phase sequence is converted into the physical relative displacement sequence of the target.

[0012] Furthermore, the spatial difference is achieved by subtracting the target observation phase sequence from the clean phase noise sequence point by point to cancel the common-mode phase error of the radar transceiver link; the temporal difference is achieved by calculating the change of the spatial difference phase at each moment relative to the initial moment to eliminate the static phase constant introduced by the initial physical path.

[0013] Furthermore, the pure dynamic deformation phase sequence is converted into the target's physical relative displacement sequence, as shown in the following formula:

[0014] in, The center wavelength of the signal transmitted by the millimeter-wave radar. This refers to the pure dynamic deformation phase value after removing hardware noise and static constants.

[0015] The beneficial effects of this invention are: This invention directly utilizes the first-path signal generated by the inherent coupling between the transmitting and receiving antennas of the millimeter-wave interferometric radar as a phase noise reference. It does not require upgrading the high-performance oscillator inside the radar, nor does it require setting up a special artificial reflector as a reference at the measurement site. It can effectively eliminate hardware phase noise without changing the existing hardware architecture of the radar. To address the issue that the first-path signal in open environments is susceptible to interference from external electromagnetic pulses and multipath clutter, this invention employs a cascaded purification strategy that combines Hampel robust pre-filtering with Gaussian mixture model statistical separation. This strategy effectively eliminates transient abnormal jumps and slowly changing residual clutter, accurately extracting a clean phase noise sequence that characterizes the true phase fluctuation of the radar local oscillator from complex environments. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating the principle of phase noise removal in millimeter-wave interferometric radar based on the first-path signal as described in this invention. Detailed Implementation

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0018] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0019] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0020] To this end, this invention proposes a phase noise reduction method for millimeter-wave interferometric radar based on the first-path signal, so as to achieve high-precision and high-fidelity monitoring of millimeter-level micro-deformations of large infrastructure.

[0021] This method identifies the first-path signal that can characterize phase noise and uses it to remove phase noise.

[0022] First, the initial path signal formed by the coupling between the radar transceiver antennas is locked in the time delay domain using the power delay spectrum (PDP) energy integration, and its original phase sequence is extracted to provide the original reference for hardware phase noise. To address the issue that the first-path signal is susceptible to non-stationary electromagnetic interference in actual monitoring environments, this paper integrates Hampel robust pre-filtering and Gaussian mixture model (GMM) statistical separation techniques to accurately remove transient pulse jumps and residual environmental clutter, reconstructing a pure phase sequence that can truly characterize the local oscillator phase fluctuation. Finally, a spatiotemporal double-difference interferometry (ST-DDI) model is constructed. By using spatial dimension difference to cancel the common-mode phase error in the target echo and the first-path signal, and combining it with temporal dimension difference to eliminate the initial static constant, the true dynamic relative displacement information of the target can be completely preserved and reconstructed while effectively suppressing hardware phase noise interference.

[0023] like Figure 1 As shown, the present invention includes the following steps: S1: Based on the raw echo data of millimeter-wave interferometric radar, the initial path signal of the radar transmitting and receiving antenna coupling is locked in the time delay domain by the power delay spectrum, and the raw initial path phase sequence is extracted. S2: The original first-path phase sequence is purified by fusing Hampel robust pre-filtering and Gaussian mixture model to remove transient pulse interference and environmental clutter, and a clean phase noise sequence characterizing the local oscillator phase fluctuation of the radar hardware is obtained. S3: Construct a spatiotemporal double-difference interferometry model, using the pure phase noise sequence as a reference, perform spatial and temporal difference processing on the target observation phase sequence, remove common-mode phase noise and static phase constant, and reconstruct the target's true dynamic micro-deformation sequence.

[0024] (1) Search and extraction of first path signal based on power delay spectrum (PDP) Acquire multi-carrier channel frequency response data of a millimeter-wave interferometric radar over a continuous observation period. Let the number of frequency points within each sweep or pulse period be... The inverse fast Fourier transform (IFFT) is applied to the original complex frequency domain sequence to transform it into the time delay domain (i.e., the range domain), obtaining the channel impulse response (CIR) sequence. The discrete expression for CIR is: (1) in, Indexing the slow time dimension of radar observations. For frequency point index, For the first Time, Number Channel frequency response at each frequency point For corresponding delay Complex impulse response at the point.

[0025] The number of frequency points K determines the range resolution and the maximum unambiguous range, and can be set to an appropriate value based on the sweep bandwidth and maximum observation range of the radar system. When the sweep bandwidth is wide enough, the range resolution can reach the millimeter level or even higher, which is sufficient to effectively separate the first path signal from adjacent multipath signals in the time delay domain.

[0026] IFFT transforms the frequency domain channel response into the time domain impulse response, enabling physical paths with different propagation delays to be distinguished in the distance dimension. The first path signal, due to its shortest path, appears at the position with the smallest delay.

[0027] To stably identify signal paths, the squared magnitude of the channel impulse response is calculated to generate a power delay spectrum (PDP) to characterize the signal energy distribution along different physical paths in space. (2) Considering that the distance between radar transceiver antennas is fixed and the physical distance is the shortest, the first path signal generated by its internal spatial coupling is the peak value that arrives first and has concentrated energy in the time delay domain. PDP reflects the echo signal power of each time delay unit. For the first path signal, its power is usually much higher than the noise floor and environmental multipath, so reliable positioning can be achieved through peak detection.

[0028] To avoid interference from multipath propagation in the far-field environment, a near-field time delay observation window is set. Within this window, a local extremum search is performed to pinpoint the precise time delay location of the first-path signal. : (3) After locking the first path position, extract the phase angle of the complex value of the impulse response at that time delay to obtain the original first path phase sequence containing the phase fluctuation of the system's local oscillator. : (4) in, and These represent the operations of extracting the imaginary and real parts of a complex number, respectively. The extracted original first-diameter phase sequence. It will serve as the phase noise reference signal for subsequent hardware common-mode phase error reconstruction.

[0029] (2) First-path phase cleanup combining Hampel robust filtering and Gaussian mixture model (GMM) Because the radar operates in an open environment, the acquired raw first-path phase sequence In long-term observations, external transient electromagnetic pulse interference and residual multipath clutter are inevitably coupled. In order to obtain a clean local oscillator phase noise reference, the original sequence needs to be unwrapped and cascaded for purification.

[0030] First, for those folded in The original phase of the interval is unwrapped in one dimension to restore its continuous physical change trend, resulting in an unwrapped phase sequence. .

[0031] Subsequently, Hampel robust pre-filtering was used to remove heavy-tailed transient pulse jumps (i.e., anomalous outliers) caused by external electromagnetic interference. The sliding window length was set to... For the set of phase samples within the window, calculate the value of the sample. And the absolute median deviation (MAD): (5) (6) Set the judgment threshold as ( The scaling factor is typically three times 1.4826. If the current center point phase deviates from the median by more than this threshold, it is determined to be impulse interference, and the median is used as the scaling factor. Replace the values; otherwise, leave the original values ​​unchanged. After Hampel filtering, the output is the initial filtered sequence with transient jumps removed. .

[0032] The sliding window length L can be set to an odd number based on the duration of the impulse interference and the sampling rate to ensure window symmetry; the scaling factor k can be selected as an appropriate positive value based on the desired anomaly detection sensitivity. Anomaly detection thresholds are determined using the absolute median and the scaling factor. Phase samples deviating from the median by more than the threshold are identified as impulse interference, and the median is used to replace these anomalous samples. Because the absolute median is robust to heavy-tailed distributions, this method is more suitable for handling impulse noise than the standard deviation.

[0033] To further remove slowly changing residual environmental clutter from the initial filter sequence, a Gaussian mixture model (GMM) is constructed. The statistical characteristics are used for modeling and separation. It is assumed that the phase fluctuations introduced by the pure hardware local oscillator and external clutter are statistically independent. A linear combination of Gaussian probability density functions is used to approximate the probability distribution of the mixed signal: (7) in, For the input phase sample, For the first The weights of the Gaussian components (and satisfying) ), and The first The mean and variance of each component. It is a standard Gaussian probability density function.

[0034] The GMM parameters are iteratively estimated using the Expectation-Maximization (EM) algorithm until the likelihood function converges. From the separated statistical components, based on the variance clustering characteristics of the hardware phase noise and the derived characteristics of Gaussian white noise, the dominant principal component representing the random walk of the local oscillator is extracted, and the corresponding time series is reconstructed, ultimately obtaining a clean local oscillator phase noise sequence. .

[0035] Step 3: High-fidelity reconstruction of the target signal based on spatiotemporal double difference interferometry (ST-DDI) Obtaining a pure local oscillator phase noise sequence Simultaneously, for the infrastructure targets to be monitored, the far-field time delay unit where the target is located is locked through radar echo search. And extract its original observation phase sequence. The observed phase includes the target's physical displacement information, the common-mode hardware phase noise of the transmit / receive link, and the initial static path constant.

[0036] Perform spatial dimension difference. Utilize the clean local oscillator phase noise sequence extracted in step (2). As a reference, the phase sequence of the target observation is subtracted point-by-point. Since the first-path reference signal and the target echo signal share the same local oscillator, their hardware phase fluctuations are highly correlated. Spatial differential analysis effectively decouples and cancels common-mode hardware phase noise in the echo signal. (8) in, This is the phase sequence after spatial difference, where the hardware phase noise term that fluctuates randomly over time has been eliminated.

[0037] Subsequently, time-differencing is performed. To eliminate the static phase constant caused by the initial physical path length difference between the radar antenna and the target, the observation start time is selected. Using the spatial difference phase as a reference, the changes at subsequent time points relative to the initial time are calculated to complete the spatiotemporal double-difference interferometry process. (9) in, This refers to the pure dynamic deformation phase after removing hardware noise and static constants.

[0038] The spatial difference is achieved by subtracting the target observation phase sequence from the clean phase noise sequence point by point to cancel the common-mode phase error of the radar transceiver link; the temporal difference is achieved by calculating the change of the spatial difference phase at each moment relative to the initial moment to eliminate the static phase constant introduced by the initial physical path.

[0039] Finally, based on the principle of microwave interferometry, the phase sequence after double difference is converted into a physical relative displacement sequence of the target. : (10) in, This refers to the center wavelength of the millimeter-wave radar's transmitted signal. Through the aforementioned spatiotemporal dual-differential processing, this invention can directly cancel common-mode low-frequency drift from the physical link source without relying on external high-performance hardware support. This allows for the high-fidelity reconstruction of the infrastructure's millimeter-level real dynamic micro-deformation response while suppressing hardware phase noise interference.

[0040] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for phase noise removal in millimeter-wave interferometric radar based on the first-path signal, characterized in that, Includes the following steps: S1: Based on the raw echo data of millimeter-wave interferometric radar, the initial path signal of the radar transmitting and receiving antenna coupling is locked in the time delay domain by the power delay spectrum, and the raw initial path phase sequence is extracted. S2: The original first-path phase sequence is purified by fusing Hampel robust pre-filtering and Gaussian mixture model to obtain a clean phase noise sequence characterizing the local oscillator phase fluctuation of the radar hardware. S3: Construct a spatiotemporal double-difference interferometry model, using the pure phase noise sequence as a reference, and reconstruct the target observation phase sequence by performing spatial and temporal difference processing to obtain the target's true dynamic micro-deformation sequence.

2. The method for phase noise removal in millimeter-wave interferometric radar based on the first-path signal according to claim 1, characterized in that, The step of locking the initial path signal of the radar transceiver antenna coupling in the time delay domain through the power delay spectrum and extracting the original initial path phase sequence specifically includes: S11: Acquire multi-carrier channel frequency response data within the continuous radar observation period, and convert the complex frequency domain sequence into a time-delay domain channel impulse response sequence through inverse fast Fourier transform; S12: Calculate the squared magnitude of the channel impulse response and generate a power delay spectrum to characterize the signal energy distribution of different physical paths; S13: Set the near-field delay observation window, and locate the delay position of the first path signal by jointly searching for the shortest delay and the maximum energy within the window; S14: Extract the phase angle of the channel impulse response at the corresponding time delay of the first path signal to obtain the original first path phase sequence.

3. The method for phase noise removal in millimeter-wave interferometric radar based on the first-path signal according to claim 1, characterized in that, The purification process of fusing Hampel robust pre-filtering and Gaussian mixture model to the original first-path phase sequence specifically includes: S21: Perform one-dimensional phase unwrapping on the original first-path phase sequence to restore the continuous physical phase change trend and obtain the unwrapped phase sequence; S22: The unwrapped phase sequence is processed by Hampel robust pre-filtering to remove abnormal outliers of transient pulse jumps and obtain the initial filtered phase sequence; S23: Construct a Gaussian mixture model to model the statistical characteristics of the initial filtered phase sequence, estimate the model parameters iteratively through the expectation-maximization algorithm, separate and extract the dominant principal components that characterize the local oscillator phase noise, and obtain the pure phase noise sequence.

4. The method for phase noise removal in millimeter-wave interferometric radar based on the first-path signal according to claim 3, characterized in that, The Hampel robust pre-filtering process is as follows: Set a sliding window and calculate the median and absolute median difference of the phase samples within the window; Anomaly detection thresholds are determined based on the absolute median difference and the scaling factor. Phase samples that deviate from the median by more than the threshold are identified as impulse interference, and the median is used to replace the abnormal sample.

5. The method for phase noise removal in millimeter-wave interferometric radar based on the first-path signal according to claim 3, characterized in that, The iterative estimation of model parameters using the expectation-maximization algorithm specifically includes: iteratively estimating the weights, mean, and variance of each Gaussian component until the likelihood function converges; The extraction of the dominant principal component characterizing the local oscillator phase noise specifically involves: extracting the dominant principal component from the separated components based on the variance aggregation characteristics of the hardware phase noise.

6. The method for phase noise removal in millimeter-wave interferometric radar based on the first-path signal according to claim 1, characterized in that, The target observation phase sequence is reconstructed by spatial and temporal difference processing to obtain the target's true dynamic micro-deformation sequence, specifically including... S31: Extract the original observation phase sequence of the far-field echo of the monitored target, perform sampling point-by-sampling spatial difference with the clean phase noise sequence, cancel common-mode hardware phase noise, and obtain the spatial difference phase sequence; S32: Using the spatial difference phase at the start of the observation as a reference, perform time difference on the spatial difference phase sequence to eliminate the initial static path phase constant and obtain a pure dynamic deformation phase sequence; S33: Based on the principle of microwave interferometry, the pure dynamic deformation phase sequence is converted into the physical relative displacement sequence of the target.

7. The method for phase noise removal in millimeter-wave interferometric radar based on the first-path signal according to claim 6, characterized in that, The pure dynamic deformation phase sequence is converted into the target's physical relative displacement sequence using the following formula: ； in, The center wavelength of the signal transmitted by the millimeter-wave radar. This refers to the pure dynamic deformation phase value after removing hardware noise and static constants.

Images