Airship SAR fine motion compensation method based on DSP

Abstract

The invention belongs to the technical field of airship SAR fine motion compensation. The invention provides an airship SAR fine motion compensation method based on a DSP. According to the embodiment of the invention, the PGA and MD algorithms are fused, low / high order errors are compensated through the PGA, the azimuth modulation frequency is accurately estimated through the MD, and the second-order phase residual error is corrected; and a high-frequency vibration space-variant model is constructed, and a time-varying phase error estimation module is introduced, so that the improvement of azimuth and distance resolution is ensured. A self-adaptive sub-block division strategy is designed, sub-apertures are overlapped by half length, and phase error continuity is guaranteed; a parameter self-optimization mechanism is introduced, and the PGA / MD sub-aperture length, the number of samples and the number of iterations are optimized in real time in combination with scene features. A PGA + MD refined model is constructed, the long aperture characteristic is adapted, and the DSP load is reduced; a real-time priority resource scheduling strategy is designed, high-frequency access data is stored by adopting an L2 cache to reduce delay, and power consumption is controlled while high resolution is guaranteed in combination with dynamic task allocation.

Description

Technical Field

[0001] This disclosure relates to the field of fine motion compensation technology for airship SAR, and more particularly to a DSP-based fine motion compensation method for airship SAR. Background Technology

[0002] Airship SAR has advantages such as low cost, flexible routes, wide angle measurement range, and long dwell time, and has significant application potential in fields such as disaster monitoring, infrastructure inspection, and border patrol.

[0003] However, the unique characteristics of the airship platform also present two core technical challenges that severely restrict its high-precision real-time imaging capabilities: ① Accumulation of motion errors caused by low-speed flight: The airship's cruising speed is 20-30 km / h, far lower than that of an aircraft. The synthetic aperture time is extended to 20-30 times that of an aircraft, and motion errors accumulate to the wavelength level over a long integration time. Even after basic compensation, the residual phase error still does not meet the requirements for high-resolution imaging. ② Space-varying phase errors caused by high-frequency vibrations: High-frequency angular vibrations caused by wind disturbances lead to slight changes in radar beam pointing, introducing space-varying phase errors, resulting in target defocusing and geometric distortion. Traditional compensation methods fail, and the image ISLR deteriorates by 8-12 dB, reducing imaging quality.

[0004] Traditional motion compensation mainly relies on INS / GPS combination to calculate slant range error compensation echoes through position and attitude information, but it has inherent limitations: GPS positioning error is difficult to meet the millimeter-level compensation requirements, INS sampling frequency is difficult to capture high-frequency vibrations of airships, traditional uniform linear model cannot adapt to three-axis coupled vibrations, and high-frequency spatial phase error still remains after basic compensation, resulting in image blurring.

[0005] Although the PGA algorithm can correct residual phase errors, its O(N) 3 The complexity means that processing a single frame of 2048×2048 data points on a general-purpose processor takes more than 10 seconds, which cannot meet real-time requirements. Although the DSP platform has the potential for acceleration due to its multi-core parallel architecture, existing solutions fail to fully utilize its hardware characteristics, resulting in insufficient algorithm speedup and still failing to break through the real-time threshold.

[0006] Current airship SAR imaging technology has the following drawbacks: (1) The motion compensation accuracy is insufficient and cannot meet the high-precision imaging requirements of airship SAR. Sensor accuracy bottlenecks: INS angle error ≥0.1°, GPS positioning error ≥1m, only meter-level coarse error is compensated, reducing azimuth resolution; lack of high-frequency vibration compensation: INS sampling frequency cannot capture high-frequency angular jitter and linear vibration of the airship, ISLR of the compensated image deteriorates, and defocus remains; failure of non-stationary motion modeling: traditional uniform linear model cannot handle three-axis coupled vibration, and azimuth spatial variation error is difficult to correct.

[0007] (2) The fine compensation algorithm has high computational complexity and serious real-time performance issues. Algorithm complexity skyrockets: The contrast-maximizing method has a complexity of O(N). 4 The single-frame processing time is >20 seconds; the minimum entropy method requires multiple FFTs, resulting in large memory accesses and failing to meet real-time requirements; the serial processing architecture is inefficient: the improved PGA method uses serial sub-block processing, and the computational load increases exponentially with the number of sub-blocks K. 2 The speedup is only 2-3 times, resulting in a long processing time per frame and insufficient utilization of hardware features: existing algorithms are not optimized for features such as DSP multi-core parallelism, EDMA high-speed transmission, and inline assembly instructions.

[0008] (3) The air-varying phase error has poor adaptability and cannot match the vibration characteristics of the airship. Lack of spatial variation modeling: Existing methods assume that the phase error is spatially invariant, which cannot handle target position errors caused by minute angular jitter; Mismatch in scene adaptation: The sub-block strategy of the improved PGA on the spacecraft is suitable for the slowly varying spatial variation of the ionosphere, but it is difficult to cope with the high-frequency and strong spatial variation of the airship. Moreover, the sub-block still assumes that the error is spatially invariant and lacks a mechanism for handling edge spatial variation gradients, resulting in large compensation residuals; High parameter sensitivity: The core parameters are optimized for the low-frequency and large-scale spatial variation of the spacecraft, which is not compatible with the high-frequency and small-scale spatial variation characteristics of the airship, resulting in poor adaptability.

[0009] (4) Lack of algorithm and hardware co-optimization makes it impossible to balance accuracy and real-time performance. The algorithm is not optimized for low-speed characteristics: the existing fine compensation algorithm does not take into account the long synthesis aperture time of the airship. The amount of computation and storage access increases linearly with the aperture time, which the DSP platform cannot handle; the real-time architecture design is lacking: the lack of deep scheduling of hardware resources such as DSP cache leads to long delays in fine compensation processing, which cannot meet the requirements of real-time imaging.

[0010] In summary, current airship SAR imaging faces a dual technical challenge: high-precision fine-grained compensation and real-time processing. Existing technologies, lacking both algorithmic innovation and deep collaborative optimization of hardware architecture, cannot simultaneously meet these stringent requirements.

[0011] Therefore, it is necessary to improve one or more of the problems existing in the above-mentioned related technical solutions.

[0012] It should be noted that this section is intended to provide background or context for the technical solutions of this disclosure as set forth in the claims. The description herein does not constitute an admission that it is prior art simply because it is included in this section. Summary of the Invention

[0013] The purpose of this disclosure is to provide a DSP-based method for fine motion compensation of airship SAR, thereby overcoming, to at least some extent, one or more problems caused by the limitations and defects of related technologies.

[0014] According to embodiments of this disclosure, a DSP-based method for fine motion compensation of airship SAR is provided, comprising: Step S1: Phase error compensation is performed based on the PGA phase gradient autofocus algorithm using a multi-core DSP. The full aperture data is divided into multiple sub-aperture data with a preset degree of overlap; The SAPGA sub-function algorithm is used to estimate the phase error of each sub-aperture data in parallel to obtain the local phase error estimate of each sub-aperture data. The local phase error estimates of each sub-aperture data are stitched together to generate the initial full-aperture phase error; Phase error compensation is performed on the full aperture data using the initial full aperture phase error to obtain the echo signal; Step S2, First MD Doppler frequency estimation and compensation processing; Set the sub-aperture length and sub-aperture step, use the SAMD function algorithm to estimate the modulation frequency of each sub-aperture in parallel, and interpolate to generate the first global modulation frequency; Phase compensation of the echo signal is performed using the first global frequency modulation to obtain the MD-compensated echo signal; Step S3, Second MD Doppler frequency modulation estimation and compensation processing; Increase the sub-aperture length and sub-aperture step, use the SAMD function algorithm to estimate the modulation frequency of each sub-aperture in parallel, and interpolate to generate the second global modulation frequency; Phase compensation is performed on the echo signal after the first MD compensation using the second global frequency modulation to obtain the target full aperture data.

[0015] Furthermore, the multi-core DSP has 8 cores, including core 0, core 1, core 2, core 3, core 4, core 5, core 6 and core 7; among them, core 0 is used to pre-calculate the FFT / IFFT rotation factor, the distance gate sampling position vector, the azimuth frequency modulation and the azimuth time square vector, and store them in L2.

[0016] Furthermore, the step of dividing the full aperture data into multiple sub-aperture data with a preset overlap includes: The Fork-Join model is used to divide the sub-aperture data into 8 blocks according to a preset rule, and assign them to cores 0 to 7. The preset rule is CORE_SEC = nrn / 8. Each core calculates the memory access offset based on its CoreID, and the starting offset for each core is nan. CORE_SEC Data word length CoreID, where nrn is the number of points in the distance dimension and nan is the number of points in the orientation dimension.

[0017] Furthermore, the step of using the SAPGA sub-function algorithm to perform phase error estimation on each sub-aperture data in parallel to obtain the local phase error estimate value of each sub-aperture data includes: Based on the sub-aperture data, the sub-aperture energy was calculated using the vecsum optimization function and the VecSqrt optimization function, and strong scattering point signals with energies greater than or equal to 4 times the mean were selected. After zero-padding the signal at each strong scattering point, perform an FFT with four times the aperture length. Use the inline C function and DSPLIB function to find the spectral peak and calculate the signal frequency. Generate a compensation factor to perform circular displacement and linear phase compensation. Based on the length of the window, the iteration window is decreased sequentially, and the compensated strong scattering point signal is processed cyclically. In each iteration, FFT and IFFT are performed on the current strong scattering point signal, and the DSP's fftSPxSP and ifftSPxSP functions are used for acceleration. Frequency term compensation is generated through the optimization function, and the peak value is shifted to 0 frequency and 0 phase. The phase error is calculated using the optimized vector dot product function and conjugate dot product function expanded by inline C+ loops, and the results are weighted according to the energy of the strong scattering point signal. After traversing all strong scattering point signals, the WLS-PGA strategy is used to dynamically allocate the weight of strong points according to the sample variance to handle the global non-space variable phase error. The LWLS-PGA strategy is used to divide the entire aperture into adaptive sub-blocks. The phase gradient is obtained by multiplying the azimuth to the conjugate of adjacent sampling points, and the weighted summation and integration are used to obtain the phase error estimate for this iteration. Iterate through the different windows in a loop until all windows have been iterated over; Polynomial fitting is performed on the phase of the phase error estimate to remove linear and quadratic trend terms, so as to obtain the local phase error estimate of each sub-aperture data.

[0018] Furthermore, the step of stitching together the local phase error estimates of each sub-aperture data to generate the initial full-aperture phase error includes: For the sub-aperture data of N points, discard the first and last n1 points, keep the middle points, and use the n1 points as the splicing area; For the first aperture, unwrap the phase n2 points before the phase, save the phase n3 points as the reserved area, and cache the phase n1 points as the splicing area. For the intermediate sub-aperture, unwrap points n1~n2-1 as the effective area, calculate the phase difference with the splicing area of ​​the previous sub-aperture, and after linear fitting with polyfit, subtract the corrected phase from the splicing area and take the mean value, update the splicing area and save points n1~n3-1. For the last aperture, unwrap points n1 to N-1, process the phase difference with the intermediate aperture, and retain points n1 to n2-1 to obtain the phase of the spliced ​​full aperture. The stitched full-aperture phase is sequentially subjected to first-order polynomial fitting to remove residual linear trends, remove median offset, perform boundary expansion, and phase filtering to generate the initial full-aperture phase error.

[0019] Furthermore, the step of compensating for the phase error of the full aperture data using the initial full aperture phase error to obtain the echo signal includes: Ensure all cores are ready using SYNC0_7_ACQ, distribute tasks evenly according to distance, with each core processing CORE_SEC = number of distance dimension points / 8, and define the starting address of the data as Dadr; Call the cosdp and sindp functions to generate a factor based on the initial full aperture phase error correction array; The full aperture data is moved from DDR3 to L2, and the signal and factor multiplication is performed by the optimized VectorDotproduct to obtain the echo signal, which is then written back to DDR3. SYNC0_7_RLS confirms that all cores have completed synchronization, releases the synchronization, and allows other cores to continue processing data.

[0020] Furthermore, the steps of setting the sub-aperture length and sub-aperture step, estimating the modulation frequency of each sub-aperture in parallel using the SAMD function algorithm, and interpolating to generate the first global modulation frequency include: Initialize the sub-aperture length, sub-aperture step, and overlap; calculate the distance segment. Kernel 0 is initialized, FFT / IFFT twiddle factors are generated, and MD factors are calculated; The system operates on 8 cores in parallel, with sub-aperture tasks allocated according to CORE_SEC = nrn / 8; SYNCO_7_ACQ ensures data synchronization across all cores. For each sub-aperture, the SAMD function is called to estimate the modulation frequency, and interpolation is used to generate the first global modulation frequency.

[0021] Furthermore, the SAMD function algorithm specifically includes: Read parameters and set the DMA address; The optimization function of the SAMD function is called to calculate the average energy of the distance cell, and the top KNUM high-energy samples are selected in descending order of energy. Traverse the first KNUM high-energy samples by sequence number. For each sample: remove the quadratic term of the signal using the inertial navigation velocity, divide the signal into two sub-apertures and pad with zeros to twice the length, and accelerate the FFT with DSPF_sp_fftSPxSP. After energy normalization, VectorConjDotproduct calculates the sub-aperture spectrum conjugate correlation and adds the magnitude of the correlation results to the total correlation spectrum; After the traversal is completed, for the accumulated total correlation spectrum, DSPF_sp_maxidx finds the peak position to obtain the tuning frequency deviation ka_deviate1, and the cubic spline interpolation refines the deviation. Based on ka_deviate1, correct the baseline frequency modulation value, repeat the above steps to obtain ka_deviate2, and finally take the average frequency modulation deviation:

[0022] in, The slope distance is the distance between the midpoint and the midpoint. For reference slope distance; If the absolute value of the average frequency deviation is greater than 10, then the boundary value is taken, and the first global frequency ka_est = ka0 + ka_deviate is finally obtained.

[0023] Furthermore, the step of performing phase compensation on the echo signal using the first global frequency modulation to obtain the MD-compensated echo signal includes: Core 0 collects the first global tuning frequency of all sub-apertures, and uses the inertial navigation value to compare the tuning frequencies of the first 3 and the last 3 sub-apertures. Using cubic spline interpolation, a continuous global frequency modulation array ka_global is generated that corresponds one-to-one with the full aperture azimuth sampling points based on the frequency modulation estimate of the sub-aperture center. A more accurate airship speed is calculated by back-calculating the mean of the global frequency modulation array ka_global, and the cache is refreshed. Each core generates a quadratic phase compensation factor based on the ka_global array; Phase compensation is performed on the PGA-compensated echo signal in parallel to obtain the MD-compensated echo signal.

[0024] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects: In the embodiments of this disclosure, the above-described DSP-based airship SAR fine motion compensation method firstly improves motion compensation accuracy and achieves high-resolution imaging: By integrating PGA and MD algorithms, low / high-order errors are first compensated by PGA, and then the azimuth frequency is accurately estimated by MD to correct the second-order phase residual. A high-frequency vibration spatial variation model is constructed, and a time-varying phase error estimation module is introduced to ensure a resolution of 1m in both azimuth and range directions.

[0025] Second, reduce the computational complexity of the algorithm to enable real-time processing on the DSP platform: Based on the dynamic termination of PGA / MD iteration (strong point screening) based on signal-to-noise ratio, parallel processing replaces serial sub-block partitioning, reducing the amount of computation; 8-core task scheduling combined with inline assembly + loop unrolling accelerates core operations, and EDMA enables high-speed data transmission, compressing the processing time of 8192×8192 scene to 3.1s.

[0026] Third, enhance adaptability to air-varying errors and match the vibration characteristics of the airship: An adaptive sub-block partitioning strategy is designed, with half-length overlap between sub-apertures to ensure phase error continuity; a parameter self-optimization mechanism is introduced to optimize the PGA / MD sub-aperture length, number of samples, and number of iterations in real time based on scene characteristics.

[0027] Fourth, achieve collaborative optimization of algorithms and hardware to balance accuracy and real-time performance: A refined PGA+MD model is constructed specifically for low-speed scenarios to adapt to the characteristics of long aperture and reduce DSP load. A real-time priority resource scheduling strategy is designed, and L2 cache is used to store high-frequency access data to reduce latency. Combined with dynamic task allocation, power consumption is controlled while ensuring 1m resolution. Attached Figure Description

[0028] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0029] Figure 1 The diagram illustrates the steps of a DSP-based airship SAR fine motion compensation method in an exemplary embodiment of this disclosure. Figure 2 A flowchart illustrating the algorithm for implementing PGA based on DSP in an exemplary embodiment of this disclosure is shown. Figure 3 This diagram illustrates the SAPGA sub-function algorithm for sub-aperture phase error estimation implemented using DSP in an exemplary embodiment of this disclosure. Figure 4 A flowchart illustrating the algorithm for implementing MD based on DSP in an exemplary embodiment of this disclosure is shown. Figure 5The flowchart of the SAMD function (single-core frequency modulation estimation) in an exemplary embodiment of this disclosure is shown; Figure 6 This diagram illustrates the DSP Cache structure in an exemplary embodiment of this disclosure. Figure 7 This diagram illustrates the DSP memory access sequence in an exemplary embodiment of this disclosure. Figure 8 This diagram shows a magnified view of a coarse compensation based on inertial navigation in an exemplary embodiment of this disclosure. Figure 9 A magnified view of a portion of the PGA-based compensation in an exemplary embodiment of this disclosure is shown; Figure 10 This diagram shows a partially enlarged view of compensation based on a first-order MD in an exemplary embodiment of this disclosure; Figure 11 This diagram shows a partially enlarged view of the compensation based on two MD operations in an exemplary embodiment of this disclosure; Figure 12 This diagram shows a detailed compensation section based on PGA+first-order MD in an exemplary embodiment of this disclosure. Figure 13 This diagram shows a detailed compensation detail based on PGA+ and two MD operations in an exemplary embodiment of this disclosure. Figure 14 The image shows the imaging results of an 8192×8192 scene by the DSP under fine compensation of PGA+ and two MDs in an exemplary embodiment of the present disclosure. Detailed Implementation

[0030] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that this disclosure will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0031] Furthermore, the accompanying drawings are merely illustrative diagrams of embodiments of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0032] This example implementation provides a DSP-based method for fine motion compensation of airship SAR. (Reference) Figure 1 As shown, the DSP-based airship SAR fine motion compensation method may include: Step S1: Phase error compensation is performed based on the PGA phase gradient autofocus algorithm using a multi-core DSP. The full aperture data is divided into multiple sub-aperture data with a preset degree of overlap; The SAPGA sub-function algorithm is used to estimate the phase error of each sub-aperture data in parallel to obtain the local phase error estimate of each sub-aperture data. The local phase error estimates of each sub-aperture data are stitched together to generate the initial full-aperture phase error; Phase error compensation is performed on the full aperture data using the initial full aperture phase error to obtain the echo signal; Step S2, First MD Doppler frequency estimation and compensation processing; Set the sub-aperture length and sub-aperture step, use the SAMD function algorithm to estimate the modulation frequency of each sub-aperture in parallel, and interpolate to generate the first global modulation frequency; Phase compensation of the echo signal is performed using the first global frequency modulation to obtain the MD-compensated echo signal; Step S3, Second MD Doppler frequency modulation estimation and compensation processing; Increase the sub-aperture length and sub-aperture step, use the SAMD function algorithm to estimate the modulation frequency of each sub-aperture in parallel, and interpolate to generate the second global modulation frequency; Phase compensation is performed on the echo signal after the first MD compensation using the second global frequency modulation to obtain the target full aperture data.

[0033] The above-mentioned DSP-based airship SAR fine motion compensation method firstly improves motion compensation accuracy and achieves high-resolution imaging: By integrating PGA and MD algorithms, low / high-order errors are first compensated by PGA, and then the azimuth frequency is accurately estimated by MD to correct the second-order phase residual. A high-frequency vibration spatial variation model is constructed, and a time-varying phase error estimation module is introduced to ensure a resolution of 1m in both azimuth and range directions.

[0034] Second, reduce the computational complexity of the algorithm to enable real-time processing on the DSP platform: Based on the dynamic termination of PGA / MD iteration (strong point screening) based on signal-to-noise ratio, parallel processing replaces serial sub-block partitioning, reducing the amount of computation; 8-core task scheduling combined with inline assembly + loop unrolling accelerates core operations, and EDMA enables high-speed data transmission, compressing the processing time of 8192×8192 scene to 3.1s.

[0035] Third, enhance adaptability to air-varying errors and match the vibration characteristics of the airship: An adaptive sub-block partitioning strategy is designed, with half-length overlap between sub-apertures to ensure phase error continuity; a parameter self-optimization mechanism is introduced to optimize the PGA / MD sub-aperture length, number of samples, and number of iterations in real time based on scene characteristics.

[0036] Fourth, achieve collaborative optimization of algorithms and hardware to balance accuracy and real-time performance: A refined PGA+MD model is constructed specifically for low-speed scenarios to adapt to the characteristics of long aperture and reduce DSP load. A real-time priority resource scheduling strategy is designed, and L2 cache is used to store high-frequency access data to reduce latency. Combined with dynamic task allocation, power consumption is controlled while ensuring 1m resolution.

[0037] Below, we will refer to Figures 1 to 14 The steps of the DSP-based airship SAR fine motion compensation method described in this example embodiment will be explained in more detail.

[0038] In one embodiment, refer to Figure 1 The core of this application consists of PGA and two MD modules; combined with Figure 2 and Figure 3 The specific implementation and optimization steps on the DSP platform are as follows: Step S1, as follows Figure 2 As shown, the PGA phase gradient self-focusing algorithm is implemented. 1.1 Sub-aperture division and kernel 0 parameter initialization When the synthesized aperture is too large, the phase error is prone to spatial variation, causing the standard PGA to fail. Therefore, the full aperture is divided into multiple sub-apertures to ensure seamless stitching of overlapping areas. Key parameters: sub-aperture length: 128 azimuth sampling points; step size: 64 azimuth sampling points; overlap: 50%; start / end distance gates: 1 / 4 and 3 / 4 of the number of distance dimension points, respectively; number of sub-apertures = number of azimuth dimension points / number of sub-aperture steps.

[0039] The azimuth signal phase model is as follows:

[0040] in, It is the initial phase of the signal. It is the linear phase of the signal. It is the quadratic term of the signal. This is the higher-order error term of the signal (the main target of PGA correction), and the above-mentioned errors will be calculated and eliminated in sequence.

[0041] Core 0 parameter initialization: 1) Kernel 0 pre-computes FFT / IFFT rotatability factor and distance gate sampling position vector. azimuth frequency modulation azimuth-time squared vector , stored in L2.

[0042] 2) When removing the secondary frequency modulation term, the key calculations are performed by calling the optimized vecmul_u (33.76 times faster) and VectorDotproduct (30.13 times faster) functions with inline + loop expansion.

[0043] 1.2 Multi-core parallelism: Sub-aperture phase error estimation Multi-core architecture design use Figure 2 The Fork-Join model and improved synchronization mechanism in China: a) Storage optimization: 8 cores process in parallel, and DMA moves data to L2 / MSMC. Figure 7 and Figure 8 The data shows the DSP's cache storage structure and access order, demonstrating that accessing the L2 cache is much faster than accessing DDR3 memory.

[0044] b) Multi-core synchronization: Based on the 64 hardware semaphores in the DSP, mutual exclusion and synchronization operations are achieved by calling functions in the CSL library (Chip Support Library). To overcome the limitations of traditional methods in terms of synchronization interval and operation order, the following improvement strategy is adopted: Core 0 is separated from the other cores (1~7). Core 0 is responsible for flow control and external communication, while the other cores focus on data processing. Core 0's synchronization operation: Acquire and release its own semaphore; after acquisition, check the semaphore status of other cores to confirm that all are occupied; release the semaphore after other cores have completed data processing. Other cores' synchronization operation: First, confirm that Core 0's semaphore has been released; then acquire its corresponding semaphore; release the semaphore after data processing is completed.

[0045] c) Parallel model: The data is divided into 8 blocks according to "CORE_SEC = nrn / 8". Each core moves the data from DDR3 to L2 for processing and then writes it back. After completion, it waits synchronously to ensure the core load is balanced.

[0046] like Figure 3 As shown, the core steps of sub-aperture phase error estimation (SAPGA subfunction) are as follows: 1) Screening of strong scattering points Calculate the mean of the distance cell modulus, filter sample points with energy ≥ 4 times the mean, and call the VecSqrt (SIMD parallelism + inverse square root approximation, speeding up by 70.76 times) and vecsum (8 accumulators in parallel, speeding up by 45.40 times) optimization functions to improve the filtering efficiency; 2) Circular displacement and linear phase compensation Perform an FFT with 4 times the aperture length points after zero-padding the signal to locate the spectral peaks. Calculate signal frequency A compensation factor is generated to remove the linear phase.

[0047]

[0048]

[0049]

[0050] in, It is the sampling point where the maximum peak value is located. It is the aperture length. It is a compensation linear term, multiplied in the frequency domain. That is, circular displacement. It compensates for the initial phase of the echo signal.

[0051] The key to DSP implementation is to call functions such as DSPF_sp_fftSPxSP, VectorConjDotproduct, DSPF_sp_maxidx, and VecSqrt to construct a phase compensation vector, multiply the signal by the compensation term, and then use FFTshift to center the spectrum.

[0052] 3) Multi-window iterative phase optimization The window length is reduced in multiple iterations, filtering out more noise and clutter each time, thus improving the signal-to-noise ratio.

[0053] 4) WLS-LWLS fusion weighted Traditional PGA uses maximum likelihood (ML) estimation and linear unbiased variation minimum (LUMV) estimation. [9] To address the issues of high sample requirements and slow convergence in traditional PGA, WLS-PGA is introduced, which dynamically allocates weights to strong points based on sample variance, with higher weights for smaller variances, thus handling global non-vacuum-varying phase errors. Furthermore, LWLS-PGA is proposed, which divides the entire aperture into adaptive sub-blocks, ensuring that vacuum-varying errors within each sub-block are negligible and independently handling sub-aperture vacuum-varying errors. The phase gradient is calculated by multiplying the azimuth gradient by the conjugate of adjacent sampling points, and then weighted and accumulated to obtain the phase error, reducing sample requirements and improving convergence efficiency.

[0054]

[0055]

[0056]

[0057] in, These are weighting coefficients. The phase error is obtained by summing the results.

[0058] 3) and 4) DSP Implementation Details: After initializing the phase and index arrays, iterate according to the window number: ① Copy the modulation function generated by the previous phase estimation and initialize the weight array; ② For each strong scattering point, after window processing, FFTshift, and DSPF_sp_ifftSPxSP, calculate the weights based on signal quality: call optimization functions such as VecSqrt, vecsum, vecadd_u, and DSPF_sp_vecmul to calculate the amplitude, mean, and variance; ③ After normalizing the weights, fuse the phase information, calculate the phase difference through atan2, and reconstruct the phase sequence through integration; ④ Use polynomial fitting to remove linear / quadratic trends and save the final phase estimate.

[0059] 1.3 Core 0 splicing full aperture phase error By smoothly stitching together sub-aperture phases in overlapping areas, edge discontinuities are eliminated, thus improving the accuracy of full-aperture phase estimation. 1) Sub-aperture region division A three-zone model is adopted: taking a 128-point sub-aperture as an example, the first and last 32 points are discarded (phase discontinuity), the middle 64 points are retained (reliable phase), and the last 32 points are used as the splicing area (overlapping with the next sub-aperture).

[0060] 2) Phase splicing process ① First sub-aperture (i=0): Unwrap the first 96 points of the phase, save the first 64 points (reserved area), and cache the last 32 points (stitching area); ② Intermediate sub-aperture (i=1 to SANum-2): Unwrap 32-95 points (effective area), calculate the phase difference with the stitching area of ​​the previous sub-aperture, perform linear fitting with polyfit, subtract the corrected phase from the stitching area and take the average value, update the stitching area and save 32-63 points; ③ Last sub-aperture (i=SANum-1): Unwrap 32-127 points, process the phase difference in the same way as the intermediate sub-aperture, and retain 32-95 points.

[0061] 3) Global phase optimization ① Remove linear trend: Construct a position index to associate phase with spatial location, fit with a first-order polynomial and subtract fitted terms; ② Remove median offset: Center the phase after taking the median to eliminate DC offset; ③ Boundary expansion: Mirror the expansion of both ends of the data to avoid convolution edge effects; ④ Phase filtering: Load filter coefficients and perform convdp convolution, normalize and separate high-frequency phase errors; ⑤ Generate the final phase correction amount for compensation: Subtract the low-frequency component from the original phase to obtain the high-frequency error, call WritebackInvalidCache to refresh the cache and ensure multi-core data consistency.

[0062] 1.4 Multi-core parallel processing: compensating for echo signals Based on the Fork-Join model, 8 cores synchronously compensate for the echo signal (i.e., full aperture data): 1) Synchronization and parameter allocation: Ensure all cores are ready via SYNC0_7_ACQ, distribute tasks evenly according to distance, with each core processing CORE_SEC = distance dimension points / 8, and define the starting address of the data as Dadr; 2) Generate compensation factor: Call cosdp and sindp to generate the factor based on the phase error correction array of 1.3; 3) Parallel compensation: DMA moves the echo data from DDR3 to L2, performs signal-factor multiplication using the optimized VectorDotproduct, corrects the signal, and writes it back to DDR3; 4) Synchronization Release: SYNC0_7_RLS confirms that all cores have completed synchronization, releases the synchronization, and allows other cores to continue processing data.

[0063] Step S2, as follows Figure 4 As shown, the first MD frequency modulation estimation compensation 2.1 Core 0 Initialization Parameters Sub-aperture settings: Sub-aperture length nan_part = 128, sub-aperture step nan_step = 64, ensuring 50% overlap to improve interpolation continuity; Range segment division: Only process the middle reliable region to avoid near / far false targets; Pre-calculation: Generate FFT / IFFT rotation factors, as well as range-related frequency modulation base values ​​and time square arrays.

[0064] 2.2 such as Figure 5 As shown, multi-core parallel frequency modulation estimation (SAMD function) The 8 cores are assigned sub-apertures according to i % 8 = CoreID, and SAMD functions are called to execute in parallel: 1) Screening of strong scattering points The optimization functions vecadd_u, VecSqrt, and vecsum are called to calculate the average energy of the distance cell. The top KNUM strong points are selected in descending order of energy to reduce the amount of computation.

[0065] 2) Frequency modulation estimation in two iterations ① First iteration: Remove the quadratic term of the signal using the inertial navigation velocity to reduce the number of iterations. Divide the signal into two sub-apertures and pad them with zeros to twice the length. Accelerate the FFT using DSPF_sp_fftSPxSP. After energy normalization, calculate the conjugate correlation of the sub-aperture spectrum using VectorConjDotproduct. Find the peak position using DSPF_sp_maxidx to obtain the tuning frequency deviation ka_deviate1. Refine the deviation using cubic spline interpolation. ② Second iteration: Based on ka_deviate1, correct the base value of the frequency modulation, repeat the above steps to obtain ka_deviate2, and finally take the average:

[0066] The slope distance is the distance between the midpoint and the midpoint. The reference slant distance is used for distance correlation correction.

[0067] ③ Bad value limit: If the absolute value of the deviation is >10, then the boundary value is taken, and the final tuning frequency ka_est = ka0 + ka_deviate.

[0068] 2.3 Core 0 interpolation to generate global frequency modulation The tuning frequencies for the first three and last three sub-apertures are calculated using inertial navigation values ​​to avoid unreliable edge estimations and interpolation boundary effects. Kernel 0 collects all estimation results and generates the full-aperture tuning frequencies through cubic spline interpolation. The airship velocity is recalculated based on the mean tuning frequencies, and the cache is refreshed to ensure data consistency.

[0069]

[0070] 2.4 Multi-core parallel processing for phase compensation factor to compensate echo data The echo signal is compensated in parallel by 8 cores according to the distance segment. The process is as follows: 1) SYNC0_7_ACQ ensures kernel readiness.

[0071] 2) Use cosdp and sindp to generate phase compensation factors.

[0072] 3) Parallel compensation echo signal: DMA reads the distance segment data into L2, VectorDotproduct performs signal and factor multiplication, removes frequency modulation error, and writes it back to DDR3.

[0073] 4) SYNC0_7_RLS confirms that all cores have completed synchronization and releases the synchronization.

[0074] Step S3, Second MD Frequency Estimation and Compensation To improve estimation accuracy, a larger sub-aperture (nan_part=256, nan_step=128) is used, and the process is exactly the same as step S2.

[0075] Through two MD iterations, the secondary phase deviation of the signal caused by inertial navigation measurement error is effectively corrected, providing high-quality input for subsequent imaging processing and improving image resolution.

[0076] Furthermore, such as Figure 6 The diagram shows the DSP cache structure; as shown... Figure 7 The diagram shown is a sequence diagram of DSP memory access.

[0077] In a specific embodiment, experiments were conducted using traditional methods and the method of this application; Table 1 shows the inline functions called by this application; Table 2 shows the efficiency comparison of key functions after optimization of inline functions + loop unrolling (8192 complex points); Table 3 shows the time consumption of four measures jointly optimized in the 8192×8192 scenario.

[0078] Table 1. Inline functions called

[0079] Table 2. Efficiency comparison of key functions after optimization with inline functions and loop expansion (8192 complex points)

[0080] Table 3. Four measures jointly optimize time consumption in the 8192×8192 scenario.

[0081] The results are as follows Figures 8 to 14 As shown, Figure 8 This is a magnified view of the coarse compensation based on inertial navigation. Figure 9 This is a magnified view of the compensation area based on PGA. Figure 10 This is a magnified view of the compensation area based on a single MD (Mechanical Design) step. Figure 11 This is a magnified view of the compensated area based on two MD operations; Figure 12 This is a fine-compensation local map based on PGA+first-order MD; Figure 13 This is a detailed compensation local map based on PGA and two MD operations; Figure 14 The image shows the imaging results of an 8192×8192 scene by DSP under fine compensation of PGA+ and two MDs.

[0082] In summary, the beneficial effects of this application are as follows: First, improve motion compensation accuracy to achieve high-resolution imaging: By integrating PGA and MD algorithms, low / high-order errors are first compensated by PGA, and then the azimuth frequency is accurately estimated by MD to correct the second-order phase residual. A high-frequency vibration spatial variation model is constructed, and a time-varying phase error estimation module is introduced to ensure a resolution of 1m in both azimuth and range directions.

[0083] Second, reduce the computational complexity of the algorithm to enable real-time processing on the DSP platform: Based on the dynamic termination of PGA / MD iteration (strong point screening) based on signal-to-noise ratio, parallel processing replaces serial sub-block partitioning, reducing the amount of computation; 8-core task scheduling combined with inline assembly + loop unrolling accelerates core operations, and EDMA enables high-speed data transmission, compressing the processing time of 8192×8192 scene to 3.1s.

[0084] Third, enhance adaptability to air-varying errors and match the vibration characteristics of the airship: An adaptive sub-block partitioning strategy is designed, with half-length overlap between sub-apertures to ensure phase error continuity; a parameter self-optimization mechanism is introduced to optimize the PGA / MD sub-aperture length, number of samples, and number of iterations in real time based on scene characteristics.

[0085] Fourth, achieve collaborative optimization of algorithms and hardware to balance accuracy and real-time performance: A refined PGA+MD model is constructed specifically for low-speed scenarios to adapt to the characteristics of long aperture and reduce DSP load. A real-time priority resource scheduling strategy is designed, and L2 cache is used to store high-frequency access data to reduce latency. Combined with dynamic task allocation, power consumption is controlled while ensuring 1m resolution.

[0086] It should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise" in the above description indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing the embodiments of this disclosure and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the embodiments of this disclosure.

[0087] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of embodiments of this disclosure, "a plurality of" means two or more, unless otherwise explicitly specified.

[0088] In the embodiments of this disclosure, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this disclosure according to the specific circumstances.

[0089] In embodiments of this disclosure, unless otherwise expressly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature being directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature being directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0090] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0091] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the appended claims.

Claims

1. A DSP-based method for fine motion compensation of airship SAR, characterized in that, include: Step S1: Based on multi-core DSP, PGA phase gradient autofocus algorithm is used to perform phase error compensation; The full aperture data is divided into multiple sub-aperture data with a preset degree of overlap; The SAPGA sub-function algorithm is used to estimate the phase error of each sub-aperture data in parallel to obtain the local phase error estimate of each sub-aperture data. The local phase error estimates of each sub-aperture data are stitched together to generate the initial full-aperture phase error; Phase error compensation is performed on the full aperture data using the initial full aperture phase error to obtain the echo signal; Step S2, First MD Doppler frequency estimation and compensation processing; Set the sub-aperture length and sub-aperture step, use the SAMD function algorithm to estimate the modulation frequency of each sub-aperture in parallel, and interpolate to generate the first global modulation frequency; Phase compensation of the echo signal is performed using the first global frequency modulation to obtain the MD-compensated echo signal; Step S3, Second MD Doppler frequency modulation estimation and compensation processing; Increase the sub-aperture length and sub-aperture step, use the SAMD function algorithm to estimate the modulation frequency of each sub-aperture in parallel, and interpolate to generate the second global modulation frequency; Phase compensation is performed on the echo signal after the first MD compensation using the second global frequency modulation to obtain the target full aperture data.

2. The DSP-based airship SAR fine motion compensation method according to claim 1, characterized in that, The multi-core DSP has 8 cores, including core 0, core 1, core 2, core 3, core 4, core 5, core 6 and core 7; among them, core 0 is used to pre-calculate the FFT / IFFT twitch factor, the distance gate sampling position vector, the azimuth frequency modulation and the azimuth time square vector, and store them in L2.

3. The DSP-based airship SAR fine motion compensation method according to claim 2, characterized in that, The step of dividing full-aperture data into multiple sub-aperture data with a preset overlap includes: The Fork-Join model is used to divide the sub-aperture data into 8 blocks according to a preset rule, and assign them to cores 0 to 7. The preset rule is CORE_SEC = nrn / 8. Each core calculates the memory access offset based on its CoreID, and the starting offset for each core is nan. CORE_SEC Data word length CoreID, where nrn is the number of points in the distance dimension and nan is the number of points in the orientation dimension.

4. The DSP-based airship SAR fine motion compensation method according to claim 3, characterized in that, The step of using the SAPGA sub-function algorithm to perform phase error estimation on each sub-aperture data in parallel to obtain the local phase error estimate value of each sub-aperture data includes: Based on the sub-aperture data, the sub-aperture energy was calculated using the vecsum optimization function and the VecSqrt optimization function, and strong scattering point signals with energies greater than or equal to 4 times the mean were selected. After zero-padding the signal at each strong scattering point, perform an FFT with four times the aperture length. Use the inline C function and DSPLIB function to find the spectral peak and calculate the signal frequency. Generate a compensation factor to perform circular displacement and linear phase compensation. Based on the length of the window, the iteration window is decreased sequentially, and the compensated strong scattering point signal is processed cyclically. In each iteration, FFT and IFFT are performed on the current strong scattering point signal, and the DSP's fftSPxSP and ifftSPxSP functions are used for acceleration. Frequency term compensation is generated through the optimization function, and the peak value is shifted to 0 frequency and 0 phase. The phase error is calculated using the optimized vector dot product function and conjugate dot product function expanded by inline C+ loops, and the results are weighted according to the energy of the strong scattering point signal. After traversing all strong scattering point signals, the WLS-PGA strategy is used to dynamically allocate the weight of strong points according to the sample variance to handle the global non-space variable phase error. The LWLS-PGA strategy is used to divide the entire aperture into adaptive sub-blocks. The phase gradient is obtained by multiplying the azimuth to the conjugate of adjacent sampling points, and the weighted summation and integration are used to obtain the phase error estimate for this iteration. Iterate through the different windows in a loop until all windows have been iterated over; Polynomial fitting is performed on the phase of the phase error estimate to remove linear and quadratic trend terms, so as to obtain the local phase error estimate of each sub-aperture data.

5. The DSP-based airship SAR fine motion compensation method according to claim 4, characterized in that, The step of stitching together the local phase error estimates of each sub-aperture data to generate the initial full-aperture phase error includes: For the sub-aperture data of N points, discard the first and last n1 points, keep the middle points, and use the n1 points as the splicing area; For the first aperture, unwrap the phase n2 points before the phase, save the phase n3 points as the reserved area, and cache the phase n1 points as the splicing area. For the intermediate sub-aperture, unwrap points n1~n2-1 as the effective area, calculate the phase difference with the splicing area of ​​the previous sub-aperture, and after linear fitting with polyfit, subtract the corrected phase from the splicing area and take the mean value, update the splicing area and save points n1~n3-1. For the last aperture, unwrap points n1 to N-1, process the phase difference with the intermediate aperture, and retain points n1 to n2-1 to obtain the phase of the spliced ​​full aperture. The stitched full-aperture phase is sequentially subjected to first-order polynomial fitting to remove residual linear trends, remove median offset, perform boundary expansion, and phase filtering to generate the initial full-aperture phase error.

6. The DSP-based airship SAR fine motion compensation method according to claim 5, characterized in that, The step of compensating for the phase error of the full aperture data using the initial full aperture phase error to obtain the echo signal includes: Ensure all cores are ready using SYNC0_7_ACQ, distribute tasks evenly according to distance, with each core processing CORE_SEC = number of distance dimension points / 8, and define the starting address of the data as Dadr; Call the cosdp and sindp functions to generate a factor based on the initial full aperture phase error correction array; The full aperture data is moved from DDR3 to L2, and the signal and factor multiplication is performed by the optimized VectorDotproduct to obtain the echo signal, which is then written back to DDR3. SYNC0_7_RLS confirms that all cores have completed synchronization, releases the synchronization, and allows other cores to continue processing data.

7. The DSP-based airship SAR fine motion compensation method according to claim 6, characterized in that, The steps of setting the sub-aperture length and sub-aperture step, estimating the modulation frequency of each sub-aperture in parallel using the SAMD function algorithm, and interpolating to generate the first global modulation frequency include: Initialize the sub-aperture length, sub-aperture step, and overlap; calculate the distance segment. Kernel 0 is initialized, FFT / IFFT twiddle factors are generated, and MD factors are calculated; The system operates on 8 cores in parallel, with sub-aperture tasks allocated according to CORE_SEC = nrn / 8; SYNCO_7_ACQ ensures data synchronization across all cores. For each sub-aperture, the SAMD function is called to estimate the modulation frequency, and interpolation is used to generate the first global modulation frequency.

8. The DSP-based airship SAR fine motion compensation method according to claim 7, characterized in that, The SAMD function algorithm specifically includes: Read parameters and set the DMA address; The optimization function of the SAMD function is called to calculate the average energy of the distance cell, and the top KNUM high-energy samples are selected in descending order of energy. Traverse the first KNUM high-energy samples by sequence number. For each sample: remove the quadratic term of the signal using the inertial navigation velocity, divide the signal into two sub-apertures and pad with zeros to twice the length, and accelerate the FFT with DSPF_sp_fftSPxSP. After energy normalization, VectorConjDotproduct calculates the sub-aperture spectrum conjugate correlation and adds the magnitude of the correlation results to the total correlation spectrum; After the traversal is completed, for the accumulated total correlation spectrum, DSPF_sp_maxidx finds the peak position to obtain the tuning frequency deviation ka_deviate1, and the cubic spline interpolation refines the deviation. Based on ka_deviate1, correct the baseline frequency modulation value, repeat the above steps to obtain ka_deviate2, and finally take the average frequency modulation deviation: in, The slope distance is the midpoint. For reference slope distance; If the absolute value of the average frequency deviation is > 10, then the boundary value is taken, and the first global frequency ka_est = ka0 + ka_deviate is finally obtained.

9. The DSP-based airship SAR fine motion compensation method according to claim 8, characterized in that, The step of performing phase compensation on the echo signal using a first global frequency modulation to obtain the MD-compensated echo signal includes: Core 0 collects the first global tuning frequency of all sub-apertures, and uses the inertial navigation value to compare the tuning frequencies of the first 3 and the last 3 sub-apertures. Using cubic spline interpolation, a continuous global frequency modulation array ka_global is generated that corresponds one-to-one with the full aperture azimuth sampling points based on the frequency modulation estimate of the sub-aperture center. A more accurate airship speed is calculated by back-calculating the mean of the global frequency modulation array ka_global, and the cache is refreshed. Each core generates a quadratic phase compensation factor based on the ka_global array; Phase compensation is performed on the PGA-compensated echo signal in parallel to obtain the MD-compensated echo signal.

Images