Acoustic simulation method and system based on low-frequency waveguide digital mesh and geometric modeling

Abstract

This invention discloses an acoustic simulation method and system based on low-frequency waveguide digital mesh and geometric modeling, belonging to the field of acoustic simulation technology. It imports a three-dimensional scene model composed of triangular facets and initializes sound source parameters and receiver positions. Based on the three-dimensional scene model, sound source parameters, and receiver positions, multiple sound field propagation paths are obtained through geometric acoustic pre-analysis, and initial perception importance parameters are calculated for each sound field propagation path. According to the initial perception importance parameters, each sound field propagation path is assigned to a first calculation strategy or a second calculation strategy for processing. This invention, through geometric acoustic pre-analysis and real-time dynamic evaluation, assigns high-precision calculation strategies only to key sound field propagation paths with high perception contribution, while employing efficient geometric methods for a large number of secondary paths. This resource scheduling strategy based on auditory perception allows for highly focused computational resources, avoiding unnecessary consumption on secondary components.

Description

Technical Field

[0001] This invention relates to the field of acoustic simulation technology, specifically to an acoustic simulation method and system based on low-frequency waveguide digital mesh and geometric modeling. Background Technology

[0002] The ultimate goal of audible acoustic simulation is to generate sound signals that are highly faithful to subjective hearing and indistinguishable from real sounds. This goal places extreme demands on the efficiency of simulation technology, because audibility, especially real-time or interactive audibility, must complete all calculations within a limited time window. While the traditional low-frequency ripple (DWM) + high-frequency geometry hybrid modeling framework has laid the foundation for cross-frequency band simulation, its resource allocation strategy is fundamentally contradictory to the core requirement of audibility—maximizing efficiency guided by perceptual accuracy.

[0003] Current hybrid methods allocate resources based on fixed frequency divisions, leading to a globally homogenized computational pattern in low-frequency digital waveguide mesh (DWM) simulations. Regardless of the actual contribution of acoustic events to the final auditory experience, as long as their frequencies are below the dividing line, they are all simulated with full precision using computationally expensive DWM. However, in complex architectural sound fields, the low-frequency energy distribution is extremely uneven: early strong reflection paths (perceptual critical paths), crucial for establishing spatial awareness and clarity, are mixed with a large amount of weak, late-stage diffraction and modal sound (perceptual minor components) that are masked by reverberation tails. Existing methods fail to recognize this significant difference in perceptual importance, wasting substantial computational resources on minor components with negligible impact on auditory experience, resulting in low overall computational efficiency and severely limiting the feasibility of achieving real-time audibility in large-scale scenes. Summary of the Invention

[0004] The purpose of this invention is to provide an acoustic simulation method and system based on low-frequency waveguide digital mesh and geometric modeling to address the shortcomings of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an acoustic simulation method based on low-frequency waveguide digital mesh and geometric modeling, comprising:

[0006] S1. Import the 3D scene model composed of triangular facets and initialize the sound source parameters and receiver position;

[0007] S2. Based on the three-dimensional scene model, sound source parameters and receiver position, multiple sound field propagation paths are obtained through geometric acoustic pre-analysis, and initial perception importance parameters are calculated for each sound field propagation path;

[0008] S3. Based on the initial perceived importance parameters, each sound field propagation path is assigned to a first calculation strategy or a second calculation strategy for processing; wherein, the simulation accuracy of the first calculation strategy is higher than that of the second calculation strategy.

[0009] S4. During the simulation calculation, the real-time acoustic characteristics of each sound field propagation path are monitored in real time, and the real-time perception contribution is dynamically evaluated based on the real-time acoustic characteristics.

[0010] S5. Based on the real-time perceived contribution, dynamically adjust the calculation strategy used for the corresponding sound field propagation path;

[0011] S6. Integrate the sound field data processed by different calculation strategies to generate acoustic simulation results.

[0012] In a preferred embodiment, the import of a 3D scene model composed of triangular facets in S1 includes support for directly importing defect models containing gaps or overlapping surfaces, without the need to pre-convert them into closed volume meshes.

[0013] In a preferred embodiment, the initial perceived importance parameter in S2 is obtained by calculating a quantified initial score for each sound field propagation path in the geometric acoustic pre-analysis, based at least on its reflection order and cumulative propagation attenuation.

[0014] In a preferred embodiment, S3 includes:

[0015] The initial perceived importance parameter is compared with an initial threshold;

[0016] Sound field propagation paths with an initial perceived importance parameter higher than the initial threshold are assigned to the first calculation strategy, and the rest are assigned to the second calculation strategy.

[0017] In a preferred embodiment, the first calculation strategy is a time-domain wave equation solving method based on digital waveguide grids, used to accurately simulate the physical phenomena of sound wave interference and diffraction.

[0018] The second calculation strategy is a geometric acoustic method based on ray tracing or mirror sound source method, which is used to efficiently simulate the reflection, scattering and energy attenuation of sound rays.

[0019] In a preferred embodiment, within the first calculation strategy, the local calculation accuracy of the associated digital waveguide mesh is dynamically adjusted based on the real-time perceived contribution.

[0020] Within the second calculation strategy, the corresponding reflection or scattering model is dynamically selected based on the acoustic scattering properties of the surfaces interacting with the sound field propagation path.

[0021] In a preferred embodiment, in S4, the real-time acoustic characteristics of each sound field propagation path are monitored in real time, including at least the sound pressure level, propagation direction, and arrival time.

[0022] The dynamic evaluation of the real-time perceived contribution based on the real-time acoustic characteristics involves calculating a dynamic correction amount based at least on the monitored sound pressure level changes, the arrival time delay relative to the direct sound, and the angle of the propagation direction relative to the listener, and using this dynamic correction amount to update the real-time perceived contribution.

[0023] In a preferred embodiment, the calculation strategy used in S5 to dynamically adjust the corresponding sound field propagation path specifically includes:

[0024] S51. Compare the real-time perception contribution with a preset upgrade threshold, and recalculate the sound field propagation path whose contribution has been increased to the upgrade threshold from the second calculation strategy to the first calculation strategy.

[0025] S52. Compare the real-time perceived contribution with a preset degradation threshold, and recalculate the sound field propagation path whose contribution drops below the degradation threshold from the first calculation strategy to the second calculation strategy.

[0026] The upgrade threshold and downgrade threshold are adaptive thresholds that are dynamically adjusted based on the simulation calculation load or target accuracy.

[0027] In a preferred embodiment, the fusion step in S6 is performed in the frequency domain. For the low-frequency data from the first calculation strategy and the high-frequency data from the second calculation strategy, within an overlapping frequency band, weights that change smoothly with frequency are assigned to the two types of data and weighted superposition is performed to generate a continuous full-band frequency response.

[0028] The acoustic simulation results include binaural or monoaural impulse responses generated based on the full-band frequency response, and at least one acoustic parameter among reverberation time, speech intelligibility, and musical clarity calculated from the impulse response.

[0029] This invention also provides an acoustic simulation system based on low-frequency waveguide digital mesh and geometric modeling, comprising:

[0030] The model input module is used to import 3D scene models composed of triangular facets;

[0031] The initialization module is used to configure the sound source parameters and receiver position;

[0032] The perceptual scheduling engine further includes:

[0033] The pre-analysis unit is used to obtain the sound field propagation path and calculate the initial perception importance parameters through geometric acoustic pre-analysis;

[0034] The strategy allocation unit is used to allocate an initial calculation strategy for the sound field propagation path;

[0035] The dynamic evaluation unit is used to monitor real-time acoustic features and dynamically evaluate the contribution of real-time perception during the simulation process.

[0036] The strategy adjustment unit is used to dynamically adjust the calculation strategy based on the real-time perceived contribution.

[0037] The computing engine includes a first computing unit for executing the first computing strategy and a second computing unit for executing the second computing strategy;

[0038] The fusion output module is used to fuse data and generate acoustic simulation results.

[0039] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0040] This invention employs geometric acoustic pre-analysis and real-time dynamic evaluation to allocate high-precision computational strategies only to key sound field propagation paths with high perceptual contribution, while using efficient geometric methods for a large number of secondary paths. This resource scheduling strategy based on auditory perception allows for highly focused computational resources, avoiding unnecessary consumption on secondary components.

[0041] Through a dynamic scheduling mechanism, spatial auditory cues that are difficult to accurately simulate using traditional geometric methods can be processed with high precision by the wave method when necessary. This enables more realistic and reliable predictions of subjective auditory elements such as spatial sense and localization, thereby improving the auditory fidelity of the simulation results. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0043] Figure 1 This is a flowchart of the method of the present invention.

[0044] Figure 2 This is a system block diagram of the present invention.

[0045] Figure 3 This is a comparison chart of the experimental results of the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Perceived importance parameters refer to quantitative scores calculated based on physical properties such as the reflection order and cumulative propagation attenuation of the sound field propagation path, combined with psychoacoustic principles, to characterize the potential influence of the path on the final subjective listening experience.

[0048] Real-time perception contribution refers to the index after dynamic correction of the initial perception importance parameters based on the real-time acoustic characteristics (sound pressure level, propagation direction, etc.) of the path during the simulation process, which is used to trigger the adjustment of the calculation strategy.

[0049] Example 1, please refer to Figure 1 As shown in this embodiment, the acoustic simulation method based on low-frequency waveguide digital mesh and geometric modeling includes:

[0050] S1. Import the 3D scene model composed of triangular facets and initialize the sound source parameters and receiver position;

[0051] S2. Based on the three-dimensional scene model, sound source parameters and receiver position, multiple sound field propagation paths are obtained through geometric acoustic pre-analysis, and initial perception importance parameters are calculated for each sound field propagation path;

[0052] S3. Based on the initial perceived importance parameters, each sound field propagation path is assigned to a first calculation strategy or a second calculation strategy for processing; wherein, the simulation accuracy of the first calculation strategy is higher than that of the second calculation strategy.

[0053] S4. During the simulation calculation, the real-time acoustic characteristics of each sound field propagation path are monitored in real time, and the real-time perception contribution is dynamically evaluated based on the real-time acoustic characteristics.

[0054] S5. Based on the real-time perceived contribution, dynamically adjust the calculation strategy used for the corresponding sound field propagation path;

[0055] S6. Integrate the sound field data processed by different calculation strategies to generate acoustic simulation results.

[0056] The method will be described in detail below with reference to specific implementation methods.

[0057] In one possible implementation, step S1 specifically includes:

[0058] S11 can directly read 3D scene model files composed of triangular facets from architectural design software or common 3D formats.

[0059] Specifically, the system's built-in model input module is directly compatible with the native or exchange formats of mainstream engineering and modeling software, such as .skp files from SketchUp, .obj files from 3ds Max or Blender, and .stl files for 3D printing. This direct reading method avoids the step of importing the model into intermediate software for secondary conversion or reconstruction, ensuring the integrity of the original geometric accuracy and material property information.

[0060] For example, when designing a concert hall, a user might use SketchUp to create a model that includes seating, railings, and intricate ceiling decorations, and then directly drag the .skp file into this simulation system. The system then reconstructs the complete 3D structure of the concert hall without loss of quality by parsing the triangular mesh data, layer groups, and associated material names within the file.

[0061] S12 automatically analyzes and lightweights the imported triangular patch model, while being compatible with common engineering defects that may exist in the model, such as gaps, overlapping patches, or non-closed surfaces, without requiring pre-repair and conversion into a closed watertight volume mesh. A watertight volume mesh refers to a seamless, non-overlapping, geometrically closed volume mesh, which is a mandatory requirement for models in traditional wave simulations. This invention does not require such preprocessing.

[0062] Specifically, the system automatically constructs spatial acceleration structures (such as BVH hierarchical bounding boxes) for all triangular faces in the scene to optimize the efficiency of subsequent ray intersection detection. For minute gaps in the model that should not exist in the physical world (such as gaps less than 1 cm between faces) or slight overlaps between faces, the system treats them as acceptable defects when constructing the acceleration structure, without forcing users to perform tedious model repairs to make it a perfect watertight entity. This feature significantly reduces the lead time from CAD design to acoustic simulation.

[0063] For example, when importing a complex steel roof model of a large stadium, due to modeling accuracy issues or software export problems, some steel component connections contain millimeter-level micro-gaps. Traditional finite element or volume mesh-based wave methods require these gaps to be repaired; otherwise, an effective mesh cannot be generated. This system, however, allows direct use of the defective model. During the pre-analysis phase, the system treats these gaps as special acoustic boundaries, rather than as errors hindering the simulation.

[0064] S13 allows users to interactively navigate through a 3D scene or preset one or more sound source points using coordinate files and configure their core acoustic parameters.

[0065] Specifically, the initialization of sound source parameters includes: the precise three-dimensional coordinates (X, Y, Z) of the sound source in the scene; the frequency characteristics of the sound source, such as specifying it as a broadband noise source covering the entire frequency range of 20Hz to 20kHz, or supporting the definition of its spectrum according to 27 standard 1 / 3 octave band center frequencies; the radiation directivity pattern of the sound source (such as omnidirectional, cardioid, etc.); and the emitted sound power level or sound pressure level of the sound source. All parameters can be entered through the graphical interface or loaded from preset configuration files.

[0066] For example, in the theater model, the main sound source is initialized to the center of the stage, 1.2 meters above the ground, and is set as an omnidirectional point sound source with a sound power level of 100dB and a frequency response of pink noise spectrum, to simulate the main sound energy radiation during an actual performance.

[0067] S14, preset the positions of one or more receivers (or listening points) in the listening area of ​​interest (such as the audience seating area).

[0068] Specifically, receiver locations are also defined using three-dimensional coordinates. Individual receivers can be deployed to evaluate the sound effects at a specific seat, or a grid of receivers or a series of receivers can be deployed along a specific path to comprehensively evaluate the sound field distribution across an entire area or a specific flow path. Each receiver can be independently configured with its reception properties, such as whether to calculate binaural hearing signals (BRIR).

[0069] For example, in classroom acoustic simulation, 20 receivers were arranged in a grid pattern at 1.5-meter intervals in the student seating area in front of the podium (sound source location) to systematically evaluate the speech intelligibility distribution of the entire teaching area and ensure that each seat can obtain a good listening effect.

[0070] In one possible implementation, step S2 specifically includes:

[0071] S21, an efficient geometric acoustic algorithm is used to perform ray tracing pre-analysis on the three-dimensional scene model, and batch generate sound field propagation paths from the sound source to the designated receiver.

[0072] Specifically, the system emits a large number of virtual sound rays (e.g., 1 million sound rays) uniformly or selectively from an initialized sound source location into the entire hemispherical space. Each sound ray propagates within the scene according to the geometric boundaries defined by triangular facets and the specified material reflection / scattering coefficients, tracking each interaction (reflection, scattering) with the surface. A valid propagation path is considered found when a sound ray or its derived secondary sound rays (generated during scattering events) enters a capture sphere with a preset radius centered on the receiver location. The system records all physical information of the path, including: the total length of the path, the sequence of all reflection points encountered and their corresponding reflective surface materials, the incident angle and reflection angle of each reflection, and the reflection order to which the path belongs (e.g., direct sound is order 0, the first reflection is order 1, and so on). The capture sphere is a spherical region with a preset radius (preferably 0.5 meters, adjustable according to scene size; 0.8-1.0 meters for large scenes) centered on the receiver location, used to determine valid propagation paths.

[0073] For example, during the pre-analysis of a conference room model, the sound rays emitted from the lectern source, in addition to the direct sound rays, part of the sound rays reach the receiver in the audience area after being reflected once by the ceiling, forming a first-order reflection path; another part of the sound rays reach the receiver after being reflected twice by the side walls and the ground, forming a second-order reflection path. The system will record the geometry of these different paths separately.

[0074] S22, Based on the path physical information obtained from the pre-analysis, calculate the initial acoustic energy intensity estimate of each sound field propagation path at the current stage.

[0075] Specifically, the initial acoustic energy intensity of each path An estimation using a simplified energy decay model can be performed, and the calculation formula can be expressed as:

[0076] = · · ;

[0077] in, The reference sound intensity of the sound source is denoted by ; N is the number of reflections along the path. Let be the energy reflection coefficient of the surface corresponding to the i-th reflection (related to the material absorption coefficient); This represents the total path length. The air absorption coefficient is a frequency-dependent factor (an average value representing mid-to-high frequencies can be used initially). This estimation prioritizes geometric diffusion attenuation and material absorption attenuation, providing an energy scale directly related to the path's physical properties for subsequent importance assessments.

[0078] For example, the estimated sound intensity of a 10-meter-long path that undergoes two reflections (both reflecting surfaces are plasterboard with a reflection coefficient of approximately 0.9) will be significantly lower than that of a 5-meter-long path that undergoes only one reflection (the reflecting surface is marble with a reflection coefficient of approximately 0.98).

[0079] S23, for each identified sound field propagation path, calculate a quantified initial perception importance score based on its multi-dimensional physical properties, and use it as the initial perception importance parameter.

[0080] Specifically, the score is the result of a multi-factor weighted function. The system calculates the score based on at least two core factors:

[0081] Reflection order factor: To assign higher weights to early reflection paths, an exponential decay model is used. = The calculation is performed, where order represents the reflection order. After extensive testing in typical scenarios, the optimal range for the coefficient k is [0.2, 0.5].

[0082] In a preferred implementation, k = 0.3. For example, the weight of direct sound (order = 0) is 1, the weight of first-order reflection is approximately 0.74, and the weight of second-order reflection is approximately 0.55.

[0083] Cumulative propagation attenuation factor: The initial acoustic energy intensity estimate calculated in S22 is used directly. Or its logarithmic form (i.e., sound pressure level) The higher the energy and the smaller the attenuation of a path, the greater its potential perceptual importance.

[0084] Initial perceived importance score The quantification calculation uses the following formula:

[0085] = · ;

[0086] in:

[0087] This is the sound pressure level estimated based on path geometric attenuation and material absorption.

[0088] For reference sound pressure level.

[0089] For example, in a single pre-analysis, the system may identify hundreds of valid paths. Calculations show that a path with a high reflection order and long propagation distance typically has a lower initial perceived importance score, while a lateral reflection path with a low reflection order and low attenuation has a significantly higher score. Based on this, the system initially determines that the latter is far more likely to require priority attention in subsequent simulations, reflecting the basic logic of perceived importance screening.

[0090] In one possible implementation, step S3 specifically includes:

[0091] S31, Based on the path list generated by pre-analysis and its initial perceived importance score, set a dynamic initial threshold for the initial traffic split.

[0092] Specifically, the initial threshold is not a fixed value, but is adaptively determined based on the path analysis results of the current scenario.

[0093] Initial threshold The statistical percentile method is used. Typically, the [80th, 95th] percentile of the initial ranking of all paths is used as the threshold range. For example, using the 85th percentile score as the initial threshold ensures that approximately 15% of the top important paths are included in the first calculation strategy.

[0094] S32, compare the initial perceived importance score of each sound field propagation path with the initial threshold, and perform a binary allocation decision.

[0095] Specifically, the system iterates through all sound field propagation paths generated in step S2 and executes the following judgment logic: if the initial perceived importance score of a certain path is... Greater than or equal to the initial threshold If the score is below the threshold, the path is marked as a critical path and assigned the first calculation strategy; if the score is below the threshold, the path is marked as a non-critical path and assigned the second calculation strategy. All assignment decisions are recorded in a global path-strategy mapping table.

[0096] For example, assuming that the current scene is calculated by S31, The score is 7.5. A first-order strong reflection path from the sidewall with a score of 8.3 is assigned to the first calculation strategy (i.e., a detailed wave simulation will be performed using a digital waveguide mesh (DWM)) because its score is above the threshold. A fourth-order weak reflection path from the far-field diffuser with a score of 5.1 is assigned to the second calculation strategy (i.e., an efficient ray energy tracing method will be performed using a geometric acoustic method).

[0097] S33 performs differentiated simulation task initialization and parameter configuration for the two computational strategies. Specifically, for the path set assigned to the first computational strategy (high-precision fluctuation calculation), the system will:

[0098] Constructing a local waveguide mesh: Instead of building a globally uniform mesh for the entire scene, a local, regular digital waveguide mesh (DWM) is adaptively generated based on the spatial distribution of these critical paths, within their propagation corridors and major reflection regions. The spatial step size of the mesh is determined according to the highest frequency to be simulated (determined by the boundary frequency, e.g., 500Hz), ensuring that the Nyquist sampling theorem is satisfied to accurately capture low-frequency ripple effects.

[0099] Initial state of injection path: The geometric information of each critical path (such as the starting direction and the expected location of the first reflection point) is converted into the initial excitation conditions in the local DWM mesh, in preparation for solving the time-domain wave equation.

[0100] For the set of paths assigned to the second computational strategy (efficient geometric computation), the system will:

[0101] Packaged geometric computation task: These paths themselves are used as the acoustic skeleton that needs to be verified and refined. Subsequent geometric acoustic algorithms (such as ray tracing) will perform more intensive secondary sampling or energy diffusion calculations around these skeleton paths to simulate high-frequency scattering and obtain more accurate energy attenuation.

[0102] Configure materials and reflection models: associate the material properties of the surfaces along the path, and pre-set the corresponding specular reflection, diffuse scattering or mixed reflection models and coefficients for geometric calculations.

[0103] In one possible implementation, step S4 specifically includes:

[0104] S41, in the parallel simulation calculation process, periodically performs multi-dimensional acoustic feature sampling on each sound field propagation path being calculated.

[0105] Specifically, the system establishes a global simulation clock and sampling period (e.g., a snapshot sampling is performed every 1 millisecond of simulation progression). For paths employing the first computational strategy (DWM), the system extracts a set of physical quantities from the associated local waveguide mesh of the region where the current acoustic wavefront of the path is located, including: instantaneous sound pressure value, used to calculate sound pressure level (SPL); particle velocity vector, used to determine the main direction of energy propagation; and the current main frequency components obtained through short-time Fourier transform (STFT) analysis. For paths employing the second computational strategy (geometric method), the system statistically analyzes the energy, direction, and timestamps of all acoustic rays arriving at the end of the path (near the receiver) within the most recent sampling period from the acoustic ray bundle it represents, thereby estimating the equivalent sound pressure level, average direction of arrival, and time distribution of energy arrival for the path.

[0106] For example, at the 50th millisecond of the simulation, the system samples a ceiling reflection path being calculated using DWM. The sampling shows that, at the current moment, due to constructive interference with the sound waves from another path, the local sound pressure level of this path is 6 dB higher than the expected value; at the same time, its energy is mainly concentrated in the frequency band below 300 Hz. These real-time characteristics are accurately recorded.

[0107] S42 dynamically calculates the real-time perception contribution of each path based on the real-time acoustic features obtained from sampling and combined with a psychoacoustic model.

[0108] Specifically, real-time perception of contribution It is a composite index that integrates multidimensional real-time features. Its calculation not only relies on the static factors used in S2 (such as reflection order), but more importantly, it introduces a dynamic correction factor derived from real-time monitoring. A basic computational framework can be expressed as:

[0109] = +Δ ;

[0110] in, It is the initial score in step S2, Δ It is a dynamic correction amount. Δ It is determined by the following real-time factors:

[0111] Dynamic correction amount Δ The calculation introduces the following factors:

[0112] Spatial saliency factor: For lateral reflection paths whose direction of arrival forms an angle between 60° and 120° with the listener's front, an additional weighting factor of 1.2 to 1.8 is assigned. Preferably, it is 1.5.

[0113] Energy spurious factor: ΔE=( - ) / 10, where, To monitor sound pressure levels in real time, To estimate the sound pressure level for preliminary analysis, this factor is directly linearly superimposed onto the contribution level.

[0114] Temporal masking factor: Analysis of the arrival time of the path. According to the Haas effect (priority effect) and the temporal masking model, strong early reflections arriving very shortly after the direct sound (e.g., 5-35 milliseconds) contribute greatly to speech intelligibility and fullness of sound and should receive a high dynamic score; while later reflections arriving later than 100 milliseconds with weaker energy may be masked by earlier sounds, and their dynamic contribution will be reduced.

[0115] For example, the aforementioned path where the interference sound pressure level surges by 6 dB, its Δ This will result in a significant positive bonus. However, if the system detects another initially high-scoring second-order reflection path whose energy consistently falls below expectations in the simulation due to unfavorable phase cancellation, then its Δ... The value will be negative, causing its real-time contribution ranking to drop.

[0116] S43, update the global path contribution ranking and determine if dynamic scheduling conditions are triggered. Specifically, after each evaluation cycle, the system will update the ranking based on the latest path contribution ranking. The system reorders all paths. Simultaneously, it compares the real-time contribution of each path with a set of dynamic thresholds. This comparison determines not only whether the upgrade criteria are met but also whether the downgrade conditions are satisfied. The system generates a list of paths to be scheduled, marking those whose real-time contribution has significantly deviated from the importance level corresponding to their current calculation strategy, providing clear input instructions for precise scheduling in step S5.

[0117] For example, after this round of evaluation, an early lateral reflection path that was originally in the slow class (the slow class is the second calculation strategy) jumped to the top 10% of the overall ranking due to its significantly increased dynamic contribution and was marked as awaiting upgrade; while a late reflection path that was originally in the fast class (the fast class is the first calculation strategy) was marked as awaiting downgrade because its contribution was consistently lower than expected and its ranking fell to the bottom 50%.

[0118] In one possible implementation, step S5 specifically includes:

[0119] S51 parses the scheduling instructions and determines the path to be adjusted and its target strategy.

[0120] Specifically, the system receives a list of paths to be scheduled from the output of step S43. Each record in this list explicitly includes: a unique path identifier, the current calculation strategy, and the latest evaluated real-time perceived contribution. And based on the preset upgrade threshold and downgrade threshold The target strategy derived from the comparison.

[0121] Upgrade / downgrade threshold / These two thresholds form a hysteresis interval to prevent the path from oscillating frequently between policies. The initial setting can be: = +δ, = -2δ, where δ is a buffer value set according to the variance of the rating distribution, usually 2%-5% of the total rating range.

[0122] For example, the list indicates that path ID_1024 is currently using the second calculation strategy (geometric method). =8.7, which has exceeded =8.0, target strategy is to upgrade to the first computing strategy (DWM); path ID_2048, currently using the first computing strategy, its =4.2, already lower than =5.0, the target strategy is to downgrade to the second computation strategy.

[0123] S52 executes the computation strategy migration and completes the context switching and inheritance of the simulation state.

[0124] This is crucial to ensuring that dynamic adjustments do not compromise simulation accuracy or cause abrupt numerical changes. The system has designed sophisticated state transition procedures for both upgrade and downgrade operations:

[0125] For the path from strategy two to strategy one:

[0126] State extraction: Freeze the acoustic beam corresponding to the path at the current simulation moment from the current geometric acoustic calculation engine. The key information includes: the energy distribution envelope of the sound beam in space, the average propagation direction, and the energy spectrum of the main frequency components.

[0127] Waveguide mesh excitation initialization: The extracted energy envelope and direction information are transformed into the local digital waveguide mesh (DWM) at time t in the first computational strategy through spatial interpolation and projection algorithms. The initial conditions (i.e., the initial displacement and velocity fields on the mesh nodes). For example, a beam of lateral acoustic energy representing path ID_1024 is precisely injected into the node of the corresponding spatial region in the DWM mesh, serving as a new starting point for this path in wave calculations.

[0128] Historical state inheritance (optional but preferred advanced operation): For a smoother transition, the system not only initializes the current state but also attempts to retrieve the path from the geometric calculation cache. The simplified energy history of a short period of time (such as the last 10 milliseconds) is converted into the historical state array of the corresponding node in the DWM grid, which enables the wave solver to start up hot and simulate the existing part of the sound field establishment process of the path, avoiding the auditory discontinuity caused by starting the calculation from an absolutely silent state.

[0129] For the path from strategy 1 to strategy 2:

[0130] Fluctuation Feature Extraction: Extract the grid region associated with the path at time [time value missing] from the DWM calculation. The wave field characteristics, such as those obtained through beamforming or spatial Fourier analysis, can be used to obtain the path's current main propagation direction, average wave number, and energy intensity.

[0131] Geometric acoustic characterization reconstruction: The extracted wave features are reconstructed into a backbone acoustic line (or a bundle) with equivalent direction, energy, and spectral characteristics, and then incorporated into the geometric acoustic tracking queue of the second computational strategy. Simultaneously, based on the geometric information and material properties of this path, a subsequent reflection / scattering computational model is configured for it.

[0132] S53, update global scheduling mapping and load balancing monitoring.

[0133] Specifically, after the path policy migration is completed, the system immediately updates the global path-policy mapping table. Simultaneously, the system's resource scheduler monitors the real-time computational load of both computing strategies and dynamically adjusts the upgrade threshold accordingly. This is to achieve load balancing and maximize efficiency. Specifically, the system sets a load safety threshold (e.g., 85% CPU / GPU utilization) for system health monitoring.

[0134] Based on this, a more refined dynamic adjustment strategy is implemented: when the real-time computing load of the first computing unit (DWM) exceeds 75% of the total available resources, the system automatically... Temporarily increase δ to slow down the upgrade speed and prevent overload; when the load is below 30%, appropriately decrease it. This encourages more paths to use high-precision computing in order to make full use of idle computing power to improve accuracy.

[0135] In one possible implementation, step S6 specifically includes:

[0136] S61, aligning and preprocessing sound field data from two computational strategies.

[0137] Specifically, the system spatiotemporally aligns the sound field responses calculated by the two strategies that correspond to the same sound source-receiver pair.

[0138] Time alignment: Based on the simulated global clock, ensure that the starting points of the time axes (the excitation time of the sound source) of the two data streams are completely synchronized.

[0139] Data format conversion: The output from the first computational strategy (DWM) (typically a time-domain grid node stress sequence) is spatially summarized and Fourier transformed into a low-frequency (e.g., 20Hz-500Hz) complex frequency response centered on the receiver location. While preserving its phase information .

[0140] The outputs from the second computational strategy (geometric method) (usually a list of particles or rays recorded in time-energy-direction format) are statistically summarized to generate the energy decay curve (ETC) for the entire frequency band (e.g., 20Hz-20kHz). Furthermore, the frequency response amplitude in the high-frequency band (e.g., 500Hz-20kHz) is estimated using methods such as Hilbert transform. |, but its phase information is usually considered random or not retained.

[0141] S62 performs confidence-based adaptive weighted fusion in the frequency domain to generate a full-band impulse response.

[0142] Specifically, this is the core algorithm of the fusion process. Instead of simply switching at a fixed frequency point, the system assigns frequency-dependent dynamic weights to data from both strategies within an overlapping frequency band (e.g., 300Hz-700Hz). The frequency response after fusion... Calculated by the following formula:

[0143] = + ;in:

[0144] Weighting function and The sum of the two is 1. It dominates at low frequencies and gradually attenuates towards higher frequencies. Dominated by high frequencies, it smoothly increases towards lower frequencies. The attenuation / growth curve uses a cosine-square roll-off function to ensure a smooth transition within overlapping frequency bands, without abrupt changes in frequency response.

[0145] Mixed phase In the low-frequency band, use directly To maintain the phase accuracy of wave interference effects; in overlapping and high-frequency bands, A smooth transition to a phase based on geometric path delay estimation or optimized by a psychoacoustic model. To maintain the spatial impression rationality in hearing at high frequencies.

[0146] For example, when integrating the sound field of a concert hall, the precise phase of the 80Hz room mode from DWM is preserved, ensuring the fullness and accuracy of the low frequencies; while in the frequency range above 1kHz, the fine energy attenuation provided by the geometric approach, combined with the optimized phase, generates clear, spatially defined high-frequency reverberation tails.

[0147] S63, based on the fused full-band data, generates the final acoustic simulation results. Specifically, the system performs the following post-processing:

[0148] Generation of impulse response: frequency response after fusion An inverse Fourier transform is performed to generate the final binaural room impulse response (BRIR) or single-point impulse response. This impulse response also exhibits smooth characteristics in the time domain and will not produce non-physical jitter or echoes due to fusion.

[0149] Calculate acoustic parameters: Based on the above impulse response, automatically calculate a series of standard acoustic evaluation indicators:

[0150] Reverberation time (RT60): The time required for energy to decay by 60 dB at different frequencies (e.g., octaves of 125 Hz, 250 Hz, ... 4 kHz).

[0151] Early Decay Time (EDT): Calculates the time it takes for the first 10 dB of reverberation to decay, and is more relevant to subjective reverberation.

[0152] Clarity (C50) and Intelligibility (C80): The ratio of energy to total energy in the first 50 milliseconds and 80 milliseconds, respectively, are used to evaluate speech clarity and music clarity.

[0153] Spatial perception indicators such as lateral energy fraction (LF).

[0154] Results Output and Visualization: The system packages and outputs impulse response files (such as WAV format) and structured acoustic parameter reports (such as JSON and PDF formats). It can simultaneously generate visual charts such as sound field energy distribution cloud maps and attenuation curves for designers to analyze intuitively.

[0155] Example 2, please refer to Figure 2 As shown in this embodiment, the acoustic simulation system based on low-frequency waveguide digital mesh and geometric modeling includes:

[0156] The model input module is used to import 3D scene models composed of triangular facets;

[0157] This module is responsible for transforming 3D geometric data from various sources into a unified representation that can be processed internally by the system. Its purpose is to maximize engineering compatibility and minimize user preprocessing costs.

[0158] Implementation details: This module incorporates dedicated parsers for various mainstream 3D file formats (such as .skp, .obj, .stl). When a user imports a SketchUp model of a concert hall, the parser directly reads the triangular mesh data, layer structure, and associated material names from the file, without requiring prior format conversion. The module's key innovation lies in its fault-tolerant geometry processor. When constructing the spatial acceleration structure of a scene (such as a BVH tree), this processor proactively identifies and accommodates minute gaps (such as gaps less than 1cm between faces) or slight overlaps in the model that shouldn't exist in the physical world, marking them as acceptable defects rather than fatal errors that would halt the simulation. For example, when processing a theater box model with intricate carvings, even if the triangular faces of the decorative elements have minor intersections or are not completely closed, the module can successfully incorporate them into the acceleration structure, preparing for subsequent sound line intersection detection, thereby reducing the user's model repair time by more than 80%.

[0159] The initialization module is used to configure the sound source parameters and receiver position;

[0160] This module is responsible for setting clear physical scenarios and observation targets for simulations, and for converting all user intentions into precise, computable parameters.

[0161] Implementation details: The module provides a graphical interface and / or script interface, allowing users to:

[0162] Define the sound source: Click or enter coordinates in the 3D view to place the sound source, and configure its frequency spectrum (such as selecting a standard noise spectrum or uploading a custom spectrum curve), radiation directivity (select from omnidirectional, cardioid, supercardioid, etc. or import measurement data) and emitted sound power in detail in the properties panel.

[0163] Receiver Placement: Users can place individual listening points or use array generation tools to automatically generate a grid-like or linear array of receivers in the audience area or along the aisles. Each receiver can be set to its own type, such as a virtual head receiver for calculating binaural room impulse response (BRIR), in which case the system will automatically associate it with head-related transfer function (HRTF) data.

[0164] Associating material properties: Binding the surface material names in the model (such as wooden walls and fabric seats) with the acoustic properties (sound absorption coefficient and scattering coefficient) in the system's material database to provide a physical basis for subsequent calculations;

[0165] The perceptual scheduling engine further includes:

[0166] The pre-analysis unit is used to obtain the sound field propagation path and calculate the initial perception importance parameters through geometric acoustic pre-analysis;

[0167] Before the core simulation calculations begin, this unit initiates a highly parallel, rapid pre-computation of the geoacoustic geometry. It invokes the second computational unit (Geometric Mode) within the computational engine to fire millions of test sound rays into the scene, rapidly constructing a potential sound field propagation path map from the sound source to each receiver. For each path, it integrates its reflection order, estimated cumulative attenuation (based on geometric diffusion and material absorption), and propagation direction, applying an initial importance scoring model to calculate a quantified initial perceived importance parameter. For example, it might identify several low-order, high-energy lateral reflection paths from the stage to the front row of the stalls and assign them high initial scores.

[0168] The strategy allocation unit is used to allocate an initial calculation strategy for the sound field propagation path;

[0169] Strategy Allocation Unit: This unit receives the path list and scores output by the pre-analysis unit. Internally, it maintains a dynamic initial threshold algorithm that automatically determines the threshold based on the score distribution of all paths in the current scene (e.g., taking the top 15% quantile). The unit compares the score of each path with this threshold and performs a binary decision: paths scoring above the threshold have their metadata marked and sent to the first computation unit queue of the computation engine; paths scoring below the threshold are sent to the second computation unit queue. This completes the initial optimized allocation of resources.

[0170] The dynamic evaluation unit is used to monitor real-time acoustic features and dynamically evaluate the contribution of real-time perception during the simulation process.

[0171] Dynamic Evaluation Unit: After the simulation starts, this unit runs continuously as an independent monitoring service. It periodically (e.g., every 10 milliseconds of simulation) sniffs out real-time data of the critical path from the two computation units via the system bus, including instantaneous sound pressure level, main frequency components, and energy flow direction. It has a built-in real-time contribution evaluation model that dynamically adjusts the initial score by incorporating psychoacoustic knowledge (such as the Haas effect and auditory masking). For example, it might discover a path with a mediocre initial score that experiences a sudden energy surge due to constructive interference with another path, thus significantly increasing its real-time contribution.

[0172] The strategy adjustment unit is used to dynamically adjust the calculation strategy based on the real-time perceived contribution.

[0173] Strategy Adjustment Unit: This unit is the executor of dynamic scheduling. It listens for instructions issued by the dynamic evaluation unit. When the real-time contribution of a path crosses a preset upgrade or downgrade threshold, this unit initiates a strategy migration process. It coordinates two computing units in the computing engine to perform fine-grained export, transformation, and import of computing states. For example, for a path that needs to be upgraded from geometric computing to wave computing, its current acoustic energy distribution state is precisely initialized to the excitation conditions in the DWM mesh through mathematical transformation, ensuring the continuity of the physical process.

[0174] The computing engine includes a first computing unit for executing the first computing strategy and a second computing unit for executing the second computing strategy;

[0175] It adopts a heterogeneous parallel computing architecture, which includes two relatively independent but collaborative computing units.

[0176] The first computational unit (high-precision wave calculation): The core of this unit is a time-domain wave equation solver based on a digital waveguide mesh (DWM). It receives a set of critical paths assigned by the scheduling engine, and instead of creating a global mesh for the entire hall, it adaptively generates a regular DWM mesh for the local spatial regions where these paths reside. This solver is highly optimized, supporting large-scale parallel computation using GPUs. The state update of each mesh node is completed in a GPU thread, specifically designed for accurately simulating the interference, diffraction, and room resonance modes of low-frequency sound waves.

[0177] The second computational unit (high-efficiency geometric calculation): At its core is a highly optimized ray tracing / mirror source hybrid calculator. It handles a large number of non-critical paths and post-reverberation. This unit employs a delay shading approach: first, it rapidly traces a large number of sound rays or calculates mirror sources to determine the geometric information of the propagation path; then, in the later stages, it uniformly calculates energy attenuation and filtering based on material properties and distance, greatly improving computational efficiency. It also supports GPU acceleration and can trace millions of sound rays simultaneously.

[0178] The fusion output module is used to fuse data and generate acoustic simulation results.

[0179] This module is responsible for combining data calculated using two different principles into a unified, high-fidelity final product.

[0180] Implementation details: The module receives raw data streams from two computing units: a low-frequency (e.g., 20-500Hz) complex frequency response (including phase information) from the first unit, and a full-band energy decay curve and mid-to-high frequency amplitude response from the second unit. Its core is a frequency-domain adaptive weighted fusion algorithm. Within an overlapping frequency band (e.g., 300-700Hz), this algorithm assigns weights that smoothly vary with frequency to the two data sets, then performs an inverse Fourier transform to generate a seamless full-band binaural impulse response (BRIR). Subsequently, the module automatically analyzes the BRIR, batch-calculates and generates standardized reports containing metrics such as reverberation time (RT60), intelligibility (C50, C80), and lateral energy factor (LF), while also outputting WAV files for direct playback by the audio engine and various visualization charts.

[0181] The system's modules do not operate in isolation, but rather collaborate closely through a unified data bus and scheduling instructions to form an intelligent processing pipeline from data input to result generation.

[0182] Example 3: Effect Verification Experiment;

[0183] To verify the technical effectiveness of the method of this invention, a three-dimensional triangular facet model of a well-known concert hall was used as the test scene. The sound source was placed in the center of the stage, and three typical receiving points (R1, R2, R3) were set up in the front row of the stalls, the balcony, and the back row, respectively. The comparative experiment setup is as follows:

[0184] Control group: The fixed frequency boundary (500Hz) hybrid method described in the background technology was used to simulate the entire scene using a uniform digital waveguide mesh (DWM) in the low frequency band (below 500Hz).

[0185] Experimental group: The hybrid modeling method based on dynamic scheduling of perceived importance described in this invention is adopted.

[0186] The simulation was performed on the same hardware platform (Intel i9-13900K, NVIDIA RTX4090);

[0187] The main results are as follows: Figure 3 As shown;

[0188] Experimental Analysis:

[0189] This invention significantly reduces the low-frequency DWM simulation range, which dominates computational costs, from the entire domain to local key areas through perception-based importance screening and dynamic scheduling. This results in a reduction of more than 75% in both total simulation time and memory usage, laying a solid foundation for achieving real-time interactive audible simulation.

[0190] Because high-precision computing resources are dynamically and precisely allocated to the propagation path (such as early reflections) that contributes most to the auditory experience, the simulation accuracy of the method in this invention on key acoustic indicators (such as RT60 and C80) is improved by about 50% compared with the traditional static mixing method, and is closer to the reference solution of the high-precision finite element method (FEM).

[0191] Traditional geometric methods struggle to accurately simulate spatial auditory cues (such as lateral energy). This invention addresses this by employing a dynamic scheduling mechanism to enable high-precision simulation of key lateral reflection paths using the Dynamic Wave Method (DWM) when necessary, thereby achieving more reasonable predictions of spatial perception parameters.

[0192] This embodiment demonstrates through objective data that the present invention effectively solves the core problem of the mismatch between computing resources and the importance of perception as pointed out in the background art, and achieves an order-of-magnitude leap in computing efficiency while ensuring or even improving simulation accuracy.

[0193] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An acoustic simulation method based on low-frequency waveguide digital mesh and geometric modeling, characterized in that, include: S1. Import the 3D scene model composed of triangular facets and initialize the sound source parameters and receiver position; S2. Based on the three-dimensional scene model, sound source parameters and receiver position, multiple sound field propagation paths are obtained through geometric acoustic pre-analysis, and initial perception importance parameters are calculated for each sound field propagation path; S3. Based on the initial perceived importance parameters, each sound field propagation path is assigned to a first calculation strategy or a second calculation strategy for processing; wherein, the simulation accuracy of the first calculation strategy is higher than that of the second calculation strategy. S4. During the simulation calculation, the real-time acoustic characteristics of each sound field propagation path are monitored in real time, and the real-time perception contribution is dynamically evaluated based on the real-time acoustic characteristics. S5. Based on the real-time perceived contribution, dynamically adjust the calculation strategy used for the corresponding sound field propagation path; S6. Merge the sound field data processed by different calculation strategies to generate acoustic simulation results; The initial perceived importance parameter in S2 is obtained by the following method: in the geometric acoustic pre-analysis, for each sound field propagation path, at least based on its reflection order and cumulative propagation attenuation, a quantitative initial score is calculated. The first calculation strategy is a time-domain wave equation solution method based on digital waveguide grids, which is used to accurately simulate the physical phenomena of sound wave interference and diffraction; The second calculation strategy is a geometric acoustic method based on ray tracing or mirror sound source method, which is used to efficiently simulate the reflection, scattering and energy attenuation of sound rays; In S4, the real-time acoustic characteristics of each sound field propagation path are monitored in real time, including at least the sound pressure level, propagation direction and arrival time. The dynamic evaluation of the real-time perceived contribution based on the real-time acoustic features is based at least on the monitored sound pressure level change, the time of arrival delay relative to the direct sound, and the angle of the propagation direction relative to the listener. A dynamic correction amount is calculated and used to update the real-time perceived contribution. The calculation strategy used in S5 to dynamically adjust the corresponding sound field propagation path specifically includes: S51. Compare the real-time perception contribution with a preset upgrade threshold, and recalculate the sound field propagation path whose contribution has been increased to the upgrade threshold from the second calculation strategy to the first calculation strategy. S52. Compare the real-time perceived contribution with a preset degradation threshold, and recalculate the sound field propagation path whose contribution drops below the degradation threshold from the first calculation strategy to the second calculation strategy. The upgrade threshold and downgrade threshold are adaptive thresholds that are dynamically adjusted based on the simulation calculation load or target accuracy.

2. The acoustic simulation method based on low-frequency waveguide digital mesh and geometric modeling according to claim 1, characterized in that: The S1 imports a 3D scene model composed of triangular facets, including the ability to directly import defect models containing gaps or overlapping surfaces without pre-converting them into closed volume meshes.

3. The acoustic simulation method based on low-frequency waveguide digital mesh and geometric modeling according to claim 1, characterized in that: S3 includes: The initial perceived importance parameter is compared with an initial threshold; Sound field propagation paths with an initial perceived importance parameter higher than the initial threshold are assigned to the first calculation strategy, and the rest are assigned to the second calculation strategy.

4. The acoustic simulation method based on low-frequency waveguide digital mesh and geometric modeling according to claim 1, characterized in that: Within the first calculation strategy, the local calculation accuracy of the associated digital waveguide mesh is dynamically adjusted based on the real-time perceived contribution. Within the second calculation strategy, the corresponding reflection or scattering model is dynamically selected based on the acoustic scattering properties of the surfaces interacting with the sound field propagation path.

5. The acoustic simulation method based on low-frequency waveguide digital mesh and geometric modeling according to claim 1, characterized in that: The fusion step in S6 is performed in the frequency domain. For the low-frequency data from the first calculation strategy and the high-frequency data from the second calculation strategy, within an overlapping frequency band, weights that change smoothly with frequency are assigned to the two types of data and weighted superposition is performed to generate a continuous full-band frequency response. The acoustic simulation results include binaural or monoaural impulse responses generated based on the full-band frequency response, and at least one acoustic parameter among reverberation time, speech intelligibility, and musical clarity calculated from the impulse response.

6. An acoustic simulation system based on low-frequency waveguide digital mesh and geometric modeling, used to implement the acoustic simulation method based on low-frequency waveguide digital mesh and geometric modeling as described in any one of claims 1-5, characterized in that, include: The model input module is used to import 3D scene models composed of triangular facets; The initialization module is used to configure the sound source parameters and receiver position; The perceptual scheduling engine further includes: The pre-analysis unit is used to obtain the sound field propagation path and calculate the initial perception importance parameters through geometric acoustic pre-analysis; The strategy allocation unit is used to allocate initial calculation strategies for the sound field propagation path; The dynamic evaluation unit is used to monitor real-time acoustic features and dynamically evaluate the contribution of real-time perception during the simulation process. The strategy adjustment unit is used to dynamically adjust the calculation strategy based on the real-time perceived contribution. The computing engine includes a first computing unit for executing the first computing strategy and a second computing unit for executing the second computing strategy; The fusion output module is used to fuse data and generate acoustic simulation results.

Images