A multi-sampling rate room correction method and system for embedded platforms

By employing a two-stage decoupled architecture of offline design and online execution and a self-describing parameter set on an embedded platform, the problems of high computational complexity and signal distortion in multi-sampling-rate room acoustic correction on embedded platforms are solved, achieving efficient and reliable signal correction results.

CN123075339APending Publication Date: 2026-11-10深圳远虑科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611569226.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-12
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

On embedded platforms, existing technologies struggle to efficiently and reliably achieve room acoustic correction at multiple sampling rates, resulting in issues such as high computational load, signal distortion, and poor parameter design reusability.

Method used

A two-stage decoupled architecture of offline design and online execution is adopted. Parameter design and verification are completed through a general computing platform to generate a self-describing parameter set. Fixed-coefficient filtering operations are performed on an embedded platform. The signal is restored without distortion by using complementary frequency division and time delay alignment mechanisms, and signal quality is ensured by a four-level progressive verification mechanism.

Benefits of technology

It significantly reduces the amount of real-time computation, improves design efficiency and signal quality in multi-sampling rate scenarios, and ensures the stable and reliable operation of the embedded platform.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123075339A_ABST
    Figure CN123075339A_ABST
Patent Text Reader

Abstract

This application relates to the technical field of digital audio signal processing, and discloses a multi-sampling-rate room correction method and system for embedded platforms. The method includes: calculating reference correction parameters based on the room impulse response acquired at a reference sampling rate; generating frequency division parameters and time delay alignment parameters for each target playback sampling rate, which together form a self-describing parameter set; loading the self-describing parameter set matching the current playback sampling rate onto the embedded platform; dividing the input audio by the cutoff frequency; directly transmitting the high-band signal; downsampling the low-band signal to the reference sampling rate, performing room equalization, and then upsampling it back to the current playback sampling rate; adjusting the time delay of the high-band signal, and superimposing it with the processed low-band signal for output; using a complementary frequency division structure; and the superposition of the high and low band signals is equivalent to the original input signal with pure delay. This application can efficiently and reliably achieve room acoustic correction in multi-sampling-rate playback scenarios under the limited computing power of embedded platforms.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the technical field of digital audio signal processing, and in particular to a multi-sampling-rate room correction method and system for embedded platforms. Background Technology

[0002] Room acoustic calibration is used to compensate for frequency response distortion introduced by both the loudspeakers and the room acoustic system. The conventional implementation involves acquiring the room impulse response through test signal acquisition, calculating the equalization curve through frequency domain analysis, and then converting it into a digital filter that is fed into the audio playback chain. With the development of high-definition audio technology, audio playback sampling rates are constantly increasing, and the computational load of full-band equalization processing increases synchronously with the sampling rate. For embedded playback platforms with limited computing power and storage resources, full-band calibration schemes often cannot meet the computing power requirements for real-time operation.

[0003] To reduce computational load, existing technologies often employ subband decomposition equalization, dividing the audio signal into multiple frequency bands for separate processing. However, these solutions typically perform equalization operations on each subband at the same sampling rate, requiring independent design of corresponding filters for different playback sampling rates, resulting in poor design reusability. Furthermore, the frequency division and sampling rate conversion introduced by subband processing can easily cause signal distortion. When parameters are directly deployed to embedded fixed-point platforms, quantization errors and execution deviations are prone to occur. The lack of a full-link verification mechanism from algorithm to hardware makes it impossible to guarantee the reliability of the output signal.

[0004] As can be seen from the above, how to efficiently and reliably achieve room sound correction in multi-sampling rate playback scenarios under the limited computing power of embedded platforms has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] To efficiently and reliably achieve room acoustic correction in multi-sampling rate playback scenarios under the limited computing power of embedded platforms, this application provides a multi-sampling rate room correction method and system for embedded platforms.

[0006] Firstly, this application provides a multi-sampling-rate room correction method for embedded platforms, employing the following technical solution: A multi-sampling-rate room calibration method for embedded platforms is proposed. Offline parameter design is completed on a general computing platform. The benchmark calibration parameters are calculated based on the room impulse response collected at the benchmark sampling rate. Then, frequency division parameters and time delay alignment parameters are generated for each target playback sampling rate. Together with the benchmark calibration parameters, they form a self-describing parameter set for the corresponding sampling rate. Each parameter set can be loaded and used independently without the design environment. Load a self-describing parameter set that matches the current playback sampling rate on the embedded platform. Divide the input audio signal according to the cutoff frequency corresponding to the division parameter. The cutoff frequency is set according to the high-frequency response flatness characteristics of the room acoustics. After division, a low-band signal below the cutoff frequency and a high-band signal above the cutoff frequency are obtained. The high-band signal is transmitted directly without equalization processing. The low-band signal is downsampled to the reference sampling rate, and then the reference correction parameter is called to complete the room equalization processing. Then it is upsampled to restore the current playback sampling rate. The high-band signal is time-delayed to align with the time delay of the processed low-band signal. The two signals are then superimposed and output. The frequency division adopts a complementary construction method so that the superposition of the low-band and high-band signals is equivalent to the original input signal with pure delay.

[0007] Optionally, the step of dividing the input audio signal according to the cutoff frequency corresponding to the division parameter includes: The cutoff frequency was determined based on the test statistics of the actual room acoustic impulse response. 24kHz was selected as the frequency division point. The room frequency response fluctuation amplitude above this cutoff frequency meets the flatness requirement and no room equalization processing is required. A linear-phase low-pass filter is designed to extract low-band signals below the cutoff frequency. A high-pass filter is obtained by subtracting the impulse response of the low-pass filter from the pure delay signal, and then the high-band signal above the cutoff frequency is extracted. The sum of the impulse responses of the low-pass filter and the high-pass filter is equivalent to a pure delay signal of fixed duration. The original input signal can be restored without distortion by directly superimposing the two signals.

[0008] Optionally, the step of calling the reference correction parameters after downsampling the low-band signal to the reference sampling rate includes: The sampling rate of the low-band signal is reduced to the reference sampling rate, and the pre-generated reference correction parameters under the reference sampling rate are directly called to complete the equalization process. All different target playback sampling rates share the same set of reference correction parameters. After equalization processing, the sampling rate of the low-band signal is restored to the current playback sampling rate. The reference correction parameters only need to be designed once based on the reference sampling rate, without the need for separate calculations for different playback sampling rates.

[0009] Optionally, the step of generating the self-describing parameter set corresponding to the sampling rate includes: Each generated parameter set undergoes a four-level progressive verification. The first level verifies the frequency domain synthesis flatness of the frequency divider, corresponding to the frequency divider design stage. The second level verifies the degree of restoration of the signal time domain superposition after frequency division, corresponding to the frequency division time domain synthesis stage. The third level verifies the signal transparency after sampling rate conversion and time delay alignment, corresponding to the sampling rate conversion and time delay alignment stage. The fourth level verifies the execution accuracy of the parameters under fixed-point arithmetic on the embedded platform, corresponding to the hardware deployment stage. If the preceding stage verification fails, the system returns to the corresponding stage to readjust the parameters. If the final stage verification fails, the system automatically adjusts the parameters and re-verifies. If the system still fails after adjustment, a degradation parameter package for the corresponding level is generated.

[0010] Optionally, the step of automatically adjusting parameters and re-verifying after the final-level verification fails includes: Within a preset number of iterations, parameter adjustments are performed sequentially. First, the optimal alignment value is searched within a preset range of adjacent samples of the time delay alignment parameter to eliminate subsample-level phase deviation introduced by fixed-point operations. Further fine-tuning the order of the frequency divider filter optimizes the synthesis flatness of the transition band under fixed-point accuracy; Finally, the coefficients of the correction parameters are dynamically scaled to avoid overflow or loss of precision in fixed-point calculations. After each adjustment is completed, the final-level verification is re-executed until the verification passes or the preset number of iterations is reached.

[0011] Optionally, the step of generating a degradation parameter package for the corresponding level if the adjustment still fails includes: Generate a mild degradation parameter package, close the high-band pass-through path, downsample the full-band input signal to the reference sampling rate, call the reference correction parameters to complete the equalization process, and then upsample to restore the original playback sampling rate, while retaining the basic functions of room correction. Generate a heavily degraded parameter package, disable all equalization processing steps, and directly pass the input signal to the output to avoid degraded audio signals at the output; Each degradation parameter package contains a corresponding mode identifier, which allows the embedded platform to identify the current operating mode.

[0012] Optionally, the step of loading a self-describing parameter set matching the current playback sampling rate and performing signal processing on the embedded platform includes: Each set of self-describing parameters corresponding to the target playback sampling rate reuses the same set of reference correction parameters, and only additionally stores the frequency division filter coefficients and time delay alignment values ​​corresponding to the sampling rate; After the embedded platform loads the parameter set, it only performs convolution and recursive operations with fixed coefficients, without performing any parameter calculations or optimization iterations. The computational complexity is determined solely by the filter order and the current playback sampling rate, and the resource consumption can be estimated in advance.

[0013] Optionally, the step of determining the cutoff frequency based on the test statistics of the actual room acoustic impulse response includes: Acoustic impulse response data of various typical rooms with different volumes and different interface decoration materials were collected, and the fluctuation amplitude of room frequency response in each frequency band was statistically analyzed. The lowest frequency point at which the room frequency response fluctuation amplitude meets the flatness requirement is taken as the general critical value. This general critical value is 24kHz, which is the common dividing point for the high frequency response of various typical rooms to enter the flat range. The frequency division logic is determined based on the inherent high-frequency attenuation characteristics of room acoustics. Room equalization is performed only on the low-frequency band with uneven response, while the original signal is retained and transmitted directly on the high-frequency band with flat response.

[0014] Optionally, the step of performing a four-level progressive validation on each generated parameter set includes: The first level of verification addresses coefficient matching deviations introduced by the design of complementary frequency dividers; the second level of verification addresses phase mismatch deviations introduced by frequency division time-domain synthesis; the third level of verification addresses signal distortion and aliasing introduced by sampling rate conversion and time delay alignment; and the fourth level of verification addresses computational errors introduced by embedded fixed-point quantization and block processing. The four-level verification is set up sequentially along with the sub-band processing architecture. The verification objects correspond one-to-one with the inherent risks of each stage of the architecture, and are executed synchronously with the generation of the sub-band multi-sampling rate architecture.

[0015] Secondly, this application provides a multi-sampling-rate room correction system for embedded platforms, employing the following technical solution: A multi-sampling-rate room correction system for embedded platforms includes: The offline parameter design module completes the offline parameter design on a general computing platform. It calculates the benchmark correction parameters based on the room impulse response collected at the benchmark sampling rate, and then generates frequency division parameters and time delay alignment parameters for each target playback sampling rate. Together with the benchmark correction parameters, they form a self-describing parameter set for the corresponding sampling rate. Each parameter set can be loaded and used independently without the design environment. The online subband correction module loads a self-describing parameter set that matches the current playback sampling rate on the embedded platform. It divides the input audio signal according to the cutoff frequency corresponding to the division parameter. This cutoff frequency is set according to the high-frequency response flatness characteristics of the room acoustics. After division, a low-band signal below the cutoff frequency and a high-band signal above the cutoff frequency are obtained. The high-band signal is transmitted directly without equalization processing. The low-band signal is downsampled to the reference sampling rate, and then the reference correction parameter is called to complete the room equalization processing. Finally, it is upsampled to restore the current playback sampling rate. The alignment and synthesis output module adjusts the time delay of the high-band signal to align the time delay of the high-band signal with that of the processed low-band signal. The two signals are then superimposed and output. The frequency division adopts a complementary construction method so that the superposition of the low-band and high-band signals is equivalent to the original input signal with pure delay.

[0016] In summary, this application includes at least one of the following beneficial technical effects: By employing a two-stage decoupled architecture of offline design and online execution, computationally intensive tasks such as spectrum analysis, equalization calculation, and parameter optimization are all transferred to a general-purpose computing platform. The embedded platform only needs to load a pre-generated parameter set to perform fixed-coefficient filtering operations, making the computational complexity fully predictable and adaptable to the computing power constraints of the embedded platform. Simultaneously, a sub-band processing architecture is constructed based on the flat high-frequency response of room acoustics. High-frequency signals are transmitted directly without equalization, eliminating the overhead of high-frequency equalization operations. Low-frequency signals are downsampled and processed using unified reference correction parameters, eliminating the need to repeatedly design filters for different playback sampling rates. This significantly reduces real-time computation while improving design efficiency in multi-sampling-rate scenarios.

[0017] This solution employs a complementary frequency division method and a time delay alignment mechanism, ensuring that the two divided signals can be superimposed without distortion to restore the original signal. This avoids signal distortion introduced by sub-band processing and guarantees signal quality before and after correction. The accompanying progressive verification logic corresponds one-to-one with each stage of the sub-band processing architecture, verifying parameter accuracy step-by-step from the frequency domain to the time domain, and from floating-point arithmetic to embedded fixed-point execution. Combined with automatic parameter adjustment and a graded degradation mechanism, it fundamentally prevents signal degradation from the embedded platform, achieving stable and reliable room correction in multi-sampling-rate playback scenarios. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the two-stage overall architecture of this embodiment. It is divided into an offline design stage and an online execution stage. The offline design stage runs on a general computing platform, takes the room impulse response as input, and outputs multiple sets of self-describing parameter sets corresponding to different sampling rates. The online execution stage runs on an embedded platform, loads the parameter sets corresponding to the sampling rates, and completes the real-time correction and output of the audio signal.

[0019] Figure 2 This is a flowchart illustrating a multi-sampling rate room correction method for an embedded platform according to an exemplary embodiment.

[0020] Figure 3 The flowchart for the 48kHz reference core design in this embodiment illustrates the complete process from the reference sampling rate room impulse response to the generation of reference correction parameters.

[0021] Figure 4 The subband decomposition signal flow diagram in this embodiment illustrates the complete processing flow of dividing a signal into low-band and high-band paths at a high sampling rate.

[0022] Figure 5 The frequency response diagram of the complementary frequency divider in this embodiment shows the frequency response curves of the low-pass filter, the high-pass filter, and the combination of the two at a sampling rate of 96kHz.

[0023] Figure 6 This is a flowchart of the four-level closed-loop verification process in this embodiment, which shows the progressive order of the four levels of verification, the verification objects at each level, the judgment criteria, and the exception handling process.

[0024] Figure 7 This is a block diagram illustrating a multi-sampling-rate room correction system for an embedded platform according to an exemplary embodiment. Detailed Implementation

[0025] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0026] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are included in at least one embodiment or example of this application. 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.

[0027] This application discloses a multi-sampling-rate room correction method for embedded platforms, primarily applicable to various embedded audio playback devices, especially suitable for embedded DSP power amplifiers with limited computing and storage resources. In applications such as home theater, indoor sound reinforcement, and professional audio playback, room acoustic correction is used to compensate for frequency response distortion introduced by both the speaker system and the room's acoustic environment, optimizing sound reproduction in the listening area and improving the perceived quality of audio playback. With the popularization of high-definition audio technology, higher sampling-rate audio formats are gradually becoming the market mainstream, and users' demands for sound quality at high sampling rates are constantly increasing. In traditional full-band room correction solutions, the computational load of digital filters increases synchronously with the playback sampling rate. The computational load of FIR filters with a fixed order is linearly positively correlated with the sampling rate. Doubling the sampling rate doubles the computational load. The order of the FIR filter is proportional to the sampling rate. To maintain the same frequency resolution, doubling the sampling rate doubles the order, and the number of samples that need to be processed per second also doubles. Therefore, the real-time computational load increases quadratically with the sampling rate, far exceeding the real-time computing power limit of most embedded DSP platforms, making it impossible to achieve full-band real-time correction at high sampling rates.

[0028] Meanwhile, audio products supporting multi-sampling rate playback typically require separate design of corresponding correction filters for each playback sampling rate. Correction parameters for different sampling rates are designed and stored independently, resulting in repetitive and redundant design processes, poor parameter reusability, long development cycles, and high mass production costs. More importantly, correction parameters designed in a floating-point environment are susceptible to signal distortion when directly deployed to fixed-point hardware platforms due to factors such as coefficient quantization errors, streaming block processing deviations, and finite word length effects. Existing sub-band equalization solutions generally suffer from three common limitations: First, all sub-bands participate in equalization design; even if the room response of some sub-bands is sufficiently flat, correction filters still need to be designed and run for them, wasting computing power. Second, each sub-band has its own independently designed filters; at high sampling rates, the filters for each sub-band need to be redesigned at the corresponding downsampling rate, making it impossible to reuse the design results at the baseline sampling rate. Third, there is a lack of deployment-level verification; after parameter design, the system directly enters the execution phase without verification under the fixed-point / streaming execution model of the target hardware, making it impossible to detect quantization errors and block effects in advance. Currently, there is a lack of a closed-loop verification mechanism from the design stage to the hardware execution stage, which means that when parameters are wrong, only a degraded signal can be silently output, ultimately affecting the user experience.

[0029] like Figure 1 As shown, this solution breaks down the room calibration process into two computationally decoupled stages: offline design and online execution. All computationally intensive tasks are concentrated in the offline stage, while the online stage performs only fixed calculations, fully adapting to the resource constraints of the embedded platform. This embodiment adopts a two-stage decoupled architecture of offline design and online execution, migrating all computationally intensive tasks to a general-purpose computing platform. The embedded end only performs filtering operations with fixed coefficients, making the computational load completely controllable and predictable. Based on the physical characteristic that the high-frequency response of a room tends to be flat, a sub-band processing framework is constructed. High-frequency signals are passed through without calibration, and low-frequency signals are downsampled and reused using a unified reference calibration core, significantly reducing real-time computation and improving the efficiency of multi-sampling rate design. A four-level progressive closed-loop verification method, corresponding one-to-one with each stage of the sub-band architecture, is constructed, combined with automatic parameter adjustment and graded degradation methods, to ensure the reliability of the embedded platform deployment. The three problems cannot be solved independently. Computing power optimization depends on the sub-band architecture of "high-band bypass + low-band core reuse". Core reuse depends on the perfect reconstruction guarantee of complementary frequency dividers. The frequency division / downsampling / upsampling / quantization links introduced by the entire architecture make closed-loop verification necessary. The three form a functionally coupled technical solution.

[0030] In this application embodiment, a multi-sampling rate room correction method for embedded platforms is described, referring to... Figure 2 It includes the following steps: S100, this stage executes on a general-purpose computing platform, such as a personal computer or server, where computing power and storage resources are unlimited. This stage uses the room impulse response acquired at a single reference sampling rate as the sole input to complete the design, verification, and encapsulation of all correction parameters. The final output is multiple sets of self-describing parameters corresponding to different playback sampling rates, each of which can be loaded and used independently of the design environment. This stage covers all computationally intensive aspects, including spectrum analysis, equalization calculation, filter design, subband decomposition, delay estimation, step-by-step verification, and parameter encapsulation. The embedded platform does not need to undertake any of these computational tasks.

[0031] In this embodiment of the application, S100 specifically includes the following steps: S101, this step generates reference correction parameters based on the room impulse response at the reference sampling rate, serving as the core correction parameter shared by all high sampling rate scenarios. In this embodiment, the reference sampling rate is selected from the standard sampling rate commonly used in the audio field, balancing correction accuracy and computational efficiency, while also conforming to the conventional design benchmarks of most audio algorithms.

[0032] First, frequency domain analysis is performed on the acquired room impulse response, and the amplitude spectrum of the room transfer function is obtained through Fast Fourier Transform. The conventional method for acquiring the room impulse response is as follows: a measurement microphone is placed at the listening position, a known test signal, such as a sinusoidal sweep signal or a maximum-length sequence signal, is played, and the signal picked up by the microphone is recorded. Then, the room impulse response is obtained through deconvolution. The deconvolutioned impulse response contains all the acoustic characteristics of the entire acoustic link from the speaker to the microphone, including frequency distortion caused by factors such as speaker frequency response, room reflections, reverberation, and standing waves. It forms the basis for room calibration.

[0033] A Fast Fourier Transform (FFT) is performed on the room impulse response, followed by 1 / 6 octave band smoothing to eliminate comb ripples, resulting in the smoothed room transfer function amplitude spectrum. Subsequently, a target curve decision is made, automatically selecting either a shelf or flat target frequency response curve. The target curve type can be matched based on the room's acoustic characteristics and the product's listening positioning. The equalization curve calculation satisfies the following relationship: in, For the target frequency response, it is usually set to a flat or slightly tilted frequency response curve, which can be adjusted according to listening preferences or product positioning; The measured amplitude spectrum of the room transfer function; The value represents the equalization curve to be achieved, expressed in dB. A positive value indicates that the gain of the corresponding frequency band needs to be increased, while a negative value indicates that the gain of the corresponding frequency band needs to be decreased.

[0034] During the calculation process, out-of-band The equalization curve in the interval is set to zero to limit the effective frequency range of the correction. After obtaining the equalization curve, an inverse fast Fourier transform is performed on it, and a Tukey window is added for truncation to obtain the impulse response coefficients of the FIR filter. Then, IIR fitting is performed on the residual spectrum after FIR correction to obtain the SOS matrix in second-order section form. .

[0035] Among them, the FIR filter is used to achieve high-precision linear phase equalization to ensure that the signal phase is not distorted; the IIR filter is used to achieve low-order high-order gain adjustment, achieving a large range of gain adjustment with low computational load. The two work together to balance correction accuracy and computational efficiency.

[0036] The final generated benchmark calibration parameters are stored in the form of a parameter set, denoted as the benchmark kernel parameter set: ,in, Let be the impulse response coefficients of the FIR filter, which is a sequence of real numbers; This is the set of second-order section coefficients for the IIR filter, containing the numerator and denominator coefficients of each second-order section.

[0037] This baseline kernel parameter set will be directly reused in subsequent high-sampling-rate paths. Only one design based on the baseline sampling rate is needed; all subsequent high-sampling-rate playback scenarios can directly call upon it without redesigning. For example... Figure 3 As shown, the design process of the reference core sequentially involves impulse response acquisition, spectrum analysis, equalization curve calculation, and filter design, ultimately outputting the reference core parameters. This design approach allows the parameter set to be distributed and reused across devices. During mass production, only one room measurement and one parameter design are required, and all devices of the same model can use the same set of parameters, significantly reducing mass production calibration costs.

[0038] S102, for each target playback sampling rate, complete the crossover design and sub-band core multiplexing architecture construction. This embodiment supports multiple target playback sampling rates, which correspond to different integer multiples of the base sampling rate, and can cover the current mainstream high-definition audio sampling rate specifications.

[0039] For playback sampling rate ,in This is the sampling rate multiplier. =2 corresponds to a 96kHz playback sampling rate. =4 corresponds to a 192kHz playback sampling rate, which can cover the current mainstream high-definition audio sampling rate specifications. The core idea is based on the following physical prior: the room response in the frequency band above 24kHz is flat enough that no correction is needed. Therefore, the signal is divided into a low band (<24kHz, requires correction) and a high band (≥24kHz, pass-through without any equalization). The low band is downsampled to 48kHz and processed using a reference core.

[0040] This scheme differs from existing subband equalization schemes in three key ways: First, high-band equalization is not performed; the signal passes directly through. The room response above 24kHz is already quite flat, eliminating the need for additional equalization, saving computational resources and avoiding phase distortion caused by unnecessary processing. Second, cross-sampling rate multiplexing is used; the calibration core is designed only once. After downsampling to 48kHz in the low band, the pre-designed calibration core at the 48kHz reference sampling rate is used directly, eliminating the need for redesign for high sampling rates. A single design is applicable to all high sampling rate scenarios. Third, complementary frequency division ensures lossless synthesis; the high-pass filter is... The complementary structure results in a low-pass plus high-pass equaling a pure delay. Even without any correction in the high-pass, the combined signal can be perfectly restored without frequency domain dips. All three are indispensable: complementary frequency division is a prerequisite for high-pass pass-through; otherwise, the synthesis will be distorted. Core reuse is the source of computing power advantage; and high-pass pass-through is the significance of core reuse.

[0041] In this embodiment, the crossover cutoff frequency is not arbitrarily selected, but rather obtained through statistical analysis of the acoustic characteristics of various typical rooms. The general pattern of room acoustics is as follows: in the low-frequency band, due to factors such as room standing waves and boundary reflections, the frequency response fluctuates significantly and distortion is obvious; as the frequency increases, the sound wave wavelength shortens, the phase randomness of room reflections increases, and the overall response tends to flatten after superposition, with the fluctuation amplitude gradually decreasing. To determine the accurate crossover cutoff point, acoustic impulse response data of various typical rooms with different volumes and different interface decoration materials were collected, and the fluctuation amplitude of the room's frequency response in each frequency band was statistically analyzed. The tests covered common residential and commercial acoustic scenarios of different sizes, and the decoration materials covered common interior decoration types, fully covering different acoustic environments. Statistical analysis was performed on the frequency response curves of all test data, and the maximum deviation of the amplitude response at each frequency point was calculated. Statistical results show that in most indoor environments, above a certain frequency, the attenuation of room acoustic reflections accelerates, the frequency response fluctuation amplitude decreases significantly, the overall frequency response tends to flatten, and the fluctuation amplitude can be controlled within a very small range, without the need for equalization correction. Therefore, the lowest frequency point where the room's frequency response fluctuation amplitude meets the flatness requirement is used as a universal critical value. This critical frequency is selected as the crossover point, which is the common boundary point for the high-frequency response of various typical rooms to enter the flat range. Based on this physical characteristic, the crossover logic is determined, performing room equalization processing only on the low-frequency band with uneven response, while retaining the original signal for direct transmission in the high-frequency band with flat response, minimizing the computational load while ensuring the correction effect. This crossover point selection is a targeted design based on the acoustic characteristics of the room, which differs from the conventional approach of dividing frequency bands according to the unit's frequency response in loudspeaker crossover, and is specifically adapted to the characteristics of room correction scenarios.

[0042] Frequency division is achieved using a complementary linear-phase FIR filter bank to ensure the distortion-free synthesized signal after frequency division. The specific design process is as follows: At the target playback sampling rate Design a linear phase FIR low-pass filter. The cutoff frequency is set to the aforementioned critical frequency, and a Hamming window is used to ensure that the passband flatness and stopband attenuation meet the design requirements. The filter order... Based on the preset group delay target Confirmed, among which It can be configured based on system delay budget and transition band performance requirements, while setting a minimum order lower limit to ensure that the transition band performance of the frequency division meets the requirements. The order calculation formula is: in, The preset group delay target, in seconds, can be adjusted according to the system's total delay budget. Higher order filters have narrower transition bands, cleaner frequency separation, and greater stopband attenuation. However, they also require more computational resources for frequency division filtering, resulting in a longer group delay. Adjustments can be made to achieve this. The value can achieve a balance between frequency division performance and computational overhead and system latency.

[0043] The selection of the frequency divider order can be referenced as follows: When the group delay target is 1ms, the corresponding order at 96kHz is approximately 193, and the transition band width is approximately 4.5kHz. The transition band is too wide, the attenuation near 24kHz is insufficient, and the high band may be mixed with low-frequency components that should be corrected; When the group delay target is 2.5ms, the corresponding order at 96kHz is approximately 481, and the transition band width is approximately 1.8kHz. The transition band is narrow enough, the computational load is moderate, and it can pass all self-test thresholds in actual tests; When the group delay target is 5ms, the corresponding order at 96kHz is approximately 961, and the transition band width is approximately 0.9kHz. The transition band is extremely narrow, but the order is doubled, the computational load of frequency division filtering increases, and the total system delay increases; When the group delay target is 10ms, the corresponding order at 96kHz is approximately 1921, and the transition band width is approximately 0.5kHz. This is an over-design, with diminishing returns and significant costs.

[0044] The high-pass filter is not designed independently; it is constructed complementaryly by subtracting the impulse response of the low-pass filter from the pure delay signal, satisfying the following relationship: in, For unit impulse signal, This is the group delay of the low-pass filter, measured in samples. Since the low-pass filter is linear in phase, its group delay is a fixed value, equal to half the filter order minus one.

[0045] The core characteristic of this complementary construction method is that the sum of the impulse responses of the low-pass filter and the high-pass filter is equivalent to a pure time-delay signal of fixed duration, that is: This characteristic means that after the signal is divided by a low-pass and a high-pass filter, the two signals can be directly added together to perfectly restore the original input signal, introducing only a fixed delay and avoiding problems such as frequency domain dips, phase distortion, or amplitude overshoot. Figure 5 As shown, the frequency response of the complementary frequency divider exhibits complementary low-pass and high-pass characteristics. The combined frequency response curve is nearly flat, with only a very small fluctuation near the cutoff frequency, verifying the distortion-free characteristics of the complementary frequency divider.

[0046] This complementary frequency division construction method is a necessary prerequisite for achieving high-band pass-through. If conventionally designed independent low-pass and high-pass filters are used, the frequency response superposition of the two filters will result in dips or bulges near the cutoff frequency, making distortion-free synthesis impossible. In this case, if the high-band pass-through is not processed, the synthesized signal will exhibit significant distortion at the frequency division point. The complementary construction, in principle, ensures the flatness of the synthesis, allowing the high-band signal to pass through directly without any processing, while simultaneously guaranteeing the quality of the synthesized signal. The two form a complementary technical whole.

[0047] After completing the crossover design, a core multiplexing logic for sub-band processing is established to achieve the reuse of reference correction parameters across sampling rates. For low-band signals below the cutoff frequency, their sampling rate is reduced to the reference sampling rate, and then the pre-designed reference correction parameters are directly called to complete the equalization process. All different target playback sampling rates share the same set of reference correction parameters. After equalization, the sampling rate of the low-band signal is then increased back to the current playback sampling rate. Specifically, for different high playback sampling rates, the low-band signal is downsampled by the corresponding factor to the reference sampling rate, equalization is completed by calling the reference core parameters, and then upsampled by the corresponding factor back to the original sampling rate. In this way, the reference correction parameters only need to be designed once based on the reference sampling rate, without the need for separate calculations for different playback sampling rates, which greatly improves the design efficiency of multi-sampling rate products, while ensuring the consistency of the correction effect under different sampling rates and avoiding differences in listening experience under different sampling rates.

[0048] like Figure 4As shown, the signal flow of subband decomposition is divided into two parallel paths: low-band processing and high-band processing. The low-band path sequentially undergoes four processing steps: low-pass filtering, downsampling, reference core correction, and upsampling; the high-band path sequentially undergoes two processing steps: high-pass filtering and time delay alignment; finally, the two signals are superimposed to obtain the output signal. This subband architecture has a significant advantage in computational complexity. In high sampling rate scenarios, only the frequency divider operates at the native high sampling rate, while the low-band portion is downsampled to the reference sampling rate before running the reference core. The overall computational complexity is significantly reduced compared to the full-band solution. The higher the sampling rate, the more significant the reduction in computational complexity, which can greatly alleviate the computational burden on embedded DSPs.

[0049] In the S103 subband processing architecture, the low-band path undergoes multiple processing stages such as frequency division, downsampling, equalization, and upsampling, while the high-band path only undergoes frequency division filtering. The total group delay of the two paths differs. If the two signals are directly superimposed, amplitude ripple will occur near the frequency division point due to the phase difference, affecting the output signal quality. Therefore, it is necessary to calculate the time delay difference between the two paths and perform alignment and compensation.

[0050] This step completes the time delay estimation during the design phase, using the impulse response cross-correlation method to accurately measure the group delay of the two paths. Unit impulse signals of length 65536 are input to both the low-band and high-band paths. Cross-correlation calculations are used to locate the peak positions of the two output signals; the number of samples corresponding to each peak position is the group delay of that path. Let the group delay of the low-band path be denoted as... The group delay of the high-band path is Then the delay that the high-band needs to compensate for is: During operation, high-band signal front-end compensation Two-way alignment and overlay can be achieved with just one zero. The calculated... The delay alignment parameter is stored in the parameter set corresponding to the sampling rate. During online operation, this value is directly called to perform delay alignment on the high-band signal. Since the low-band path has more processing stages and usually a larger group delay, it is necessary to add delay in the high-band path to ensure that the two signals are aligned in time before being superimposed.

[0051] After completing the design of all the above parameters, a set of self-describing parameters is generated for each target playback sampling rate. Each set of parameters contains all the coefficients and metadata required for real-time processing at the corresponding sampling rate. It can be loaded and executed independently of the design environment without relying on any computing resources on the design side.

[0052] For each high sampling rate, the corresponding sub-band parameter set is denoted as: in, The playback sampling rate corresponding to the current parameter set; and The parameters are completely consistent with the reference core, and the design results of the reference core are directly reused. These are the coefficients of the low-pass filter. These are the coefficients of the high-pass filter; This is the group delay of the low-pass filter; The order of the frequency divider filter; This is the delay alignment value for the high-band path.

[0053] In this embodiment, the self-describing parameter set corresponding to each target playback sampling rate reuses the same set of reference correction parameters, only additionally storing the frequency division filter coefficients and delay alignment values ​​for the corresponding sampling rate, resulting in extremely low additional storage overhead. Compared to complete full-band correction parameters, storage usage is significantly reduced, making it suitable for embedded platforms with limited storage resources. Simultaneously, the parameter set is self-describing, automatically identifying key information such as sampling rate and filter order during loading, adapting to different hardware platforms.

[0054] S104 performs a four-level progressive verification on each generated parameter set. The four levels of verification are set progressively according to the stages of the sub-band processing architecture. The verification object corresponds one-to-one with the inherent deviation of each stage of the architecture, and is executed synchronously with the generation of the sub-band multi-sampling rate architecture. The verification range expands step by step from the frequency domain to the time domain, and from floating-point arithmetic to fixed-point execution, to ensure the correctness of parameter execution on the target hardware.

[0055] The subband processing architecture introduces multiple processing stages, each of which may introduce deviations: frequency divider design may result in coefficient mismatch, leading to uneven synthesis; frequency division and time-domain synthesis may result in phase mismatch, causing distortion in time-domain reconstruction; sampling rate conversion and delay alignment may introduce signal distortion and aliasing; and embedded deployment may introduce computational errors due to fixed-point quantization and block processing. Therefore, four levels of verification correspond to these four stages, systematically checking for problems at each stage. If a previous stage fails verification, it does not proceed to the next stage but directly returns to the corresponding stage for correction, ensuring accurate problem localization and high efficiency in problem resolution.

[0056] like Figure 6 As shown, the four-level closed-loop verification proceeds progressively from the frequency domain to the time domain, and from floating-point to fixed-point. The first three levels are completed in floating-point precision to verify the mathematical correctness of the algorithm, while the fourth level is completed under the target hardware model to verify the reliability of the actual deployment. Each level of verification has a clear judgment criterion; if it passes, it proceeds to the next level; if it fails, the subsequent process is terminated and the process returns to the corresponding stage for adjustment.

[0057] The hierarchical relationship of the four-level verification is as follows: Level 1 verification targets the frequency divider coefficients, verified in a frequency domain floating-point environment; failure results in termination and redesign of the frequency divider. Level 2 verification targets the frequency divider and time domain superposition, verified in a time domain floating-point environment; failure results in termination and checking for coefficient truncation. Level 3 verification targets the complete path of frequency division, sampling rate conversion, and delay alignment, verified in a floating-point environment; failure results in termination and checking the sampling rate conversion module. Level 4 verification targets fixed-point quantization and streaming block execution, verified under the target hardware model; failure results in automatic parameter adjustment or a degradation package being issued. The first three levels ensure the mathematical correctness of the algorithm, while the fourth level ensures the reliability of the parameters under the actual execution conditions of the target hardware. The four levels progressively expand the verification scope, forming a complete closed loop of "parameter self-description → floating-point mathematical verification → hardware execution verification → automatic adjustment / degradation".

[0058] The first level of verification is the frequency division flatness check, corresponding to the design phase of the complementary frequency divider. It verifies the frequency domain synthesis flatness of the divider and identifies matching deviations introduced by the divider coefficient design. This verification is performed in the frequency domain to verify whether the sum of the frequency responses of the low-pass and high-pass dividers is sufficiently flat. The specific method is as follows: [The text abruptly ends here, likely due to an incomplete sentence or missing information.] and high-pass filter coefficients Perform 8192-point Fast Fourier Transform on each to obtain their respective frequency responses. and Calculate the combined frequency response of the two. Check the maximum deviation within the effective frequency band: The effective frequency band is set from 20Hz to 0.49 times the sampling rate, covering the effective audio frequency band audible to the human ear and avoiding aliasing areas near half the sampling rate. The pass / fail criterion for this level of verification is... dB. If the verification fails, the subsequent verification process is terminated, and the process returns to the divider design stage to readjust the filter parameters, such as adjusting the filter order and optimizing the window function parameters.

[0059] The second-level verification is a correlation test for frequency division reconstruction, corresponding to the time-domain synthesis stage of the frequency-divided signal. It verifies the degree of reconstruction after superimposing the two signals and investigates phase mismatch bias introduced by time-domain superposition. This verification is performed in the time domain, verifying whether the signal, after being split into high and low bands by the frequency divider and then added together, is consistent with the original signal and has only a fixed delay. The specific method is as follows: a pseudo-random white noise signal with a fixed seed and a duration of 2 seconds is generated, and its amplitude is normalized to ensure uniform signal energy distribution across the entire frequency band; the white noise signal is input into the frequency divider, and the low-band and high-band signals are obtained separately and then directly added together to obtain the reconstructed signal; the known group delay is compensated. Then, the normalized correlation coefficient between the reconstructed signal and the original input signal is calculated: in, The original input white noise signal, This is the signal after frequency division and reconstruction. The closer the normalized correlation coefficient is to 1, the higher the consistency between the reconstructed signal and the original signal. Under ideal complementary frequency division, this coefficient is 1. The pass / fail criterion for this level of verification is... ≥0.95. If the verification fails, terminate the subsequent verification process, check and correct the quantization truncation error of the filter coefficients, or adjust the coefficient length of the filter.

[0060] The third level of verification is the full-chain bypass reconstruction test, corresponding to the sampling rate conversion and delay alignment processing stage. It verifies the signal transparency of the complete signal link and investigates signal distortion and aliasing introduced by downsampling, upsampling, and delay adjustment. This verification checks whether the entire sub-band processing link remains transparent in floating-point precision after the equalization function is disabled; that is, after the input signal is processed by the complete sub-band link, it is completely consistent with the original signal except for a fixed delay. The specific method is as follows: First, a unit impulse signal is input into the complete sub-band link to obtain the peak position of the output signal, which is the total delay Dtotal of the link. Then, the equalization function of the reference core is disabled, i.e., the FIR coefficient is set to unit impulse and the IIR coefficient is set to pass-through. The same fixed-seed pseudo-random white noise is input, and the signal is sequentially processed through frequency division, downsampling, bypassing, upsampling, delay alignment, and superposition to obtain the full-chain bypass output signal. After compensating for the total delay Dtotal, the normalized correlation coefficient between the bypass output signal and the original input signal is calculated. in, This is the output signal for the entire chain bypass. The pass / fail criterion for this level of verification is... If the verification fails, terminate the subsequent verification process, check the design of the sampling rate conversion filter and the accuracy of the delay alignment parameters, and optimize the performance of the anti-aliasing filters for downsampling and upsampling.

[0061] The third level of verification is the full-chain bypass reconstruction test, corresponding to the complete floating-point data path of the sub-band processing link. It verifies whether the entire sub-band processing link remains transparent in floating-point precision after equalization is disabled, and investigates signal distortion and aliasing introduced by sampling rate conversion and delay alignment. This verification is performed in floating-point precision, verifying whether the signal, after passing through the complete sub-band processing link (frequency division → downsampling → bypass → upsampling → delay alignment → superposition), is consistent with the original signal, with only a fixed delay. The specific method is as follows: First, disable the reference core equalization function, input a unit impulse signal to the complete sub-band link, and let the output be... Subsequently, fixed-seed pseudo-random white noise was used. The output undergoes frequency division, downsampling, bypassing, upsampling, delay alignment, and superposition in sequence to obtain the full-chain bypass output. Compensation for total delay Then, calculate the normalized correlation coefficient: The closer the value is to 1, the less impact the complete subband link has on the input signal after equalization is disabled. In other words, the better the transparency of the subband processing chain to the signal. The pass / fail criterion for this level of verification is γ ≥ 0.95. If the verification fails, the subsequent verification process is terminated. The design of the sampling rate conversion module and the accuracy of the delay alignment parameters are checked. The performance of the downsampling and upsampling filters is optimized to ensure that no distortion or aliasing components exceeding the threshold are introduced during the sampling rate conversion process. The fourth level of verification is the target hardware fixed-point / streaming execution verification, corresponding to the deployment and execution phase of the embedded platform. It verifies the parameter execution accuracy under fixed-point arithmetic and investigates computational errors introduced by quantization and block processing. The first three levels of verification are all performed in floating-point precision, verifying the mathematical correctness of the algorithm. However, embedded DSPs typically use fixed-point arithmetic format and streaming block processing. Floating-point parameter quantization and block execution introduce additional errors, such as coefficient quantization errors, truncation errors, and discontinuities at block boundaries. This level of verification specifically verifies the correctness of parameters under the actual execution conditions of the target hardware, simulating the real hardware's computational environment. The specific verification process consists of three steps: The first step is fixed-point quantization, where the floating-point coefficients in the parameter set are quantized according to the word length of the target hardware. This embodiment uses a 32-bit Q30 fixed-point format to obtain a fixed-point parameter set. During the quantization process, the quantization error is recorded to ensure that the error is within an acceptable range. The second step is streaming block execution. The test signal is input in blocks according to the actual processing block size of the target hardware, simulating the streaming operation of a DSP that reads and processes blocks one by one. After each block is processed, the result of that block is output. Blocks are processed continuously, simulating the buffer connection of real hardware, to obtain the fixed-point execution output signal. The third step is multi-dimensional joint judgment. The fixed-point execution output is compared with the floating-point reference output, and a comprehensive judgment is made from multiple dimensions, including frequency band error, phase error, and aliasing energy. Frequency band error refers to the amplitude deviation between the fixed-point output and the floating-point reference in each frequency band, with a passing criterion of no more than 0.5dB; phase error refers to the phase offset after superposition of two paths near the frequency division point, with a passing criterion of no more than 5°; aliasing energy refers to the aliasing component energy in the passband during downsampling, with a passing criterion of no more than -60dBFS. When all indicators simultaneously meet the criteria, the parameter set passes hardware-level verification.

[0062] The existence of the aforementioned four-level verification system stems directly from the subband decomposition architecture itself. The introduction of multiple processing stages, such as complementary frequency division, downsampling / upsampling, core reuse, and path delay alignment, makes step-by-step verification necessary. The fourth level, hardware execution verification and the degradation package mechanism, arises because the subband architecture faces additional risks from fixed-point quantization and streaming block partitioning when deployed on embedded platforms. Therefore, closed-loop verification is not an independent additional function, but rather a functionally coupled technical solution with the subband decomposition framework: the subband architecture generates verification requirements, and the verification system ensures the reliable deployment of the subband architecture.

[0063] S105, if the fourth-level hardware verification fails, the system will automatically execute the parameter adjustment process, and perform multiple parameter adjustments in sequence within the preset maximum number of iterations. After each adjustment is completed, the fourth-level verification will be re-executed until the verification passes or the maximum number of iterations is reached.

[0064] In this embodiment, parameter adjustments are performed in a progressive order from easy to difficult and from local to global, namely, delay fine-tuning, frequency division parameter adjustment, and coefficient rescaling. Delay fine-tuning is performed first, adjusting the delay alignment parameter... The optimal alignment value is searched within a preset range of adjacent samples, with the search range set to ±1 to 4 samples. Each sample is tested individually to find the alignment value that minimizes the output error. This step is used to eliminate sub-sample-level phase deviation introduced by fixed-point quantization, because the ideal delay in floating-point calculations may have a deviation of less than half a sample in fixed-point implementations. Fine-tuning the number of samples can effectively reduce phase error. Next, the frequency division parameters are adjusted. Adjacent odd-numbered orders are selected within ±10% of the original frequency divider order, and the transition band width is fine-tuned to optimize the synthesized flatness of the transition band under fixed-point accuracy. Order adjustment changes both frequency division performance and computational load; the optimal order is selected within the allowable computational power. Finally, coefficient rescaling is performed, dynamically scaling the coefficients of the FIR and IIR filters. The quantization step size of the coefficients is adjusted, scaling the coefficients by a certain proportion so that the dynamic range of the intermediate results during fixed-point calculations matches the representation range of the fixed-point format, avoiding overflow or precision loss in fixed-point calculations. The scaling ratio is determined statistically based on the intermediate results of fixed-point arithmetic, ensuring that the maximum calculated value does not overflow and that the precision of small signals is sufficient. After each parameter adjustment, the fourth-level hardware verification is re-executed to determine whether it meets the acceptance criteria. If it passes, the adjustment stops and the adjusted parameter set is output; if it still fails after reaching the maximum number of iterations, the hierarchical degradation process is initiated.

[0065] If the hardware-level verification still fails after reaching the maximum number of iterations, the system automatically generates a corresponding degradation parameter package, divided into two levels: mild degradation and severe degradation. This ensures that the device can still function normally under extreme conditions and avoids outputting severely degraded signals. In this embodiment, in mild degradation mode, the high-band pass-through path is closed, the entire input signal across the frequency band is downsampled to the reference sampling rate, the reference correction parameters are called to complete the equalization process, and then upsampling is performed to restore the original playback sampling rate. This mode sacrifices the high-frequency response above the cutoff frequency but retains the basic function of room correction, ensuring the correction effect in the mid-low frequency band and maintaining a basic listening experience. This mode is equivalent to a full-band downsampling correction method, with a computational load slightly higher than the normal sub-band mode, but still far lower than the computational load of the native full-band high sampling rate correction. In severe degradation mode, all equalization processing stages are closed, all filters are degraded to unit pulses, the input signal is directly passed through to the output, completely bypassing the correction link, avoiding the output of degraded audio signals, and ensuring the device's most basic playback function. This mode serves as a last resort, ensuring that the device will not experience serious malfunctions such as silence or popping sounds regardless of parameter anomalies. Each degradation parameter package embeds a corresponding mode identifier, allowing the embedded platform to identify the current operating mode. This facilitates upper-level software informing the user of the current operating status or triggering a process to re-calibrate the room acoustics. The mode identifier is stored in a fixed field within the parameter set and can be directly read when the device loads parameters.

[0066] S200 loads a self-describing parameter set matching the current playback sampling rate on the embedded platform. It divides the input audio signal according to the cutoff frequency corresponding to the division parameter. The cutoff frequency is set according to the high-frequency response flatness characteristics of the room acoustics. After division, a low-band signal below the cutoff frequency and a high-band signal above the cutoff frequency are obtained. The high-band signal is transmitted directly without equalization processing. The low-band signal is downsampled to the reference sampling rate, and then the reference correction parameter is called to complete the room equalization processing. Finally, it is upsampled to restore the current playback sampling rate.

[0067] S201: The embedded platform detects the current playback sampling rate and loads a self-describing parameter set that matches the current playback sampling rate. If the playback sampling rate changes, the loaded parameter set is switched accordingly, achieving seamless switching between different sampling rates.

[0068] In this embodiment, after the parameter set is loaded, all filter coefficients, delay values, and other parameters are fixed. No parameter calculations or optimization iterations are performed during operation; only convolution and recursive operations with fixed coefficients are executed. The computational complexity is determined solely by the filter order and the current playback sampling rate, allowing for pre-estimation of runtime resource consumption and adapting to the computing power planning of embedded platforms. This pure execution mode is highly suitable for fixed-point DSP implementations, accurately predicting the worst-case CPU utilization, avoiding audio stuttering caused by computational overload, and simplifying the development of embedded software.

[0069] S202, Input the audio signal to be processed The sampling rate is the current playback sampling rate. The signal length is L. The input signal is simultaneously fed into a low-pass filter and a high-pass filter to extract the low-band signal below the cutoff frequency and the high-band signal above the cutoff frequency, respectively. The low-band signal is denoted as . High-band signal is denoted as The symbol "*" indicates a convolution operation. High-band signals undergo no equalization processing and are transmitted directly to the next signal, preserving the original high-frequency signal's sound quality while saving on high-frequency equalization computation overhead.

[0070] Execute on low-band signals Downsampling, where This is the ratio of the current playback sampling rate to the reference sampling rate. Before downsampling, the signal undergoes anti-aliasing filtering to remove frequency components higher than half the downsampled sampling rate, thus preventing aliasing distortion. After downsampling, the sampling rate of the low-band signal is reduced to the reference sampling rate, and the signal length is correspondingly shortened to L / .

[0071] S203: The downsampled low-band signal is processed sequentially by an FIR filter and an IIR filter, and room equalization correction is performed by calling the reference correction parameters. The correction process is as follows: first, an FIR convolution operation is performed to achieve linear phase amplitude equalization; then, an IIR recursive operation is performed to achieve higher-order gain adjustment. The two work together to obtain the equalized low-band signal. Since the correction parameters at the reference sampling rate are directly reused, there is no need to redesign the filter for high sampling rates, ensuring the consistency of the correction effect at different sampling rates.

[0072] After equalization is completed, the low-band signal is processed. Upsampling uses interpolation to restore the signal's sampling rate from a reference sampling rate to the current playback sampling rate. After inserting zero values ​​during upsampling, the signal is smoothed and image spectra are removed through interpolation filtering to obtain the processed low-band signal. The signal length is restored to its original length L.

[0073] The S300 performs time delay adjustment on the high-band signal to align the time delay of the high-band signal with that of the processed low-band signal. The two signals are then superimposed and output. The frequency division adopts a complementary construction method so that the superposition of the low-band and high-band signals is equivalent to the original input signal with pure time delay.

[0074] S301 performs time delay adjustment on the high-band signal, supplementing the high-band signal with... Using zero samples, the group delay difference between the high-band and low-band paths is filled, aligning the time delay of the high-band signal with that of the processed low-band signal, thus obtaining the time-delay-aligned high-band signal. Delay alignment ensures that the two signals are synchronized in time, avoiding amplitude dips caused by phase cancellation at the frequency division point during superposition, and ensuring the flatness of the frequency response near the frequency division point.

[0075] S302 will process the low-band signal High-band signal aligned with time delay The corresponding sample points are added one by one to obtain the final output signal: Because the frequency divider uses a complementary structure, the superposition of the low-band and high-band signals is equivalent to the original input signal with pure delay. No additional distortion is introduced due to the frequency division and synthesis process. While achieving room correction in the low-frequency band, the original signal quality of the high-frequency band is completely preserved.

[0076] In this embodiment, the application of an embedded DSP power amplifier product is verified in a practical application scenario. This product uses a fixed-point DSP, supports multi-sampling rate playback, and has a built-in room correction function. When using a traditional full-band FIR room correction scheme, the computational load in high-sampling-rate scenarios far exceeds the DSP's computing power limit, making real-time operation impossible. The only solution is to reduce the filter order, leading to a decrease in correction accuracy. With the solution in this embodiment, high-frequency bands are directly passed at high sampling rates, while low-band bands are downsampled to the reference sampling rate before running the reference core. The overall computational load is significantly reduced, fully meeting real-time operation requirements, and the correction accuracy is comparable to the full-band scheme at the reference sampling rate. Regarding reliability, the four-level verification system can identify most parameter problems during the design phase, significantly reducing the error rate of parameter deployment. The graded degradation method ensures basic playback functionality under abnormal conditions, avoiding serious failures caused by parameter anomalies. During product development, traditional solutions require independent design of correction filters for each sampling rate, resulting in a long design cycle. With this solution, only one reference core design is needed, followed by separate design and verification of the crossovers for each high sampling rate, significantly shortening the overall design cycle and greatly improving development efficiency. In the mass production stage, only one standard room measurement and offline parameter design need to be completed, and the parameter set can be distributed to all devices in batches, significantly reducing the cost of mass production calibration.

[0077] In this embodiment, the reference sampling rate can also be selected according to actual needs, and the corresponding high sampling rate will be adjusted accordingly to adapt to different audio standards. The crossover cutoff frequency can also be adjusted according to the acoustic characteristics of the target application scenario. For example, the crossover point can be appropriately increased for professional spaces with better acoustic conditions, and appropriately decreased for small acoustic spaces, all based on actual room acoustic test results. The embedded platform is not limited to DSP; other embedded processors such as microcontrollers and ARM can also be used. The fixed-point format can also be adjusted according to the processor word length, and the corresponding verification standards and adjustment methods can be adapted accordingly. The judgment threshold of the four-level verification can also be flexibly adjusted according to the performance requirements of the product. The threshold can be appropriately tightened for high-fidelity products, and appropriately relaxed for entry-level products, balancing performance and cost.

[0078] Based on a real-world deployment case of an embedded DSP power amplifier product, this solution fundamentally reduces the computational load at high sampling rates through subband decomposition and reference core reuse architecture, adapting to the computational constraints of embedded platforms. In traditional full-band FIR correction schemes, the filter order and computational load increase synchronously with the sampling rate. At high sampling rates, the computational load far exceeds the computational limit of most embedded DSPs, making real-time operation impossible. This solution, based on the physical characteristics of flat high-frequency acoustics in a room, sets the frequency band above 24kHz as a direct path requiring no correction. Only the low-frequency band is downsampled to the reference rate and then processed by a unified correction core. At high sampling rates, only the crossover operates at the native sampling rate. In mainstream high-definition sampling rate scenarios, the real-time computational load is significantly reduced compared to the full-band solution, falling entirely within the computational capacity of the embedded DSP. This achieves room correction functionality at high sampling rates under limited computational power conditions.

[0079] In the R&D and mass production of multi-sampling rate products, the cross-sampling rate core reuse mechanism of this solution significantly improves design efficiency and reduces mass production costs. Traditional solutions require the independent design of a complete set of calibration filters for each sampling rate, resulting in redundant parameter design and long R&D cycles. In this solution, the reference calibration core only needs to be designed once based on the reference sampling rate. For high sampling rate scenarios, only the corresponding frequency divider and delay parameters need to be designed separately to directly reuse the reference core. The design cycle of the entire set of multi-sampling rate parameters is significantly shortened compared to traditional solutions. In the mass production stage, only one room measurement and parameter design are required to cover all sampling rate products. The parameter set can be distributed to all devices in batches, greatly reducing the R&D and calibration costs of multi-sampling rate products.

[0080] To address the deployment risks of embedded fixed-point platforms, this solution employs a four-level progressive verification and graded degradation mechanism to construct a complete reliability closed loop, ensuring the stable and reliable operation of the multi-sampling rate correction function. Traditional solutions deploy directly after parameter design, making it impossible to pre-screen errors introduced by floating-point parameter fixed-point quantization and streaming block execution, easily leading to silent output of degraded signals. This solution's four-level verification corresponds one-to-one with each stage of the sub-band architecture, verifying parameter correctness step-by-step from the frequency domain to the time domain, and from floating-point to fixed-point. Most parameter issues can be identified during the design phase. Combined with automatic parameter adjustment and a two-level degradation fallback mechanism, the parameter deployment error rate is significantly reduced. Even in extreme cases, basic playback functionality is guaranteed, and serious failures such as popping or silence caused by parameter anomalies have never occurred. Highly reliable multi-sampling rate room correction has been implemented on embedded platforms.

[0081] This application discloses a multi-sampling-rate room correction method for embedded platforms, referring to... Figure 7 ,include: The offline parameter design module 001 completes the offline parameter design on a general computing platform. It calculates the benchmark correction parameters based on the room impulse response collected at the benchmark sampling rate, and then generates frequency division parameters and time delay alignment parameters for each target playback sampling rate. Together with the benchmark correction parameters, they form a self-describing parameter set for the corresponding sampling rate. Each parameter set can be loaded and used independently without being separated from the design environment. The online subband correction module 002 loads a self-describing parameter set that matches the current playback sampling rate on the embedded platform. It divides the input audio signal according to the cutoff frequency corresponding to the division parameter. The cutoff frequency is set according to the flat high-frequency response characteristics of the room acoustics. After division, a low-band signal below the cutoff frequency and a high-band signal above the cutoff frequency are obtained. The high-band signal is transmitted directly without equalization processing. The low-band signal is downsampled to the reference sampling rate, and then the reference correction parameter is called to complete the room equalization processing. Finally, it is upsampled to restore the current playback sampling rate. The alignment and synthesis output module 003 adjusts the time delay of the high-band signal to align the time delay of the high-band signal with that of the processed low-band signal. The two signals are then superimposed and output. The frequency division adopts a complementary construction method so that the superposition of the low-band and high-band signals is equivalent to the original input signal with pure delay.

[0082] This application also discloses an electronic device, including a processor, wherein the processor runs a program for the multi-sampling rate room correction method for embedded platforms as described in any one of the above embodiments.

[0083] This application also discloses a storage medium storing a program for the multi-sampling rate room correction method for embedded platforms as described in any one of the above embodiments.

[0084] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A multi-sampling-rate room correction method for embedded platforms, characterized in that, include: Offline parameter design is completed on a general computing platform. The baseline correction parameters are calculated based on the room impulse response collected at the baseline sampling rate. Then, frequency division parameters and time delay alignment parameters are generated for each target playback sampling rate. Together with the baseline correction parameters, they form a self-describing parameter set for the corresponding sampling rate. Each parameter set can be loaded and used independently without being separated from the design environment. Load a self-describing parameter set that matches the current playback sampling rate on the embedded platform. Divide the input audio signal according to the cutoff frequency corresponding to the division parameter. The cutoff frequency is set according to the high-frequency response flatness characteristics of the room acoustics. After division, a low-band signal below the cutoff frequency and a high-band signal above the cutoff frequency are obtained. The high-band signal is transmitted directly without equalization processing. The low-band signal is downsampled to the reference sampling rate, and then the reference correction parameter is called to complete the room equalization processing. Then it is upsampled to restore the current playback sampling rate. The high-band signal is time-delayed to align with the time delay of the processed low-band signal. The two signals are then superimposed and output. The frequency division adopts a complementary construction method so that the superposition of the low-band and high-band signals is equivalent to the original input signal with pure delay.

2. The multi-sampling-rate room correction method for embedded platforms according to claim 1, characterized in that, The step of dividing the input audio signal according to the cutoff frequency corresponding to the division parameter includes: The cutoff frequency was determined based on the test statistics of the actual room acoustic impulse response. 24kHz was selected as the frequency division point. The room frequency response fluctuation amplitude above this cutoff frequency meets the flatness requirement and no room equalization processing is required. A linear-phase low-pass filter is designed to extract low-band signals below the cutoff frequency. A high-pass filter is obtained by subtracting the impulse response of the low-pass filter from the pure delay signal, and then the high-band signal above the cutoff frequency is extracted. The sum of the impulse responses of the low-pass filter and the high-pass filter is equivalent to a pure delay signal of fixed duration. The original input signal can be restored without distortion by directly superimposing the two signals.

3. The multi-sampling-rate room correction method for embedded platforms according to claim 1, characterized in that, The step of calling the reference correction parameters after downsampling the low-band signal to the reference sampling rate includes: The sampling rate of the low-band signal is reduced to the reference sampling rate, and the pre-generated reference correction parameters under the reference sampling rate are directly called to complete the equalization process. All different target playback sampling rates share the same set of reference correction parameters. After equalization processing, the sampling rate of the low-band signal is restored to the current playback sampling rate. The reference correction parameters only need to be designed once based on the reference sampling rate, without the need for separate calculations for different playback sampling rates.

4. The multi-sampling-rate room correction method for embedded platforms according to claim 1, characterized in that, The step of generating the self-describing parameter set corresponding to the sampling rate includes: Each generated parameter set undergoes a four-level progressive verification. The first level verifies the frequency domain synthesis flatness of the frequency divider, corresponding to the frequency divider design stage. The second level verifies the degree of restoration of the signal time domain superposition after frequency division, corresponding to the frequency division time domain synthesis stage. The third level verifies the signal transparency after sampling rate conversion and time delay alignment, corresponding to the sampling rate conversion and time delay alignment stage. The fourth level verifies the execution accuracy of the parameters under fixed-point arithmetic on the embedded platform, corresponding to the hardware deployment stage. If the preceding stage verification fails, the system returns to the corresponding stage to readjust the parameters. If the final stage verification fails, the system automatically adjusts the parameters and re-verifies. If the system still fails after adjustment, a degradation parameter package for the corresponding level is generated.

5. The multi-sampling-rate room correction method for embedded platforms according to claim 4, characterized in that, The step of automatically adjusting parameters and re-verifying after the final-level validation fails includes: Within a preset number of iterations, parameter adjustments are performed sequentially. First, the optimal alignment value is searched within a preset range of adjacent samples of the time delay alignment parameter to eliminate subsample-level phase deviation introduced by fixed-point operations. Further fine-tuning the order of the frequency divider filter optimizes the synthesis flatness of the transition band under fixed-point accuracy; Finally, the coefficients of the correction parameters are dynamically scaled to avoid overflow or loss of precision in fixed-point calculations. After each adjustment is completed, the final-level verification is re-executed until the verification passes or the preset number of iterations is reached.

6. The multi-sampling-rate room correction method for embedded platforms according to claim 5, characterized in that, The step of generating the corresponding level's degradation parameter package if the adjustment still fails includes: Generate a mild degradation parameter package, close the high-band pass-through path, downsample the full-band input signal to the reference sampling rate, call the reference correction parameters to complete the equalization process, and then upsample to restore the original playback sampling rate, while retaining the basic functions of room correction. Generate a heavily degraded parameter package, disable all equalization processing steps, and directly pass the input signal to the output to avoid degraded audio signals at the output; Each degradation parameter package contains a corresponding mode identifier, which allows the embedded platform to identify the current operating mode.

7. The multi-sampling-rate room correction method for embedded platforms according to claim 1, characterized in that, The steps of loading a self-describing parameter set matching the current playback sampling rate and performing signal processing on an embedded platform include: Each set of self-describing parameters corresponding to the target playback sampling rate reuses the same set of reference correction parameters, and only additionally stores the frequency division filter coefficients and time delay alignment values ​​corresponding to the sampling rate; After the embedded platform loads the parameter set, it only performs convolution and recursive operations with fixed coefficients, without performing any parameter calculations or optimization iterations. The computational complexity is determined solely by the filter order and the current playback sampling rate, and the resource consumption can be estimated in advance.

8. The multi-sampling-rate room correction method for embedded platforms according to claim 2, characterized in that, The steps for determining the cutoff frequency based on the statistical results of the actual room acoustic impulse response include: Acoustic impulse response data of various typical rooms with different volumes and different interface decoration materials were collected, and the fluctuation amplitude of room frequency response in each frequency band was statistically analyzed. The lowest frequency point at which the room frequency response fluctuation amplitude meets the flatness requirement is taken as the general critical value. This general critical value is 24kHz, which is the common dividing point for the high frequency response of various typical rooms to enter the flat range. The frequency division logic is determined based on the inherent high-frequency attenuation characteristics of room acoustics. Room equalization is performed only on the low-frequency band with uneven response, while the original signal is retained and transmitted directly on the high-frequency band with flat response.

9. The multi-sampling-rate room correction method for embedded platforms according to claim 4, characterized in that, The step of performing a four-level progressive validation on each generated parameter set includes: The first level of verification addresses coefficient matching deviations introduced by the design of complementary frequency dividers; the second level of verification addresses phase mismatch deviations introduced by frequency division time-domain synthesis; the third level of verification addresses signal distortion and aliasing introduced by sampling rate conversion and time delay alignment; and the fourth level of verification addresses computational errors introduced by embedded fixed-point quantization and block processing. The four-level verification is set up sequentially along with the sub-band processing architecture. The verification objects correspond one-to-one with the inherent risks of each stage of the architecture, and are executed synchronously with the generation of the sub-band multi-sampling rate architecture.

10. A multi-sampling-rate room correction system for embedded platforms, characterized in that, include: The offline parameter design module completes the offline parameter design on a general computing platform. It calculates the benchmark correction parameters based on the room impulse response collected at the benchmark sampling rate, and then generates frequency division parameters and time delay alignment parameters for each target playback sampling rate. Together with the benchmark correction parameters, they form a self-describing parameter set for the corresponding sampling rate. Each parameter set can be loaded and used independently without the design environment. The online subband correction module loads a self-describing parameter set that matches the current playback sampling rate on the embedded platform. It divides the input audio signal according to the cutoff frequency corresponding to the division parameter. This cutoff frequency is set according to the high-frequency response flatness characteristics of the room acoustics. After division, a low-band signal below the cutoff frequency and a high-band signal above the cutoff frequency are obtained. The high-band signal is transmitted directly without equalization processing. The low-band signal is downsampled to the reference sampling rate, and then the reference correction parameter is called to complete the room equalization processing. Finally, it is upsampled to restore the current playback sampling rate. The alignment and synthesis output module adjusts the time delay of the high-band signal to align the time delay of the high-band signal with that of the processed low-band signal. The two signals are then superimposed and output. The frequency division adopts a complementary construction method so that the superposition of the low-band and high-band signals is equivalent to the original input signal with pure delay.