An audio scene-aware power amplifier power supply system adaptive control method and system

Abstract

The application discloses a power amplifier power supply system adaptive control method and system based on audio scene perception, which extracts multi-dimensional characteristic parameters such as dynamic range, low-frequency energy proportion and short-time power change rate of an audio signal to perform scene classification, and simultaneously predicts future audio dynamic trend according to the change trend of the characteristic parameters; generates an execution schedule containing bus voltage transition total time length, voltage change acceleration curve, auxiliary bus intervention envelope curve and charging power step-down node according to the scene type and the prediction result; performs bus voltage gradual adjustment with variable acceleration according to the prediction result, and performs smooth cut-in of the auxiliary bus with an S-shaped intervention envelope curve; and simultaneously updates the power amplifier static bias compensation coefficient to the power amplification stage at a synchronous rate. Optimal timing matching of power supply scheduling and audio content dynamic demand is realized.

Description

Technical Field

[0001] This invention relates to the field of audio power amplifier power management technology, specifically to a power amplifier power supply bus voltage variable acceleration scheduling method and system that uses a timetable as the control carrier to sense the dynamic characteristics of the audio playback content in real time and predict its future trend. Background Technology

[0002] The power supply solution of a professional audio power amplifier directly affects the system's sound quality, energy efficiency, and operational reliability. With the increasing diversification of professional audio applications, a single fixed installation system may serve multiple purposes within a single day, such as conference voice reinforcement, background music playback, pop music performances, classical music concerts, and even movie screenings. This places higher demands on the adaptability and flexibility of the power amplifier's power supply system.

[0003] Existing power amplifier power management solutions mainly suffer from the following technical defects: Firstly, there is a mismatch between power supply scheduling and dynamic audio demands. Traditional power amplifier systems typically employ a fixed bus voltage supply scheme, maintaining a constant voltage on the main power supply bus regardless of the type of content being played. For example, a typical 500-watt Class AB amplifier, in a conference audio broadcast scenario, typically only delivers around 15 watts of audio output power. However, because the bus voltage remains constant, the static voltage drop loss on the power amplifier stage output transistors remains high, resulting in a total system power consumption of 45 to 50 watts—more than two-thirds of the electrical energy is converted into useless heat. While some technologies attempt to adjust the supply voltage by detecting the audio signal amplitude (such as a servo power supply voltage tracking the audio source signal amplitude), these solutions essentially respond to changes in signal amplitude that have already occurred. When transitioning from low-dynamic to high-dynamic content, the power supply adjustment always lags behind the changes in audio content, causing the bus voltage to not have reached its optimal level by the time the high-dynamic peak arrives.

[0004] Secondly, the control method for voltage adjustment is simplistic and lacks targeted optimization. Existing dynamic voltage adjustment schemes typically employ a fixed-rate gradual change or simple first-order hysteresis filtering to control the change in bus voltage from its current value to the target value. Regardless of whether the current scene transition is urgent (e.g., a sudden shift from background music to a live sound reinforcement climax) or gradual (e.g., a slow transition from pop music to background music), the voltage adjustment rate and acceleration use the same preset values. This "one-size-fits-all" control mode results in insufficient response speed in scenarios requiring rapid response, and prolonged transition periods due to excessively slow adjustment speeds in scenarios where rapid response is not required. It fails to optimize the temporal characteristics of the voltage adjustment process according to the actual dynamic needs of the audio content.

[0005] Thirdly, there is the issue of power amplifier nonlinear distortion degradation during the voltage adjustment transition period. When the power supply strategy switches due to changes in the scenario, the adjustment process of the bus voltage will change the operating point state of the power amplifier stage, causing the nonlinear characteristics of the power amplifier (such as transconductance curve and saturation characteristics) to drift. For professional audio applications that pursue high fidelity, this drift will introduce measurable harmonic distortion and intermodulation distortion. Existing research (such as Korean Patent KR102550079B1) discloses a power amplifier nonlinear distortion compensation device based on deep learning, but such solutions are aimed at distortion compensation under steady-state operating conditions and do not address the synchronous compensation problem of continuous drift of nonlinear characteristics during dynamic changes in bus voltage.

[0006] Fourth, the intervention of the auxiliary DC bus lacks a smooth transition plan. During the switch from a disabled state to a fully intervened state, if the change in intervention depth is too drastic, it can introduce voltage spikes or current surges on the DC bus, potentially triggering overcurrent protection or causing electromagnetic interference to the audio signal path. Traditional solutions typically employ simple step-like or linear changes for auxiliary bus intervention, lacking precise control over the transient characteristics of the intervention process.

[0007] In the Chinese patent database, CN112671226A discloses a scheme for pre-feedback of audio signals to the power supply control stage to adjust the output voltage in advance. However, its control target is the voltage / current loop regulation inside the power converter, and it does not involve the generation of power supply strategies and execution schedule planning based on audio scene classification. CN117544118A discloses a scheme for determining the control level based on target audio data and controlling the boost circuit to boost the voltage. However, its boost control is based on discrete level switching of audio data frames and does not involve the dynamic configuration of acceleration during continuous gradual changes. CN103607176B discloses a scheme for tracking the amplitude of the audio source signal by the servo power supply voltage. However, its tracking is continuous tracking in the analog domain and does not involve timing planning and execution schedule generation in the digital domain. Summary of the Invention

[0008] To overcome the shortcomings of the prior art, this invention provides an adaptive control method and system for a power amplifier power supply system based on audio scene perception. By using an execution schedule as a carrier, the scene recognition result is deeply coupled with the power supply scheduling process. Variable acceleration control is used to achieve timing matching between the voltage adjustment process and the dynamic audio demand. An S-shaped intervention envelope curve is used to achieve smooth entry of the auxiliary bus. Synchronous updates of the bias compensation coefficient are used to eliminate distortion and degradation during the transition period.

[0009] To achieve the above objectives, the following technical solution is provided: An adaptive control method for a power amplifier power supply system based on audio scene awareness includes the following steps: S1 acquires audio input signals in real time and extracts audio feature parameters to characterize the dynamic properties and energy distribution of audio content.

[0010] S2, based on the extracted audio feature parameters, classify the current audio content into a scene and identify it as one of multiple preset audio scene types.

[0011] S3, based on the changing trend of audio feature parameters within a preset time window, predicts the dynamic changing trend of audio content within a preset duration in the future. The dynamic changing trend includes at least a high dynamic changing trend and a low dynamic changing trend.

[0012] S4. Based on the identified scenario type and predicted dynamic change trend, generate the execution schedule of the power amplifier power supply system. The execution schedule includes at least: the total transition time of the main power supply bus voltage of the power amplifier from the current value to the target value, the acceleration curve of the voltage change acceleration at each moment within the total transition time as a function of time, the intervention envelope curve of the auxiliary DC bus intervention depth as a function of time, and the step-down nodes of the energy storage unit charging power as a function of time.

[0013] S5, according to the execution schedule, performs a gradual adjustment of the bus voltage from the current value to the target value using an acceleration curve, and performs smooth intervention and withdrawal of the auxiliary DC bus using an intervention envelope curve.

[0014] Among them, the acceleration of voltage change is higher when the predicted dynamic change trend is higher than that when the predicted dynamic change trend is lower, and a positive acceleration value indicates that the bus voltage is increasing, while a negative acceleration value indicates that the bus voltage is decreasing.

[0015] In some embodiments, the audio characteristic parameters include at least dynamic range, low-frequency energy proportion, and short-time power change rate. Dynamic range is used to distinguish between high-dynamic and low-dynamic content; low-frequency energy proportion is used to identify content containing bass components; the calculation window for the short-time power change rate is no more than 10 milliseconds, used to capture the transient impact characteristics of the audio signal.

[0016] In some embodiments, trend prediction is determined based on the slope of the short-term power change rate within a preset window and the cumulative change in the proportion of low-frequency energy within the preset window: when the slope of the short-term power change rate exceeds a first slope threshold and the duration exceeds a preset duration, it is predicted as a high dynamic change trend; when the slope of the short-term power change rate is lower than a second slope threshold, it is predicted as a low dynamic change trend; the first slope threshold is greater than the second slope threshold.

[0017] In some embodiments, in the acceleration curve: when a high dynamic change trend is predicted, the absolute value of the voltage change acceleration is configured as a first acceleration value; when a low dynamic change trend is predicted, the absolute value of the voltage change acceleration is configured as a second acceleration value; the first acceleration value is greater than the second acceleration value. More preferably, the first acceleration value does not exceed 5% per square second of the rated bus voltage value, and the second acceleration value does not exceed 1% per square second of the rated bus voltage value.

[0018] In some embodiments, the intervention envelope curve is an S-shaped curve, with the slopes of the initial and final segments being less than the slope of the middle segment, and the time constant is configured to be 200 milliseconds to 800 milliseconds depending on the scenario type.

[0019] In some embodiments, the step-down node includes at least: the charging power is reduced to a first preset power ratio at a first preset time after the scene switching is triggered, the charging power is reduced to a second preset power ratio at a second preset time, and the charging power is reduced to a third preset power ratio at a third preset time, and the charging power change rate between each adjacent node does not exceed 20% of the rated charging power per second.

[0020] In some embodiments, during the gradual adjustment of the bus voltage, the static bias compensation coefficient of the power amplifier stage is adjusted synchronously. The static bias compensation coefficient is obtained by looking up a table through a pre-established bus voltage-power amplifier nonlinear characteristic mapping model, and its update period does not exceed the duration corresponding to 1% of the bus voltage change caused by the voltage change acceleration.

[0021] The present invention also provides an adaptive control system for a power amplifier power supply system based on audio scene perception to implement the above method, comprising: an audio scene perception and prediction module, an execution schedule generation module, a bus voltage variable acceleration control module, an auxiliary bus envelope control module, and an offset synchronization compensation module.

[0022] In some embodiments, the system interacts with an external hybrid energy peak compensation system via a standard communication bus, synchronously transmitting at least one timing parameter in the execution schedule to the external system so that the external system can perform pre-adjustment of the state of charge of the energy storage unit in advance.

[0023] By adopting the above technical solution, compared with the prior art, the present invention has the following advantages: (1) The present invention maps the scene recognition results into a control scheme containing multi-dimensional timing parameters such as voltage change acceleration curve and intervention envelope curve by executing a time schedule, thereby achieving optimal timing matching between power supply scheduling and dynamic audio demand.

[0024] (2) The voltage change acceleration of the present invention is dynamically configured according to the prediction results: a high acceleration (not exceeding 5% / s²) is used under high dynamic change trend so that the bus voltage can quickly approach the target value within a limited time; a low acceleration (not exceeding 1% / s²) is used under low dynamic change trend so as to complete the adjustment in the smoothest way. The variable acceleration control realizes the differentiated matching between the voltage adjustment process and the urgency of the scenario.

[0025] (3) The present invention uses an S-shaped envelope curve to control the time trajectory of the intervention depth. The rate of change of the initial and final segments is small, while the rate of change of the middle segment is large, so that the second derivative of the intervention depth is continuous and the sudden acceleration is avoided. The S-shaped intervention envelope curve eliminates the electrical shock of the auxiliary bus cut-in.

[0026] (4) By synchronously adjusting the static bias compensation coefficient, the present invention makes the power amplifier as close as possible to the minimum distortion operating point at every moment of voltage change, and the synchronous update of the bias compensation coefficient eliminates the distortion degradation during the transition period.

[0027] After adopting the solution of this invention, during the 15-second transition from 80% rated voltage to 110% rated voltage, THD+N remains below 0.05%. Compared with the traditional fixed-rate solution, the bus voltage arrival rate in high dynamic scenarios is increased from less than 60% to more than 95%. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the system module composition and information flow of the present invention; Figure 2 This is a schematic diagram of the execution schedule generation process of the present invention; Figure 3 This is a schematic diagram comparing acceleration curves under high dynamic prediction and low dynamic prediction in this invention; Figure 4 This is a schematic diagram comparing the S-shaped interventional envelope curve of the present invention with the traditional linear interventional curve; Figure 5 This is a timing diagram of the step-down node of the charging power in this invention; Figure 6 This is a schematic diagram illustrating the principle of synchronous update of the bias compensation coefficient in this invention. Figure 7 This is a comparison chart of the bus voltage response timing of the present invention and the traditional solution in a high dynamic scenario. Detailed Implementation

[0029] The present invention will be further described in detail below with reference to the embodiments. The embodiments are only intended to provide a clearer understanding of the technical features, objectives and effects of the present invention.

[0030] This application provides an adaptive control method for a power amplifier power supply system based on audio scene awareness, including the following steps: Real-time acquisition of audio input signals, and extraction of audio feature parameters to characterize the dynamic characteristics and energy distribution of audio content; Based on the extracted audio feature parameters, the current audio content is classified into a scene and identified as one of several preset audio scene types; Based on the changing trend of audio feature parameters within a preset time window, predict the dynamic changing trend of audio content within a preset duration in the future. The dynamic changing trend includes at least a high dynamic changing trend and a low dynamic changing trend. Based on the identified scenario type and the predicted dynamic change trend, an execution schedule for the power amplifier power supply system is generated. The execution schedule includes at least the following: the total transition time for the main power supply bus voltage of the power amplifier to reach the target value from the current value, the acceleration curve of the voltage change acceleration at each moment within the total transition time as a function of time, the intervention envelope curve of the auxiliary DC bus intervention depth as a function of time, and the step-down nodes of the energy storage unit charging power as a function of time. According to the execution schedule, the bus voltage is gradually adjusted from the current value to the target value using an acceleration curve, and the auxiliary DC bus is smoothly intervened and de-intercepted using an intervention envelope curve. Among them, the acceleration of voltage change is higher when the predicted dynamic change trend is higher than that when the predicted dynamic change trend is lower, and a positive acceleration value indicates that the bus voltage is increasing, while a negative acceleration value indicates that the bus voltage is decreasing.

[0031] In some embodiments, the audio characteristic parameters include at least dynamic range, low-frequency energy proportion, and short-time power change rate; the dynamic range is the difference between the maximum and minimum loudness of the audio signal within a preset time window, used to distinguish between high-dynamic content and low-dynamic content; the low-frequency energy proportion is the proportion of energy below a preset frequency band to the total energy, used to identify content containing bass components; the calculation window for the short-time power change rate does not exceed 10 milliseconds, used to capture the transient impact characteristics of the audio signal.

[0032] In some embodiments, trend prediction is determined based on the slope of the short-time power change rate within a preset window and the cumulative change in the proportion of low-frequency energy within the preset window: When the slope of the short-term power change rate exceeds the first slope threshold and the duration exceeds the preset duration, it is predicted to be a high dynamic change trend. When the slope of the short-term power change rate is lower than the second slope threshold, it is predicted to be a low dynamic change trend; the first slope threshold is greater than the second slope threshold.

[0033] In some embodiments, the acceleration curves show: When the predicted trend is one of high dynamic change, the absolute value of the voltage change acceleration is set as the first acceleration value; When the predicted trend is low dynamic change, the absolute value of the voltage change acceleration is configured as the second acceleration value; the first acceleration value is greater than the second acceleration value.

[0034] The first acceleration value shall not exceed 5% of the rated bus voltage value per square second, and the second acceleration value shall not exceed 1% of the rated bus voltage value per square second.

[0035] In some embodiments, the intervention envelope curve is an S-shaped curve, wherein the slope of the initial and final segments of the S-shaped curve is less than the slope of the middle segment, and the time constant of the S-shaped curve is configured to be 200 milliseconds to 800 milliseconds depending on the scenario type.

[0036] In some embodiments, the step-down node includes at least: the charging power is reduced to a first preset power ratio at a first preset time after the scene switching is triggered, the charging power is reduced to a second preset power ratio at a second preset time, and the charging power is reduced to a third preset power ratio at a third preset time, and the charging power change rate between each adjacent node does not exceed 20% of the rated charging power per second.

[0037] In some embodiments, during the gradual adjustment of the bus voltage, the static bias compensation coefficient of the power amplifier stage is adjusted synchronously. The static bias compensation coefficient is obtained by looking up a table through a pre-established bus voltage-power amplifier nonlinear characteristic mapping model. The update period of the static bias compensation coefficient does not exceed the duration corresponding to 1% of the bus voltage change caused by the voltage change acceleration.

[0038] The present invention also provides an adaptive control system for a power amplifier power supply system based on audio scene perception to implement the above method, comprising: The audio scene perception and prediction module is used to extract audio feature parameters, classify scenes, and predict dynamic change trends. The execution schedule generation module is used to generate an execution schedule based on scenario classification results and trend prediction results; The bus voltage variable acceleration control module is used to perform controlled gradual adjustment of the bus voltage according to the acceleration curve in the execution schedule; The auxiliary bus envelope control module is used to perform smooth intervention and withdrawal of the auxiliary DC bus according to the intervention envelope curve; The bias synchronization compensation module is used to synchronously adjust the static bias compensation coefficient of the power amplifier stage during the gradual adjustment of the bus voltage.

[0039] The system interacts with an external hybrid energy peak compensation system via a standard communication bus, synchronously transmitting at least one timing parameter from the execution schedule to the external hybrid energy peak compensation system. The external system then performs pre-adjustment of the energy storage unit's state of charge based on the received timing parameter.

[0040] Example 1: Audio feature parameter extraction and scene classification. This example describes in detail the specific implementation of steps S1 and S2.

[0041] The audio input signal is digitized by an analog-to-digital converter at a sampling rate of at least 48kHz and a bit depth of 24 bits. The system processes the audio signal using a sliding time window method.

[0042] The specific calculation methods for each feature parameter are as follows: Dynamic Range (DR): Calculates the difference between the maximum and minimum values ​​of the equivalent sound level (Leq) of the audio signal within a 500-millisecond window. In practice, to prevent a single outlier sampling point from affecting the results, a robust estimate of DR is obtained by taking the difference between the 95th percentile and the 5th percentile after sorting. A larger DR value indicates more dramatic dynamic fluctuations in the audio signal.

[0043] Low-Frequency Energy Ratio (LFR): Calculate the proportion of total energy in the 20Hz to 120Hz frequency band to the total energy of the entire frequency band (20Hz to 20kHz) after applying a Hanning window to the audio signal within a 500ms window and performing a Fast Fourier Transform (FFT) with at least 2048 points. The LFR value reflects the richness of the bass response in the audio content.

[0044] Short-Time Power Change Rate (STPCR): Using a 10-millisecond short-time window (corresponding to 480 sampling points at a 48kHz sampling rate), the root mean square power of the audio signal within the window is calculated. The absolute value of the power difference between adjacent short-time windows is taken, and the arithmetic mean of this difference is calculated within a 500-millisecond window. The 10-millisecond calculation window for STPCR is determined based on the minimum perceptible temporal resolution of the human ear for transient responses; an excessively large window would smooth out crucial transient information.

[0045] The scenario classification adopts a two-level decision architecture: the first-level decision is executed based on a preset threshold rule with a delay of no more than 10 milliseconds; the second-level decision is based on the minimum duration verification (configurable from 500 milliseconds to 2000 milliseconds) to prevent frequent switching.

[0046] Example 2: Trend Prediction Model This embodiment elaborates on the specific implementation of dynamic trend prediction in step S3.

[0047] The trend prediction model is based on the observation that changes in the dynamic characteristics of audio content are not instantaneous abrupt changes, but rather exhibit predictable evolutionary trends over a timescale of several seconds. For example, before the chorus of pop music, there is usually a short-term "energy accumulation" process with a gradually increasing rate of power change, typically lasting between 1 and 3 seconds.

[0048] Trend Feature Extraction: The system maintains a 2-second circular buffer to store the STPCR values ​​calculated frame by frame. At the arrival of each new audio frame, the slope of STPCR change (in W / ms²) within the most recent 500 milliseconds window and the cumulative change in LFR within the most recent 1-second window are calculated.

[0049] Trend Judgment Logic: When the STPCR slope > 3.0 W / ms² and persists for more than 300 milliseconds, it is predicted as a "high dynamic range," indicating that the audio content is about to enter a period of dense transient impact. This threshold was determined through statistical analysis of a large number of music and sound effect samples: the STPCR slope of typical speech playback is usually below 1.0 W / ms², the chorus of pop music is usually between 2.5 and 5.0 W / ms², and explosive sound effects can reach above 8.0 W / ms². Using 3.0 W / ms² as the threshold can ensure prediction sensitivity while avoiding misjudgment of ordinary music segments.

[0050] When the STPCR slope is less than 1.0 W / ms² and the cumulative change in LFR is less than 8%, the prediction is "low dynamic trend," indicating that the audio content will remain stable. This threshold ensures that the system does not trigger unnecessary pre-actions in voice broadcasting and background music scenarios.

[0051] When the STPCR slope is between 1.0 W / ms² and 3.0 W / ms², the current prediction state is maintained to avoid frequent jumps in the prediction results.

[0052] Example 3: Execution Schedule Generation This embodiment describes in detail the method for generating the execution schedule in step S4.

[0053] The execution schedule is one of the core innovations of this invention. It transforms the results of scene recognition and trend prediction into a complete control scheme containing multiple time-series parameters. The generation process of the execution schedule is as follows: Step S4-1: Determine the target bus voltage value. Based on the scene classification results, look up the target voltage value in the scene-voltage mapping table. Typical mappings are: voice broadcast 70%, background music 80%, pop music 100%, classical music 100% (transient can be increased to 105%), movie sound effects 105%, and live sound reinforcement 110%.

[0054] Step S4-2: Determine the total transition time. The total transition time T_total is determined by both the voltage change ΔV and the prediction result: When the prediction is "high dynamic change trend", T_total = max(10 seconds, |ΔV| / (first acceleration value × 2)). The physical meaning of dividing by 2 × acceleration is: in uniformly accelerated motion, starting from zero initial velocity, the relationship between displacement and acceleration is d = ½·a·t², therefore the time required to reach the target displacement is t = sqrt(2d / a). Setting the lower limit of max to 10 seconds ensures sufficient transition time even with small voltage changes to avoid excessive acceleration.

[0055] When the prediction is "low dynamic change trend", T_total = max(20 seconds, |ΔV| / (second acceleration value × 2)).

[0056] Step S4-3: Generate the acceleration curve a(t). The acceleration curve adopts a three-segment planning of "acceleration-uniform speed-deceleration" within the total transition time: Taking the increase in bus voltage as an example (target value is greater than current value): First segment (acceleration segment): Starting from t=0, the acceleration is +A (A is the first acceleration value or the second acceleration value), lasting for T_acc=T_total / 4.

[0057] Second segment (uniform speed segment): acceleration is 0, lasting T_const = T_total / 2.

[0058] The third stage (deceleration stage): the acceleration is -A, and the duration is T_dec = T_total / 4.

[0059] The voltage reaches the target value at time T_total, and the rate of change at that time is 0 (smooth arrival).

[0060] When the bus voltage decreases (the target value is less than the current value), the acceleration curve is a mirror image of the above curve (the acceleration sign is reversed).

[0061] The technical significance of the three-stage acceleration programming lies in the following: the initial acceleration stage allows the voltage to quickly move away from its initial value; the uniform velocity stage in the middle stage maintains a stable trend of change; and the deceleration stage in the final stage allows the voltage to smoothly approach the target value without overshoot or shock. Compared to a simple uniform velocity or first-order response, the three-stage programming achieves an optimal balance between response speed and transition smoothness.

[0062] The first acceleration value (high dynamic prediction) does not exceed 5% per square second of the rated value, and the second acceleration value (low dynamic prediction) does not exceed 1% per square second of the rated value. These upper limits are determined based on the charging current surge limit of the internal energy storage filter capacitor and the stability analysis of the voltage regulation loop.

[0063] Step S4-4: Generate the intervention envelope curve E(t). The trajectory of the auxiliary bus intervention depth from 0% to 100% adopts an S-shaped curve: E(t) = 1 / (1 + e^{-k·(t-t_mid)}) Where t_mid is the midpoint of the S-curve (configured at the middle of the intervention process), k is a parameter controlling the steepness of the curve, and the time constant τ = 1 / k. By adjusting k, the rise time of the S-curve from 10% to 90% can be configured between 200 milliseconds and 800 milliseconds.

[0064] The advantage of the S-curve over the linear curve is that the slope of the initial and final segments is smaller, which makes the rate of change of the intervention depth (first derivative) continuous, and there is no abrupt slope change at the starting and ending points of the linear curve; at the same time, the rate of change of the intervention depth (second derivative) is also continuous, ensuring the smoothness of the rate of change of current.

[0065] Step S4-5: Generate charging power step-down nodes. When scene switching requires reducing or pausing charging (such as entering on-site sound reinforcement mode), the charging power gradually decreases according to the following step nodes:

[0066] The charging power change rates between adjacent nodes are 20% / second, 30% / second, 30% / second, and 20% / second, respectively, all of which do not exceed the safety limit of 20% per second of the rated charging power. The technical significance of the step-like rather than non-linear descent is that it provides a stable reference node for the charging control circuit, avoiding tracking errors that may be introduced by the control loop tracking continuously changing signals.

[0067] Example 4: Variable Acceleration Bus Voltage Regulation This embodiment details the specific implementation of bus voltage acceleration control in step S5.

[0068] The bus voltage regulation module drives the voltage setpoint of the switching power supply according to the acceleration curve a(t) in the execution schedule. Within each control cycle (100 microseconds), the module calculates the target voltage for the next moment based on its current position on the acceleration curve. V(t+Δt)=V(t)+v(t)·Δt+½·a(t)·Δt² v(t+Δt)=v(t)+a(t·Δt Where v(t) is the current voltage change rate, and the initial state is v(0) = 0. The actual voltage tracks the target voltage through the voltage control loop of the switching power supply, and the tracking error is controlled within ±0.5% of the rated value.

[0069] A typical operational case under high dynamic prediction: Assuming the bus voltage needs to increase from 80% to 110% (ΔV = +30%), the prediction shows a high dynamic change trend with an acceleration A = 5% / s². Under a three-stage planning approach: acceleration phase 0-3.75s (velocity increases from 0 to 18.75% / s, displacement 35.16%), constant velocity phase 3.75-11.25s (velocity remains constant at 18.75% / s, displacement 140.63%), and deceleration phase 11.25-15s (velocity decreases from 18.75% / s to 0, displacement 35.16%). A 30% voltage change is completed within a total transition time of 15 seconds. The voltage exceeds 100% in approximately 10 seconds, meaning it is essentially in place before the high dynamic peak arrives.

[0070] Comparative case under low dynamic prediction: For the same voltage change (30%), predicting a low dynamic trend, with acceleration A = 1% / s², the total transition time T_total = sqrt(2×30 / 1)×2 ≈ 30.98 seconds. The voltage change is smooth throughout, minimizing the impact of THD+N. The comparison shows that, for the same voltage change, the choice of acceleration directly affects the total transition time and voltage trajectory shape, achieving a match between voltage adjustment speed and the dynamic requirements of the scenario.

[0071] Example 5: S-shaped interventional envelope control This embodiment describes in detail the S-shaped envelope control with the assistance of a DC bus.

[0072] The auxiliary bus arbitration module controls the intervention depth of the auxiliary DC bus according to the intervention envelope curve E(t) in the execution schedule. The intervention depth is achieved by adjusting the duty cycle of the auxiliary power supply's power switch transistor via PWM, and the duty cycle setting is updated every 100 microseconds.

[0073] For an S-shaped curve with a time constant τ = 500 ms, the rise time from 10% to 90% is approximately 2.2 × τ = 1.1 seconds. During the entire switching process from disabled (0%) to fully intervened (100%), the rate of change of intervention depth gradually increases from 0 to a peak and then gradually decreases back to 0, while the rate of change of current (di / dt) remains continuous.

[0074] Comparative experiments with the traditional linear intervention scheme (linearly increasing from 0 to 100% within 1 second) show that the linear scheme experiences a step change of di / dt of approximately 100% / second at the initial moment, generating a voltage spike of about 50mV on the DC bus side; while the S-shaped scheme has a di / dt of 0 at the initial moment, gradually increasing to the peak value and then gradually decreasing, resulting in a bus voltage spike of less than 10mV. In conducted emission tests, the peak conducted interference of the S-shaped scheme is reduced by approximately 15dBμV compared to the linear scheme.

[0075] Example 6: Synchronous Update of Offset Compensation Coefficients This embodiment details the synchronous update mechanism of the static bias compensation coefficient of the power amplifier stage.

[0076] During changes in the bus voltage, the quiescent operating point of the power amplifier stage output transistor drifts accordingly. Specifically, when the bus voltage increases, the Vce of the output transistor increases, and under the same quiescent bias current, the junction temperature of the transistor rises faster, changing its transconductance characteristics. When the bus voltage decreases, the linear operating region of the output transistor shrinks, making it easier to enter the saturation region under large signals. If these drifts are not compensated, they will lead to a significant increase in the THD+N of the power amplifier during the transition period.

[0077] This invention achieves compensation through the following mechanism: Mapping model establishment: During the system factory calibration phase, the THD+N performance of the power amplifier output stage under different static bias currents was measured for different bus voltage values ​​(with a step size of 5% of the rated value, a total of 9 calibration points from 70% to 110%). At each voltage point, the optimal static bias current I_bias_opt that minimizes THD+N was recorded, and the compensation coefficient K_bias = I_bias_opt / I_bias_default (default bias current) was calculated.

[0078] The above measurement results are stored in non-volatile memory in the form of a lookup table. The inputs to the lookup table are the bus voltage V_bus and the current output power P_out (estimated by sampling the output current), and the output is K_bias.

[0079] Synchronous update execution: During the bus voltage gradual adjustment, the bias synchronization compensation module performs the following operations in each control cycle (10 milliseconds): (1) Read the current bus voltage value V_bus_current and the predicted next-time voltage value V_bus_next from the bus voltage control module; (2) Using V_bus_current and the current output power P_out as indices, perform bilinear interpolation in the lookup table to obtain the compensation coefficient K_bias_current at the current time. (3) Using V_bus_next and P_out as indices, obtain the prediction compensation coefficient K_bias_next for the next time step; (4) Gradually change the static bias current of the power amplifier stage from I_bias_current = I_bias_default × K_bias_current to I_bias_next = I_bias_default × K_bias_next, and match the rate of change of the bus voltage.

[0080] The update period of the compensation coefficient, 10 milliseconds, is much smaller than the time constant (seconds) of the bus voltage change. Therefore, the change of bias current is continuous and smooth for the power amplifier stage.

[0081] Experimental data: During the complete transition period of the bus voltage rising from 80% to 110% (lasting 15 seconds), with synchronous bias compensation, THD+N remained between 0.04% and 0.05%. In contrast, when synchronous compensation was turned off, THD+N increased to 0.15% to 0.20% during the middle segment of the same transition period, when the voltage change rate was the highest.

[0082] Example 7: Complete System Integration and Operation Process Taking a multi-functional venue as an example, the venue is equipped with a power amplifier system with a rated power of 500 watts and a matching 24V auxiliary DC bus energy storage unit.

[0083] Scene switching trigger: The system detects that the audio features are changing from background music mode to live sound reinforcement mode, and the trend prediction model outputs "high dynamic change trend".

[0084] Execution schedule generation: Scenario-strategy mapping determines the target value of the bus voltage to be 110% (current value is 80%, ΔV = +30%). The prediction is high dynamics, with acceleration A = 5% / s² and total transition time T_total = 15 seconds. The acceleration curve adopts a three-stage structure of acceleration-constant speed-deceleration. The intervention envelope curve time constant τ = 400ms. The charging power decreases in steps at 1-second, 2-second, 3-second, and 5-second intervals.

[0085] Variable acceleration control: The bus voltage quickly drops below 80% during the acceleration phase, rises steadily during the constant speed phase, and smoothly approaches 110% during the deceleration phase. The voltage reaches 105% in about 10 seconds, meeting the early power supply requirements of high dynamic scenarios, and accurately reaches 110% in 15 seconds.

[0086] S-type intervention: The auxiliary bus intervention depth smoothly transitions from 0% to 100% in about 1.8 seconds, the bus voltage spike is less than 10mV, and there is no overcurrent trigger.

[0087] Bias synchronization compensation: During the entire 15-second transition period, the static bias compensation coefficient smoothly transitions from the calibration value corresponding to 80% voltage to the calibration value corresponding to 110% voltage, and THD+N remains below 0.05%.

[0088] The above embodiments fully illustrate the feasibility and technical advantages of the present invention. Those skilled in the art can make various modifications and variations based on the principles of this invention, and all such modifications should be included within the scope of protection of this invention.

Claims

1. An adaptive control method for an audio scene-aware power amplifier power supply system, characterized in that, Includes the following steps: Real-time acquisition of audio input signals, and extraction of audio feature parameters to characterize the dynamic characteristics and energy distribution of audio content; Based on the extracted audio feature parameters, the current audio content is classified into a scene and identified as one of multiple preset audio scene types; Based on the changing trend of the audio feature parameters within a preset time window, predict the dynamic changing trend of the audio content within a preset duration in the future. The dynamic changing trend includes at least a high dynamic changing trend and a low dynamic changing trend. Based on the identified scenario type and the predicted dynamic change trend, an execution schedule for the power amplifier power supply system is generated. The execution schedule includes at least: the total transition time for the main power supply bus voltage of the power amplifier to reach the target value from the current value, the acceleration curve of the voltage change acceleration at each moment within the total transition time, the intervention envelope curve of the auxiliary DC bus intervention depth changing with time, and the step-down nodes of the energy storage unit charging power changing with time. According to the execution schedule, the bus voltage is gradually adjusted from the current value to the target value using the acceleration curve, and the auxiliary DC bus is smoothly intervened and de-intercepted using the intervention envelope curve. The voltage change acceleration is higher when the predicted dynamic change trend is higher than when the predicted dynamic change trend is lower, and a positive acceleration indicates an increase in bus voltage, while a negative acceleration indicates a decrease in bus voltage.

2. The method of claim 1, wherein, The audio characteristic parameters include at least dynamic range, low-frequency energy proportion, and short-time power change rate; the dynamic range is the difference between the maximum and minimum loudness of the audio signal within a preset time window, used to distinguish between high-dynamic content and low-dynamic content; the low-frequency energy proportion is the proportion of energy below a preset frequency band to the total energy, used to identify content containing bass components; the calculation window for the short-time power change rate does not exceed 10 milliseconds, used to capture the transient impact characteristics of the audio signal.

3. The method of claim 1, wherein, Trend prediction is also determined based on the slope of the short-term power change rate within a preset window and the cumulative change in the proportion of low-frequency energy within the preset window: When the slope of the short-term power change rate exceeds the first slope threshold and the duration exceeds the preset duration, it is predicted to be a high dynamic change trend. When the slope of the short-term power change rate is lower than the second slope threshold, it is predicted to be a low dynamic change trend; the first slope threshold is greater than the second slope threshold.

4. The method of claim 1, wherein, In the acceleration curve: When the predicted trend is one of high dynamic change, the absolute value of the voltage change acceleration is configured as the first acceleration value; When the predicted trend is low dynamic change, the absolute value of the voltage change acceleration is configured as the second acceleration value; The first acceleration value is greater than the second acceleration value.

5. The method of claim 4, wherein, The first acceleration value does not exceed 5% of the rated bus voltage value per square second, and the second acceleration value does not exceed 1% of the rated bus voltage value per square second.

6. The method of claim 1, wherein, The intervention envelope curve is an S-shaped curve, with the slope of the starting and ending segments of the S-shaped curve being less than the slope of the middle segment. The time constant of the S-shaped curve is configured to be between 200 milliseconds and 800 milliseconds depending on the scenario type.

7. The method of claim 1, wherein, The stepped descent nodes include at least the following: the charging power drops to a first preset power ratio at a first preset time after the scene switching is triggered, the charging power drops to a second preset power ratio at a second preset time, and the charging power drops to a third preset power ratio at a third preset time, with the charging power change rate between each adjacent node not exceeding 20% ​​of the rated charging power per second.

8. The method of claim 1, wherein, During the gradual adjustment of the bus voltage, the static bias compensation coefficient of the power amplifier stage is adjusted synchronously. The static bias compensation coefficient is obtained by looking up a table through a pre-established bus voltage-power amplifier nonlinear characteristic mapping model. The update period of the static bias compensation coefficient does not exceed the duration corresponding to 1% of the bus voltage change caused by the voltage change acceleration.

9. An adaptive control system for a power amplifier power supply system based on audio scene perception, implementing the method of any one of claims 1-8, characterized in that, include: The audio scene perception and prediction module is used to perform the extraction of the audio feature parameters, the scene classification, and the prediction of the dynamic change trend; The execution schedule generation module is used to generate the execution schedule based on the scenario classification results and trend prediction results; The bus voltage variable acceleration control module is used to perform controlled gradual adjustment of the bus voltage according to the acceleration curve in the execution schedule; The auxiliary bus envelope control module is used to perform smooth intervention and withdrawal of the auxiliary DC bus according to the intervention envelope curve; The bias synchronization compensation module is used to synchronously adjust the static bias compensation coefficient of the power amplifier stage during the gradual adjustment of the bus voltage.

10. The system according to claim 9, characterized in that, The system interacts with an external hybrid energy peak compensation system via a standard communication bus, synchronously transmitting at least one timing parameter from the execution schedule to the external hybrid energy peak compensation system. The external system then pre-adjusts the state of charge of the energy storage unit based on the received timing parameters.

Images