Modular digital audio multi-channel IO control system and method for aeronautical communications

Abstract

The application belongs to the technical field of data processing, and relates to a modular digital audio multi-channel IO control system and method for aviation communication, aiming to solve the problems of poor environmental adaptability, static synchronization parameters, calibration lag and insufficient intelligent fault tolerance of synchronization technology in the aviation field. The application comprises: a scene perception module fusing multi-domain data to generate a scene vector, generating synchronization parameters through a forward control module, calibrating clock deviation of multiple IO channels through a synchronous execution module, and realizing synchronization control; a health monitoring module quantifying the health degree of the whole link, and dynamically adapting a self-healing strategy through a self-healing learning module. The application balances between high-precision synchronization and high-robustness fault tolerance through closed-loop cooperation of forward-looking prediction and responsive self-healing, and endows the system with intelligent adaptive evolution capability, thereby improving the synchronization performance and task reliability of the aviation audio system in a complex dynamic environment.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of data processing, and particularly relates to a modular digital audio multi-channel IO control system and method for aviation communication. BACKGROUND

[0002] In modern avionics systems, audio input / output systems play a crucial role, carrying key tasks such as cockpit intercom, air-ground communication, and warning prompts. These systems usually contain multiple distributed input / output nodes, which must maintain microsecond-level or even higher precision time synchronization between them to ensure the integrity, real-time performance, and clarity of audio streams.

[0003] Existing synchronization technologies, such as Phase-Locked Loop (PLL), Kalman filtering, or Network Time Protocol (NTP) schemes, can provide good synchronization results under stable and ideal working conditions. However, the aviation environment has its inherent complexity and harshness:

[0004] Dynamic environment: changes in flight status (such as takeoff, cruising, and landing) and electromagnetic environment can unpredictably impact the quality of communication links (latency, jitter, and packet loss rate), leading to synchronization parameter mismatch and performance degradation;

[0005] Limitations of passive response: traditional methods are mostly "post-correction" mode, i.e., compensation is performed only after detecting synchronization errors; this passive response mechanism has inherent delays and cannot foresee and avoid synchronization jitter caused by predictable scenario changes (such as entering a strong electromagnetic interference zone);

[0006] Insufficient fault tolerance: in the face of link interruptions, hardware failures, and other sudden and structural problems, traditional systems rely on preset and fixed redundancy switching logic. This logic lacks intelligence and cannot make optimal recovery decisions based on real-time fault context, resulting in long recovery time and reduced system reliability.

[0007] Therefore, there is an urgent need for a synchronization control technology that can actively perceive complex aviation environments, predictively adjust synchronization strategies, and intelligently and quickly recover from faults, to meet the higher standards of future aviation systems for audio communication. SUMMARY

[0008] To solve the above problems in the prior art and address the poor environmental adaptability, static synchronization parameters, calibration lag, and insufficient intelligent fault tolerance in aviation audio synchronization technology, the present application proposes, in a first aspect, a modular digital audio multi-channel IO control system for aviation communication, which includes:

[0009] The scene perception module, the forward control module, the synchronous execution module, the health monitoring module and the self-healing learning module are connected through a bus communication to form a closed-loop control architecture.

[0010] The scene perception module collects multi-domain heterogeneous data and performs dynamic weight fusion processing, and outputs a scene vector representing the current flight scene, wherein the dynamic weight is adaptively adjusted according to the electromagnetic interference intensity.

[0011] The forward control module receives the scene vector as input, and a pre-trained neural network inference model generates core synchronization parameters adapted to the current scene.

[0012] The synchronous execution module generates a reference clock according to the core synchronization parameters, adopts a prediction compensation mechanism to calibrate the clock deviation of the digital audio multi-IO channel, and realizes multi-channel synchronous control.

[0013] The health monitoring module collects the full-link running state, evaluates the system health degree and outputs the evaluation result.

[0014] The self-healing learning module dynamically selects a self-healing strategy according to the evaluation result and the scene vector, feeds back the self-healing result to the forward control module, and regenerates the adapted parameters based on the changed state to realize the control strategy closed loop.

[0015] In some preferred embodiments, the scene perception module comprises:

[0016] An interface adaptation unit configured with interfaces connected to a flight management system, a traffic warning and collision avoidance system and an engine control system respectively, to collect three-dimensional route data, multi-target threat data and engine real-time state parameters respectively.

[0017] A dynamic weight fusion unit with an electromagnetic interference detection subunit, which adjusts the weight proportion of each domain data according to the electromagnetic interference intensity, specifically:

[0018] When the electromagnetic interference intensity is higher than -80dBm, the weight proportion of the engine real-time state parameter is increased; when the electromagnetic interference intensity is lower than -100dBm, the weight proportion of the multi-target threat data is increased.

[0019] A scene vector generation unit for quantizing the fused data into a four-dimensional scene vector containing flight phase, threat density, engine state and electromagnetic interference intensity.

[0020] In some preferred embodiments, the forward control module comprises:

[0021] The neural network processing unit is configured to receive four-dimensional parameters of a scene vector, extract scene depth features through a ReLU activation function, and output core synchronization parameters from a layer output.

[0022] The parameter cache unit is configured to store the generated core synchronization parameters, including a synchronization accuracy threshold, a sampling rate, and a PI controller adjustment coefficient, and a parameter update delay is less than or equal to 2 ms.

[0023] The parameter feedback adjustment unit is in communication connection with the self-healing learning module, configured to receive optimization instructions and dynamically correct output parameters of the neural network inference model.

[0024] In some preferred embodiments, the synchronization execution module comprises:

[0025] The reference clock generation unit is configured to generate a reference clock signal according to the sampling rate in the core synchronization parameters.

[0026] The prediction compensation calibration unit comprises a bias prediction subunit and a compensation execution subunit, configured to analyze historical clock bias data through the bias prediction subunit to calculate a bias change trend, and generate a compensation amount according to the predicted bias through the compensation execution subunit to realize advance calibration of the clock bias.

[0027] The calibration period adjustment unit is configured to dynamically adjust a calibration period according to the scene vector, and the calibration period in an emergency scenario is one fifth of a regular period, and the calibration period in a cruising scenario is twice the regular period.

[0028] In some preferred embodiments, the method for realizing advance calibration of the clock bias comprises:

[0029] The timestamps of local clocks of each IO channel are collected, and a reference timestamp corresponding to the reference clock signal is taken as a reference to calculate current clock biases of each channel.

[0030] The historical bias data of each channel is analyzed by using a sliding window method, and a current bias change rate is calculated in combination with a sampling period.

[0031] A compensation value is generated in combination with the current clock bias of each channel, the bias change rate, and a prediction coefficient corresponding to the scene; and the prediction coefficient is dynamically adjusted according to a flight phase.

[0032] The compensation value is written into a clock controller of each IO channel through a software interrupt, the frequency of the local clock is adjusted, and calibration is completed; the calibration period is consistent with the sampling period, and the clock bias after calibration does not exceed the synchronization accuracy threshold.

[0033] In some preferred embodiments, the synchronous execution module further comprises a link dynamic configuration unit, which is in communication connection with the prospective control module and dynamically adjusts the link working mode according to the synchronization precision threshold in the core synchronization parameter, and the method is as follows:

[0034] When the synchronization precision threshold is lower than the preset precision threshold, a low-delay mode is enabled, the link cache depth is reduced, and the CPU resource allocation proportion is increased;

[0035] When the synchronization precision threshold is higher than the preset precision threshold, an energy efficiency mode is enabled, the link cache depth is adjusted, and a lightweight transmission coding is adopted;

[0036] The link dynamic configuration unit comprises a priority determination subunit, which dynamically adjusts the link priority according to the threat collision time in the multi-target threat data, and sets the link priority to the highest level in an emergency scenario.

[0037] In some preferred embodiments, the link dynamic configuration unit further comprises a spatial audio enhancement subunit, which is used to calculate the left-right channel gain difference through a head-related transfer function when the link is at the highest level, and adds delay compensation according to the threat azimuth angle to ensure that the pilot's identification error of the threat azimuth does not exceed the error threshold.

[0038] In some preferred embodiments, the health monitoring module comprises:

[0039] A state acquisition unit is configured to acquire three types of indexes, including synchronization stability, link smoothness and data integrity, wherein the synchronization stability is the time proportion of the clock deviation meeting the synchronization precision threshold, the link smoothness is the time proportion of the link delay being less than or equal to the delay threshold, and the data integrity is 1 minus the data packet loss rate.

[0040] A health degree calculation unit is configured to calculate the system health degree through a weighted algorithm based on the synchronization stability, the link smoothness and the data integrity.

[0041] A threshold dynamic adjustment unit is configured to correct the health degree threshold according to the historical fault times to realize adaptive optimization of the self-healing triggering condition.

[0042] In some preferred embodiments, the self-healing learning module comprises:

[0043] A self-healing strategy library stores three types of self-healing strategies, including a non-key link sacrifice strategy, an anti-interference coding enhancement strategy and a minimum feature set start strategy, and each self-healing strategy is in a mapping relationship with a scene vector.

[0044] A strategy selection unit is configured to call the adaptive self-healing strategy according to the health degree evaluation result and the current scene vector, and call the anti-interference coding enhancement strategy in a strong electromagnetic interference scene and call the non-key link release strategy in a multi-threat scene.

[0045] A Q-learning unit is configured to update the mapping relationship between the scene and the strategy according to the self-recovery result.

[0046] The second aspect of the application provides a modular digital audio multi-channel IO control method for aviation communication, which comprises the following steps:

[0047] Obtain and fuse multi-domain heterogeneous data to construct a scene vector for quantifying a real-time aviation scene; the scene vector comprises a flight phase, a threat density, an engine state and an electromagnetic interference intensity;

[0048] Based on the scene vector, a pre-trained neural network inference model is used to infer and generate core synchronization parameters matched with the current scene; the core synchronization parameters at least comprise a synchronization accuracy threshold for calibrating clock bias;

[0049] According to the core synchronization parameters, clock calibration and link configuration are performed on multiple IO channels of aviation audio to realize synchronization control between channels;

[0050] The running state of the synchronization control is monitored to evaluate the system health, and the neural network inference model is feedback optimized based on the evaluation result of the system health.

[0051] The application has the following beneficial effects:

[0052] 1. Through dynamic weight fusion of three types of core domain data, the forward control module performs fine prediction optimization, and the self-recovery learning module performs rapid and robust recovery in an abnormal state, realizing closed-loop cooperation of forward-looking prediction and responsive self-recovery, changing passive error correction to active prediction, thereby solving the contradiction between accuracy and robustness of the traditional system, and endowing the system with intelligent self-recovery and adaptive evolution ability, and realizing significant improvement of the synchronization performance and task reliability of the aviation audio system in a complex dynamic environment;

[0053] 2. The system can predict the potential influence of environmental changes on synchronization performance by using neural networks to deeply understand multi-domain information, and can adjust the cache strategy and switch the best communication link in advance to kill the synchronization jitter in the bud, change passive error correction to active prediction, realize more smooth and more forward-looking control than the traditional feedback system, and significantly improve the audio communication quality;

[0054] 3. When the system is healthy, the forward-looking module is responsible for fine and forward-looking optimization to ensure that the synchronization accuracy reaches the theoretical upper limit; when facing serious faults such as link interruption, the self-recovery module based on reinforcement learning can break free from the shackles of fixed rules, make dynamic optimal decisions according to real-time fault context, greatly shorten the fault recovery time, improve the task reliability of the system under extreme conditions, and endow the system with fast and intelligent autonomous fault tolerance ability;

[0055] 4. Through the state information feedback link, the fault handling experience of the self-healing module can make the forward-looking module adapt to the new normal of the system structure; this makes the system have the ability to learn from faults and continuously evolve, so that the whole system forms an intelligent closed loop that can learn from faults and adapt to the new normal, and is more stable and adaptable. BRIEF DESCRIPTION OF DRAWINGS

[0056] Other features, objects, and advantages of the application will become more apparent from the following detailed description of non-limiting embodiments, when read in conjunction with the accompanying drawings:

[0057] Figure 1 is a structural diagram of a modular digital audio multi-channel IO control system for aviation communication in an embodiment of the application;

[0058] Figure 2 is a flowchart of a modular digital audio multi-channel IO control method for aviation communication in an embodiment of the application. DETAILED DESCRIPTION

[0059] The application will be further described in conjunction with the drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related application, and not to limit the application. In addition, it should be noted that only the parts related to the application are shown in the drawings for ease of description.

[0060] It should be noted that the embodiments and features in the embodiments in the application can be combined with each other without conflict. The application will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0061] The application provides a modular digital audio multi-channel IO control system for aviation communication, which constructs a multi-modal, predictive response adaptive control architecture, through the closed-loop cooperation of forward-looking prediction and responsive self-healing, changes passive error correction to active prediction, balances between high-precision synchronization and high-robustness fault tolerance, and endows the system with intelligent self-healing and adaptive evolution ability, significantly improving the synchronization performance and task reliability of the aviation audio system in a complex dynamic environment.

[0062] In order to more clearly illustrate the application, a modular digital audio multi-channel IO control module for aviation communication will be described below in conjunction with Figure 1 The system in the embodiment of the application will be described in detail, which includes a scene perception module, a forward-looking control module, a synchronization execution module, a health monitoring module and a self-healing learning module, each module is connected through bus communication, forming a closed-loop control architecture;

[0063] The scene perception module collects multi-domain heterogeneous data and performs dynamic weight fusion processing, and outputs a scene vector representing the current flight scene, wherein the dynamic weight is adaptively adjusted according to the electromagnetic interference intensity.

[0064] The forward control module receives the scene vector as input, and a pre-trained neural network inference model generates core synchronization parameters adapted to the current scene.

[0065] The synchronization execution module generates a reference clock according to the core synchronization parameters, adopts a prediction compensation mechanism to calibrate the clock deviation of the digital audio multi-IO channel, and realizes multi-channel synchronization control.

[0066] The health monitoring module collects the full-link operating state, evaluates the system health degree, and outputs the evaluation result.

[0067] The self-healing learning module dynamically selects a self-healing strategy according to the evaluation result and the scene vector, feeds back the self-healing result to the forward control module, and regenerates adapted parameters based on the changed state to realize a control strategy closed loop.

[0068] Preferably, the scene perception module comprises:

[0069] An interface adaptation unit configured with interfaces respectively connected to a flight management system (FMS), a traffic collision avoidance system (TCAS), and an engine control system (ECU) to respectively collect three-dimensional route data, multi-target threat data, and engine real-time state parameters.

[0070] A dynamic weight fusion unit with an electromagnetic interference detection subunit, which adjusts the weight proportion of each domain data according to the electromagnetic interference intensity, specifically:

[0071] When the electromagnetic interference intensity is higher than -80dBm, the weight proportion of the engine real-time state parameters is increased; when the electromagnetic interference intensity is lower than -100dBm, the weight proportion of the multi-target threat data is increased.

[0072] A scene vector generation unit for quantizing the fused data into a four-dimensional scene vector containing flight phase, threat density, engine state, and electromagnetic interference intensity.

[0073] Further preferably, in some embodiments, the interface adaptation unit directly collects the three types of core data through a dedicated interface:

[0074] Flight Management System (FMS) data: three-dimensional route data, including longitude, latitude, altitude, and approach speed; ARINC429 interface, baud rate 1 Mbps;

[0075] Traffic Alert and Collision Avoidance System (TCAS) data: multi-target threat data, including threat azimuth, distance, altitude difference, alert level, and time to collision (TTC); ARINC664p7 interface, rate 100 Mbps;

[0076] Engine Control Unit (ECU) data: real-time state parameters, including speed, exhaust temperature, vibration value, and engine status code; CAN bus interface, rate 500 kbps;

[0077] Environmental data: electromagnetic interference intensity I (collected by an electromagnetic interference detection subunit, sampling frequency 1 MHz).

[0078] Specifically, the collected data needs to be preprocessed, and the method is as follows:

[0079] 1) Outlier rejection: using the criterion to calculate the mean and standard deviation x of the data sequence x 1, x 2,..., n , and rejecting data with μ |>3 σ ; x ᵢ- μ σ

[0080] 2) Time stamp alignment: aligning with the GPS synchronous clock as the reference, alignment accuracy ≤1μs;

[0081] 3) Aviation verification: FMS stage coding needs to be ∈{1 (takeoff), 2 (cruise), 3 (approach), 4 (emergency)}, otherwise trigger interface reconnection; TCAS data needs to satisfy "threat number N≥0 and TTC>0", otherwise marked as invalid data;

[0082] 4) Get preprocessed structured data (e.g. "flight phase = approach, N = 5, I = -75dBm, TTC = 25s").

[0083] Preferably, the dynamic weight fusion unit adjusts the weight proportion of each domain data according to the electromagnetic interference intensity, and the method is as follows:

[0084] Construct a multi-domain data confidence calculation model, and dynamically adjust the weight coefficient according to the electromagnetic interference intensity:

[0085] ;

[0086] wherein, ​The confidence degree (0-1) of FMS, TCAS, and ECU data, respectively; .

[0087] Further preferably, the confidence degree is inversely deduced from the data error rate, and the specific inverse deduction rule is: confidence degree S = 1-data error rate, data error rate = number of error codes in the data transmission process / total transmission times (for example, if the TCAS data error rate is 0.5%, then STCAS = 0.995); if the error rate is > 20%, the confidence degree is forced to be 0.1 (marked as low reliable data).

[0088] Further preferably, the fused data output is “scene feature data (including flight phase, threat density, and electromagnetic interference intensity)” and “business-driven data (including threat coordinates, air route parameters, and ECU alarm)”. Double bus parallel transmission is adopted, scene feature data is transmitted to the synchronous scheduling module through the ARINC664p7 safety bus, the transmission delay is ≤1 ms, and the error code rate is ≤10 - ¹²; business-driven data is transmitted to the dynamic link module through the PCIe (Peripheral Component Interconnect Express) 4.0 high-speed bus, the bandwidth is ≥8 GB / s, and the delay is ≤0.8 ms. Bus transmission adopts “frame priority marking”, threat alarm data is marked as “emergency frame”, and transmission resources are preferentially occupied.

[0089] Further preferably, in the embodiment, the initial value of the weight is: 、 ; the dynamic adjustment rule is: when the electromagnetic interference intensity I > -80 dBm, , the weight is increased by 30% (i.e. =0.52), and α is reduced to 0.28, so as to ensure that the weight proportion of reliable data (such as ECU) under strong interference is ensured.

[0090] Preferably, the scene vector generation unit quantizes the fused data into a four-dimensional scene vector containing flight phase, threat density, engine state, and electromagnetic interference intensity, and the method is as follows:

[0091] The synchronous scheduling module maps the unstructured data into discrete values according to the preset quantization rule based on the fused scene feature data, and constructs a four-dimensional scene vector:

[0092] ;

[0093] Among them, is the flight phase ∈ {1 (takeoff), 2 (cruise), 3 (approach), 4 (emergency)}, and the subscript “ph” is the abbreviation of “phase”, representing the flight phase; N is the threat density, i.e. the number of threat targets detected by TCAS. Engine state; I is electromagnetic interference intensity, unit is dBm.

[0094] Further preferably, the quantization rules of each dimension are as follows:

[0095] N is an integer from 0 to 10, no threat N = 0, multiple threats N ≥ 5, threat density N ≥ 10 is forced to take 10 (upper limit protection), N < 0 takes 0 (lower limit protection); = 0, indicating that the engine is running normally (no alarm or failure); = 1, indicating that the engine enters an alarm state (such as sensor abnormalities, parameter over-limit); = 2, indicating that the engine has a serious failure (such as mechanical damage, power loss);

[0096] In order to ensure the unity of the description, I in the scene vector V adopts the original dBm value of electromagnetic interference intensity (for example, -90), and the continuous value is directly taken as the input of the neural network. Table 1 below provides a reference method for mapping the original dBm value to a discrete level, which is used to assist in understanding the interference intensity level, rather than as a direct input value of the scene vector.

[0097]

[0098] In this embodiment, when the electromagnetic interference intensity I < -120 dBm, -120 dBm is taken (lower limit protection), and when I > -60 dBm, -60 dBm is taken (upper limit protection).

[0099] Preferably, the forward control module comprises:

[0100] A neural network processing unit is configured to receive four-dimensional parameters of the scene vector by using the trained neural network inference model, extract scene depth features by using a ReLU (Rectified Linear Unit) activation function, and output core synchronization parameters from the output layer.

[0101] A parameter caching unit is configured to store the generated core synchronization parameters, wherein the core synchronization parameters include a synchronization accuracy threshold, a sampling rate, and a PI controller adjustment coefficient, and a parameter update delay is less than or equal to 2 ms.

[0102] A parameter feedback adjustment unit is in communication connection with the self-healing learning module, configured to receive optimization instructions and dynamically correct the output parameters of the neural network inference model.

[0103] Further preferably, the forward control module receives the scene vector as input, and the pre-trained neural network inference model generates core synchronization parameters adapted to the current scene, and the method is as follows:

[0104] The scene vector V is received, and a pre-trained neural network inference model maps V to synchronization parameters, outputting the synchronization execution result / core synchronization parameters. Link operating status (latency) Resource utilization rate R used ).

[0105] Specifically, the core synchronization parameters include:

[0106] 1) Synchronization accuracy threshold As the maximum allowable value for multi-channel IO clock skew, it directly determines the audio synchronization effect. The smaller the value, the higher the synchronization accuracy. High accuracy is a key prerequisite for achieving high-fidelity three-dimensional spatial audio alarms (enabling drivers to intuitively determine the location of threats through hearing).

[0107] 2) The operating sampling frequency f of the Software Phase-Locked Loop (SW-PLL) serves as the "frequency reference" for generating the reference clock signal, directly determining the accuracy of the reference clock and the period of synchronous calibration. The higher the sampling rate, the higher the time resolution of the reference clock, and the shorter the period of synchronous calibration (i.e., the time interval between two calibrations), enabling a faster response to changes in clock deviation.

[0108] 3) PI controller coefficients ( , ), used to adjust the locking speed and stability of the SW-PLL ( The larger the value, the faster the lock-on. The larger the value, the stronger the anti-interference capability.

[0109] In some embodiments, the neural network inference model employs a "4-input-2-hidden-layer-4-output" neural network structure to construct the scene vector V and core synchronization parameters. The mathematical mapping relationship between them solidifies expert knowledge into a deterministic model, ensuring the accuracy and real-time performance of parameter predictions; specifically:

[0110] The input layer consists of 4 neurons and receives scene vectors. The first hidden layer consists of 16 neurons and uses the ReLU activation function to extract low-dimensional features of the scene (such as the combined features of "approach + multiple threats"). The second hidden layer consists of 16 neurons and uses the ReLU activation function to extract high-dimensional feature maps (such as emergency features of "fault + strong interference"). The output layer consists of 4 neurons and outputs the synchronization core parameters. .

[0111] Preferably, the mapping relationship includes:

[0112] Synchronization accuracy threshold Negative correlation with threat density N (the more threats, the higher the precision required), negative correlation with electromagnetic interference intensity I (the stronger the interference, the higher the precision required):

[0113] ;

[0114] wherein, is the scene correction coefficient, = 1 (takeoff) -> = 1.0, = 2 (cruise) -> = 1.2, = 3 (approach) -> = 1.0, = 4 (emergency) -> = 0.9; sampling rate f is positively correlated with electromagnetic interference intensity I (the stronger the interference, the higher the sampling rate required to resist interference), and is associated with flight phase (the highest sampling rate is required for emergency scenarios); the baseline is adjusted based on the flight phase, the dynamic enhancement basis is the electromagnetic interference intensity, and the fault superposition coefficient is the engine state, and through the two-level mapping of initial calculation by formula and final optimization by neural network, the precise matching of sampling rate and scene demand is realized, which is specifically expressed as follows:

[0115] ;

[0116] wherein, is the ECU state coefficient, = 0 -> = 1.0, = 1 -> = 1.2, = 2 -> = 1.5; is the absolute value of interference intensity (dBm), is the emergency compensation item, which reduces the sampling rate redundancy of the cruise scene;

[0117] is associated with the threat density N in the scene vector and the engine state , and the formula is as follows:

[0118] ;

[0119] ;

[0120] wherein, N is the threat density, and the base value 0.3 / 0.05 is the coefficient of the cruise threat-free scene (to ensure stability).

[0121] Preferably, in the present embodiment, the threat density ≥ 5 Forced to 0.3 us ; the constant term is 0.3 to achieve the minimum accuracy of the emergency scene to ensure the synchronization of the alarm sound; the interference strength increases by 10 dBm, f 20MHz (range 10MHz~250MHz; ≤60 take 60 (no additional sampling requirements), ≥80 take 80 (upper limit of sampling rate), increase the sampling rate in strong interference scenarios; emergency scenarios =0.9, =0.25, cruise scenario =0.3, =0.05; N≥3 take 3 (upper limit protection) to avoid too large to cause oscillation; =2 (fault) to additionally increase , enhance anti-interference ability.

[0122] Specifically, the pre-training process is:

[0123] 1) Sample generation: using the neural network inference model, input the scene vector V covering all 7428 scene combinations, generate more than 100,000 groups of high-precision "(input V, output )" training sample pairs. The training data is derived from high-fidelity flight simulators and real flight data records.

[0124] 2) Network training: adopt the neural network structure of "4 input-2 hidden layer-4 output" for training until the model prediction error does not exceed the preset error threshold.

[0125] 3. Model solidification: solidify the trained neural network weight as a deterministic software component and deploy it to the onboard "audio synchronization control module". No online training is performed during flight to meet the deterministic requirements of DO-178C Level A.

[0126] Further preferably, the parameter feedback adjustment unit receives the optimized parameters fed back by the reinforcement learning module, such as the synchronization accuracy threshold obtained after self-healing, and generates adaptive parameters based on the changed state to realize the closed loop of the control strategy.

[0127] In the embodiments of the present application, a neural network inference model is adopted; however, those skilled in the art can understand that other types of machine learning models can also be adopted, such as gradient boosting decision trees, support vector machines, etc. In an alternative embodiment, the model can not rely on a preset explicit mathematical formula, but directly learn the complex nonlinear mapping relationship from the scene vector to the core synchronization parameter through an end-to-end deep learning method from a large amount of flight simulator or real running data, to adapt to more complex scenarios.

[0128] Preferably, the synchronization execution module comprises:

[0129] a reference clock generation unit for generating a reference clock signal according to a sampling rate in the core synchronization parameter;

[0130] a prediction compensation calibration unit, which internally comprises a bias prediction subunit and a compensation execution subunit, for calculating a bias change trend by analyzing historical clock bias data through the bias prediction subunit, and generating a compensation amount according to the predicted bias through the compensation execution subunit, to realize advanced calibration of clock bias;

[0131] a calibration period adjustment unit for dynamically adjusting the calibration period according to the scene vector, wherein the calibration period in an emergency scenario is one fifth of the regular period, and the calibration period in a cruising scenario is twice the regular period.

[0132] Further preferably, the method for realizing advanced calibration of clock bias is as follows:

[0133] collecting the time stamp of the local clock of each IO channel, taking the reference time stamp corresponding to the reference clock signal as a reference, and calculating the current clock bias of each channel;

[0134] analyzing the historical bias data of each channel by using a sliding window method, and combining the sampling period to calculate the current bias change rate;

[0135] combining the current clock bias, the bias change rate of each channel, and the prediction coefficient corresponding to the scene to generate a compensation value; the prediction coefficient is dynamically adjusted according to the flight phase;

[0136] writing the compensation value into the clock controller of each IO channel through a software interrupt, respectively adjusting the local clock frequency, and completing the calibration; wherein the calibration period is consistent with the sampling period, and the clock bias after calibration does not exceed the synchronization accuracy threshold.

[0137] In some embodiments, the method for realizing advanced calibration of clock bias is as follows:

[0138] collecting the time stamp of the local clock of each IO channel, taking the reference time stamp corresponding to the reference clock signal as a reference, and calculating the current clock bias of each channel; i ​​​

[0139] ;

[0140] Calculate the deviation rate of change by using the sliding window method :

[0141] ;

[0142] Combine the current deviation and the rate of change to generate a predicted compensation value:

[0143] ;

[0144] Wherein, is the channel timestamp, is the reference timestamp; M is the window size, is the sampling period, t is the current time, is the clock deviation of the i IO channel at the k M sampling period ago; k is the prediction coefficient, which is dynamically adjusted according to the scene vector, and the dynamic adjustment rule is: in the emergency scene k =0.8, in the approach / take-off scene k =0.6, in the cruise scene k =0.3;

[0145] Write the compensation value into the channel clock controller through software interruption, adjust the local clock frequency, and complete the calibration.

[0146] Further preferably, the synchronization execution module further comprises a link dynamic configuration unit, which is in communication connection with the forward-looking control module, and dynamically adjusts the link working mode according to the synchronization precision threshold in the core synchronization parameter, and the method is:

[0147] When the synchronization precision threshold is lower than the preset precision threshold, enable the low delay mode, reduce the link cache depth and increase the CPU resource allocation ratio;

[0148] When the synchronization precision threshold is higher than the preset precision threshold, enable the energy efficiency mode, adjust the link cache depth and adopt the lightweight transmission coding;

[0149] The link dynamic configuration unit comprises a priority determination subunit, which dynamically adjusts the link priority according to the threat collision time in the multi-target threat data, and sets the link priority to the highest level in the emergency scene.

[0150] In this embodiment, the precision threshold is preferably 0.5μs:

[0151] When ≤0.5μs (high synchronization requirement, such as multi-threat scenario): enable "low latency mode", link cache depth ≤ 32 bytes, CPU resources are preferentially allocated (accounting for ≥ 70%), unnecessary data compression is disabled;

[0152] When > 0.5μs (regular synchronization requirement, such as cruise scenario): enable "energy efficiency mode", link cache depth is dynamically adjusted:

[0153] The specific adjustment rule is: cache depth D = ceil(data amount x 1.2) (ceil is rounding up), and 64 ≤ D ≤ 256; data amount ≤ 53 bytes / ms → D = 64 bytes; 54 bytes / ms < data amount ≤ 213 bytes / ms → D = ceil(data amount x 1.2); data amount > 213 bytes / ms → D = 256 bytes, enable lightweight compression (compression ratio 2:1).

[0154] Preferably, the transmission encoding mode includes Turbo encoding, CRC check or no encoding.

[0155] In the embodiment, based on the threat collision time TTC and the engine state , the link priority P is divided into 0-2 levels (0 level is the highest):

[0156] ;

[0157] The link configuration table (including unit connection order, parameter value, and validity period) is automatically generated, the configuration takes effect within 3 minutes, and no code modification is required.

[0158] In some embodiments, the link dynamic configuration unit further includes a spatial audio enhancement subunit configured to, when the link is the highest level, calculate the left and right channel gain difference through the head-related transfer function, and add delay compensation according to the threat azimuth to ensure that the pilot's identification error of the threat azimuth is not more than an error threshold.

[0159] Specifically, the link configuration of the spatial audio enhancement subunit is adjusted through the head-related transfer function (HRTF), including the following contents:

[0160] Azimuth mapping: threat azimuth (-180°~180°) corresponds to the left and right channel gain difference:

[0161] ;

[0162] Left channel gain: (when <0, left threat); right channel gain: (when > 0, right threat).

[0163] Delay compensation: different azimuth corresponds to different propagation delay, need to add compensation value in the link: ; for example: = 30° (right threat), right channel delay reduction 0.03ms, ensure binaural synchronization.

[0164] As the left threat azimuth threat, left channel gain +3dB, right channel gain -3dB, ensure pilot azimuth identification error ≤ error threshold, the error threshold of the embodiment is preferably .

[0165] Further preferably, based on the link priority P and the scene vector V, a multi-dimensional resource allocation model is constructed to ensure core link resource priority, and the resource scheduler allocates CPU and bandwidth resources according to the balance model and priority. The specific method is:

[0166] 1) CPU resource allocation:

[0167] ;

[0168] Wherein, is the CPU resource of the current link priority, is the CPU priority coefficient of the current link priority (link priority P=0→ =0.7, P=1→ =0.2, P=2→ =0.1); is the scene emergency coefficient ( =4→ =1.2, other scenes→ =1.0);

[0169] When the sum of the calculation values of all links is >100%, reduce according to the priority ladder:

[0170] P=0 level link remains the calculation value unchanged, P=1 level link is allocated according to 80% of the calculation value, P=2 level link is allocated according to 50% of the calculation value; if still over 100%, then suspend P=2 level link;

[0171] 2) Bus bandwidth allocation:

[0172] ;

[0173] Wherein, is the bandwidth resource of the current link priority, is the bandwidth priority weight of the current link priority (link priority P=0→ =5, P=1→ =3, P=2→ =2), meet ;

[0174] 3) Allocate memory:

[0175] ;

[0176] Memory of the current link priority and the cache depth D and the number of links linearly related.

[0177] In this embodiment, for example:

[0178] 0-level link: CPU ratio ≥ 70%, bandwidth ratio ≥ 50%, transmission delay ≤ 1.5ms;

[0179] 1-level link: CPU ratio 20%~30%, bandwidth ratio 30%~40%;

[0180] 2-level link: CPU ratio ≤ 10%, bandwidth ratio ≤ 20%;

[0181] Further preferably, reserve 20% redundancy space to avoid memory overflow; resource allocation adjustment delay ≤ 1ms, avoid link jam caused by resource contention.

[0182] Preferably, the health monitoring module comprises:

[0183] a state acquisition unit for acquiring three types of indexes of synchronization stability, link smoothness and data integrity, wherein the synchronization stability is the time proportion of clock deviation meeting the synchronization accuracy threshold, the link smoothness is the time proportion of link delay less than or equal to the delay threshold, and the data integrity is 1 minus the data packet loss rate;

[0184] a health degree calculation unit for calculating the system health degree based on the synchronization stability, link smoothness and data integrity through a weighted algorithm;

[0185] a threshold dynamic adjustment unit for correcting the health degree threshold according to the historical fault times to realize adaptive optimization of self-healing triggering condition.

[0186] Further preferably, the state acquisition unit is deployed in the cross-domain fusion, synchronization scheduling and dynamic link layer to acquire indexes:

[0187] synchronization scheduling layer: SW-PLL locking state, IO channel synchronization deviation compliance rate;

[0188] cross-domain fusion layer: data packet loss rate (≤0.1%), fusion model confidence (≥0.8);

[0189] Dynamic link layer: link total delay (≤3ms), unit fault identification.

[0190] Further preferably, the system health degree is calculated by a weighting algorithm, and the method is:

[0191] H=0.4×synchronization stability+0.3×link smoothness+0.3×data integrity;

[0192] Wherein, the synchronization stability is the time proportion of clock deviation less than the synchronization accuracy threshold, the link smoothness is the time proportion of link delay less than or equal to the delay threshold, and the data integrity is 1 minus the data packet loss rate.

[0193] In the embodiment, the synchronization stability statistical window is 1 second, that is, the sampling number / total sampling number within 1 second×100% (sampling frequency 1 kHz, total sampling number 1000); the link smoothness statistical window is 10 seconds, and in the embodiment, the delay threshold is preferably 3 ms, that is, the transmission frame number / total transmission frame number within 10 seconds×100% of Tlink≤3 ms.

[0194] Further preferably, the unit fault identification is obtained based on a fault code mapping, and the fault code mapping relationship is shown in Table 2.

[0195]

[0196] In the embodiment, according to the health degree H (such as H=0.55, lower than the threshold 0.6) and the fault type F (such as F=0) obtained above, it is judged that when H≤0.6, an interrupt signal is triggered, and [V, F, H, KPI] is packaged and sent to the reinforcement learning module, and the response delay is ≤1 ms.

[0197] Further preferably, in order to cope with the potential unexpected behavior of the intelligent model and meet the stringent requirements of Level A, the application designs an independent safety monitoring and degradation mechanism:

[0198] Output boundary monitoring: the hardware monitoring unit in the FPGA continuously checks whether the synchronization parameters (δ, f, Kp, Ki) output by the BP neural network are within the preset safety boundary (for example, );

[0199] Heartbeat and timeout monitoring: the running state of the intelligent algorithm module is monitored, and if there is no response or calculation timeout, it is determined that the module is faulty.

[0200] Degradation mode triggering: once the model output is out of bounds or the module is faulty, the system will immediately trigger the degradation mode to bypass the intelligent decision module.

[0201] ​Deterministic strategy takeover: in degraded mode, the system will load a set of preset, flight phase-based, simple but absolutely reliable fixed synchronization parameter table (for example: fixed =1.5μs, emergency phase fixed =0.5μs), to ensure the most core audio warning function uninterrupted.

[0202] Preferably, the self-healing learning module comprises:

[0203] A self-healing strategy library storing three types of self-healing strategies, i.e. non-critical link sacrifice strategy, anti-interference coding enhancement strategy, and minimum function set startup strategy, each of which is mapped to a scenario vector;

[0204] A strategy selection unit for calling an adaptive self-healing strategy according to the health degree evaluation result and the current scenario vector, calling the anti-interference coding enhancement strategy in a strong electromagnetic interference scenario, and calling the non-critical link release strategy in a multi-threat scenario;

[0205] A Q-learning unit for updating the mapping relationship between scenarios and strategies according to the self-healing result.

[0206] Further preferably, a control strategy closed loop is realized, and the method is as follows:

[0207] 1) Dynamically selecting a self-healing strategy according to the evaluation result and the scenario vector:

[0208] Based on a multi-dimensional scenario vector , two key system state indicators F and H are added to form a complete state vector S;

[0209] Defining an action space:

[0210] Action A1: non-critical link sacrifice strategy;

[0211] Triggering condition: threat density ≥ 3 (multi-threat scenario), fault type is link delay exceeding standard;

[0212] Operation content: immediately cut off priority level 2 link (such as normal navigation sound), release CPU / bandwidth resources; preferentially repair level 0 / 1 link (TCAS warning, approach instruction), restart link driver; record the sacrificed link information, and automatically restore after the fault is removed.

[0213] Action A2: anti-interference enhancement strategy;

[0214] Triggering condition: electromagnetic interference intensity > -70dBm (strong interference scenario), fault type is data packet loss rate exceeding standard;

[0215] Operation content: enable Turbo coding (code rate from 1 / 3 to 1 / 2), enhance data anti-interference ability; switch to backup transmission link (such as switching from PCIe bus to ARINC664p7 bus); increase SW-PLL sampling rate by 20MHz, enhance clock synchronization stability.

[0216] Action A3: Minimum function set start strategy

[0217] Trigger condition: engine state = failure (emergency scenario), health degree H≤0.6;

[0218] Operation content: only keep fire warning, TCAS threat warning 2 types of core links, other links are suspended; load light software image (remove unnecessary functions, start time≤8ms); lock synchronization parameters (fixed ), avoid risks introduced by dynamic adjustment.

[0219] Design reward function:

[0220] ;

[0221] Wherein, is the scene emergency coefficient, when the emergency scene (flight phase = 4) or engine failure (state = 2) =2, double the reward, and =1 in other scenarios; is the fault recovery time, is the self-healing success rate (value range is 0~1, 1 represents complete success, 0 represents complete failure); is the number of fault recurrence within 2 minutes, fault recurrence = the same link appears the same type of fault again within 2 minutes (such as P=0 level link t=0 F=0, t=50 seconds F=0 again, then it is recorded as 1 recurrence); Different links or different types of faults are not counted as recurrence;

[0222] Adopt ε-greedy strategy to select self-healing strategy from action space A={A1,A2,A3}, Q value update:

[0223] ;

[0224] Wherein, =0.1 is the learning rate, which controls the update amplitude to avoid large fluctuations; =0.9 is the discount factor; is the previous state, is the new state after executing action ;

[0225] After multiple iterations, the optimal self-healing strategy (such as A1) is obtained, and the optimized synchronization / link parameters are obtained.

[0226] Total state number = 4 (flight phase) x 11 (threat density) x 3 (engine state) x 7 (interference intensity) x 3 (fault type) x 3 (health degree) = 4 x 11 x 3 x 7 x 3 x 3 = 7428 kinds, covering all typical scenarios and fault combinations.

[0227] Preferably, the number of iterations is the number of self-healing operations (i.e. after each trigger H≤0.6 and perform self-healing action, recorded as 1 iteration); in this embodiment, the initial = 0.3, every 100 iterations Decrease 0.05, The minimum is 0.1.

[0228] 2) Execute the self-healing strategy according to the scenario vector and obtain the self-healing result, for example:

[0229] Threat density ≥ 3 (multi-threat scenario): execute the "quick sacrifice non-critical link" strategy, preferentially repair 0 / 1 level link, temporarily cut off 2 level link (such as ordinary navigation sound), self-healing time ≤ 15ms;

[0230] Electromagnetic interference intensity > -70dBm (strong interference scenario): automatically enable Turbo anti-interference encoding (code rate 1 / 2), while restarting the data transmission link affected by interference, self-healing time ≤ 20ms;

[0231] Engine state = fault (emergency scenario): trigger "minimum feature set", only keep fire warning, TCAS warning link, other links are suspended, self-healing time ≤ 8ms.

[0232] 3) Feedback the self-healing result to the forward control module, use the Q-learning reinforcement learning model, take the self-healing result (recovery time, link stability) as the reward signal, update the scenario-self-healing strategy mapping relationship, and regenerate the adaptive parameters based on the changed state:

[0233] State space: discretized value of scenario vector V;

[0234] Action space: 3 types of self-healing strategies (sacrifice non-critical link, anti-interference encoding, minimum feature set);

[0235] Reward function: R = 100 - 5 x recovery time - 10 x fault recurrence number.

[0236] The model training and deployment are both completed offline, an initial Q table of the Q-learning model is generated through large-scale simulation training on the ground, and covers all typical fault scenarios; and the mature Q table is solidified and deployed to the airborne system. Learning update in flight is only performed during non-critical flight stages and when system resources are sufficient, or learning logs are recorded for iterative optimization of the ground model.

[0237] It should be noted that the above embodiment provides a modular digital audio multi-channel IO control system for aviation communication, which is only exemplified by the division of the above functional modules. In actual application, the above functions can be completed by different functional modules according to needs, that is, the modules or steps in the embodiment of the application are further decomposed or combined, for example, the modules of the above embodiment can be combined into one module, or can be further split into multiple sub-modules to complete all or part of the functions described above. The names of the modules and steps involved in the embodiments of the application are only for distinguishing the modules and steps, and should not be considered as an improper limitation of the application.

[0238] The second embodiment of the application provides a modular digital audio multi-channel IO control method for aviation communication, which comprises:

[0239] S1, acquiring and fusing multi-domain heterogeneous data to construct a scene vector for quantifying a real-time aviation scene; the scene vector includes a flight stage, a threat density, an engine state, and an electromagnetic interference intensity;

[0240] S2, based on the scene vector, inferring and generating core synchronization parameters matched with the current scene through a pre-trained neural network inference model; the core synchronization parameters at least include a synchronization accuracy threshold for calibrating clock bias;

[0241] S3, according to the core synchronization parameters, performing clock calibration and link configuration on multiple IO channels of aviation audio to realize synchronization control between channels;

[0242] S4, monitoring the running state of the synchronization control to evaluate system health, and performing feedback optimization on the neural network inference model based on the evaluation result of the system health.

[0243] Although the steps in the above embodiments are described in the above order, those skilled in the art can understand that, in order to achieve the effect of the embodiments, the steps do not have to be executed in such an order, and can be executed simultaneously (in parallel) or in a reversed order, and these simple changes are within the protection scope of the application.

[0244] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related description of the above-described method can refer to the corresponding process in the foregoing system embodiment, and will not be repeated here.

[0245] The electronic device of the third embodiment of the application comprises:

[0246] at least one processor; and

[0247] a memory in communication with the at least one processor; wherein

[0248] The memory stores instructions executable by the processor, and the instructions are used to be executed by the processor to implement the above-described modular digital audio multi-channel IO control method for aviation communication.

[0249] The computer readable storage medium of the fourth embodiment of the application stores computer instructions, and the computer instructions are used to be executed by a computer to implement the above-described modular digital audio multi-channel IO control method for aviation communication.

[0250] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process and related description of the above-described electronic device and computer readable storage medium can refer to the corresponding process in the foregoing method embodiment, and will not be repeated here.

[0251] Those skilled in the art should appreciate that the modules and method steps of each example described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. The software modules, method steps corresponding to the program can be placed in random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. In order to clearly illustrate the interchangeability of electronic hardware and software, the components and steps of each example have been generally described in the foregoing description. Whether the functions are performed in electronic hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the application.

[0252] Computer program code for carrying out operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0253] The computer program instructions can also be loaded onto a computer or other programmable information processing apparatus to cause a series of operations to be performed on the computer or other programmable information processing apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable information processing apparatus implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0254] The terms "first", "second", etc. are used to distinguish between similar objects, rather than to denote or imply a particular order or sequence.

[0255] The terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0256] The technical scheme of the present application has been described in combination with the preferred embodiments shown in the drawings, but it is easy for those skilled in the art to understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical schemes after the changes or replacements will all fall within the protection scope of the present application.

Claims

1. A modular digital audio multi-channel IO control system for avionics communications, characterized in that, The system comprises: A scene perception module, a forward control module, a synchronous execution module, a health monitoring module and a self-healing learning module, each module is connected through a bus communication, forming a closed-loop control architecture; The scene perception module collects multi-domain heterogeneous data and performs dynamic weight fusion processing, outputs a scene vector representing the current flight scene, and the dynamic weight is adaptively adjusted according to the electromagnetic interference intensity; The forward control module receives the scene vector as input, and a pre-trained neural network inference model generates core synchronization parameters adapted to the current scene; The synchronous execution module generates a reference clock according to the core synchronization parameters, uses a prediction compensation mechanism to calibrate the clock deviation of the digital audio multi-IO channel, and realizes multi-channel synchronous control; The health monitoring module collects the whole link running state, evaluates the system health degree and outputs the evaluation result; The self-healing learning module dynamically selects a self-healing strategy according to the evaluation result and the scene vector, feeds back the self-healing result to the forward control module, and regenerates the adaptive parameters based on the changed state to realize the control strategy closed loop; The synchronous execution module further comprises a link dynamic configuration unit, which is connected with the forward control module and dynamically adjusts the link working mode according to the synchronization precision threshold in the core synchronization parameters, and the method is: When the synchronization precision threshold is lower than the preset precision threshold, the low delay mode is enabled, the link cache depth is reduced and the CPU resource allocation ratio is increased; When the synchronization precision threshold is higher than the preset precision threshold, the energy efficiency mode is enabled, the link cache depth is adjusted and the lightweight transmission coding is used; The link dynamic configuration unit comprises a priority determination subunit, which dynamically adjusts the link priority according to the threat collision time in the multi-target threat data, and sets the link priority to the highest level in the emergency scene; The multi-target threat data is obtained by the traffic warning and collision avoidance system.

2. The modular digital audio multi-channel IO control system for aeronautical communications of claim 1, wherein, The scene perception module comprises: An interface adaptation unit configured with interfaces connected with a flight management system, a traffic warning and collision avoidance system and an engine control system respectively, to collect three-dimensional route data, multi-target threat data and engine real-time state parameters respectively; A dynamic weight fusion unit, which internally comprises an electromagnetic interference detection subunit, adjusts the weight ratio of each domain data according to the electromagnetic interference intensity, specifically: When the electromagnetic interference intensity is higher than-80dBm, the weight ratio of the engine real-time state parameters is increased; When the electromagnetic interference intensity is lower than-100dBm, the weight ratio of the multi-target threat data is increased; A scene vector generation unit for quantizing the fused data into a four-dimensional scene vector containing flight phase, threat density, engine state and electromagnetic interference intensity.

3. The modular digital audio multi-channel IO control system for aeronautical communications of claim 1, wherein, The forward control module comprises: A neural network processing unit for receiving the four-dimensional parameters of the scene vector by using the trained neural network inference model, extracting the scene depth feature through the ReLU activation function, and outputting the core synchronization parameters by the output layer; A parameter cache unit for storing the generated core synchronization parameters, the core synchronization parameters including synchronization precision threshold, sampling rate and PI controller adjustment coefficient, and the parameter update delay is ≤2ms; The parameter feedback adjustment unit is in communication connection with the self-healing learning module, and is configured to receive an optimization instruction and dynamically correct an output parameter of the neural network inference model.

4. The modular digital audio multi-channel IO control system for aeronautical communications of claim 3, wherein, The synchronous execution module comprises: A reference clock generation unit configured to generate a reference clock signal according to a sampling rate in the core synchronization parameter; A prediction compensation calibration unit comprising a bias prediction subunit and a compensation execution subunit, configured to analyze historical clock bias data to calculate a bias change trend by the bias prediction subunit, and to generate a compensation amount according to the predicted bias by the compensation execution subunit, so as to realize advanced calibration of clock bias; A calibration period adjustment unit configured to dynamically adjust a calibration period according to the scene vector, wherein the calibration period in an emergency scene is one fifth of a regular period, and the calibration period in a cruising scene is twice the regular period.

5. The modular digital audio multi-channel IO control system for aeronautical communications of claim 4, wherein, The method for realizing advanced calibration of clock bias comprises: Collecting timestamps of local clocks of each IO channel, taking a reference timestamp corresponding to the reference clock signal as a reference, and calculating current clock biases of each channel; Using a sliding window method to analyze historical bias data of each channel, and combining a sampling period to calculate a current bias change rate; Combining the current clock bias of each channel, the bias change rate, and a prediction coefficient corresponding to the scene to generate a compensation value; the prediction coefficient is dynamically adjusted according to the flight phase; Writing the compensation value into a clock controller of each IO channel through a software interrupt, respectively adjusting the local clock frequency, and completing the calibration; wherein the calibration period is consistent with the sampling period, and the clock bias after calibration does not exceed a synchronization accuracy threshold.

6. The modular digital audio multi-channel IO control system for aeronautical communications of claim 4, wherein, The link dynamic configuration unit further comprises a spatial audio enhancement subunit configured to, when the link is the highest level, calculate left and right channel gain differences through a head-related transfer function, and simultaneously add delay compensation according to a threat azimuth angle.

7. The modular digital audio multi-channel IO control system for aeronautical communications of claim 5, wherein, The health monitoring module comprises: A state collection unit configured to collect three types of indexes, including synchronization stability, link smoothness, and data integrity, wherein the synchronization stability is a time proportion of clock bias meeting a synchronization accuracy threshold, the link smoothness is a time proportion of link delay being less than or equal to a delay threshold, and the data integrity is 1 minus a data packet loss rate; A health degree calculation unit configured to calculate a system health degree through a weighting algorithm based on the synchronization stability, the link smoothness, and the data integrity; A threshold dynamic adjustment unit configured to correct a health degree threshold according to a historical fault number, so as to realize adaptive optimization of a self-healing triggering condition.

8. The modular digital audio multi-channel IO control system for aeronautical communications of claim 1, wherein, The self-healing learning module comprises: A self-healing strategy library storing three types of self-healing strategies, including a non-key link sacrifice strategy, an anti-interference coding enhancement strategy, and a minimum feature set start strategy, each self-healing strategy being in a mapping relationship with a scene vector; A strategy selection unit configured to call an adaptive self-healing strategy according to a health degree evaluation result and a current scene vector, call the anti-interference coding enhancement strategy in a strong electromagnetic interference scene, and call a non-key link release strategy in a multi-threat scene; A Q learning unit configured to update a mapping relationship between a scene and a strategy according to a self-healing result.

9. A modular digital audio multi-channel IO control method for aeronautical communications, characterized by, The method comprises the following steps: Acquiring and fusing multi-domain heterogeneous data to construct a scene vector for quantifying real-time aviation scenes; the scene vector comprises a flight phase, a threat density, an engine state, and an electromagnetic interference intensity; Based on the scene vector, a pre-trained neural network inference model is used to infer and generate core synchronization parameters matching the current scene; the core synchronization parameters at least include a synchronization accuracy threshold for clock bias calibration; According to the core synchronization parameters, clock calibration and link configuration are performed on multiple IO channels of the aviation audio to achieve synchronization control between channels; According to the synchronization accuracy threshold in the core synchronization parameters, the link working mode is dynamically adjusted, and the method is: When the synchronization accuracy threshold is lower than the preset accuracy threshold, a low delay mode is enabled, the link cache depth is reduced, and the CPU resource allocation ratio is increased; When the synchronization accuracy threshold is higher than the preset accuracy threshold, an energy efficiency mode is enabled, the link cache depth is adjusted, and a lightweight transmission coding is adopted; According to the threat collision time in the multi-target threat data, the link priority is dynamically adjusted, and the link priority is set to the highest level in the emergency scene; the multi-target threat data is obtained by the traffic warning and collision avoidance system; The running state of the synchronization control is monitored to evaluate the system health, and the neural network inference model is feedback optimized based on the evaluation result of the system health.

Images