Audio integrated signal processing system based on VPX architecture

Abstract

The application provides an audio integrated signal processing system based on a VPX architecture, which comprises: a case conforming to an OpenVPX standard, wherein a backboard is arranged in the case, the backboard is integrated with a high-speed serial RapidIO bus and an Ethernet interface; a plurality of heterogeneous computing board cards are connected to the case through VPX slots, the board cards are in point-to-point communication through the high-speed serial RapidIO bus, the heterogeneous computing board cards contain different types of computing chips and perform different types of audio signal processing tasks; and a multi-machine stacking expansion module is connected to the case through the Ethernet interface of the backboard, when it is detected that the computing capacity of any of the heterogeneous computing board cards cannot meet the audio signal processing demand, an external processing machine is connected through the Ethernet interface of the backboard to perform distributed computing. Through implementation of the scheme, the signal processing efficiency of the whole system can be improved, and the fault tolerance and reliability of the system are improved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of audio signal processing, and particularly relates to an audio integrated signal processing system based on a VPX architecture. BACKGROUND

[0002] As the computing center and software running base of a sonar system, the audio integrated signal processing system receives sonar large-scale array sensing signal data streams, performs signal processing calculation, and completes functions such as water surface and underwater target detection, identification and positioning, underwater acoustic communication, underwater environment measurement, and the like in a real-time mode combined with intelligent processing. The signal processing in the underwater acoustic field is carried out around the underwater acoustic sensing array and the beam domain real-time data stream processing procedure, and many algorithms are closely connected. Therefore, multiple computing nodes need to be integrated internally to achieve efficient processing requirements.

[0003] Traditional audio signal processing systems usually adopt a relatively single computing architecture, and their computing capacity is limited, making it difficult to simultaneously process multiple complex audio signal processing tasks. In addition, the expandability of traditional systems is poor. When facing more complex tasks or needing to process larger-scale data, it is difficult to improve the computing capacity of the system through simple means. If new functional modules or upgraded computing chips are needed, the entire system often needs to be overhauled on a large scale, which is not only costly but also time-consuming and laborious.

[0004] In order to solve these problems, people began to seek a more efficient, flexible and scalable audio signal processing system architecture. The emergence of the VPX standard provides a new way of thinking for the development of audio signal processing systems. The VPX standard has the advantages of high-speed data transmission, good expandability and compatibility, and can meet the requirements of modern audio signal processing systems for high performance and flexibility. However, there are still some deficiencies in the actual application of the audio signal processing system based on the VPX standard. For example, although the VPX standard provides high-speed data transmission channels, in actual systems, how to fully utilize these channels to achieve efficient communication between heterogeneous computing boards to improve the processing efficiency of the overall system is still a problem that needs to be solved. At the same time, how to reasonably allocate tasks of heterogeneous computing boards according to different audio signal processing requirements, and how to effectively expand when the computing capacity is insufficient, also need further research and optimization. SUMMARY

[0005] Therefore, the present application aims to provide an audio integrated signal processing system and device based on a VPX architecture to meet the needs of improving the processing efficiency of the overall system and the reliability of the system.

[0006] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0007] The application provides an audio integrated signal processing system based on a VPX architecture, which comprises: a case conforming to an OpenVPX standard, a backboard arranged in the case, the backboard integrating a high-speed serial RapidIO bus and an Ethernet interface; a plurality of heterogeneous computing board cards, the heterogeneous computing board cards being connected to the case through VPX slots, the board cards being in point-to-point communication through the high-speed serial RapidIO bus, the heterogeneous computing board cards containing different types of computing chips and being used for executing different types of audio signal processing tasks; and a multi-machine stacking expansion module, which is connected to the case through the backboard Ethernet interface and is used for connecting an external processing machine through the backboard Ethernet interface when it is detected that the computing capacity of any of the heterogeneous computing board cards cannot meet the audio signal processing demand, and performing distributed computing, the external processing machine being stacked by a plurality of modules with audio signal processing functions.

[0008] The application provides an audio integrated signal processing system based on a VPX architecture, which selects an OpenVPX integrated specification as a bus integrated architecture and has an open feature, can support various loads to be inserted into the same case, has strong expansibility, the case backboard integrates a high-speed serial RapidIO bus, so that the heterogeneous computing board cards can be in point-to-point communication through the bus, supports an addressing model and supports message transmission to ensure efficient and rapid data transmission, can meet the demand for rapid transmission of a large amount of data in audio signal processing, and simultaneously integrates the high-speed serial RapidIO bus and a gigabit Ethernet in a data plane to ensure low-latency super bandwidth and independent links, thereby improving the overall signal processing efficiency of the system. In addition, the heterogeneous computing board cards contain different types of computing chips, can be optimized for different types of audio signal processing tasks, give full play to the advantages of various chips, and realize efficient audio signal processing. Finally, the 40G interface is led out through the data plane for multi-machine stacking, the computing power of the elastic expansion computing platform is expanded, and the distributed computing architecture enables the system to continue to perform audio signal processing through other normal board cards or modules when some of the board cards or modules fail, thereby improving the fault tolerance and reliability of the system.

[0009] Other advantages, objects, and features of the application will be set forth in the following specification and will be apparent to those skilled in the art from the following specification and the attached claims, or can be learned by practice of the application. The objects and other advantages of the application can be realized and attained by the structure particularly pointed out in the specification. BRIEF DESCRIPTION OF DRAWINGS

[0010] In order to make the objects, technical solutions and advantages of the application clearer, the application provides the following drawings for description:

[0011] Figure 1 It is a virtual structure schematic view of the audio integrated signal processing system based on the VPX architecture in the application.

[0012] Figure 2 A flowchart of a process for distributed computing by connecting an external processor through the backplane Ethernet interface in the present application;

[0013] Figure 3 A connection diagram of requirements 1-5 under the correlation of audio signal processing requirements of the modules in the present application. DETAILED DESCRIPTION

[0014] The technical solutions of the present application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0015] In the description of the present application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection" and "linking" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be direct connection, or indirect connection through an intermediate medium, or internal communication of two elements; it can be wireless connection, or wired connection. For those skilled in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0016] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as there is no conflict.

[0017] The present embodiment provides an audio integrated signal processing system based on VPX architecture, as shown in Figure 1 , comprising:

[0018] A case 101, which conforms to the OpenVPX standard, is provided with a backplane inside the case, and the backplane integrates a high-speed serial RapidIO bus and an Ethernet interface;

[0019] A plurality of heterogeneous computing board cards 102, which are connected to the case through VPX slots, communicate point-to-point through the high-speed serial RapidIO bus between the board cards, and the heterogeneous computing board cards contain different types of computing chips for executing different types of audio signal processing tasks;

[0020] A multi-machine stacking expansion module 103, which is connected to the case through the backplane Ethernet interface, is used to connect an external processor through the backplane Ethernet interface when it is detected that the computing capability of any of the heterogeneous computing board cards cannot meet the audio signal processing requirements, and to perform distributed computing, and the external processor is stacked by a plurality of modules with audio signal processing functions.

[0021] Exemplarily, the OpenVPX standard is an open, VPX architecture-based standard formulated by the VITA organization, aiming to provide a common hardware platform standard for high-reliability and high-performance embedded computing systems. The chassis strictly follows the OpenVPX standard specification and supports 3U or 6U size heterogeneous computing board card plugging. The chassis body is made of high-strength aluminum alloy material, has good heat dissipation performance and electromagnetic shielding effect. The chassis is equipped with a redundant fan module, which monitors the internal temperature in real time through an intelligent temperature control system and automatically adjusts the fan speed when the temperature exceeds the threshold, ensuring stable operation of the system in a wide temperature environment of -40℃ to 70℃. Taking a 6U size heterogeneous computing board card as an example, in the VPX (VITA46) protocol, P1-P6 of the 6U size heterogeneous computing board card can be customized in addition to P0.

[0022] The backplane, as the core hub of the system, integrates high-speed serial RapidIO bus and Ethernet interface. The RapidIO bus supports multiple rate modes such as 1x, 4x, and 16x, with a maximum data transmission rate of 28Gbps, supporting point-to-point, switching, and other topologies, and can flexibly configure the communication link between cards. The characteristics of the high-speed serial RapidIO (SRIO) bus include: based on SRIO physical ID addressing, using DMA, message, and doorbell for inter-processor communication, which can be based on direct connection or switching; has low-latency reliable communication features, eliminating the software overhead loss caused by traditional network packet loss and retries; has high-speed bandwidth.

[0023] The Ethernet interface conforms to the IEEE 802.3 standard and supports 10G / 40G Ethernet protocol, realizing network connection between internal cards and external devices through the backplane. The backplane also integrates a power management module to provide stable power supply for each card, with overvoltage, overcurrent, and short circuit protection functions to ensure system power safety. Based on Ethernet communication programming, the SOCKT programming interface is adopted, which has programming consistency on various operating systems and hardware platforms.

[0024] Multiple heterogeneous computing boards are connected to the case through VPX slots, and multiple types of computing chips are integrated on the boards, including general-purpose processors, digital signal processors, field programmable gate arrays, i.e., CPU boards, GPU boards, FPGA boards, etc., and the communication bandwidth between the boards is symmetrical and consistent. Different types of computing chips are optimized for specific audio signal processing tasks. For example, FPGA is good at real-time signal processing and can quickly complete audio signal filtering, modulation and demodulation, etc. GPU is suitable for large-scale parallel computing and can efficiently process deep learning algorithms of audio signals. Different types of audio signal processing tasks include underwater target detection, identification, positioning, underwater acoustic communication and underwater environment measurement. Point-to-point communication is realized between the heterogeneous computing boards through high-speed serial RapidIO bus. RapidIO controllers are integrated on the boards to support multiple communication modes such as message passing and stream transmission.

[0025] The multi-machine stacking expansion module is connected to the case through the backplane Ethernet interface. The internal module is designed in a modular manner and consists of multiple functional units. When the system detects that the computing power of any heterogeneous computing board cannot meet the audio signal processing requirements, the multi-machine stacking expansion module starts a distributed computing process. Specifically, it sends a task request to the external processing machine through the backplane Ethernet interface. The external processing machine is stacked by multiple modules with audio signal processing functions. These modules can be dynamically combined according to task requirements.

[0026] In this embodiment, full interconnection is achieved through various high-speed serial buses, making interoperation between modules very convenient. Distributed resource sharing and remote invocation provide much more convenience than in the past. Traditionally, distributed software processes based on Ethernet can be ported to VPX / OpenVPX systems in theory, greatly expanding the application field of commercial COTSA technology. At the same time, the system proposed in this embodiment has high density and flexibility in structure, allowing different structures to ensure compatibility with the original system. The pin density of the VPX high-speed serial connector is high, the transmission rate is faster, and higher computing density and higher transmission rate integration and application can be achieved. The module has more custom signal pins to support more complex boards and systems.

[0027] The embodiment of the present application provides an audio integrated signal processing system based on a VPX architecture, selects an OpenVPX integrated specification as a bus integrated architecture, and has an open feature, can insert various loads into the same cabinet independently, has strong expansibility, integrates a high-speed serial RapidIO bus on a cabinet backboard, so that point-to-point communication between heterogeneous computing boards can be performed through the bus, supports an addressing model, supports message transmission and the like to ensure efficient and rapid data transmission, can meet the requirement of rapid transmission of a large amount of data in audio signal processing, simultaneously integrates the high-speed serial RapidIO bus and a gigabit, Ethernet on a data plane, ensures low-latency super bandwidth and independent links, and therefore improves the signal processing efficiency of the whole system. Moreover, the heterogeneous computing board includes different types of computing chips, can be optimized for different types of audio signal processing tasks, fully gives play to the advantages of various chips, and realizes efficient audio signal processing. Finally, the 40G interface is led out through the data plane to perform multi-machine stacking, flexibly expands the computing platform power, and the distributed computing architecture enables the system to continue to perform audio signal processing through other normal boards or modules when partial computing boards or modules fail, and improves the fault tolerance and reliability of the system.

[0028] As an optional implementation, the multi-machine stacking expansion module comprises:

[0029] An Ethernet interface component is used to be connected with an Ethernet interface of the cabinet backboard, and realizes physical connection with an external processing machine;

[0030] A data transmission circuit comprises a high-speed signal processing chip and a transmission route, and is used for pre-processing and transmission of data;

[0031] A protocol conversion circuit is used for conversion between different protocols, so that the multi-machine stacking module can communicate with different types of external processing machines;

[0032] A computing resource detection module is used for monitoring the computing resource conditions of various heterogeneous computing boards in the cabinet;

[0033] A logic controller is used for sending a connection request and task allocation information to the external processing machine through the backboard Ethernet interface according to a preset strategy when the computing resource detection module detects insufficient computing resources.

[0034] Exemplarily, the Ethernet interface component is used as a physical bridge connecting the cabinet backboard and the external processing machine, and a high-performance 40G Ethernet connector such as a QSFP + interface is adopted.

[0035] The data transmission circuit can adopt a high-performance field programmable gate array as a core processing chip. The chip has strong parallel processing capability and rich I / O resources, supports a high-speed serial interface of up to 28 Gbps, can meet the transmission requirements of large data volume and high real-time in audio signal processing, and is internally integrated with a dedicated high-speed data processing module including a data cache, a coding and decoding unit, and an error checking module. The transmission route adopts a multi-layer PCB design and uses impedance matching technology to ensure the integrity of high-speed signals during transmission. The signal lines use differential pair wiring to reduce crosstalk and electromagnetic radiation between signals. For critical high-speed signal paths, such as 40G Ethernet data transmission lines, microstrip or stripline structures are used, and parameters such as line width, spacing, and dielectric thickness are strictly controlled to meet the requirements of high-speed signal transmission.

[0036] The protocol conversion circuit adopts a special protocol conversion chip combined with FPGA architecture. The special protocol conversion chip, such as Cavium Octeon series network processor, has strong protocol processing capability and supports multiple protocol stacks such as TCP / IP, UDP, and RapidIO. When the multi-machine stacking expansion module communicates with different types of external processors, the protocol conversion circuit first performs protocol analysis on the input data to identify the protocol type used by the data. Then, according to the format and requirements of the target protocol, the protocol conversion logic is configured and executed in the FPGA. For example, if the external processor uses the RapidIO protocol and the chassis uses the Ethernet protocol, the protocol conversion circuit will convert the Ethernet packet into a RapidIO format message packet and perform corresponding address mapping and data reorganization.

[0037] The computing resource detection module adopts a distributed monitoring architecture and deploys monitoring agents on each heterogeneous computing board in the chassis. The monitoring agents collect real-time computing resource usage data of the computing board through management interfaces such as I2C and SPI, including CPU utilization, memory occupancy, GPU computing load, and FPGA logic resource utilization. The collected data is transmitted to the main control unit of the computing resource detection module through the low-speed management bus of the backplane. The main control unit uses a high-performance ARM processor to aggregate, analyze, and process the computing resource data of each board. The main control unit uses intelligent algorithms such as historical data-based prediction algorithms and threshold judgment algorithms to evaluate the overall computing resource status of the chassis in real time. When the computing resource usage of one or more heterogeneous computing boards exceeds the preset threshold or the overall computing resource of the system cannot meet the current audio signal processing task requirements, the computing resource detection module immediately sends an alarm message to the logic controller indicating insufficient computing resources, along with a detailed resource usage report to provide a basis for decision-making by the logic controller.

[0038] The logical controller is the core control unit of the multi-machine stack expansion module. When receiving the alarm information of insufficient computing resources, the logical controller first analyzes and divides the current audio signal processing task according to the preset strategy. The strategy can be customized according to factors such as task priority, real-time requirement, and computing complexity. For example, the audio signal preprocessing task with high real-time requirement is preferentially allocated to the external processor module with strong computing performance and low delay.

[0039] The logical controller sends a connection request and task allocation information to the external processor through the backplane Ethernet interface. The connection request contains module identity authentication information, communication parameter configuration and the like, to ensure safe and reliable connection with the external processor. The task allocation information describes the type of the task to be processed, data input and output requirements, computing resource requirements and the like in detail, and the external processor receives the task and allocates resources according to the information.

[0040] In addition to the above process, during the task execution process, the logical controller also monitors the task execution state of the external processor in real time, and ensures the smooth execution of the task according to the plan through a heartbeat detection mechanism and a task progress feedback mechanism. If it is found that the external processor fails or the task execution is abnormal, the logical controller immediately starts a fault recovery strategy, such as reassigning tasks, switching to a backup processor, and the like, to ensure the continuity and reliability of the audio signal processing task.

[0041] The embodiment of the application provides an audio integrated signal processing system based on a VPX architecture. An Ethernet interface component is connected with a backplane Ethernet interface of a case, thereby providing a standard and convenient physical connection mode for the system and an external processor, facilitating system expansion. In a data transmission circuit, a high-speed signal processing chip is combined with a transmission route to realize data preprocessing and high-speed transmission, thereby ensuring the rapid flow of audio signals to meet real-time processing requirements. A protocol conversion circuit eliminates communication obstacles between different protocols, thereby enhancing the compatibility and interoperability of the system and various external processors. A computing resource detection module monitors heterogeneous computing board card resources in the case in real time. The logical controller sends a connection request and task allocation information to the external processor according to a preset strategy when the resources are insufficient, thereby realizing dynamic task allocation and improving resource utilization. The logical controller is an intelligent core, which plans the entire scheduling and management process, so that the system can flexibly adapt to different loads and task requirements, and effectively improves the system expansion, processing efficiency, versatility, resource utilization and intelligent management level.

[0042] As an optional implementation, the computing resource detection module performs the following steps, including:

[0043] The main line task of the signal processing system and the previous audio signal processing task are obtained, and the current computing resource utilization of each heterogeneous computing board card is collected; according to the current computing resource utilization, the remaining computing resources are determined; according to the main line task of the signal processing system, the previous audio signal processing task and the dynamic threshold adaptive algorithm, the resource dynamic threshold is determined; when the remaining computing resources are higher than the resource dynamic threshold, the computing resource condition is that the computing resources are sufficient; when the remaining computing resources are lower than or equal to the resource dynamic threshold, the computing resource condition is that the computing resources are insufficient.

[0044] Exemplarily, taking the audio integrated signal processing system based on the VPX architecture carried on the ship as an example, the main line task represents the task required to be completed by the audio signal processing system due to the main task of the ship when the audio signal processing system is carried on the ship, such as when a ship takes fishing as the main task, the main line task of the audio signal processing system is to identify and capture fish through the sonar.

[0045] The main line task of the signal processing system and the previous audio signal processing task can be obtained from the task management module, which is connected with the core control unit through the system bus, and records the main line task and task progress of the audio integrated signal processing system in real time. On each heterogeneous computing board card, embedded monitoring sensors are deployed, which communicate with the master control unit of the computing resource detection module through the I2C or SPI bus inside the board card. The sensors collect utilization data of CPU, GPU, FPGA and other computing chips in real time, such as CPU clock cycle occupancy, GPU stream processor workload, FPGA logic unit usage ratio; at the same time, the memory usage is collected, including used memory capacity, memory bandwidth occupancy; and the read / write rate of storage devices and other key indicators.

[0046] After receiving the collected data, a resource calculation model is established according to the hardware specification parameters of each computing board card. For CPU resources, the remaining computing resources are determined by the following formula: ;

[0047] wherein, represents the remaining CPU resources, represents the total number of CPU cores, represents the number of occupied cores, represents the unit time;

[0048] The remaining storage bandwidth is determined according to the following formula: ;

[0049] wherein, represents the remaining storage bandwidth, represents the maximum read / write rate of the storage device, represents the current read / write rate.

[0050] After the remaining computing resources are obtained by the above manner, a dynamic threshold adaptive algorithm is needed to determine the resource dynamic threshold of each kind of resource, and the resource dynamic threshold is compared with the remaining computing resources to determine whether the computing resources are sufficient, and in the case that the computing resources are insufficient, an external processing machine can be connected in advance through the backboard Ethernet interface to evaluate the computing performance, storage capacity, network connection stability and other indexes of the external device, select a suitable transmission protocol according to the external device connection mode and network environment, and realize the advance cooperation and optimized operation with the external device, so that the data transmission and processing efficiency can be effectively improved.

[0051] The embodiment of the application provides an audio integrated signal processing system based on a VPX architecture, which determines the remaining computing resources by acquiring the main line task of the signal processing system, the last audio signal processing task, and the current computing resource utilization rate of each heterogeneous computing board card, and accurately sets the resource dynamic threshold according to the main line task, the historical task and the dynamic threshold adaptive algorithm, so that the system task demand and the actual resource condition can be effectively combined for dynamic evaluation, and when the remaining computing resources and the dynamic threshold are compared to determine that the computing resources are sufficient or insufficient, a reliable basis is provided for the resource scheduling and task allocation of the system, resource waste or task execution failure caused by insufficient resources can be avoided, the utilization efficiency of the computing resources is improved, the system can be stably and efficiently operated in different task scenarios, and reasonable matching and dynamic balance of resources and tasks are realized.

[0052] As an optional implementation manner, the dynamic threshold adaptive algorithm comprises:

[0053] The main line task of the signal processing system and the last audio signal processing task are acquired, at least one next audio signal processing task is predicted according to the main line task and the last audio signal processing task, it is determined whether there is a real-time demand task in the at least one next audio signal processing task, when there is, the demand computing amount corresponding to the real-time demand task is determined, and the resource dynamic threshold is set according to the demand computing amount corresponding to the real-time demand task, when there is not, and the predicted next audio signal processing task includes multiple, the average value of the demand computing amounts of the multiple audio signal processing tasks is calculated, and the resource dynamic threshold is set according to the average value of the demand computing amounts, when there is not, and the predicted next audio signal processing task is one, the demand computing amount of the next audio signal processing task is determined, and the resource dynamic threshold is set according to the demand computing amount.

[0054] Exemplarily, according to the main task and the last audio signal processing task, the manner of predicting at least one next audio signal processing task can be that, first, a task dependency graph technology is introduced, the main task and the last task are recorded, and a time sequence dependency and a data dependency relationship between tasks are constructed, for example, the main task is target positioning, and the corresponding task dependency graph is: underwater target detection-identification-positioning, the identification task needs to rely on the target candidate area extracted by the detection task, and the target positioning task relies on the target type and position information determined by the target identification.

[0055] The task prediction is combined with a Markov chain model and an LSTM deep learning model. The Markov chain model is based on a historical task dependency graph, such as a task execution sequence of detection-positioning-identification-communication, and predicts the next task by calculating a task transition probability. Specifically, based on a historical task sequence, a state transition matrix is constructed. Assuming that the system has N types of tasks, the state transition matrix P is an N*N matrix, wherein, represents the probability of transitioning from task i to task j. The state transition matrix is calculated by counting the number of transitions between tasks in the historical task sequence. The probability of each next state is determined according to the state transition matrix. The LSTM model can process complex time series data, and takes multi-dimensional data obtained in the task context deep analysis as input, including historical task sequences, environmental parameters, resource states, etc. Specifically, the task execution in the past week, the sea water temperature change in different time periods each day, and the resource utilization fluctuation of each computing unit of the system are input into the LSTM model to learn the time dependency and complex patterns, thereby predicting the probability distribution of future tasks. The probabilities of the two models are weighted and fused to obtain the final predicted task. In addition to selecting the output with the highest weight as the prediction result, tasks with weights exceeding a threshold value can also be selected as the prediction result. Through the weighted fusion of the two models, the shortcomings of a single model in processing nonlinear, multi-factor influenced task prediction are compensated, and the prediction accuracy is improved.

[0056] When the predicted task is obtained, it is determined whether there is a real-time demand task in the task, and when there is, the demand computing amount corresponding to the real-time demand task is determined, and the resource dynamic threshold is set according to the demand computing amount corresponding to the real-time demand task. Specifically, the demand computing amount of the real-time demand task is used as the resource dynamic threshold for early judgment of whether the system can complete the real-time demand task by itself; when there is not, and the predicted next audio signal processing task includes multiple tasks, the average of the demand computing amounts of the multiple audio signal processing tasks is calculated, and the resource dynamic threshold is set according to the average of the demand computing amounts; when there is not, and the predicted next audio signal processing task is one, the demand computing amount of the next audio signal processing task is determined, and the resource dynamic threshold is set according to the demand computing amount.

[0057] The embodiment of the present application provides an audio integrated signal processing system based on a VPX architecture, which can perform reasonable task prediction based on historical task information by acquiring main line tasks of a signal processing system and last audio signal processing tasks, provides prospective guidance for subsequent resource allocation and task execution of the system, judges whether real-time requirements exist after determining predicted tasks, sets resource dynamic thresholds for different cases, and a fine processing mode can more accurately allocate system resources according to characteristics and requirements of the tasks, thereby avoiding waste caused by excessive reservation of resources and ensuring that real-time requirement tasks can obtain sufficient resource support, so that efficient and stable operation of the system is ensured, adaptability and processing capacity of the signal processing system to different types of tasks are improved, and overall resource utilization efficiency is optimized.

[0058] As an optional implementation, an audio integrated signal processing system based on a VPX architecture further comprises:

[0059] The sonar transmitters are arranged at different positions and are used for emitting acoustic wave signals.

[0060] The hydrophone array is arranged at different positions and is used for receiving echo signals reflected after the acoustic wave signals emitted by the sonar transmitters pass through a target.

[0061] As an optional implementation, the Ethernet is a 40G Ethernet, and the external processing machine is connected through a backplane Ethernet interface to perform distributed computing. Figure 2 As shown in the figure, the audio integrated signal processing system based on the VPX architecture comprises:

[0062] When the audio signal processing requirements are multiple and have an association relationship, the following steps are performed:

[0063] S201, performance parameters of each module in the external processing machine and audio signal processing task types are acquired.

[0064] S202, the audio signal processing requirements are matched with the audio signal processing task types to obtain a matching relationship between the audio signal processing requirements and the modules, and the matching relationship includes one-to-one, one-to-many, and many-to-one.

[0065] S203, the matching relationship is adjusted according to a preset recommendation mechanism, and the preset recommendation mechanism is formulated according to historical data and recommendation rules.

[0066] S204, a connection graph of the modules under the association relationship of the audio signal processing requirements is constructed according to the association relationship between the multiple audio signal processing requirements and the adjusted matching relationship.

[0067] S205, multiple groups of module combinations are formed according to the connection graph.

[0068] S206, dividing the modules in each module combination into a first module and a second module, the first module being a repeatedly occurring module and the second module being a single occurrence module;

[0069] S207, performing load performance analysis on the first module and the second module to determine performance parameters of each module in processing corresponding audio signal processing requirements, wherein the load performance analysis on the first module comprises inputting different audio signal processing requirements into a pre-established module performance prediction model of the first module to determine load performance parameters of the module in simultaneously processing different audio signal processing requirements;

[0070] S208, obtaining a score of the module combination according to the performance parameters of the first module and the performance parameters of the second module;

[0071] S209, selecting an optimal module combination according to the scores of each module combination.

[0072] Exemplarily, the performance parameters of each module and the audio signal processing task types are obtained in a manner of reading by a hardware interface or in a manner of calling by a software application interface, the audio signal processing requirements that cannot be completed by the system internally are matched with the audio signal processing task types to determine the task types corresponding to the requirements, some modules can only process one task, and some modules can complete multiple tasks simultaneously, that is, multiple audio signal processing requirements can be matched in the same module, and multiple modules can also be matched simultaneously for the same audio processing requirement, therefore, the matching relationship includes one-to-one, one-to-many, and many-to-one.

[0073] Since multiple matching relationships can result in multiple combinations, leading to excessive calculation, the embodiment proposes a recommendation mechanism to adjust the matching relationship, thereby reducing the number of combinations and reducing the calculation amount. The recommendation mechanism proposed in the embodiment can be that the system continuously records full-process data of past audio signal processing tasks, including task types, used module combinations, task execution results, and actual performance of each module in the task, such as processing speed, accuracy, and resource occupancy. Through statistical analysis of historical data, module combinations that frequently appear in the matching relationship and have good execution effects are identified, specifically, the frequency of different module combinations in successful tasks can be calculated according to an association rule mining algorithm such as the Apriori algorithm. At the same time, a performance trend model is established for each module using time series analysis methods to analyze the performance change law of the module under different task sizes and data volumes, such as the processing speed of module E being 1000 frames / second when processing small data volume audio signals, and the processing speed decreasing to 800 frames / second when the data volume doubles. A mathematical model of data volume and processing speed is established through regression analysis to predict the performance of the module in future tasks.

[0074] Then, short plate analysis is performed on the high-frequency module combination to determine the optimal processing capacity of the high-frequency module combination. According to the optimal processing capacity and the data amount corresponding to the audio signal processing requirement, the module combination satisfying the data amount of the audio signal processing requirement is screened out, and the module combination is bound as a recommended module combination. The recommended module combination can be a combination for completing two audio signal processing requirements. For other audio signal processing requirements, the matching relationship between the audio signal processing requirement and the module is determined, and therefore, the recommendation rule can be to replace the original requirement matching result with the recommended module combination. For example, there are five sequentially associated requirements, namely requirement 1-requirement 5. The modules matched with requirement 1 include module B and module C. The modules corresponding to requirement 2 include module B and module D. The modules matched with requirement 3 are module A and module E. The modules corresponding to requirement 4 are module F. The modules corresponding to requirement 5 are module G and module E. Through the above method, the recommended module combination screened out by requirement 2 and requirement 3 is module A and module B, and requirement 2 and requirement 3 are bound with module A and module B. Then, the readjusted matching relationship is: the modules matched with requirement 1 include module B and module C. Requirement 2 and requirement 3 are fixedly matched with module A and module B. Requirement 4 is matched with module F. Requirement 5 is matched with module G and module E.

[0075] According to the above adjusted matching relationship and the association relationship between the plurality of audio signal processing requirements, a connection graph between the modules is constructed, as shown in Figure 3 The connection graph of requirement 1-requirement 5 is formed. According to the connection graph, the following module combinations are formed:

[0076] Combination 1: [module B, module A, module B, module F, module G];

[0077] Combination 2: [module B, module A, module B, module F, module E];

[0078] Combination 3: [module C, module A, module B, module F, module G];

[0079] Combination 4: [module C, module A, module B, module F, module E];

[0080] The above modules are divided into a first module and a second module. Taking combination 1 as an example, the first module is module B, and the second module is module A, module F, and module G. Load analysis is performed on the first module and the second module. The load analysis method of the second module can be to input the data amount corresponding to the requirement into the performance trend model to analyze the performance of the module under the corresponding requirement, such as processing rate.

[0081] For the first module, since multiple requirements need to be processed in the first module, and although the processing of data requirements has a sequence in the process, but in the actual processing process, the data is continuous, the first module needs to have the ability to process multiple requirements at the same time, therefore, when determining the performance of the first module, different audio signal processing requirements need to be input into the pre-established module performance prediction model of the first module at the same time to determine the load performance parameters of the module when processing different audio signal processing requirements. The establishment method of the module performance prediction model can be that, first, collect historical performance data of the module when processing a large amount of different data, including processing time, accuracy, resource occupancy rate, etc. Then, use machine learning algorithms such as regression analysis, decision tree, neural network, etc. to train the historical data and establish a performance prediction model. Taking regression analysis as an example, the characteristic parameters of the data, such as data volume and data complexity, are used as input, and the performance indicators of the module, such as accuracy or processing rate, are used as output, and a regression model that can predict the performance of the module according to the input data characteristics is obtained through training. The module performance prediction model can also be constructed based on the algorithm principle, and the construction method of the module performance prediction model is not limited in the embodiment, and those skilled in the art can determine it according to the needs.

[0082] According to the performance parameters of the first module and the performance parameters of the second module, the score of the module combination can be obtained by first normalizing the performance parameters, then performing weighted summation on the normalized performance parameters to obtain the final score. For example, the performance parameters of the first module and the performance parameters of the second module are processing rates, the processing rates corresponding to all modules can be normalized, then the normalized processing rates are weighted and summed according to the importance of different modules to obtain the total processing rate of each module combination, and the module combination corresponding to the maximum total processing rate is selected as the optimal module combination.

[0083] The embodiment of the application provides an audio integrated signal processing system based on a VPX architecture, which obtains external processor module performance parameters and audio signal processing task types, matches audio signal processing requirements therewith, adjusts the matching relationship by using a preset recommendation mechanism, reduces unnecessary combinations, and reduces the amount of calculation. Based on the adjusted matching relationship and the demand association, a connection graph is constructed, which can clearly show the cooperation relationship between modules, is helpful for understanding the system structure, divides the modules and analyzes the load performance after forming multiple module combinations, obtains the module combination score in combination with the performance parameters, finally selects the optimal module combination, can improve the rationality of resource allocation of the audio signal processing system, fully develops the advantages of each module, improves the processing efficiency, at the same time ensures that the system can stably and efficiently run in different task scenarios, can also flexibly adjust the module combination according to the actual situation to adapt to changing requirements, and enhances the adaptability and scalability of the system.

[0084] Finally, it should be noted that the above preferred embodiments are merely intended to illustrate the technical solutions of the present application but not to limit the present application. Although the present application has been described in detail through the above preferred embodiments, those skilled in the art should understand that various modifications can be made in form and details without departing from the scope of the present application defined by the claims.

Claims

1. An audio integrated signal processing system based on VPX architecture, characterized in that, The application relates to a signal processing system. The signal processing system comprises: a chassis conforming to an OpenVPX standard, wherein a backboard is arranged in the chassis, and the backboard is integrated with a high-speed serial RapidIO bus and an Ethernet interface; a plurality of heterogeneous computing board cards, wherein the heterogeneous computing board cards are connected to the chassis through VPX slots, point-to-point communication is implemented between the board cards through the high-speed serial RapidIO bus, and the heterogeneous computing board cards comprise different types of computing chips and are used for executing different types of audio signal processing tasks; and a multi-machine stacking expansion module connected to the chassis through the Ethernet interface of the backboard, wherein when it is detected that the computing capacity of any of the heterogeneous computing board cards cannot meet the audio signal processing demand, the multi-machine stacking expansion module is used for connecting an external processing machine through the Ethernet interface of the backboard to perform distributed computing, and the external processing machine is stacked by a plurality of modules with audio signal processing functions. The multi-machine stacking expansion module comprises: an Ethernet interface assembly used for connecting the Ethernet interface of the backboard of the chassis to realize physical connection with the external processing machine; a data transmission circuit comprising a high-speed signal processing chip and a transmission route and used for pre-processing and transmitting data; a protocol conversion circuit used for realizing conversion between different protocols to enable the multi-machine stacking module to communicate with different types of external processing machines; a computing resource detection module used for monitoring the computing resource conditions of the heterogeneous computing board cards in the chassis; 2. The audio integrated signal processing system based on VPX architecture according to claim 1, characterized in that, and a logic controller used for sending a connection request and task allocation information to the external processing machine through the Ethernet interface of the backboard according to a preset strategy when the computing resource detection module detects that the computing resource is insufficient. The computing resource detection module performs the following steps, comprising: acquiring main line tasks of a signal processing system and a previous audio signal processing task and collecting current computing resource utilization rates of the heterogeneous computing board cards; determining residual computing resources according to the current computing resource utilization rates; determining a resource dynamic threshold value according to the main line tasks of the signal processing system, the previous audio signal processing task and a dynamic threshold value adaptive algorithm; when the residual computing resources are higher than the resource dynamic threshold value, the computing resource condition is that the computing resource is sufficient; 3. The audio integrated signal processing system based on VPX architecture according to claim 2, characterized in that, and when the residual computing resources are lower than or equal to the resource dynamic threshold value, the computing resource condition is that the computing resource is insufficient. The dynamic threshold value adaptive algorithm comprises: acquiring main line tasks of a signal processing system and a previous audio signal processing task; predicting at least one next audio signal processing task according to the main line tasks and the previous audio signal processing task; determining whether there is a real-time demand task in the at least one next audio signal processing task, determining a demand computing amount corresponding to the real-time demand task when there is the real-time demand task, and setting the resource dynamic threshold value according to the demand computing amount corresponding to the real-time demand task; when there is not the real-time demand task and the predicted next audio signal processing task comprises a plurality of audio signal processing tasks, calculating an average value of demand computing amounts of the plurality of audio signal processing tasks and setting the resource dynamic threshold value according to the average value of the demand computing amounts; and when there is not the real-time demand task and the predicted next audio signal processing task is one, determining a demand computing amount of the next audio signal processing task and setting the resource dynamic threshold value according to the demand computing amount.

4. The audio integrated signal processing system based on VPX architecture according to claim 1, characterized in that, The plurality of heterogeneous computing boards include at least two of a CPU board, a GPU board and an FPGA board, and communication bandwidths between the boards are symmetrical and consistent, and different types of acoustic signal processing tasks include underwater target detection, identification, positioning, underwater acoustic communication and underwater environment measurement.

5. The audio integrated signal processing system based on VPX architecture according to claim 1, characterized in that, Also comprising: A sonar transmitter arranged at different positions for emitting acoustic signals; An array of hydrophones arranged at different positions for receiving echo signals reflected by targets after the acoustic signals emitted by the sonar transmitter.

6. The audio integrated signal processing system based on VPX architecture according to any one of claims 1-5, characterized in that, The Ethernet is a 40G Ethernet, and an external processor is connected through a backplane Ethernet interface for distributed computing, including: When there are multiple acoustic signal processing requirements and there is a correlation relationship, the following steps are performed: Obtaining performance parameters of each module in the external processor and acoustic signal processing task types; Matching the acoustic signal processing requirements with the acoustic signal processing task types to obtain a matching relationship between the acoustic signal processing requirements and the modules, the matching relationship including one-to-one, one-to-many and many-to-one; Adjusting the matching relationship according to a preset recommendation mechanism, the preset recommendation mechanism being formulated according to historical data and recommendation rules; According to the correlation relationship between the plurality of acoustic signal processing requirements and the adjusted matching relationship, a connection graph of the modules under the correlation relationship of the acoustic signal processing requirements is constructed; According to the connection graph, a plurality of module combinations are formed; Dividing the modules in each module combination into first modules and second modules, the first modules being repeatedly appearing modules and the second modules being single appearing modules; Performing load performance analysis on the first modules and the second modules to determine performance parameters of each module when processing corresponding acoustic signal processing requirements, wherein the load performance analysis on the first modules includes inputting different acoustic signal processing requirements into a module performance prediction model previously established for the first modules to determine load performance parameters of the modules when simultaneously processing different acoustic signal processing requirements; Obtaining scores of the module combinations according to the performance parameters of the first modules and the performance parameters of the second modules; Selecting an optimal module combination according to the scores of the module combinations.

Images