Low-power wireless audio transmission methods, apparatus, devices and storage media

Abstract

The present invention provides a low-power wireless audio transmission method, device, equipment and storage medium, the method comprising: performing one-bit quantization sampling on multiple audio input signals through an audio transmitter, and obtaining a multiplexed audio data stream by time division multiplexing. Encoding the data stream according to a preset bit mapping rate and a super-Nyquist sampling rate using the encoder part of the autoencoder network to generate a coded sequence to be transmitted. Grouping the coded sequence according to a preset data packet size, setting the transmission power level according to the current battery power, and sending the grouped coded sequence to an audio receiving end. The receiving end performs multiple oversampling on the coded sequence, and reconstructs the signal through a decoder part to restore the multiple audio signals. The present invention combines super-Nyquist sampling and dynamic power regulation to achieve high-quality audio transmission under low power consumption. Through multiple oversampling and efficient decoding technology, high-fidelity restoration of the signal is ensured to meet the requirements of low power consumption and high-quality audio transmission.

Description

Technical Field

[0001] The present invention relates to the field of audio transmission, and in particular to a low-power wireless audio transmission method, device, equipment and storage medium. Background Art

[0002] With the widespread application of portable devices and the Internet of Things, low-power wireless audio transmission has become a key requirement in mobile devices, especially in battery-powered environments, where power consumption optimization is particularly important. However, in the process of achieving low-power transmission, traditional audio transmission methods often face the problem of shortened battery life while meeting the transmission distance and audio quality. In the prior art, low power consumption is usually achieved by reducing the signal sampling frequency or reducing the transmission power. However, these methods often lead to problems such as reduced audio signal quality and increased packet loss rate, and cannot take into account both audio quality and low power consumption requirements. Summary of the invention

[0003] The main purpose of the present invention is to solve the technical problem that the existing wireless audio transmission adopts reducing the signal sampling frequency or reducing the transmission power to achieve low power consumption, resulting in the degradation of audio signal quality, the increase of data packet loss rate, etc., and cannot take into account both audio quality and low power consumption requirements;

[0004] A first aspect of the present invention provides a low-power wireless audio transmission method, the low-power wireless audio transmission method comprising:

[0005] The audio transmitting end performs one-bit quantization sampling on the multiple audio input signals, and performs time division multiplexing processing on the quantized signals according to a preset time slot allocation mode to obtain a multiplexed audio data stream;

[0006] The encoder part in the preset autoencoder network encodes the multiplexed audio data stream according to a preset bit sequence mapping rate and a preset super-Nyquist sampling rate to obtain a coded sequence to be transmitted;

[0007] The to-be-transmitted coded sequence is grouped according to a preset data packet size, and a transmission power level is set based on a current battery power status of the audio transmitting end, and the grouped to-be-transmitted coded sequence is sent to the audio receiving end according to the transmission power level;

[0008] The received coding sequence is oversampled multiple times by the audio receiving end, and the decoder part of the autoencoder network is used to reconstruct the oversampled signal to obtain a restored multi-channel audio signal.

[0009] Optionally, in a first implementation of the first aspect of the present invention, the audio transmitting end performs one-bit quantization sampling on multiple audio input signals, and performs time division multiplexing processing on the quantized signals according to a preset time slot allocation mode to obtain a multiplexed audio data stream, including:

[0010] Performing one-bit analog-to-digital conversion sampling on multiple audio input signals through an audio transmitting end to obtain multiple quantized audio signals, wherein the sampling rate of the one-bit analog-to-digital conversion sampling is lower than a standard Nyquist sampling rate;

[0011] Performing noise reduction processing on the multi-channel quantized audio signals, and grading the processed signals according to preset signal priorities to obtain multi-channel noise reduction processed signals;

[0012] The multi-channel noise reduction processed signals are time-division multiplexed according to a preset time slot allocation mode to obtain the multiplexed audio data stream, wherein the preset time slot allocation mode includes a signal transmission time slot and a protection interval time slot of a fixed length, and the protection interval time slot is located between adjacent signal transmission time slots.

[0013] Optionally, in a second implementation of the first aspect of the present invention, the encoder part in the preset autoencoder network encodes the multiplexed audio data stream according to a preset bit sequence mapping rate and a preset super-Nyquist sampling rate to obtain a coded sequence to be transmitted, comprising:

[0014] Dividing the multiplexed audio data stream into a plurality of audio data blocks according to a preset block size;

[0015] Inputting the multiple audio data blocks into an encoder part in a preset autoencoder network, processing the multiple audio data blocks through each layer structure in the encoder part, and introducing soft quantization parameters to perform quantization constraints during the processing of each structure layer to obtain a soft quantization feature sequence;

[0016] Packing the soft quantization feature sequence according to the preset super-Nyquist sampling rate to obtain a time domain sampling sequence;

[0017] According to the preset spectrum efficiency constraint condition and the bit sequence mapping rate, the time domain sampling sequence is encoded to obtain a coded sequence to be transmitted.

[0018] Optionally, in a third implementation manner of the first aspect of the present invention, the inputting the multiple audio data blocks into an encoder part in a preset autoencoder network, processing the multiple audio data blocks through each layer structure in the encoder part, and introducing soft quantization parameters in the processing process of each structure layer to perform quantization constraints, and obtaining a soft quantization feature sequence includes:

[0019] Inputting the multiple audio data blocks into an encoder part in a preset autoencoder network, and performing dimension mapping on the multiple audio data blocks through an input layer of the encoder part to obtain an initial feature vector;

[0020] The initial feature vector is sequentially subjected to nonlinear transformation and normalization processing through a plurality of hidden layers in the encoder part, and a soft quantization parameter is introduced in each hidden layer processing process to obtain an intermediate feature vector;

[0021] The intermediate feature vector is dimensionally reconstructed through the output layer of the encoder part to obtain the soft quantization feature sequence.

[0022] Optionally, in a fourth implementation manner of the first aspect of the present invention, grouping the coded sequence to be transmitted according to a preset data packet size, setting a transmission power level based on a current battery power status of the audio transmitting end, and sending the grouped coded sequence to be transmitted to the audio receiving end according to the transmission power level includes:

[0023] Dividing the to-be-transmitted coded sequence into a plurality of data packets according to a preset data packet size to obtain a grouped coded sequence, wherein the data packet size includes a fixed-length data transmission segment and a header segment;

[0024] Detecting the current battery power status of the audio transmitting end, and selecting a corresponding transmission power level from a preset transmission power level group according to the detection result to obtain a current transmission power level;

[0025] The grouped coded sequence is sent to the audio receiving end in a burst transmission mode according to the transmission power level, wherein the burst transmission mode includes a data packet transmission phase and a sleep phase, and the duration of the sleep phase is determined by the processing time of the next data packet.

[0026] Optionally, in a fifth implementation of the first aspect of the present invention, the step of performing multiple oversampling on the received coding sequence by the audio receiving end, and reconstructing the oversampled signal using the decoder part of the autoencoder network to obtain the restored multi-channel audio signal includes:

[0027] Performing multiple oversampling processing on the received coding sequence through the audio receiving end to obtain an oversampled signal sequence, wherein the multiple of the multiple oversampling is greater than the preset super-Nyquist sampling rate;

[0028] Inputting the oversampled signal sequence into the decoder part of the autoencoder network for processing, and reconstructing the features of the oversampled signal sequence through the decoder part to obtain a reconstructed feature sequence;

[0029] The reconstructed characteristic sequence is subjected to time division multiplexing processing according to the preset time slot allocation mode to obtain a restored multi-channel audio signal.

[0030] Optionally, in a sixth implementation manner of the first aspect of the present invention, inputting the oversampled signal sequence into the decoder part of the autoencoder network for processing, and performing feature reconstruction on the oversampled signal sequence by the decoder part to obtain the reconstructed feature sequence includes:

[0031] Inputting the oversampled signal sequence into the decoder part of the autoencoder network, and performing dimension mapping on the oversampled signal sequence through the input layer of the decoder part to obtain an initial decoding feature vector;

[0032] Performing nonlinear transformation processing on the initial decoding feature vector in sequence through a plurality of hidden layers in the decoder part to obtain an intermediate decoding feature vector;

[0033] The intermediate decoded feature vector is dimensionally reconstructed through the output layer of the decoder part to obtain a reconstructed feature sequence.

[0034] A second aspect of the present invention provides a low-power wireless audio transmission device, the low-power wireless audio transmission device comprising:

[0035] A multiplexing module is used to perform one-bit quantization sampling on multiple audio input signals through an audio transmitter, and perform time division multiplexing processing on the quantized signals according to a preset time slot allocation mode to obtain a multiplexed audio data stream;

[0036] An encoding module, used for encoding the multiplexed audio data stream according to a preset bit sequence mapping rate and a preset super-Nyquist sampling rate through an encoder part in a preset autoencoder network to obtain a coded sequence to be transmitted;

[0037] A sending module, used for grouping the to-be-transmitted coded sequence according to a preset data packet size, setting a transmission power level based on a current battery power status of the audio sending end, and sending the grouped to-be-transmitted coded sequence to the audio receiving end according to the transmission power level;

[0038] The decoding and reconstruction module is used to perform multiple oversampling on the received coding sequence through the audio receiving end, and use the decoder part of the autoencoder network to reconstruct the oversampled signal to obtain a restored multi-channel audio signal.

[0039] A third aspect of the present invention provides a low-power wireless audio transmission device, comprising: a memory and at least one processor, wherein instructions are stored in the memory, and the memory and the at least one processor are interconnected via a line; the at least one processor calls the instructions in the memory so that the low-power wireless audio transmission device performs the steps of the above-mentioned low-power wireless audio transmission method.

[0040] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the above-mentioned low-power wireless audio transmission method.

[0041] The above-mentioned low-power wireless audio transmission method, device, equipment and storage medium perform one-bit quantization sampling on multiple audio input signals through the audio transmitter, and use time division multiplexing to obtain a multiplexed audio data stream. The encoder part of the autoencoder network is used to encode the data stream according to the preset bit mapping rate and super-Nyquist sampling rate to generate a coded sequence to be transmitted. The coded sequence is grouped according to the preset data packet size, and the transmission power level is set according to the current battery power, and the grouped coded sequence is sent to the audio receiving end. The receiving end performs multiple oversampling on the coded sequence, and reconstructs the signal through the decoder part to restore the multi-channel audio signal. The present invention combines super-Nyquist sampling and dynamic power regulation to achieve high-quality audio transmission under low power consumption. Through multiple oversampling and efficient decoding technology, high-fidelity restoration of the signal is ensured to meet the requirements of low power consumption and high-quality audio transmission.

[0042] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A schematic diagram of a first embodiment of a low-power wireless audio transmission method according to an embodiment of the present invention;

[0045] Figure 2 A schematic diagram of an embodiment of a low-power wireless audio transmission device in an embodiment of the present invention;

[0046] Figure 3 The figure is a schematic diagram of an embodiment of a low-power wireless audio transmission device in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] The terms "including" and "having" and any variations thereof mentioned in the embodiments of the present invention are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device end including a series of steps or units is not limited to the listed steps or units, but may optionally include other steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or device ends.

[0049] To facilitate understanding of this embodiment, a low-power wireless audio transmission method disclosed in an embodiment of the present invention is first described in detail. Figure 1 As shown, the method comprises the following steps:

[0050] 101. Performing one-bit quantization sampling on multiple audio input signals through an audio transmitting end, and performing time division multiplexing processing on the quantized signals according to a preset time slot allocation mode to obtain a multiplexed audio data stream;

[0051] In one embodiment of the present invention, the step of performing one-bit quantization sampling on multiple audio input signals through an audio transmitting end, and performing time-division multiplexing processing on the quantized signals according to a preset time slot allocation mode to obtain a multiplexed audio data stream includes: performing one-bit analog-to-digital conversion sampling on multiple audio input signals through an audio transmitting end to obtain multiple quantized audio signals, wherein the sampling rate of the one-bit analog-to-digital conversion sampling is lower than the standard Nyquist sampling rate; performing noise reduction processing on the multiple quantized audio signals, and grading the processed signals according to preset signal priorities to obtain multiple noise reduction processed signals; performing time-division multiplexing processing on the multiple noise reduction processed signals according to a preset time slot allocation mode to obtain the multiplexed audio data stream, wherein the preset time slot allocation mode includes a fixed-length signal transmission time slot and a guard interval time slot, and the guard interval time slot is located between adjacent signal transmission time slots.

[0052] Specifically, in the process of processing multiple audio input signals, it is first necessary to sample the multiple audio signals through a one-bit analog-to-digital converter. When sampling with one-bit analog-to-digital conversion, the sampling rate is set to a value lower than the standard Nyquist sampling rate. For example, when the highest frequency of the audio signal is 20kHz, the sampling rate is set to 16kHz. Although such a sampling rate setting is lower than the standard Nyquist sampling rate of 40kHz, it is sufficient for the transmission of voice signals. When sampling, the one-bit analog-to-digital converter only judges the positive and negative amplitude of the input signal and outputs two quantization values ​​of 1 or -1, thereby obtaining multiple quantized audio signals. This one-bit quantization method significantly reduces the power consumption of analog-to-digital conversion and simplifies the subsequent signal processing circuit. For multiple audio input signals, each signal adopts the same one-bit analog-to-digital conversion sampling method, and the sampling moment is controlled by a unified clock signal to ensure the sampling synchronization of multiple signals. When performing one-bit analog-to-digital conversion, the input analog audio signal is first sampled using a sample-and-hold circuit, and then the sampled value is compared with the zero level through a comparator, and finally a binary digital signal is output. This simple one-bit quantization structure not only reduces circuit complexity, but also ensures the low latency characteristics of the quantization process. At the same time, since the sampling rate is lower than the standard Nyquist sampling rate, the power consumption of the sampling process is further reduced.

[0053] Specifically, after obtaining multiple quantized audio signals, these signals need to be subjected to noise reduction processing and priority classification. Since one-bit quantization will introduce quantization noise, and there is background noise interference in the wireless transmission environment, it is very necessary to perform noise reduction processing on the quantized audio signals. Noise reduction processing adopts basic signal processing methods to judge and remove noise by analyzing the amplitude distribution characteristics of the signal. When performing noise reduction processing, the input quantized audio signal needs to be framed first, and the length of each frame signal is determined according to the characteristics of the audio signal. Then, noise estimation is performed on each frame signal to evaluate the noise level in the current frame. Based on the noise estimation result, the threshold judgment method is used to remove the noise component, and the setting of the threshold value needs to balance between noise removal and signal fidelity. After noise reduction, the processed signals are graded according to the preset signal priority, and the signal priority is determined by factors such as the importance of the signal and the delay requirements. The priority division adopts a multi-level structure, which is usually divided into three levels: high, medium, and low. High-priority signals are usually audio signals with high real-time requirements or high content importance. These signals will be preferentially arranged at the front end of the transmission sequence in subsequent time division multiplexing. The priority of each signal needs to be determined by comprehensively considering multiple factors such as the signal's bandwidth requirements, delay tolerance, and service type, and ultimately determining the priority of the signal. This classification mechanism provides a basis for subsequent time-division multiplexing allocation, ensuring that important signals receive priority transmission rights.

[0054] Specifically, after completing the noise reduction and classification of the signal, it is necessary to perform time division multiplexing on the multi-channel noise reduction processed signals according to the preset time slot allocation mode. Time division multiplexing adopts a fixed time slot allocation mode, and each signal is allocated to a fixed length of signal transmission time slot. When determining the length of the signal transmission time slot, factors such as the sampling rate, data volume and transmission bandwidth of the audio signal need to be considered. Signals with higher priority are usually allocated a larger time slot length to ensure the transmission quality of the signal. A guard interval time slot is set between adjacent signal transmission time slots, and the length of the guard interval time slot is set according to a fixed ratio according to the length of the signal transmission time slot, usually 10% to 20% of the length of the signal transmission time slot. The existence of the guard interval time slot avoids crosstalk between adjacent signals and improves the reliability of signal transmission. When performing time division multiplexing, the transmission order of the signal is first determined according to the priority of each signal, and then each signal is inserted into the corresponding time slot in sequence according to the preset time slot allocation mode. For each signal transmission time slot, a synchronization mark and a time slot identifier need to be added for signal identification and demultiplexing at the receiving end. The output of the time-division multiplexing process is a multiplexed data stream containing all audio signal information. This data stream has a strict timing structure, and the position and length of each time slot are fixed. This structured data organization method is conducive to subsequent encoding and transmission processing.

[0055] 102. Encoding the multiplexed audio data stream according to a preset bit sequence mapping rate and a preset super-Nyquist sampling rate through an encoder part in a preset autoencoder network to obtain a code sequence to be transmitted;

[0056] In one embodiment of the present invention, the encoding process of the multiplexed audio data stream according to a preset bit sequence mapping rate and a preset super-Nyquist sampling rate by the encoder part in the preset autoencoder network to obtain the coded sequence to be transmitted includes: dividing the multiplexed audio data stream into a plurality of audio data blocks according to a preset block size; inputting the plurality of audio data blocks into the encoder part in the preset autoencoder network, processing the plurality of audio data blocks through each layer structure in the encoder part, and introducing soft quantization parameters to perform quantization constraints in the processing of each structure layer to obtain a soft quantization feature sequence; signal packaging of the soft quantization feature sequence according to the preset super-Nyquist sampling rate to obtain a time domain sampling sequence; encoding the time domain sampling sequence according to a preset spectral efficiency constraint and a bit sequence mapping rate to obtain a coded sequence to be transmitted.

[0057] Specifically, the processing of the multiplexed audio data stream first requires data segmentation, and the continuous data stream is divided into multiple audio data blocks according to the preset block size. The setting of the block size needs to take into account the processing power and memory limitations of the autoencoder network, and the data stream is usually divided into data blocks of equal length. When performing data segmentation, a frame synchronization mark needs to be added to the boundary of each data block to ensure the integrity and boundary alignment of the data block. The data segmentation process also includes formatting the data, reorganizing the time-division multiplexed data stream into a data structure suitable for autoencoder processing. The size of the data block directly affects the complexity and delay of subsequent processing. A larger block size is conducive to improving processing efficiency but will increase system delay, while a smaller block size is conducive to reducing delay but will increase processing overhead. The multiple audio data blocks obtained by block processing constitute a data sequence, and each data block contains complete time-division multiplexing information. At the same time, block processing also needs to consider the boundary effect of the data, and padding data is added between data blocks when necessary to ensure the integrity of the data blocks.

[0058] Specifically, the divided audio data blocks are input into the encoder part of the autoencoder network for processing. This processing process needs to be completed in sequence through each structural layer in the encoder. Each layer of the encoder performs nonlinear transformation on the input data and maps the original signal space to the new feature space. During the processing, each structural layer introduces a soft quantization parameter for quantization constraint, and the value range of the soft quantization parameter is controlled between 0 and 1. The purpose of introducing the soft quantization parameter is to solve the gradient problem caused by one-bit quantization, because the derivative of one-bit quantization is undefined at zero value and zero at non-zero value, which makes it difficult to train the network. By introducing the soft quantization parameter, the quantization process becomes smooth, which helps the training convergence of the network. It should be noted that the soft quantization parameter is only used in the training stage, and the real one-bit quantization is used in actual applications. The soft quantization feature sequence obtained by this processing method retains the key features of the original signal and has the characteristics suitable for one-bit quantization.

[0059] Specifically, when the soft quantization feature sequence is packaged, the super-Nyquist sampling rate is used for processing. Super-Nyquist sampling refers to the transmission of multiple symbols within a Nyquist sampling interval, which breaks the limitations of the traditional Nyquist sampling theorem. The setting of the super-Nyquist sampling rate needs to consider the bandwidth limitation of the system and the influence of inter-symbol interference. The sampling rate is usually set to an integer multiple of the standard Nyquist sampling rate. When packaging the signal, the data needs to be reorganized and the soft quantization feature sequence is rearranged according to the requirements of the super-Nyquist sampling rate. This process will generate inter-symbol interference, which is processed by a trained autoencoder network. The time domain sampling sequence obtained after the packaging process is a high time resolution sequence containing multiple symbols transmitted within the traditional Nyquist sampling interval.

[0060] Specifically, the time domain sampling sequence needs to be encoded according to the spectrum efficiency constraint and the bit sequence mapping rate. The spectrum efficiency constraint is determined by the product of the bit sequence mapping rate and the super-Nyquist sampling rate. This product cannot exceed twice the logarithm of the multiple oversampling rate at the receiving end. The bit sequence mapping rate is a comprehensive parameter that reflects the ratio of the length of the input bit to the transmission sequence. During the encoding process, it is necessary to ensure that the encoded sequence meets the spectrum efficiency constraint. This encoding process actually maps the time domain sampling sequence to a codeword space that meets the transmission requirements. The resulting coded sequence to be transmitted not only meets the spectrum efficiency constraint, but also maintains the recoverability of the signal. This encoding method makes full use of the optimal coding strategy learned by the autoencoder network during the training process, making the encoded sequence suitable for transmission in a one-bit quantization and oversampling system.

[0061] Furthermore, the step of inputting the multiple audio data blocks into the encoder part of a preset autoencoder network, processing the multiple audio data blocks through each layer structure in the encoder part, and introducing soft quantization parameters to perform quantization constraints during the processing of each structure layer to obtain a soft quantization feature sequence includes: inputting the multiple audio data blocks into the encoder part of a preset autoencoder network, dimensionally mapping the multiple audio data blocks through the input layer of the encoder part to obtain an initial feature vector; sequentially performing nonlinear transformation and normalization processing on the initial feature vector through multiple hidden layers in the encoder part, and introducing soft quantization parameters during the processing of each hidden layer to obtain an intermediate feature vector; and dimensionally reconstructing the intermediate feature vector through the output layer of the encoder part to obtain the soft quantization feature sequence.

[0062] Specifically, when multiple audio data blocks are input into the encoder part of the autoencoder network, dimension mapping is first performed through the input layer. The input layer receives audio data blocks divided according to the preset block size, which contain time-division multiplexed multi-channel audio signal information. The dimension mapping process of the input layer is to map the input data from the original signal space to the feature space, which involves the transformation of the data dimension. During dimension mapping, the input data is first rearranged to reorganize the time series data into a format suitable for feature extraction. Then the data is mapped to a new feature space through linear transformation, and the dimension of this feature space is determined by the network structure. The result of dimension mapping is the initial feature vector, which contains the basic feature information of the original signal. When performing dimension mapping, it is necessary to ensure that the format of the input data meets the network processing requirements, including data normalization and alignment. Dimension mapping also needs to consider the balance between computational complexity and memory usage and select the appropriate mapping dimension. At the same time, the dimension mapping process should not introduce additional information loss and needs to maintain the key features of the signal. The dimension mapping of the input layer actually completes the preliminary transformation of the data, laying the foundation for subsequent feature extraction.

[0063] Specifically, after the initial feature vector is obtained, it is processed in sequence through multiple hidden layers in the encoder part. Each hidden layer performs nonlinear transformation and normalization on the input feature vector, and introduces soft quantization parameters in the processing process. In each hidden layer, the input feature vector is first linearly transformed to obtain an intermediate result. Then, activation functions such as ReLU are used to perform nonlinear transformation to enhance the representation ability of the network. While performing nonlinear transformation, soft quantization parameters are introduced to quantize the data. The soft quantization parameters are introduced by performing soft quantization processing based on the output result of the activation function through the formula Q(yl) = yl / |yl+ε|, where yl is the output value of the activation function and ε is a soft quantization parameter with a value range of 0 to 1. This soft quantization processing makes the quantization process smooth and differentiable, and solves the gradient problem of one-bit quantization during the training process. After the processing is completed, the data is normalized and scaled to a suitable range. This hierarchical feature extraction process gradually constructs a multi-level representation of the signal, and each layer extracts feature information at different levels. The serial processing of multiple hidden layers obtains an intermediate feature vector, which contains the high-level semantic features of the signal. The number of hidden layers and the number of nodes in each layer need to be determined according to the actual application requirements to balance processing performance and computational complexity. The selection of soft quantization parameters has an important impact on the training effect of the network, and the optimal value needs to be determined through experiments.

[0064] Specifically, the intermediate feature vector is reconstructed through the output layer of the encoder part, and the extracted features are remapped to the target signal space. The dimension reconstruction process of the output layer needs to consider the requirements of subsequent processing, including the requirements of super-Nyquist sampling and one-bit quantization. When reconstructing the dimension, the intermediate feature vector is first linearly transformed and mapped to a suitable dimensional space. The reconstructed dimension needs to be consistent with the original signal space to ensure the integrity of the information. In the process of dimension reconstruction, the influence of soft quantization parameters also needs to be considered so that the output signal has the characteristics suitable for one-bit quantization. The soft quantization feature sequence obtained by the output layer directly determines the effect of subsequent signal processing, so the accuracy of the reconstruction process needs to be ensured. The process of dimension reconstruction also needs to consider the timing characteristics of the signal to ensure that the reconstructed signal can correctly reflect the timing relationship of the original signal. At the same time, the reconstruction process must meet the constraints of spectral efficiency to ensure that the output sequence is suitable for transmission under a given bandwidth. The output layer is the last processing link of the encoder, and the soft quantization feature sequence output by it needs to meet both the requirements of information retention and quantization characteristics. The result of dimension reconstruction directly affects the subsequent super-Nyquist sampling and encoding processing, so it is necessary to optimize these requirements during the network training stage. The soft quantization feature sequence obtained by dimensional reconstruction not only retains the key information of the original signal, but also has characteristics suitable for subsequent processing.

[0065] 103. Grouping the code sequence to be transmitted according to a preset data packet size, setting a transmission power level based on a current battery power status of the audio transmitting end, and sending the grouped code sequence to be transmitted to the audio receiving end according to the transmission power level;

[0066] In one embodiment of the present invention, the coding sequence to be transmitted is grouped according to a preset data packet size, and a transmission power level is set based on the current battery power status of the audio transmitting end, and the grouped coding sequence to be transmitted is sent to the audio receiving end according to the transmission power level, including: dividing the coding sequence to be transmitted into multiple data packets according to a preset data packet size to obtain a grouped coding sequence, wherein the data packet size includes a data transmission segment and a header segment of a fixed length; detecting the current battery power status of the audio transmitting end, and selecting a corresponding transmission power level from a preset transmission power level group according to the detection result to obtain a current transmission power level; and sending the grouped coding sequence to the audio receiving end in a burst transmission mode according to the transmission power level, wherein the burst transmission mode includes a data packet transmission phase and a sleep phase, and the duration of the sleep phase is determined by the processing time of the next data packet.

[0067] Specifically, the processing of the transmission coding sequence first needs to be grouped according to the preset data packet size. The setting of the data packet size needs to consider the balance between transmission efficiency and system complexity. The data packet is usually divided into two parts: a data transmission segment and a header segment of fixed length. The header segment contains necessary control information, such as synchronization mark, sequence number, data packet length, etc., which are used for data packet identification and reorganization at the receiving end. The length of the data transmission segment is determined by the system's processing power and transmission bandwidth. Too long data packets will increase the probability of transmission errors, and too short data packets will reduce transmission efficiency. When dividing the data packets, it is necessary to ensure that the data boundaries are aligned, and padding bits are added at the end of the data packet if necessary. The division of the data packet also needs to consider the requirements of the burst transmission mode to ensure that the transmission time of a single data packet is within a reasonable range. The grouped coded sequence obtained by data packet division is a regular data packet sequence, and each data packet has the same structure. This unified format is conducive to subsequent transmission control.

[0068] Detecting the current battery power status of the audio transmitter is a key step in determining the transmit power level. The battery power status is detected by sampling the battery voltage and current parameters, and the current power level is calculated based on the sampled parameters. The power status detection needs to take into account the discharge characteristics of the battery, and the voltage-power relationship of different types of batteries may be different. According to the detected power status, select the appropriate transmit power level from the preset transmit power level group. The transmit power level group usually includes three levels: high, medium, and low, and each level corresponds to a different transmit power value. The selection of power level needs to balance the relationship between transmission distance and power consumption, and try to use a lower power level while ensuring communication quality. Power status detection is a continuous process. The system will regularly update the power status information and adjust the transmit power level as needed. The current transmit power level obtained in this way takes into account both communication needs and battery life.

[0069] According to the determined transmit power level, the burst transmission mode is used to send the grouped coded sequence. The burst transmission mode includes two parts: the data packet transmission stage and the sleep stage. In the data packet transmission stage, the system sends a complete data packet at the current transmit power level, and the transmission process adopts a fixed data rate. After sending a data packet, the system immediately enters the sleep stage and shuts down most of the circuits to reduce power consumption. The duration of the sleep stage is determined by the processing time of the next data packet to be sent, which includes the time required for the generation and encoding of the data packet. The method of determining the sleep duration by the processing time ensures the continuity of data transmission. After the next data packet is processed, the system is awakened and starts a new round of data packet transmission. This alternating burst transmission and sleep working mode significantly reduces the average power consumption of the system. During the burst transmission process, the system will continuously monitor the feedback information from the receiving end and adjust the transmit power level or transmission parameters when necessary. The use of the burst transmission mode enables the system to minimize power consumption while ensuring transmission performance, which is particularly suitable for battery-powered portable audio devices.

[0070] 104. The received coding sequence is oversampled multiple times by the audio receiving end, and the oversampled signal is reconstructed by using the decoder part of the autoencoder network to obtain a restored multi-channel audio signal.

[0071] In one embodiment of the present invention, the audio receiving end performs multiple oversampling on the received coding sequence, and uses the decoder part of the autoencoder network to reconstruct the oversampled signal to obtain a restored multi-channel audio signal, including: performing multiple oversampling on the received coding sequence by the audio receiving end to obtain an oversampled signal sequence, wherein the multiple of the multiple oversampling is greater than the preset super-Nyquist sampling rate; inputting the oversampled signal sequence into the decoder part of the autoencoder network for processing, and performing feature reconstruction on the oversampled signal sequence by the decoder part to obtain a reconstructed feature sequence; performing time division multiplexing on the reconstructed feature sequence according to the preset time slot allocation mode to obtain a restored multi-channel audio signal.

[0072] Specifically, the audio receiving end first needs to perform multiple oversampling on the received coded sequence. The main purpose of oversampling is to provide sufficient time domain resolution to restore the signal after one-bit quantization. The oversampling multiple needs to be set greater than the super-Nyquist sampling rate used by the transmitter. Such a setting ensures that the receiving end can obtain enough information to restore the original signal. During the oversampling process, the clock of the received coded sequence is first synchronized to ensure the accuracy of the sampling time. Then, multiple sampling points are inserted between every two adjacent coded sequence samples, and the number of inserted sampling points is determined by the oversampling multiple. The implementation of oversampling requires the use of a high-precision sampling clock to ensure the uniformity and accuracy of sampling. The signal sequence obtained by oversampling provides more detailed time domain information, which is crucial for subsequent signal reconstruction. The signal sequence after oversampling also retains the timing relationship in the original signal, which is very important for maintaining the continuity of the audio signal. The oversampling process at the receiving end actually increases the redundancy of the signal in the time domain. This redundant information helps to overcome the information loss caused by one-bit quantization.

[0073] Specifically, when reconstructing the oversampled signal sequence, it needs to be input into the decoder part of the autoencoder network. The processing of the decoder part includes multiple levels of feature extraction and reconstruction. First, the input layer of the decoder receives the oversampled signal sequence and performs preliminary feature mapping. Then, the signal is processed step by step through multiple hidden layers, each of which is responsible for extracting signal features at different scales. During the feature extraction process, the decoder will use the parameters learned in the training phase to gradually restore the features of the original signal. Nonlinear activation functions are used between the layers of the decoder, which help the network capture the nonlinear features in the signal. Finally, the extracted features are remapped to the signal space through the output layer to obtain the reconstructed feature sequence. This reconstruction process is actually the inverse process of the encoding process, and the key features of the original signal need to be accurately restored. The soft quantization parameter does not need to be used in the decoder processing because this parameter is only used in the training phase. The reconstructed feature sequence contains the complete information of the original multi-channel audio signal, which is ready for the subsequent demultiplexing process.

[0074] Specifically, the reconstructed feature sequence needs to be time-division multiplexed according to the preset time slot allocation mode. The purpose of time-division multiplexing is to restore the reconstructed feature sequence to multiple independent audio signals. The demultiplexing process first needs to identify the time slot boundaries in the feature sequence, and this identification process depends on the preset time slot allocation mode. Each time slot contains a fixed-length signal transmission segment and a protection interval segment. According to this time slot structure, the system can accurately locate the position of each audio signal in the sequence. Then, the signals in different time slots are extracted separately to obtain multiple independent audio signals. During the extraction process, the protection interval segment needs to be removed and only the valid signal transmission segment is retained. Each audio signal obtained by demultiplexing is a complete one-bit quantized audio stream, which can be directly used for audio playback or further processing. The time-division multiplexing process needs to ensure the accuracy of signal extraction and avoid signal aliasing caused by time slot identification errors. The restored multi-channel audio signals obtained in this way maintain the independence and integrity of the original signals. Time-division multiplexing is the last link of the receiving end processing, and its processing effect directly determines the overall performance of the system.

[0075] Furthermore, the step of inputting the oversampled signal sequence into the decoder part of the autoencoder network for processing, and reconstructing the features of the oversampled signal sequence through the decoder part to obtain a reconstructed feature sequence includes: inputting the oversampled signal sequence into the decoder part of the autoencoder network, and dimensionally mapping the oversampled signal sequence through the input layer of the decoder part to obtain an initial decoding feature vector; sequentially performing nonlinear transformation processing on the initial decoding feature vector through multiple hidden layers in the decoder part to obtain an intermediate decoding feature vector; and dimensionally reconstructing the intermediate decoding feature vector through the output layer of the decoder part to obtain a reconstructed feature sequence.

[0076] Specifically, when the oversampled signal sequence is input into the decoder part of the autoencoder network, it is first necessary to perform dimension mapping through the input layer. After one-bit quantization and multiple oversampling, the oversampled signal sequence is already a high-dimensional time domain signal. The dimension mapping process of the input layer first needs to reorganize this high-dimensional signal and convert the time series data into a format suitable for feature extraction. Then, the data is mapped to the feature space through a linear transformation. The weight matrix of the linear transformation has been determined in the network training phase, and these weights contain the key information required for signal reconstruction. The output of the dimension mapping is the initial decoded feature vector, which represents the initial representation of the signal in the feature space. During the processing of the input layer, it is necessary to ensure the uniformity of the data format to ensure that the subsequent hidden layer can process it correctly. The dimension mapping of the input layer is actually the process of converting the quantized and oversampled time domain signal into a feature domain representation. This process needs to consider the needs of subsequent processing to ensure that the dimension and format of the feature vector meet the processing requirements of the hidden layer. The initial decoded feature vector obtained by the input layer contains the basic feature information of the signal, laying the foundation for subsequent deep feature extraction. The processing at this stage is crucial to the success of the entire decoding process, so it is necessary to ensure the accuracy of the dimension mapping.

[0077] After obtaining the initial decoded feature vector, it is necessary to perform deep feature extraction through multiple hidden layers in the decoder part. Each hidden layer performs nonlinear transformation processing on the input feature vector. These nonlinear transformations help the network capture the complex feature relationships in the signal. The hidden layer processing adopts a hierarchical structure, and each layer is responsible for extracting features of different scales and types. In each hidden layer, the input features are first linearly transformed, and then nonlinear characteristics are introduced through activation functions. Activation functions usually take the form of ReLU, etc. These functions can effectively extract the nonlinear characteristics of the signal. The series processing of multiple hidden layers forms a hierarchical structure for feature extraction, and each layer extracts higher-level features based on the previous layer. The intermediate decoded feature vector obtained by multi-layer processing contains the multi-scale feature representation of the signal. The number of hidden layers and the number of nodes in each layer are determined during network design. The selection of these parameters requires a balance between reconstruction performance and computational complexity. The intermediate decoded feature vector is actually a complete representation of the signal in the feature space, containing all the information required to reconstruct the original signal. These features include both local details of the signal and global structural features, providing a sufficient information basis for the final signal reconstruction.

[0078] The final stage of signal reconstruction is to re-dimensionalize the intermediate decoded feature vectors through the output layer of the decoder. The main task of the output layer is to map the representation of the feature space back to the original signal space. The dimension reconstruction process first needs to determine the dimension of the output signal, which needs to match the dimension of the original audio signal. The reconstruction process uses a linear transformation to map the intermediate feature vector to the target dimensional space. The parameters of the linear transformation have been optimized during the network training phase, and these parameters can accurately map the features to signals. The reconstructed feature sequence obtained by dimension reconstruction is a time domain signal that retains the key features of the original audio signal. The processing of the output layer needs to consider the timing characteristics of the signal to ensure the continuity and smoothness of the reconstructed signal. Dimension reconstruction is a key step, and its accuracy directly determines the quality of the final reconstructed signal. The reconstructed feature sequence obtained by the output layer provides the basis for the subsequent time-decomposition multiplexing processing. The processing at this stage needs to pay special attention to the boundary effects of the signal to avoid introducing unnecessary distortion in the signal reconstruction process. The reconstructed feature sequence obtained by dimension reconstruction has the same time domain characteristics as the original signal, which is very important for maintaining the perceptual quality of the audio signal.

[0079] In this embodiment, the multi-channel audio input signal is sampled by one bit quantization through the audio transmitting end, and a multiplexed audio data stream is obtained by time division multiplexing. The encoder part of the autoencoder network is used to encode the data stream according to the preset bit mapping rate and the super-Nyquist sampling rate to generate a coded sequence to be transmitted. The coded sequence is grouped according to the preset data packet size, and the transmission power level is set according to the current battery power, and the grouped coded sequence is sent to the audio receiving end. The receiving end performs multiple oversampling on the coded sequence, and reconstructs the signal through the decoder part to restore the multi-channel audio signal. The present invention combines super-Nyquist sampling and dynamic power regulation to achieve high-quality audio transmission with low power consumption. Through multiple oversampling and efficient decoding technology, high-fidelity restoration of the signal is ensured to meet the requirements of low power consumption and high-quality audio transmission.

[0080] The low-power wireless audio transmission method in the embodiment of the present invention is described above. The low-power wireless audio transmission device in the embodiment of the present invention is described below. Figure 2 , an embodiment of the low-power wireless audio transmission device in the embodiment of the present invention includes:

[0081] A multiplexing module 201 is used to perform one-bit quantization sampling on multiple audio input signals through an audio transmitting end, and perform time division multiplexing processing on the quantized signals according to a preset time slot allocation mode to obtain a multiplexed audio data stream;

[0082] The encoding module 202 is used to encode the multiplexed audio data stream according to a preset bit sequence mapping rate and a preset super-Nyquist sampling rate through an encoder part in a preset autoencoder network to obtain a coded sequence to be transmitted;

[0083] The sending module 203 is used to group the code sequence to be transmitted according to a preset data packet size, set a transmission power level based on the current battery power status of the audio sending end, and send the grouped code sequence to be transmitted to the audio receiving end according to the transmission power level;

[0084] The decoding and reconstruction module 204 is used to perform multiple oversampling on the received coding sequence through the audio receiving end, and reconstruct the oversampled signal using the decoder part of the autoencoder network to obtain a restored multi-channel audio signal.

[0085] In an embodiment of the present invention, the low-power wireless audio transmission device runs the above-mentioned low-power wireless audio transmission method, and the low-power wireless audio transmission device performs one-bit quantization sampling on multiple audio input signals through an audio transmitter, and uses time division multiplexing to obtain a multiplexed audio data stream. The encoder part of the autoencoder network is used to encode the data stream according to a preset bit mapping rate and a super-Nyquist sampling rate to generate a coded sequence to be transmitted. The coded sequence is grouped according to the preset data packet size, and the transmission power level is set according to the current battery power, and the grouped coded sequence is sent to the audio receiving end. The receiving end performs multiple oversampling on the coded sequence, and reconstructs the signal through the decoder part to restore the multi-channel audio signal. The present invention combines super-Nyquist sampling and dynamic power regulation to achieve high-quality audio transmission under low power consumption. Through multiple oversampling and efficient decoding technology, high-fidelity restoration of the signal is ensured to meet the requirements of low power consumption and high-quality audio transmission.

[0086] above Figure 2 The low-power wireless audio transmission device in the embodiment of the present invention is described in detail from the perspective of modular functional entities. The low-power wireless audio transmission device in the embodiment of the present invention is described in detail from the perspective of hardware processing.

[0087] Figure 33 is a schematic diagram of the structure of a low-power wireless audio transmission device provided by an embodiment of the present invention. The low-power wireless audio transmission device 300 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 310 (for example, one or more processors) and a memory 320, and one or more storage media 330 (for example, one or more mass storage device terminals) storing application programs 333 or data 332. Among them, the memory 320 and the storage medium 330 can be short-term storage or permanent storage. The program stored in the storage medium 330 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations in the low-power wireless audio transmission device 300. Furthermore, the processor 310 may be configured to communicate with the storage medium 330, and execute a series of instruction operations in the storage medium 330 on the low-power wireless audio transmission device 300 to implement the steps of the above-mentioned low-power wireless audio transmission method.

[0088] The low-power wireless audio transmission device 300 may also include one or more power supplies 340, one or more wired or wireless network interfaces 350, one or more input and output interfaces 360, and / or one or more operating systems 331, such as Windows Serve, Mac OS X, Unix, Linux, FreeBSD, etc. It will be appreciated by those skilled in the art that Figure 3 The illustrated structure of the low-power wireless audio transmission device does not constitute a limitation on the low-power wireless audio transmission device provided by the present invention, and may include more or fewer components than illustrated, or a combination of certain components, or a different arrangement of components.

[0089] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the low-power wireless audio transmission method.

[0090] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device, or unit can refer to the corresponding process in the aforementioned method embodiment and will not be repeated here.

[0091] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0092] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A low-power wireless audio transmission method, characterized in that: The low-power wireless audio transmission method comprises: The audio transmitting end performs one-bit quantization sampling on the multiple audio input signals, and performs time division multiplexing processing on the quantized signals according to a preset time slot allocation mode to obtain a multiplexed audio data stream; The multiplexed audio data stream is divided into a plurality of audio data blocks according to a preset block size; the plurality of audio data blocks are input into an encoder part in a preset autoencoder network, the plurality of audio data blocks are processed through each layer structure in the encoder part, and soft quantization parameters are introduced in the processing process of each structure layer to perform quantization constraints to obtain a soft quantization feature sequence; the soft quantization feature sequence is signal packaged according to a preset super-Nyquist sampling rate to obtain a time domain sampling sequence; the time domain sampling sequence is encoded according to a preset spectrum efficiency constraint condition and a bit sequence mapping rate to obtain a coded sequence to be transmitted; Divide the to-be-transmitted coded sequence into a plurality of data packets according to a preset data packet size to obtain a grouped coded sequence, wherein the data packet size includes a data transmission segment and a header segment of a fixed length; detect the current battery power state of the audio transmitting end, and select a corresponding transmission power level from a preset transmission power level group according to the detection result to obtain a current transmission power level; and send the grouped coded sequence to the audio receiving end in a burst transmission mode according to the transmission power level, wherein the burst transmission mode includes a data packet transmission phase and a sleep phase, and the duration of the sleep phase is determined by the processing time of the next data packet; The received coding sequence is oversampled multiple times by the audio receiving end, and the decoder part of the autoencoder network is used to reconstruct the oversampled signal to obtain a restored multi-channel audio signal.

2. The low-power wireless audio transmission method according to claim 1, characterized in that: The method of performing one-bit quantization sampling on the multiple audio input signals through the audio transmitting end and performing time division multiplexing processing on the quantized signals according to a preset time slot allocation mode to obtain a multiplexed audio data stream includes: Performing one-bit analog-to-digital conversion sampling on multiple audio input signals through an audio transmitting end to obtain multiple quantized audio signals, wherein the sampling rate of the one-bit analog-to-digital conversion sampling is lower than a standard Nyquist sampling rate; Performing noise reduction processing on the multi-channel quantized audio signals, and grading the processed signals according to preset signal priorities to obtain multi-channel noise reduction processed signals; The multi-channel noise reduction processed signals are time-division multiplexed according to a preset time slot allocation mode to obtain the multiplexed audio data stream, wherein the preset time slot allocation mode includes a signal transmission time slot and a protection interval time slot of a fixed length, and the protection interval time slot is located between adjacent signal transmission time slots.

3. The low-power wireless audio transmission method according to claim 1, characterized in that: The step of inputting the multiple audio data blocks into an encoder part in a preset autoencoder network, processing the multiple audio data blocks through each layer structure in the encoder part, and introducing soft quantization parameters to perform quantization constraints during the processing of each structure layer to obtain a soft quantization feature sequence includes: Inputting the multiple audio data blocks into an encoder part in a preset autoencoder network, and performing dimension mapping on the multiple audio data blocks through an input layer of the encoder part to obtain an initial feature vector; The initial feature vector is sequentially subjected to nonlinear transformation and normalization processing through a plurality of hidden layers in the encoder part, and a soft quantization parameter is introduced in each hidden layer processing process to obtain an intermediate feature vector; The intermediate feature vector is dimensionally reconstructed through the output layer of the encoder part to obtain the soft quantization feature sequence.

4. The low-power wireless audio transmission method according to claim 1, characterized in that: The step of performing multiple oversampling on the received coding sequence by the audio receiving end, and reconstructing the oversampled signal by using the decoder part of the autoencoder network to obtain the restored multi-channel audio signal comprises: Performing multiple oversampling processing on the received coding sequence through the audio receiving end to obtain an oversampled signal sequence, wherein the multiple of the multiple oversampling is greater than the preset super-Nyquist sampling rate; Inputting the oversampled signal sequence into the decoder part of the autoencoder network for processing, and reconstructing the features of the oversampled signal sequence through the decoder part to obtain a reconstructed feature sequence; The reconstructed characteristic sequence is subjected to time division multiplexing processing according to the preset time slot allocation mode to obtain a restored multi-channel audio signal.

5. The low-power wireless audio transmission method according to claim 4, characterized in that: Inputting the oversampled signal sequence into the decoder part of the autoencoder network for processing, and reconstructing the features of the oversampled signal sequence by the decoder part to obtain the reconstructed feature sequence comprises: Inputting the oversampled signal sequence into the decoder part of the autoencoder network, and performing dimension mapping on the oversampled signal sequence through the input layer of the decoder part to obtain an initial decoding feature vector; Performing nonlinear transformation processing on the initial decoding feature vector in sequence through a plurality of hidden layers in the decoder part to obtain an intermediate decoding feature vector; The intermediate decoded feature vector is dimensionally reconstructed through the output layer of the decoder part to obtain a reconstructed feature sequence.

6. A low-power wireless audio transmission device, characterized in that: The low-power wireless audio transmission device comprises: A multiplexing module is used to perform one-bit quantization sampling on multiple audio input signals through an audio transmitter, and perform time division multiplexing processing on the quantized signals according to a preset time slot allocation mode to obtain a multiplexed audio data stream; An encoding module is used to divide the multiplexed audio data stream into multiple audio data blocks according to a preset block size; input the multiple audio data blocks into an encoder part in a preset autoencoder network, process the multiple audio data blocks through each layer structure in the encoder part, and introduce soft quantization parameters in the processing process of each structure layer to perform quantization constraints to obtain a soft quantization feature sequence; perform signal packaging on the soft quantization feature sequence according to a preset super-Nyquist sampling rate to obtain a time domain sampling sequence; encode the time domain sampling sequence according to a preset spectrum efficiency constraint condition and a bit sequence mapping rate to obtain a coded sequence to be transmitted; A sending module, configured to divide the to-be-transmitted coding sequence into a plurality of data packets according to a preset data packet size to obtain a grouped coding sequence, wherein the data packet size includes a data transmission segment and a header segment of a fixed length; detect the current battery power state of the audio transmitting end, and select a corresponding transmission power level from a preset transmission power level group according to the detection result to obtain a current transmission power level; and send the grouped coding sequence to the audio receiving end in a burst transmission mode according to the transmission power level, wherein the burst transmission mode includes a data packet transmission phase and a sleep phase, and the duration of the sleep phase is determined by the processing time of the next data packet; The decoding and reconstruction module is used to perform multiple oversampling on the received coding sequence through the audio receiving end, and use the decoder part of the autoencoder network to reconstruct the oversampled signal to obtain a restored multi-channel audio signal.

7. A low-power wireless audio transmission device, characterized in that: The low-power wireless audio transmission device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instruction in the memory to enable the low-power wireless audio transmission device to perform the steps of the low-power wireless audio transmission method according to any one of claims 1 to 5.

8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the low-power wireless audio transmission method as described in any one of claims 1 to 5 are implemented.

Images