Underwater acoustic communication adaptive modulation mode identification method giving consideration to energy collection
By using SWIPT technology and neural network models to split signals and perform adaptive modulation mode identification in underwater acoustic communication systems, the problems of energy limitation and communication link handshake difficulties in underwater acoustic communication systems are solved. This achieves efficient energy harvesting and modulation mode identification, extends equipment life, and reduces costs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SOUTH CHINA UNIV OF TECH
- Filing Date
- 2026-02-06
- Publication Date
- 2026-05-19
AI Technical Summary
In underwater acoustic communication systems, the terminal equipment is energy-constrained and blind signal processing leads to increased energy consumption. At the same time, existing technologies have failed to effectively combine energy harvesting and adaptive modulation methods, resulting in difficulties in communication link handshaking.
The SWIPT technique is used to split the signal into two channels: information decoding and energy harvesting. An adaptive modulation mode identification is performed through a neural network model. The optimal power allocation factor is used to maximize energy harvesting and identification rate. The modulation mode identification is performed by combining a convolutional neural network and a long short-term memory network (CLDNN) model.
It effectively solves the energy limitation problem of underwater acoustic communication terminal equipment, extends equipment life, reduces operating costs, and achieves accurate identification of adaptive modulation methods in complex environments.
Smart Images
Figure CN122069145A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of underwater acoustic communication technology, specifically relating to a method for identifying adaptive modulation modes in underwater acoustic communication systems that also take into account energy harvesting. Background Technology
[0002] Underwater acoustic communication technology faces problems such as multipath effect, Doppler shift, and severe channel delay spread. Therefore, adaptive modulation and coding technology is usually used at the transmitting end and blind signal processing is used at the receiving end to identify the modulation mode, so as to adapt to complex and ever-changing channel conditions and avoid communication link handshaking difficulties caused by severe channel delay spread.
[0003] Another problem with underwater acoustic communication networks is the limited energy of terminal devices due to battery power; furthermore, blind signal processing at the receiver also leads to a further increase in the energy consumption of terminal devices. Simultaneous Wireless Information and Power Transfer (SWIPT) technology can achieve simultaneous transmission of signal and energy, and is an effective solution to the energy-limited problem of underwater acoustic communication terminal devices.
[0004] Currently, only a few studies have separately proposed methods for replenishing the energy of underwater acoustic communication nodes using wireless energy harvesting technology, and for blindly identifying underwater acoustic communication modulation patterns using deep learning. Patent application CN120896823A, published on November 4, 2025, entitled "A Method and Related Device for Identifying Underwater Acoustic Signal Modulation Patterns," proposes applying an automatic modulation identification network based on deep learning to non-cooperative underwater acoustic communication networks to achieve adaptive modulation pattern identification; however, this method does not consider the energy constraints of devices in underwater acoustic communication. Patent application WO2024250990A1, published on December 12, 2024, entitled "An Underwater Acoustic Communication Device and Method with Wireless Energy Harvesting Function," proposes splitting the acoustic signal received by the underwater terminal node into two signals based on time switching for energy harvesting and information harvesting respectively; however, this method does not consider adaptive modulation of the transmitted signal and adaptive processing of the received signal. Summary of the Invention
[0005] To address at least one of the problems existing in the prior art, this invention proposes an adaptive modulation scheme identification method for underwater acoustic communication systems that takes into account energy harvesting. This method enables the receiver R in an underwater acoustic communication system to perform optimal power division based on the strength of the received signal after receiving the signal, thereby maximizing the harvested energy while ensuring the maximum recognition rate of the adaptive modulation scheme.
[0006] The objective of this invention is achieved through the following technical solution:
[0007] The SWIPT underwater acoustic communication system transmits modulated signals from the transmitter (S), and the receiver (R) receives the signals. The signals are then converted into electrical signals by an underwater acoustic transducer, and split into two signals by a power distribution circuit: one is the information decoding signal, with a power allocation factor of [missing value]. The other path is the energy harvesting signal, with a power allocation factor of [missing value]. ,in It is a constant whose value is in the interval [0,1]; the information decoding signal is decoded by the information processing circuit; the energy harvesting signal is harvested by the energy harvesting circuit, which specifically includes the following steps: Step 1: The transmitter S sends a signal to the receiver R. The signal received at receiver R is ,in The maximum number of propagation paths, For the first transmission process The amplitude attenuation caused by the path For the first Doppler frequency shift factor of the path, For the first transmission process Transmission delay on the path, The additive ambient noise during transmission has a power of .
[0008] Step 2: After receiving the signal, the receiving end R distributes the received signal using the optimal power allocation factor. Diverted to and Two signals, of which To decode the information that needs to be identified in subsequent processing, The energy harvesting signal is used for energy harvesting. This is the optimal power allocation factor.
[0009] Step 3, decode the information signal The signal is processed, filtered, and amplified to obtain the final signal used for modulation scheme identification. ,in The information signal after being processed by the information processing circuit. The power is used to process the noise generated by the information processing circuit. Energy harvesting signal After being processed by rectification and voltage regulation circuits, the battery is charged, thus achieving energy harvesting.
[0010] Step 4, send the signal The input neural network model is used to identify and classify modulation methods, which is accomplished through the following steps: Step 4.1: Sample, down-convert, and filter the input signal to obtain a discrete baseband complex signal.
[0011] Where L is the number of sampling points per frame, the real and imaginary parts are separated and stacked into a two-dimensional tensor, which is used as the network input:
[0012] Each frame of signal constitutes a sample, denoted as . .
[0013] Step 4.2, Input convolutional feature extraction module to extract local features, the first... The output of the convolutional layer is:
[0014] in For activation function, This is the bias value. For convolution kernel parameters, This represents the convolution operation. This represents the total number of convolutional layers. After multiple convolutions and downsampling, a high-dimensional spatial feature sequence is obtained. ,in This represents the number of channels in the last convolutional layer. The length is the time dimension.
[0015] Step 4.3, Transpose ,in Then, the data is input into the time-series feature extraction module, which updates the features at different time steps by selecting and discarding features to obtain the global time-series feature vector. , For this The parameters corresponding to the network, For this Number of output units of the network for Moment The final feature vector output by the network is obtained through... Complete the integration of information from the previous moment and the current moment, where for time The network output.
[0016] Step 4.4, The fully connected layer completes feature mapping and classification, and outputs the recognition result.
[0017]
[0018] For the feature mapping result, This represents matrix multiplication, where the first term represents the parameters of the fully connected layer, and the final probability mapping is achieved through an activation function. This represents the total number of categories to be identified, and of Each bit value represents the probability of taking a value in the range [0,1], and each bit corresponds to a different result. Finally, the bit containing the maximum value is determined as the recognition result.
[0019] Step 5: Based on the recognition results, demodulate the corresponding modulation scheme to complete the decoding.
[0020] Among them, the optimal power allocation factor in step 2 Obtained through the following specific steps: Step 2.1, Define the optimization problem:
[0021] in, yes The combined evaluation metric for energy collection and recognition rate can be specifically expressed as: ,in It is the signal-to-noise ratio when input into the model after analog-to-digital conversion; and This is a weighting factor for energy distribution and modulation recognition rate; This represents the power allocation factor. For the entire neural network, as the signal-to-noise ratio (SNR) of the input signal increases, the signal's features are more easily extracted, and noise interference decreases. At low SNR, the modulation scheme recognition rate significantly improves with increasing SNR. However, at high SNR, the signal features are already well extracted, and the improvement in recognition rate weakens due to the limitations of the neural network's performance. Furthermore, the highest recognition accuracy is somewhat limited. Based on this, a sigmoid logic function can be used for fitting to obtain the relationship between recognition accuracy and the input signal SNR. ,in The maximum value of this function at saturation is limited, depending on the performance of the neural network, and its maximum value is 1; where... and This is mainly determined by the network's structure and depth.
[0022] Step 2.2: The optimization problem described above is a convex problem, which can be quickly solved using the bisection method. The maximum value, at which point the corresponding value is... For the desired optimal power allocation factor .
[0023] The present invention also provides an adaptive modulation mode identification device for underwater acoustic communication systems that also takes into account energy harvesting.
[0024] The present invention also provides a computer device.
[0025] The present invention also provides a computer-readable storage medium.
[0026] Compared with existing technologies, the present invention has the following advantages: 1. This invention proposes an underwater acoustic adaptive communication scheme that also incorporates underwater wireless energy harvesting, effectively solving the problems of limited battery power and the difficulty of charging or replacing batteries. SWIPT technology allows the receiver to utilize environmental noise and interference as energy sources for harvesting, making it applicable to areas with high environmental noise, such as shallow seas and nearshore areas. It can delay or avoid the need to replace batteries or nodes due to battery depletion, extending the lifespan of underwater acoustic sensors and the underwater acoustic network to a certain extent, thereby significantly reducing the operating costs of the underwater acoustic network.
[0027] 2. This invention employs a neural network model to identify modulation schemes, exhibiting strong robustness. This allows the receiver to correctly identify the modulation scheme of the communication signal even in poor communication environments or when handshake information is lost. Furthermore, its adaptive identification feature makes it highly suitable for adaptive underwater acoustic communication nodes. Attached Figure Description
[0028] Figure 1 This is a schematic diagram of the SWIPT underwater acoustic communication system in an embodiment of the present invention.
[0029] Figure 2 This is a flowchart illustrating the specific implementation steps of the system in this embodiment of the invention.
[0030] Figure 3 This is a flowchart illustrating the steps for obtaining the optimal solution of the power allocation factor in an embodiment of the present invention.
[0031] Figure 4 This is a diagram showing the training results and parameter count of the CLDNN network in this embodiment of the invention.
[0032] Figure 5 This is a graph of the fitting function of the CLDNN network trained in this embodiment of the invention.
[0033] Figure 6 This is a diagram showing the relationship between the optimization problem function and the power allocation factor in an embodiment of the present invention. Detailed Implementation
[0034] The present invention will be further described below with reference to the accompanying drawings and embodiments, but the implementation of the present invention is not limited thereto.
[0035] The schematic diagram, implementation flowchart, and algorithm flowchart for obtaining maximum energy of the underwater acoustic communication system according to embodiments of the present invention are as follows: Figure 1 , 2 As shown in Figure 3.
[0036] In this embodiment, a specific form of underwater acoustic communication system is considered, such as... Figure 1 As shown, the receiver receives the signal through a hydrophone, and then splits the signal into two paths through a power distribution circuit. The information decoding signal enters the information processing circuit, and the energy harvesting signal enters the energy harvesting circuit, which are used for information storage and energy harvesting, respectively. The neural network example for modulation mode recognition is CLDNN (Convolutional, LSTM, FullyConnected Deep Neural Networks), which combines a convolutional neural network and a long short-term memory network.
[0037] An adaptive underwater acoustic communication system that also incorporates energy harvesting is provided in this embodiment of the invention. answer The modulation scheme identification method specifically includes the following steps: Step 1: The transmitter S sends a signal to the receiver R. The signal received at receiver R is ,in The maximum number of propagation paths, For the first transmission process The amplitude attenuation caused by the path For the first Doppler frequency shift factor of the path, For the first transmission process Transmission delay on the path, The additive ambient noise during transmission has a power of ; For time.
[0038] The transmitter S transmits signals via modulation, digital-to-analog conversion, radio frequency amplifier, and underwater acoustic transducer.
[0039] Step 2: After receiving the signal, the receiving end R allocates power to the signal according to the signal strength using an optimal power allocation factor, thus distributing the received signal... Diverted to and Two signals, of which To decode the information that needs to be identified in subsequent processing, The energy harvesting signal is used for energy harvesting. This is the optimal power allocation factor.
[0040] Among them, the optimal power allocation factor Obtained through the following specific steps: Step 2.1: Based on the traditional S-shaped logic function, construct the functional relationship between recognition accuracy and input signal-to-noise ratio. F(SNR) is the recognition accuracy. The maximum value of this function when it saturates is limited, depending on the performance of the neural network model, and its maximum value is 1; , and These are all parameters of a traditional S-shaped logistic function, and their specific values are obtained by fitting the performance of the neural network on the test set. The maximum value of this function when it saturates is limited, depending on the performance of the neural network; its maximum value is 1. and The value of is mainly related to the structure and depth of the network.
[0041] Step 2.2, Define the optimization problem:
[0042] in, It is a combined evaluation index of the energy collected by the receiver R and the recognition rate, which can be specifically expressed as: ,in It is the signal-to-noise ratio when input into a neural network model after analog-to-digital conversion; It is a weighting factor for energy allocation. It is a weighting factor for modulation recognition rate; This is the power allocation factor.
[0043] For the entire neural network model, as the signal-to-noise ratio (SNR) of the input signal increases, the features of the signal itself are easier to obtain, and the interference of noise will also decrease. Under low SNR conditions, as the SNR of the input signal increases, the recognition rate of its modulation method will significantly improve. However, when the SNR is high, the features of the signal can be extracted well. Due to the limitations of the neural network performance, the improvement in recognition rate will weaken, and the highest recognition accuracy is limited. Based on this, the relationship between recognition accuracy and input signal SNR can be obtained by fitting a sigmoid logic function.
[0044] Step 2.3: The optimization problem is a convex problem. An initial power allocation factor is set, and the result is quickly obtained using a bisection method. The maximum value obtained at this time This is the desired optimal power allocation factor.
[0045] In one embodiment, , , The results obtained through specific testing and fitting are as follows: The specific fitting results are as follows: Figure 5 As shown. According to Figure 5 The fitting results show that the model's recognition accuracy has an sigmoid logistic function relationship with the signal-to-noise ratio (SNR) of the input signal. Once a certain SNR is reached, further increases in SNR significantly reduce the improvement in recognition accuracy. According to... Figure 6 The results show that selecting the optimal power allocation factor can balance energy harvesting and modulation method recognition rate. Under a certain signal-to-noise ratio, reasonable power allocation can save energy to the greatest extent.
[0046] Step 3, decode the information signal The signal is processed through filtering, amplification, and analog-to-digital conversion to obtain the final signal used for modulation scheme identification. ,in The information signal after being processed by the information processing circuit. The power is used to process the noise generated by the information processing circuit. Energy harvesting signal After being processed by rectification and voltage regulation circuits, the battery is charged, thus achieving energy harvesting.
[0047] Step 4, send the signal The input is used to identify and classify modulation methods, and the identification results are obtained.
[0048] In one embodiment, the neural network model uses CLDNN, with the specific structure as follows: Figure 4 As shown. In other embodiments, other neural networks may also be used.
[0049] The steps involved in neural network model recognition and classification include: Step 4.1, for the input signal The discrete baseband complex signal is obtained by performing frequency mixing down-conversion, low-pass filtering, and A / D sampling. :
[0050] in, The imaginary unit, The number of sampling points per frame; in one embodiment, it is taken as... Discrete baseband complex signal real part With the imaginary part Separate and stack them into two-dimensional tensors, which are then used as input to a neural network model:
[0051] Each frame signal This constitutes a sample, denoted as .
[0052] Step 4.2: Prepare the sample Input convolutional feature extraction module (convolutional network unit), extract local features, the first... Convolutional output results for:
[0053] in For the activation function, use here. function; This is the bias value. For convolution kernel parameters, This represents the convolution operation. This represents the total number of convolutional layers. A Dropout layer is added after each convolutional layer. This layer only operates during network training, preventing overfitting by masking some neuron outputs. A reshape layer then reassembles the output dimensions to obtain a high-dimensional spatial feature sequence. ,in This represents the number of channels in the last convolutional layer. The length is the time dimension. In one embodiment, the convolutional layers use convolutional kernels of size (1,3) with a stride of 1. , , .
[0054] Step 4.3, spatial feature sequence Transpose ,in , Refers to the sequence order; then input. The network (circular network unit) is formed by forwarding each... Learning integration yields global temporal feature vectors. , For this The parameters corresponding to the network, For this Number of output units of the network for Moment The final feature vector output by the network is obtained through... Complete the integration of information from the previous moment and the current moment, where for time The network's output. In one embodiment, .
[0055] Step 4.4, Global Temporal Feature Vector The fully connected layer completes feature mapping and classification, and outputs the recognition result.
[0056]
[0057] For the feature mapping result, Represents matrix multiplication. Here are the weight parameters of the fully connected layer, and the activation function is... function, For the bias term of the fully connected layer; then through The function normalizes the feature mapping results into a probability display, where and These are the network weight parameters and bias terms for the fully connected layer, respectively. This represents the total number of categories to be identified. The recognition result output by the model, its Each bit value represents a probability of being in the range [0,1], and each corresponds to a different outcome. The bit containing the highest probability is then determined as the recognition result. In one embodiment, .
[0058] Step 5: Based on the modulation recognition result, input the corresponding demodulation circuit to complete demodulation and store the information.
[0059] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.
Claims
1. A method for adaptive modulation scheme identification in an underwater acoustic communication system that also considers energy harvesting, characterized in that, Includes the following steps: The transmitter sends a signal to the receiver. The receiving end allocates power to the signal based on the strength of the received signal using an optimal power allocation factor, splitting the received signal into two signals: an information decoding signal and an energy harvesting signal. The information decoding signal is preprocessed to obtain the final signal used for modulation mode identification; The signal involved in modulation mode identification is input into a neural network model for modulation mode identification and classification to obtain modulation identification results. In the neural network model, spatial information of the signal is extracted through convolutional network units to obtain spatial feature sequences. The spatial feature sequences are integrated through recurrent network units to obtain global temporal feature vectors. The global temporal feature vectors are then mapped and classified to obtain modulation identification results. Based on the modulation identification results, demodulation of the corresponding modulation method is performed to complete the decoding.
2. The adaptive modulation scheme identification method for an underwater acoustic communication system that also considers energy harvesting, as described in claim 1, is characterized in that... The signal received by the receiver Represented as: In the formula, The maximum number of propagation paths, For the first transmission process The amplitude attenuation caused by the path For the first Doppler frequency shift factor of the path, For the first transmission process Transmission delay on the path, Additive environmental noise during transmission. For time, The signal transmitted by the transmitter.
3. The adaptive modulation scheme identification method for an underwater acoustic communication system that also considers energy harvesting, as described in claim 1, is characterized in that... The information decoding signal and energy harvesting signal obtained after the received signal is split are represented as follows: In the formula, For information decoding signals, For energy harvesting signals, The optimal power allocation factor is... The signal received by the receiver For time.
4. The adaptive modulation scheme identification method for an underwater acoustic communication system that also considers energy harvesting, as described in claim 1, is characterized in that... The steps for determining the optimal power allocation factor include: Define the optimization problem: In the formula, It is a combined evaluation index of the energy collected at the receiving end and the recognition rate, expressed as: In the formula, It is the power allocation factor. It is a weighting factor for energy allocation. It is a weighting factor for modulation recognition rate. It is the recognition accuracy. It is the signal-to-noise ratio when input into a neural network model after analog-to-digital conversion; Set an initial power allocation factor, solve the optimization problem, and obtain the optimal power allocation factor.
5. The adaptive modulation scheme identification method for an underwater acoustic communication system that also considers energy harvesting, as described in claim 4, is characterized in that... The expression is: In the formula, , and All of these are function parameters, obtained through fitting.
6. The adaptive modulation scheme identification method for an underwater acoustic communication system that also considers energy harvesting, as described in claim 1, is characterized in that... The neural network model used is CLDNN.
7. The adaptive modulation scheme identification method for an underwater acoustic communication system that also considers energy harvesting, as described in any one of claims 1-6, is characterized in that... The neural network model is configured to perform the following operations: The signals involved in modulation mode identification are processed to obtain discrete baseband complex signals; The real and imaginary parts of the discrete baseband complex signal are separated and stacked into a two-dimensional tensor; Two-dimensional tensors are input into convolutional network units to extract local features, resulting in a high-dimensional spatial feature sequence. The spatial feature sequence is transposed and input into a recurrent network unit for processing to obtain a global temporal feature vector. The global temporal feature vector is passed through a fully connected layer to complete feature mapping and classification, and the modulation recognition result is output.
8. An adaptive modulation mode identification device for an underwater acoustic communication system that also considers energy harvesting, characterized in that, For implementing the method according to any one of claims 1-7, the apparatus comprises the following modules: The transmitting module sends signals from the transmitting end to the receiving end; The signal distribution module uses a power distribution factor to distribute the power of the received signal based on the strength of the received signal at the receiving end, splitting the received signal into two signals: an information decoding signal and an energy harvesting signal. The signal processing module is used to preprocess the information decoding signal to obtain the final signal used for modulation mode identification; The identification module is used to input the signals involved in modulation mode identification into the neural network model for modulation mode identification and classification, and obtain modulation identification results. In the neural network model, the spatial information of the signal is extracted by the convolutional network unit to obtain the spatial feature sequence, and the spatial feature sequence is integrated by the recurrent network unit to obtain the global temporal feature vector. The global temporal feature vector is then mapped and classified to obtain the modulation identification results. The demodulation module is used to demodulate the corresponding modulation method based on the modulation identification result, and complete the decoding.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1-6.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-6.