A method and device for adapting a molecular communication channel based on transfer learning
By employing a transfer learning-based molecular communication channel adaptation method, utilizing the diffusion-adsorption-reaction equation and a CNN-BiLSTM hybrid neural network, the dynamic adaptation problem of molecular communication channels is solved, improving the accuracy of symbol detection and communication reliability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGZHOU UNIVERSITY
- Filing Date
- 2026-01-27
- Publication Date
- 2026-07-07
Smart Images
Figure CN121603134B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of molecular communication technology, specifically a molecular communication channel adaptation method and apparatus based on transfer learning. Background Technology
[0002] The Internet of Bio-Nano Things (IoBNT) is an interconnected network of nanoscale devices, with the potential applications in precision medicine, targeted drug delivery, and health monitoring. Molecular communication (MC), inspired by biological systems, uses molecules as information carriers to facilitate data transmission between nanonodes and is a key enabling technology for building IoBNT.
[0003] However, the practical deployment of molecular communication still faces challenges. The random movement of information molecules leads to a long-tail effect in the channel response, causing severe inter-symbol interference (ISI) and limiting the performance of symbol detection. When large-scale networks need to be built to complete complex monitoring tasks, inter-link interference (ILI) will occur between multiple parallel communication links, further exacerbating the complexity of signal detection. Both types of interference distort the concentration distribution and time-series characteristics of the received signal, further increasing the difficulty of symbol decision. Convolutional Neural Networks (CNNs) have powerful local feature extraction capabilities and can effectively capture interference patterns and concentration abrupt changes in signals; Bidirectional Long Short-Term Memory (Bi-LSTM) networks are good at modeling long-term dependencies in time-series data and can accurately characterize the delay effects and cumulative effects of ISI during the propagation of signal molecules. The CNN-BiLSTM hybrid model combines the advantages of both, which can extract both interference and local signal features and capture the temporal dynamic changes of the signal, providing an ideal model architecture for simultaneously suppressing ISI and ILI and improving symbol detection accuracy in dynamic channels.
[0004] The molecular communication channel environment within the human body exhibits highly dynamic characteristics, with parameters changing significantly depending on the individual's health level. In a healthy state, physiological parameters such as blood flow velocity and tissue fluid viscosity remain within stable ranges, while core channel parameters like diffusion coefficient and advection velocity show regularity. However, in pathological states such as inflammation and tumors, metabolic disturbances in local tissues lead to abnormal fluctuations in the diffusion coefficient, vascular remodeling may alter the advection velocity distribution, and increased enzyme concentrations accelerate signal molecule degradation. These changes directly induce time-varying and distorted channel impulse response (CIR). In the nano-Internet of Things (IoT), the mobility of nanonodes (such as nanorobots flowing with blood) and the dynamic adjustment of inter-node distances further exacerbate the uncertainty of channel parameters, making it difficult for traditional symbol detection methods based on fixed channel models to accurately capture signal features and thus limiting communication reliability.
[0005] Data-driven methods, with their strong fitting and adaptive capabilities to complex dynamic systems, have demonstrated unique advantages in molecular communication symbol detection. These methods learn the nonlinear mapping relationship between channel features and symbols from observational data, without relying on precise channel modeling. In recent years, machine learning-driven symbol detection technology has received widespread attention. Researchers have proposed detection schemes based on models such as support vector machines and neural networks, achieving symbol decision by mining the time-series features and concentration distribution features of received signals. With the support of ideal datasets, these methods have achieved performance superior to traditional detection methods. However, these methods still face significant limitations: nanonodes are limited by size and energy supply, resulting in extremely limited computational and storage resources, making it difficult to collect sufficient high-quality observational data; the dynamic nature of in vivo channels leads to significant differences in dataset distribution under different health states and transmission scenarios, easily resulting in data scarcity or distribution shift problems, which makes the generalization ability of trained models insufficient. Summary of the Invention
[0006] To address the shortcomings of the existing technologies, this invention provides a molecular communication channel adaptation method and apparatus based on transfer learning, which can quickly achieve molecular communication channel adaptation and provides an effective way to solve the symbol detection problem in resource-constrained scenarios.
[0007] To achieve the above objectives, this invention provides a molecular communication channel adaptation method based on transfer learning, comprising the following steps:
[0008] Step 1: Consider the three key behaviors of information molecules in molecular communication: diffusion, advection, and reaction. Use the diffusion-advection-reaction equation as the channel impulse response model, and construct a molecular communication system simulator based on the channel impulse response model.
[0009] Step 2: Collect the original channel temperature and the target channel temperature, and obtain the original channel environment parameters and the target channel environment parameters based on the original channel temperature, the target channel temperature and the channel environment calculation model;
[0010] Step 3: Simulate the transmission of information in the molecular communication system under the original channel environment. Randomly simulate the original channel sequence that needs to be transmitted. Considering that the original channel dataset is easy to collect in practical applications, construct an original channel dataset containing 4000 samples based on the original channel sequence and the original channel environment parameters using the molecular communication system simulator.
[0011] Step 4: Simulate the transmission of information in the molecular communication system under the target channel environment. Randomly simulate the target channel sequence that needs to be transmitted. Considering that the target channel data is difficult to collect in practical applications, the molecular communication system simulator is used to construct a target channel dataset containing 500 samples based on the target channel sequence and the target channel environment parameters.
[0012] Step 5: Construct a domain alignment transfer learning model based on a hybrid neural network, and train the domain alignment transfer learning model based on the original channel dataset to obtain a molecular communication symbol detection model adapted to the target channel. Then, perform molecular communication in the target channel based on the molecular communication symbol detection model.
[0013] Compared with the prior art, the present invention has the following beneficial technical effects:
[0014] 1. This invention innovatively constructs a cross-channel environmental knowledge transfer framework based on domain adaptation. It simultaneously captures the temporal features of interference patterns and diffusion processes through a CNN-BiLSTM hybrid neural network and designs a transfer learning mechanism for dynamic channels. It achieves efficient model adaptation with only a small number of target domain samples, and can quickly realize molecular communication channel adaptation, providing an effective way to solve the symbol detection problem in resource-constrained scenarios.
[0015] 2. This invention can effectively transfer knowledge learned from the source domain to the target domain. Compared with retraining from the target domain dataset alone, it has better accuracy, overcomes the problem of the model not being able to fully adapt to the dynamic changes of the channel, and overcomes the problem of insufficient model performance under limited data conditions. It enables nanonodes to adapt to the dynamic characteristics of the channel caused by changes in the body temperature. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.
[0017] Figure 1 This is a flowchart of the molecular communication channel adaptation method based on transfer learning in Embodiment 1 of the present invention;
[0018] Figure 2 This is a schematic diagram of the classical physical model based on molecular communication in Embodiment 1 of the present invention;
[0019] Figure 3 This is a schematic diagram of the transfer learning model framework in Embodiment 1 of the present invention;
[0020] Figure 4 This is a structural block diagram of the molecular communication channel adaptation device based on transfer learning in Embodiment 2 of the present invention;
[0021] Figure 5 This is a structural block diagram of the terminal device in Embodiment 3 of the present invention.
[0022] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0024] Furthermore, the technical solutions of the various embodiments of the present invention can be combined with each other, but only if they are feasible for those skilled in the art. If the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.
[0025] Example 1
[0026] like Figure 1 The figure shown is a molecular communication channel adaptation method based on transfer learning disclosed in this embodiment, which mainly includes the following steps:
[0027] Step 1: Consider the three key behaviors of information molecules in molecular communication: diffusion, advection, and reaction. Use the diffusion-advection-reaction equation as the channel impulse response model, and build a molecular communication system simulator based on the channel impulse response model.
[0028] Step 2: Collect the original channel temperature and the target channel temperature, and obtain the original channel environment parameters and the target channel environment parameters based on the original channel temperature, the target channel temperature and the channel environment calculation model;
[0029] Step 3: Simulate the transmission of information in the molecular communication system under the original channel environment. Randomly simulate the original channel sequence that needs to be transmitted. Considering that the original channel dataset is easy to collect in practical applications, a molecular communication system simulator is used to construct an original channel dataset containing 4000 samples based on the original channel sequence and original channel environment parameters.
[0030] Step 4: Simulate the transmission of information in the molecular communication system under the target channel environment. Randomly simulate the target channel sequence that needs to be transmitted. Considering that the target channel data is difficult to collect in practical applications, a molecular communication system simulator is used to construct a target channel dataset containing 500 samples based on the target channel sequence and target channel environment parameters.
[0031] Step 5: Construct a domain alignment transfer learning model based on a hybrid neural network, and train the domain alignment transfer learning model based on the original channel dataset and the target channel dataset to obtain a molecular communication symbol detection model adapted to the target channel. Then, perform molecular communication in the target channel based on the molecular communication symbol detection model.
[0032] like Figure 2 The image shows a classic physical model based on molecular communication, where the transmitter uses on-off keying (OOK) modulation. When the binary symbol... When the value is 1, the transmitter fires. The molecular concentration (molecular concentration, such as a communication carrier or calcium ion) is zero; no molecules are transmitted. The receiver is a passive receiver; under the assumption of uniform concentration, the molecular concentration within the receiver volume can be approximated as the concentration at the receiver center. In the channel section, in an unbounded three-dimensional environment, the molecular concentration... The evolution over time and space can be described by Fick's second law, namely:
[0033]
[0034] in, To represent the spatial position at time t Molecular concentration at that location The diffusion coefficient is... This is the Laplacian operator in a three-dimensional Cartesian coordinate system. This equation only describes a pure diffusion scenario. However, in the actual deployment of nanonetworks within the human body, signal molecule transmission in molecular communication is often accompanied by directional advection (such as blood flow) and degradation (such as enzymatic reactions). Therefore, considering that molecules will migrate directionally with the fluid when there is uniform advection in the environment, and that signal molecules will undergo first-order irreversible degradation, this embodiment supplements the original diffusion term with advection and reaction terms to obtain the diffusion-advection-reaction equation as a channel impulse response model:
[0035]
[0036] in, The channel impulse response represents Time and space location Molecular concentration at that location; This represents the total number of molecules instantaneously released by the transmitter. The initial moment of molecule release. For degradation rate, The diffusion coefficient is... For uniform advection velocity vectors, This is the transmitter's initial position vector.
[0037] According to the channel impulse response model in this embodiment, its unknown parameters include the degradation rate. diffusion coefficient Uniform advection velocity vector The total number of molecules released instantaneously by the transmitter and the distance between the target's spatial location and the transmitter. Therefore, in this embodiment, both the original channel environment and the target channel environment include the degradation rate. diffusion coefficient Uniform advection velocity vector Total number of molecules With distance Among them, the total number of molecules in the original channel environment and the target channel environment. and distance They are all the same, but the degradation rate is the same. diffusion coefficient Uniform advection velocity vector The results are obtained based on the original channel temperature, the target channel temperature, and the channel environment calculation model. The channel environment calculation model includes a diffusion coefficient calculation model, a degradation rate calculation model, and a uniform advection calculation model.
[0038] In practical applications, an increase in human body temperature reduces the intermolecular forces between water molecules in blood plasma, leading to internal friction and a decrease in viscosity. Therefore, this embodiment uses the Stokes-Einstein equation as the diffusion coefficient calculation model, namely:
[0039]
[0040] in, Boltzmann's constant, For temperature, For dynamic viscosity, denoted as , where is the hydrodynamic radius of the solute molecule.
[0041] Based on the diffusion coefficient calculation model, using the normal human body temperature of 37°C as a baseline, it can be deduced that when the human body temperature rises by 1°C from the normal body temperature of 37°C, the reaction rate... Approximately 2%-3% more; when the human body temperature rises by 2°C from the normal body temperature of 37°C, the reaction rate... It will increase by approximately 5%-12%.
[0042] In practical applications, many degradation processes in the human body are catalyzed by enzymes. Within the heating range, enzyme activity typically increases, accelerating the breakdown of signaling molecules. Therefore, this embodiment uses the Arrhenius equation as the degradation rate calculation model, namely:
[0043]
[0044] in, As a pre-exponential factor (or frequency factor), it can be approximated as a constant independent of temperature; The activation energy represents the energy barrier that reactant molecules must overcome in order to undergo a chemical reaction; it can be approximated as a constant independent of temperature. This is the universal gas constant.
[0045] The degradation rate calculation model establishes a quantitative relationship between temperature and chemical reaction rate. Based on the model and using the normal human body temperature of 37°C as a baseline, it can be deduced that when the human body temperature rises by 1°C from 37°C, the reaction rate... Approximately 5%-10% more; when the human body temperature rises by 2°C from the normal body temperature of 37°C, the reaction rate... It will increase by approximately 20%-50%.
[0046] In practical applications, when the human body has a fever, its metabolic rate increases, tissue oxygen demand increases, and the heart increases cardiac output by increasing heart rate and stroke volume. Clinical observations show that for every 1°C increase in body temperature, the heart rate increases by approximately 8 beats per minute. Assuming that stroke volume (the amount of blood pumped out with each heartbeat) remains constant, cardiac output = heart rate. The stroke volume of the heart. Therefore, this embodiment uses the law of conservation of volumetric flow rate as the uniform advection calculation model, that is:
[0047]
[0048] in, For cardiac output, This refers to the cross-sectional area of the blood vessel.
[0049] According to the uniform advection calculation model, when the human body temperature rises by 1°C from the normal body temperature of 37°C, the heart rate increases by approximately 11% per minute. Considering the heart rate... The cardiac output per heartbeat equals the total cardiac output. It can be deduced that if cardiac output increases by about 11%, uniform advection velocity increases by about 11%. Similarly, when the human body temperature rises by 2°C from the normal body temperature of 37°C, cardiac output increases by about 22%, and uniform advection velocity increases by about 22%.
[0050] Assuming the original channel temperature is 37°C, based on the channel environment calculation model described above, the degradation rate in the original channel environment can be set accordingly. For 100s -1 diffusion coefficient 6.0×10 -10 m 2 / s, uniform advection velocity vector 1.0×10 - 3 m / s. When the target channel temperature is 38°C, the degradation rate in the target channel environment can be set. 110s -1 diffusion coefficient It is 6.2×10 -10 m 2 / s, uniform advection velocity vector 1.1×10 -3 m / s. If the target channel temperature is 39°C, the degradation rate in the target channel environment can be set. 150s -1 diffusion coefficient It is 6.7×10 -10 m 2 / s, uniform advection velocity vector 1.2×10 -3 m / s. By analogy, the channel environment under various target channels can be obtained. As for the total number of elements in the original channel environment and the target channel environment... With distance This can be set according to actual conditions and needs, such as setting the total number of molecules. =1×10 4 ,distance =4×10 -7 m, etc.
[0051] After determining the original channel environment parameters and the target channel environment parameters, the molecular communication system simulator can be constructed by combining the channel impulse response model to generate the corresponding original channel dataset and target channel dataset.
[0052] In practice, both the original channel sequence and the target channel sequence are randomly generated binary sequences. When the binary symbol in either the original or target channel sequence is 1, the transmitter transmits. The target channel dataset is generated as follows: A molecular communication system simulator constructed in step 1 is used. The original channel environment parameters are input, and the concentration of signal molecules received by the receiver under the original channel environment is simulated. The original channel dataset is obtained based on the original channel sequence and the concentration of signal molecules received by the receiver under the original channel environment. The target channel dataset is generated as follows: A molecular communication system simulator constructed in step 1 is used. The target channel environment parameters are input, and the concentration of signal molecules received by the receiver under the target channel environment is simulated. The target channel dataset is obtained based on the target channel sequence and the concentration of signal molecules received by the receiver under the target channel environment.
[0053] refer to Figure 3 In this embodiment, the domain alignment transfer learning model based on a hybrid neural network includes a two-layer one-dimensional convolutional neural network (CNN), a two-layer bidirectional long short-term memory (BiLSTM) network, and a multi-kernel maximum mean difference (MK-MMD) module. The two-layer CNN is used to extract features from the preprocessed original and target channel datasets, sharing parameters during the extraction process. The extracted features are reconstructed and input into the two-layer BiLSTM network to capture time dependencies, outputting a high-dimensional feature vector. MK-MMD is introduced to pair features from the original and target channel datasets, reducing the data distribution differences between the two datasets. The loss function during the training process of the domain alignment transfer learning model is... for:
[0054]
[0055] in, The classification loss is used to fit the mapping relationship between feature vectors and transmitted symbol labels; This is the maximum mean difference domain alignment loss; The entropy minimization loss is used to improve the prediction confidence in the target domain; , The weighting coefficients are determined through a grid search.
[0056] In practical implementation, the calculation process for the maximum mean difference domain alignment loss is as follows:
[0057]
[0058] in, This represents a mapping function used to map feature vectors to the reproducing kernel Hilbert space; The first one extracted from the original channel environment dataset represents the... Features, total One feature; The first one extracted from the new channel environment dataset represents the first one. Features, total One feature; Represents the kernel function; Let denote the norm in Hilbert space.
[0059] It is worth noting that, although this embodiment Figure 1 The steps are shown sequentially as indicated by the arrows, but they are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order in which these steps are performed; they can be executed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.
[0060] Example 2
[0061] Based on the molecular communication channel adaptation method based on transfer learning in Example 1, this example discloses a molecular communication channel adaptation device based on transfer learning. (Refer to...) Figure 4 The molecular communication channel adaptation device includes a model building unit, a channel environment acquisition unit, a dataset construction unit, and an adaptation communication unit. Specifically:
[0062] The model building unit is used to construct a molecular communication system simulator with the diffusion-advection-reaction equation as the channel impulse response;
[0063] The channel environment acquisition unit is used to acquire the original channel temperature and the target channel temperature, and to obtain the original channel environment parameters and the target channel environment parameters based on the original channel temperature, the target channel temperature and the channel environment calculation model.
[0064] The dataset construction unit obtains the original channel sequence and the target channel sequence, and obtains the original channel dataset based on the original channel sequence, the original channel environment parameters and the molecular communication system simulator, and obtains the target channel dataset based on the target channel sequence, the target channel environment parameters and the molecular communication system simulator;
[0065] The adaptive communication unit is used to construct a domain alignment transfer learning model based on a hybrid neural network, and to train the transfer learning model based on the original channel dataset and the target channel dataset to obtain a molecular communication symbol detection model adapted to the target channel. Molecular communication is then performed on the target channel based on the molecular communication symbol detection model.
[0066] In this embodiment, the specific working process and working principle of the model building unit, channel environment acquisition unit, dataset building unit, and adaptive communication unit are the same as those in Embodiment 1, and therefore will not be described again in this embodiment. Each unit module can be implemented entirely or partially through software, hardware, or a combination thereof. Each unit module can be embedded in or independent of the processor in the computer device in hardware form, or it can be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to the above unit modules.
[0067] Example 3
[0068] like Figure 5 The diagram illustrates a terminal device disclosed in this embodiment, comprising a transmitter, a receiver, a memory, and a processor. The transmitter transmits instructions and data, the receiver receives instructions and data, the memory stores computer-executed instructions, and the processor executes the computer-executed instructions stored in the memory to implement the method described in Embodiment 1 above.
[0069] It is important to note that the aforementioned memory can be either standalone or integrated with the processor. When the memory is set up independently, the terminal device also includes a bus for connecting the memory and the processor.
[0070] Example 4
[0071] This embodiment discloses a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the method in Embodiment 1 above.
[0072] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
[0073] The above description is only a preferred embodiment of the present invention and does not limit the scope of protection of the present invention. All equivalent structural transformations made under the inventive concept of the present invention using the contents of the present invention specification and drawings, or direct / indirect applications in other related technical fields, are included within the scope of protection of the present invention.
Claims
1. A molecular communication channel adaptation method based on transfer learning, characterized in that, Includes the following steps: Step 1: The diffusion-advection-reaction equation is used as the channel impulse response model, and a molecular communication system simulator is constructed based on the channel impulse response model. Step 2: Collect the original channel temperature and the target channel temperature, and obtain the original channel environment parameters and the target channel environment parameters based on the original channel temperature, the target channel temperature and the channel environment calculation model; Step 3: Simulate the transmission of information in the molecular communication system under the original channel environment. Randomly simulate the original channel sequence that needs to be transmitted in the original channel. Based on the original channel sequence and the original channel environment parameters, use the molecular communication system simulator to construct the original channel dataset. Step 4: Simulate the transmission of information in the molecular communication system under the target channel environment. Randomly simulate the target channel sequence that needs to be transmitted in the target channel. Based on the target channel sequence and the target channel environment parameters, use the molecular communication system simulator to construct the target channel dataset. Step 5: Construct a domain alignment transfer learning model based on a hybrid neural network, and train the domain alignment transfer learning model based on the original channel dataset and the target channel dataset to obtain a molecular communication symbol detection model adapted to the target channel, and perform molecular communication under the target channel based on the molecular communication symbol detection model.
2. The molecular communication channel adaptation method based on transfer learning according to claim 1, characterized in that, The channel impulse response model is as follows: in, The channel impulse response represents Time and space location Molecular concentration at that location; This represents the total number of molecules instantaneously released by the transmitter. The initial moment of molecule release. For degradation rate, Where is the diffusion coefficient. For uniform advection velocity vectors, This is the transmitter's initial position vector.
3. The molecular communication channel adaptation method based on transfer learning according to claim 2, characterized in that, The channel environment calculation model includes a diffusion coefficient calculation model, a degradation rate calculation model, and a uniform advection calculation model. The diffusion coefficient calculation model is as follows: in, Boltzmann's constant, For temperature, For dynamic viscosity, The hydrodynamic radius of the solute molecule; The degradation rate calculation model is as follows: in, Pre-exponential factor, For activation energy, This is the universal gas constant; The uniform advection calculation model is as follows: in, For cardiac output, This refers to the cross-sectional area of the blood vessel.
4. The molecular communication channel adaptation method based on transfer learning according to claim 3, characterized in that, Both the original channel environment parameters and the target channel environment parameters include the degradation rate. diffusion coefficient Uniform advection velocity vector Total number of molecules With distance ; The total number of numerators in the original channel environment parameters and the target channel environment parameters and distance They are all the same.
5. The molecular communication channel adaptation method based on transfer learning according to any one of claims 1 to 4, characterized in that, In step 3, both the original channel sequence and the target channel sequence are randomly generated binary sequences. When a binary symbol in either the original channel sequence or the target channel sequence is 1, the transmitter transmits. One molecule; when the binary symbol in the original channel sequence or the target channel sequence is 0, no molecule is sent; The process of generating the original channel dataset is as follows: using the molecular communication system simulator, inputting the original channel environment parameters, simulating the concentration of signal molecules received by the receiver under the original channel environment, and obtaining the original channel dataset based on the original channel sequence and the concentration of signal molecules received by the receiver under the original channel environment; The process of generating the target channel dataset is as follows: using the molecular communication system simulator, inputting the target channel environment parameters, simulating the concentration of signal molecules received by the receiver under the target channel environment, and obtaining the target channel dataset based on the target channel sequence and the concentration of signal molecules received by the receiver under the target channel environment.
6. The molecular communication channel adaptation method based on transfer learning according to any one of claims 1 to 4, characterized in that, In step 5, the domain alignment transfer learning model based on hybrid neural networks is a CNN-BiLSTM hybrid neural network. It uses two layers of one-dimensional CNN to extract features from the preprocessed original channel dataset and target channel dataset. During the extraction process, parameters are shared. The extracted features are reconstructed and input into a two-layer bidirectional BiLSTM network to capture time dependencies and output high-dimensional feature vectors. MK-MMD is introduced to pair the features of the original channel dataset and the target channel dataset, reducing the data distribution differences between the original channel dataset and the target channel dataset.
7. The molecular communication channel adaptation method based on transfer learning according to claim 6, characterized in that, The loss function during the training process of the domain alignment transfer learning model is: in, For loss function, For classifying losses, The maximum mean difference domain alignment loss, To minimize the loss due to entropy, , These are the weighting coefficients.
8. A molecular communication channel adaptation device based on transfer learning, characterized in that, The molecular communication channel adaptation device, using the method according to any one of claims 1 to 7, comprises: The model building unit uses the diffusion-adsorption-reaction equation as the channel impulse response and constructs a molecular communication system simulator; The channel environment acquisition unit is used to acquire the original channel temperature and the target channel temperature, and to obtain the original channel environment parameters and the target channel environment parameters based on the original channel temperature, the target channel temperature and the channel environment calculation model. The dataset construction unit simulates the sequence information transmitted through the original channel and the sequence information transmitted through the target channel, and obtains the original channel dataset and the target channel dataset based on the molecular communication system simulator. An adaptive communication unit is used to construct a domain alignment transfer learning model based on a hybrid neural network, and to train the domain alignment transfer learning model based on the original channel dataset and the target channel dataset to obtain a molecular communication symbol detection model adapted to the target channel, and to perform molecular communication under the target channel based on the molecular communication symbol detection model.