Frequency domain data feature-based molecular communication neural network signal detection scheme
The molecular communication neural network signal detection scheme addresses noise robustness and adaptability issues by employing frequency domain data features and a double-layer LSTM network, enhancing detection accuracy and reducing bit error rates in dynamic environments.
Patent Information
- Application Number
- CN202411197730.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-29
- Publication Date
- 2024-12-03
AI Technical Summary
Existing molecular communication signal detection methods suffer from poor noise robustness, real-time performance, adaptability, and insufficient utilization of frequency domain characteristics, leading to inaccurate and unstable signal detection in dynamically changing environments.
A molecular communication neural network signal detection scheme based on frequency domain data features, utilizing a single-input single-output system model, DCT transformation, and a double-layer LSTM neural network to process and analyze signal data, enhancing adaptability and robustness.
The scheme significantly reduces bit error rate and improves system performance by capturing long-term and short-term dependencies, offering high computing efficiency and adaptability in real-time processing.
Smart Images

Figure CN119070902A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of signal detection, and in particular to a molecular communication neural network signal detection scheme based on frequency domain data features. Background Art
[0002] Molecular communication is an emerging communication technology that uses chemical molecules as information carriers to transmit information through mechanisms such as molecular diffusion, convection, and reaction. Molecular communication is suitable for communication scenarios at the microscale due to its low power consumption and biocompatibility, and is widely used in nanobiomedicine, indicator environmental monitoring and other fields. Traditional communication technology based on electromagnetic waves as carriers is also expected to be applied in microscale scenarios, but its inherent biocompatibility and energy attenuation problems restrict its feasibility. For example, electromagnetic waves will show significant energy attenuation in channel media such as liquids, which limits their effective transmission distance, and may have adverse effects on biological tissues at certain frequencies and intensities, while molecular communication mechanisms are widely used by organisms. Therefore, in terms of biocompatibility and energy efficiency, molecular communication shows greater advantages than electromagnetic wave communication. In a molecular communication system, the transmitter encodes information by releasing specific chemical molecules. These information molecules are transmitted in the channel medium and further detected and decoded by the receiver.
[0003] Since the molecular diffusion process is greatly affected by factors such as noise and interference, the accuracy and reliability of signal detection and decoding become one of the key technical challenges. In molecular communication, accurate signal detection is the basis for ensuring system reliability and effectiveness, and directly affects the quality and performance of communication. Traditional signal detection methods mainly rely on time domain analysis and detection methods based on statistical features. These methods perform detection by analyzing the time series characteristics and basic statistics (such as mean, variance, etc.) of the signal. Common technical solutions include signal detection methods based on time domain and signal detection methods based on statistical features. The former, such as correlation detection method and matched filter, judge the existence of the signal by calculating the correlation between the received signal and the known template signal; the latter, such as Bayesian detection and maximum likelihood detection, judge the existence of the signal by the statistical characteristics of the signal.
[0004] Although the existing molecular communication signal detection methods have achieved certain results, there are still some significant defects and deficiencies in practical applications. First, poor noise robustness is a prominent problem. The performance of existing time domain analysis methods is significantly reduced in high noise environments, and it is difficult to accurately distinguish between signals and noise. The statistical feature method relies on the statistical properties of the signal, and the detection accuracy is low when the noise interference is strong. Secondly, poor real-time performance is also a major problem. Many traditional methods require a lot of calculations, such as correlation operations and Bayesian inference calculations, resulting in poor real-time performance and unable to meet the needs of rapid response. For example, correlation detection and matched filters require a large number of convolution operations in the signal processing process, which has high computational complexity; Bayesian detection and maximum likelihood detection require complex probability calculations and parameter estimation, and also face the problem of insufficient real-time performance.
[0005] In addition, poor adaptability is also a significant disadvantage of existing methods. Traditional methods often rely on fixed templates or statistical models, which are difficult to adapt to the dynamically changing molecular communication environment. The lack of adaptability makes it impossible to adjust the detection strategy according to environmental changes, resulting in unstable detection effects in different environments. For example, matched filters and correlation detection methods rely on pre-set signal templates. When the signal pattern changes, the detection performance will drop significantly; Bayesian detection and maximum likelihood detection rely on the prior statistical model of the signal. In the face of complex and changeable channel environments, it is difficult to maintain a stable detection effect.
[0006] Finally, the lack of input features in the existing technical solutions is also a problem that needs to be solved in the process of molecular communication signal detection. Traditional methods mainly rely on time domain or simple statistical features, and fail to fully utilize the frequency domain characteristics of the signal. The significant characteristics of the signal in the frequency domain are ignored, resulting in low effectiveness and accuracy of signal detection. For example, the time domain analysis method only focuses on the time series characteristics of the signal and fails to capture important information in the frequency domain; while the statistical feature method mainly relies on basic statistics such as the mean and variance of the signal, which cannot fully reflect the frequency domain characteristics and structural information of the signal.
[0007] In view of the shortcomings of the above-mentioned prior art solutions, the present invention proposes a molecular communication neural network signal detection solution based on frequency domain data features to solve the problem. Summary of the invention
[0008] The purpose of the present invention is to provide a molecular communication neural network signal detection solution based on frequency domain data characteristics, which solves the problems in the above-mentioned background technology.
[0009] To achieve the above object, the present invention provides the following technical solution: a molecular communication neural network signal detection solution based on frequency domain data features, comprising the following steps:
[0010] S1: Establish a molecular communication single-input single-output system model;
[0011] S2: Compute the channel impulse response of a passive receiver in a three-dimensional unbounded space;
[0012] S3: Based on step S2, an expression for the number of molecules at the receiver is obtained when considering inter-symbol interference and signal-correlated noise;
[0013] S4: Perform DCT transformation on the data set generated by the formula and perform standardization;
[0014] S5: Generate a dataset using the formula and use it for LSTM-based detector training and testing.
[0015] Preferably, the system model of step S1 consists of a point source transmitter, a diffusion channel and a passive spherical receiver. The modulation method uses on-off keying modulation. The transmitter releases messenger molecules, which propagate in the medium by random diffusion. The receiver has a radius of R and a volume of Its distance from the transmitter is d. The messenger molecules move freely in the medium. When they diffuse to the receiver position, some molecules are absorbed by the receiver. The receiver decodes the transmitted information by detecting the changes in the molecular concentration on its surface or inside. This system model realizes information transmission through the diffusion and reception of messenger molecules. Its performance is affected by the medium characteristics and system geometric parameters.
[0016] Preferably, in step S2, in a three-dimensional solution-free space, molecular diffusion follows the following diffusion equation:
[0017]
[0018] Among them, n(r, t) represents the concentration of signal molecules, α is the diffusion coefficient, and it is assumed that at time t1, M molecules are released instantly from position r1. Combining the diffusion equation, the concentration equation is obtained:
[0019]
[0020] To further analyze the intensity of the signal at the receiver, the time it takes for the molecule to diffuse from the release position to the receiver position is considered. This time is related to the distance ‖r-r1‖ that the molecule travels from the source position to the receiver position and can be combined with the diffusion coefficient α to determine the time evolution of the signal intensity;
[0021] According to the nature of the diffusion process, we can define a critical time t p , which corresponds to the time when the signal reaches its peak at the receiver, t p It can be expressed by the following relationship:
[0022]
[0023] At time t p At t, the signal strength received by the receiver reaches its maximum value, so the concentration distribution n(r,t) can be rewritten as p Related forms:
[0024]
[0025] On this basis, CIR represents the signal intensity received at the receiver at a certain point in time, which can be regarded as the molecular concentration in the receiver volume V. R The integral within , so CIR can be expressed as:
[0026]
[0027] Where: V R is the volume of the receiver,
[0028] Preferably, the signal at the molecular receiver in step S3 is expressed as:
[0029]
[0030] Among them, S i is the i-th transmitted symbol value, I(t) and n(t) correspond to the influence of ISI and noise respectively, which can be expanded by the following formula:
[0031]
[0032] T in the formula b is the symbol duration of the system. In molecular communication, the received signal is not only the current symbol sent, but also includes the influence of the previous and next symbols. The formula shows that the interference signal I(t) received at time t is due to the previous symbol S i-l The channel impulse response h(t+iT b ) The result of the joint action;
[0033] Since molecular diffusion is a random process, the noise exhibits the characteristics of random fluctuations. This fluctuation is modeled as a zero-mean normal distribution, which means that the expected value of the noise is zero, that is, n(t)~N(0,σ 2 ).
[0034] Preferably, the data set features generated by the formula in step S4 are time domain features. In order to use the frequency domain features of the signal, the sampled signal is subjected to DCT transform. The use of discrete cosine transform can concentrate most of the signal's energy on fewer DCT coefficients, which is conducive to coefficient compression and reduces data redundancy. At the same time, discrete cosine transform processes real number sequences, avoiding the complexity of complex number operations. We thus obtain the frequency domain data set of the signal, which is defined as:
[0035]
[0036] N represents the number of sampling points for each symbol, which is 100 in this model. y(n) is the signal received at the molecular receiver. k is the frequency domain index, which ranges from 0≤k<100. In order to facilitate analysis and processing, the transformed data set is normalized using Min-Max, and the formula is updated as follows:
[0037]
[0038] Preferably, step S5 is based on an LSTM neural network model for processing low received signal noise data, and through multiple training and verification, the optimal model and threshold are found to improve the accuracy of prediction and reduce the bit error rate. The LSTM neural network includes an input layer, an LSTM layer, a fully connected layer, a sigmoid activation layer, and a regression layer. During the training process, the Adam optimization algorithm is used to optimize the network parameters, the maximum training round is 1000 rounds, the initial learning rate is 0.0005, the data set covers two categories, a training set and a test set, and is divided in a ratio of 8:2. The specific steps include data loading and division, data preprocessing, network training, model prediction, threshold adjustment, binary conversion, and calculation of accuracy and BER;
[0039] In model evaluation, the performance of the model is evaluated through multiple training runs, and the average accuracy and average BER are recorded. In each run, the symbol index of the prediction error and the corresponding actual label and predicted value are recorded for subsequent analysis. The formula
[0040]
[0041] Calculate the loss function, where y i is the actual value, is the predicted value, N is the number of samples. During the training process, the model parameters are optimized by minimizing the loss function L, and the BER is optimized by dynamically adjusting the threshold. The results of each training are recorded, and an LSTM model with the best performance in processing low received signal noise data is successfully found, and its stability and reliability are verified.
[0042] The present invention provides a molecular communication neural network signal detection scheme based on frequency domain data features. The molecular communication neural network signal detection scheme based on frequency domain data features has the following beneficial effects:
[0043] 1. Compared with the traditional MMSE-FDE and ZF-FDE methods, the present invention shows significant advantages in bit error rate, adaptability and robustness. The double-layer LSTM structure can effectively capture the long-term and short-term dependencies of signals, overcome the strong dependence of traditional methods on channel prior information, and enhance the flexibility of the system.
[0044] 2. The present invention also has high computing efficiency in real-time processing performance, making it more suitable for dynamically changing communication environments. Experimental verification shows that the frequency domain-based double-layer LSTM detection method not only reduces the bit error rate, but also improves the overall performance of the system, providing new ideas and solutions for the development of molecular communication technology. BRIEF DESCRIPTION OF THE DRAWINGS
[0045] Figure 1 A schematic diagram of receiving signals of the present invention;
[0046] Figure 2 It is a schematic diagram of the process flow of the present invention;
[0047] Figure 3 It is a schematic diagram of the double-layer LSTM structure of the present invention;
[0048] Figure 4 It is a schematic diagram of the slow-propagation molecular communication system of the present invention;
[0049] Figure 5 It is a schematic diagram of the rapid propagation molecular communication system of the present invention. DETAILED DESCRIPTION
[0050] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described with reference to the accompanying drawings.
[0051] A preferred embodiment of the molecular communication neural network signal detection scheme based on frequency domain data features provided by the present invention is as follows: Figures 1 to 5 As shown:
[0052] A molecular communication neural network signal detection scheme based on frequency domain data features includes the following steps:
[0053] S1: Establish a molecular communication single-input single-output system model. The system model consists of a point source transmitter, a diffusion channel, and a passive spherical receiver. The modulation method uses on-off keying modulation. The transmitter releases messenger molecules, which propagate in the medium by random diffusion. The receiver radius is R and the volume is The distance between it and the transmitter is d. The messenger molecules move freely in the medium. When they diffuse to the receiver position, some molecules are absorbed by the receiver. The receiver decodes the transmitted information by detecting the change of molecular concentration on its surface or inside. The system model realizes information transmission through the diffusion and reception of messenger molecules. Its performance is affected by the medium characteristics and system geometric parameters.
[0054] S2: Calculate the channel impulse response of a passive receiver in a three-dimensional unbounded space. In a three-dimensional unsolvable space, molecular diffusion follows the following diffusion equation:
[0055]
[0056] Among them, n(r, t) represents the concentration of signal molecules, α is the diffusion coefficient, and it is assumed that at time t1, M molecules are released instantly from position r1. Combining the diffusion equation, the concentration equation is obtained:
[0057]
[0058] To further analyze the intensity of the signal at the receiver, the time it takes for the molecule to diffuse from the release position to the receiver position is considered. This time is related to the distance ‖r-r1‖ that the molecule travels from the source position to the receiver position and can be combined with the diffusion coefficient α to determine the time evolution of the signal intensity;
[0059] According to the nature of the diffusion process, we can define a critical time t p , which corresponds to the time when the signal reaches its peak at the receiver, t p It can be expressed by the following relationship:
[0060]
[0061] At time t p At t, the signal strength received by the receiver reaches its maximum value, so the concentration distribution n(r,t) can be rewritten as p Related forms:
[0062]
[0063] On this basis, CIR represents the signal intensity received at the receiver at a certain point in time, which can be regarded as the molecular concentration in the receiver volume V. R The integral within , so CIR can be expressed as:
[0064]
[0065] Where: V R is the volume of the receiver,
[0066] S3: Based on step S2, an expression for the number of molecules at the receiver is obtained when considering inter-symbol interference and signal-correlated noise. Here, the signal at the molecular receiver is expressed as:
[0067]
[0068] Among them, S i is the i-th transmitted symbol value, I(t) and n(t) correspond to the influence of ISI and noise respectively, which can be expanded by the following formula:
[0069]
[0070] T in the formula b is the symbol duration of the system. In molecular communication, the received signal is not only the current symbol sent, but also includes the influence of the previous and next symbols. The formula shows that the interference signal I(t) received at time t is due to the previous symbol S i-l The channel impulse response h(t+iT b ) The result of the joint action;
[0071] Since molecular diffusion is a random process, the noise exhibits the characteristics of random fluctuations. This fluctuation is modeled as a zero-mean normal distribution, which means that the expected value of the noise is zero, that is, n(t)~N(0,σ 2 );
[0072] S4: The data set generated by the formula is subjected to DCT transformation and standardized. The data set generated by the formula is characterized by time domain features. In order to use the frequency domain features of the signal, the sampled signal is subjected to DCT transformation. The use of discrete cosine transform can concentrate most of the signal's energy on fewer DCT coefficients, which is conducive to coefficient compression and reduces data redundancy. At the same time, discrete cosine transform processes real number sequences, avoiding the complexity of complex number operations. We thus obtain the frequency domain data set of the signal, which is defined as:
[0073]
[0074] N represents the number of sampling points for each symbol, which is 100 in this model. y(n) is the signal received at the molecular receiver. k is the frequency domain index, which ranges from 0≤k<100. In order to facilitate analysis and processing, the transformed data set is normalized using Min-Max, and the formula is updated as follows:
[0075]
[0076] S5: Generate a data set using the formula and use it for LSTM-based detector training and testing. The LSTM-based neural network model is used to process low received signal noise data. Through multiple training and verification, the optimal model and threshold are found to improve the prediction accuracy and reduce the bit error rate. The LSTM neural network contains an input layer, an LSTM layer, a fully connected layer, a sigmoid activation layer, and a regression layer. During the training process, the Adam optimization algorithm is used to optimize the network parameters. The maximum number of training rounds is 1000 rounds, and the initial learning rate is 0.0005. The data set covers two categories: training set and test set, and is divided into 8:2 ratios. The specific steps include data loading and division, data preprocessing, network training, model prediction, threshold adjustment, binary conversion, and calculation of accuracy and BER;
[0077] In model evaluation, the performance of the model is evaluated through multiple training runs, and the average accuracy and average BER are recorded. In each run, the symbol index of the prediction error and the corresponding actual label and predicted value are recorded for subsequent analysis. The formula
[0078]
[0079] Calculate the loss function, where y i is the actual value, is the predicted value, N is the number of samples. During the training process, the model parameters are optimized by minimizing the loss function L, and the BER is optimized by dynamically adjusting the threshold. The results of each training are recorded, and an LSTM model with the best performance in processing low received signal noise data is successfully found, and its stability and reliability are verified.
[0080] in, Figure 2 The program flow chart of the method proposed in this design is shown.
[0081] The technical concept of the present invention is to solve the problems of poor noise robustness, poor real-time performance, poor adaptability and insufficient feature utilization by introducing normalization processing, DCT transform feature extraction and neural network learning. First, the original signal is normalized to eliminate amplitude differences; then, the frequency domain features are extracted by DCT to highlight the main features of the signal and suppress noise; finally, the powerful learning ability of the neural network is used to identify and adaptively adjust the frequency domain features.
[0082] Figure 3The dual-layer LSTM structure is demonstrated, which effectively captures the complex features in time series data through multi-level processing. First, the input layer receives the original signal data and passes it to the first layer of LSTM, which uses its unique gating mechanism and memory units to identify and store long-term dependencies in the input data. Then, the second layer of LSTM further extracts features at a higher level of abstraction to enhance the expressiveness of the model. The output passes through a fully connected layer, which maps the output of the LSTM layer to a lower dimension, and applies a sigmoid activation function to convert the result into a probability value between 0 and 1, which is suitable for classification tasks. The final regression layer generates the prediction results of the model, enabling it to perform robust signal processing in complex noise environments, thereby effectively reducing BER and improving the accuracy of signal recognition.
[0083] Figure 4 The BER comparison of the double-layer LSTM detector, minimum mean square error frequency domain equalization (MMSE-FDE) and zero-forcing frequency domain equalization (ZF-FDE) in a slow-propagating molecular communication system is shown. From the figure, we can see that with the increase of molecular concentration, the bit error rate shows an overall downward trend. Among them, the double-layer LSTM detector always shows the lowest bit error rate, which is better than MMSE-FDE and ZF-FDE, which perform similarly but slightly higher. It shows its advantage in processing complex signals. This difference may be due to LSTM's ability to capture long-term dependencies in time series data, allowing it to adapt to dynamically changing environments more effectively.
[0084] Figure 5 The BER comparison of the double-layer LSTM detector, MMSE-FDE and ZF-FDE in a fast-transmitting molecular communication system is shown. In a fast-transmitting molecular communication system, since the signal propagates quickly in liquid or gas, and there may be multiple interference factors in the environment, resulting in signal attenuation and distortion, classic algorithms such as MMSE-FDE and forced ZF-FDE can reduce the bit error rate to a certain extent when facing high-speed transmission, but their effects are still limited under high-speed transmission conditions. This is because these methods usually rely on prior information and assumptions and are difficult to adapt to dynamically changing signal environments. The double-layer LSTM detector shows significant advantages under fast transmission conditions, and its BER is always lower than the other two methods in the entire concentration range. This result shows that the double-layer LSTM detector has stronger adaptability and robustness, can effectively cope with rapidly changing signal environments, and significantly reduce the bit error rate.
[0085] The above description is only an illustrative specific implementation mode of the present invention and is not intended to limit the scope of the present invention. Any equivalent changes and modifications made by any technician in the field without departing from the concept and principle of the present invention should fall within the scope of protection of the present invention. It should also be noted that the various components of the present invention are not limited to the above-mentioned overall application. The various technical features described in the specification of the present invention can be selected one by one or multiple ones can be selected and used in combination according to actual needs. Therefore, the present invention should naturally cover other combinations and specific applications related to the invention point of this case.
Claims
1. A molecular communication neural network signal detection scheme based on frequency domain data features, characterized by: The following steps are involved: S1: Establish a molecular communication single-input single-output system model; S2: Compute the channel impulse response of a passive receiver in a three-dimensional unbounded space; S3: Based on step S2, an expression for the number of molecules at the receiver is obtained when considering inter-symbol interference and signal-correlated noise; S4: Perform DCT transformation on the data set generated by the formula and perform standardization; S5: Generate a dataset using the formula and use it for LSTM-based detector training and testing.
2. According to claim 1, a molecular communication neural network signal detection scheme based on frequency domain data features is characterized by: The system model of step S1 consists of a point source transmitter, a diffusion channel and a passive spherical receiver. The modulation method uses on-off keying modulation. The transmitter releases messenger molecules, which propagate in the medium by random diffusion. The receiver has a radius of R and a volume of Its distance from the transmitter is d. The messenger molecules move freely in the medium. When they diffuse to the receiver position, some molecules are absorbed by the receiver. The receiver decodes the transmitted information by detecting the changes in the molecular concentration on its surface or inside. This system model realizes information transmission through the diffusion and reception of messenger molecules. Its performance is affected by the medium characteristics and system geometric parameters.
3. According to claim 1, a molecular communication neural network signal detection scheme based on frequency domain data features is characterized in that: In the three-dimensional solution-free space, the molecular diffusion follows the following diffusion equation: Among them, n(r, t) represents the concentration of signal molecules, α is the diffusion coefficient, and it is assumed that at time t1, M molecules are released instantly from position r1. Combining the diffusion equation, the concentration equation is obtained: To further analyze the intensity of the signal at the receiver, the time it takes for the molecule to diffuse from the release position to the receiver position is considered. This time is related to the distance ‖r-r1‖ that the molecule travels from the source position to the receiver position and can be combined with the diffusion coefficient α to determine the time evolution of the signal intensity; According to the nature of the diffusion process, we can define a critical time t p , which corresponds to the time when the signal reaches its peak at the receiver, t p It can be expressed by the following relationship: At time t p At t, the signal strength received by the receiver reaches its maximum value, so the concentration distribution n(r,t) can be rewritten as p Related forms: On this basis, CIR represents the signal intensity received at the receiver at a certain point in time, which can be regarded as the molecular concentration in the receiver volume V. R The integral within , so CIR can be expressed as: Where: V R is the volume of the receiver, 4. According to claim 1, a molecular communication neural network signal detection scheme based on frequency domain data features is characterized by: The signal at the molecular receiver in step S3 is expressed as: Among them, S i is the i-th transmitted symbol value, I(t) and n(t) correspond to the influence of ISI and noise respectively, which can be expanded by the following formula: T in the formula b is the symbol duration of the system. In molecular communication, the received signal is not only the current symbol sent, but also includes the influence of the previous and next symbols. The formula shows that the interference signal I(t) received at time t is due to the previous symbol S i-l The channel impulse response h(t+iT b ) The result of the joint action; Since molecular diffusion is a random process, the noise exhibits the characteristics of random fluctuations, which are modeled as zero-mean normal distribution, meaning that the expected value of the noise is zero, that is, n(t)~N(0,σ 2 ).
5. According to claim 1, a molecular communication neural network signal detection scheme based on frequency domain data features is characterized by: The data set features generated by the formula in step S4 are time domain features. In order to use the frequency domain features of the signal, the sampled signal is transformed by DCT. The use of discrete cosine transform can concentrate most of the signal's energy on fewer DCT coefficients, which is conducive to coefficient compression and reduces data redundancy. At the same time, discrete cosine transform processes real number sequences, avoiding the complexity of complex number operations. We thus obtain the frequency domain data set of the signal, which is defined as: N represents the number of sampling points for each symbol, which is 100 in this model. y(n) is the signal received at the molecular receiver. k is the frequency domain index, which ranges from 0≤k<100. In order to facilitate analysis and processing, the transformed data set is normalized using Min-Max, and the formula is updated as follows:
6. The molecular communication neural network signal detection scheme based on frequency domain data features according to claim 1 is characterized in that: Step S5 is based on the LSTM neural network model, which is used to process low received signal noise data. Through multiple training and verification, the optimal model and threshold are found to improve the prediction accuracy and reduce the bit error rate. The LSTM neural network includes an input layer, an LSTM layer, a fully connected layer, a sigmoid activation layer, and a regression layer. During the training process, the Adam optimization algorithm is used to optimize the network parameters. The maximum number of training rounds is 1000 rounds, and the initial learning rate is 0.0005. The data set covers two categories: training set and test set, and is divided into 8:2 ratios. The specific steps include data loading and division, data preprocessing, network training, model prediction, threshold adjustment, binary conversion, and calculation of accuracy and BER; In model evaluation, the performance of the model is evaluated through multiple training runs, and the average accuracy and average BER are recorded. In each run, the symbol index of the prediction error and the corresponding actual label and predicted value are recorded for subsequent analysis. The formula Calculate the loss function, where y i is the actual value, is the predicted value, N is the number of samples. During the training process, the model parameters are optimized by minimizing the loss function L, and the BER is optimized by dynamically adjusting the threshold. The results of each training are recorded, and an LSTM model with the best performance in processing low received signal noise data is successfully found, and its stability and reliability are verified.
Citation Information
Cited By
Molecular communication channel adaptation method and device based on transfer learning
CN121603134A
A method and device for adapting a molecular communication channel based on transfer learning
CN121603134B
