Method for optimizing the number of molecules released in a multiple-input multiple-output molecular communication system

Abstract

This invention discloses a method for optimizing the number of molecules released in a multi-input multi-output (MIMO) molecular communication system. The method includes: acquiring simulation data and establishing a time-varying channel impulse response function (TMR); minimizing the error between the predicted value of the TMR and the simulation data using a nonlinear fitting method to obtain a well-fitted TMR; deriving the average bit error rate (BER) of the MIMO system in the m-th time slot based on the fitted TMR, and establishing a multivariate optimization objective function for the current time slot; establishing and training a deep neural network; and inputting the initial distance vectors formed by each link into the trained deep neural network to obtain the optimal molecule release number vector for each sending nanomachine. This method can optimize the number of molecules released by the sending nanomachines in less time and with lower computational complexity, significantly reducing the average BER.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical fields of Internet of Things, nanotechnology and communication technology, and particularly relates to a molecule release quantity optimization method of a multiple-input multiple-output molecule communication system. BACKGROUND

[0002] With the development of science and technology and the further improvement of human communication demand, molecule communication begins to occupy a place in the communication field and becomes an important communication means in biology, environmental protection and microcosmic communication, etc. Due to its good biocompatibility, it can meet the communication demand of people in the biological environment, thereby realizing important applications in biological research, medical rescue, environmental protection and other fields. Molecule communication is a communication form using molecules to transmit information, which is used to transmit information between nanomachines and nanorobots, between nanomachines and large machines, and between nanorobots and cells. One of the most important applications of molecule communication networks in biology and medicine is drug delivery in the human body, which lays the foundation for the networking of biological nanomaterials. In order to increase the performance of molecule communication and improve the efficiency of the communication system, further meet the effect of molecule communication in drug delivery application in the human body, a multiple-input multiple-output mode communication system structure can be used. The multiple-input multiple-output mode molecule communication system can satisfy multiple users to communicate with the receiver at the same time, meet the communication demand of multiple links, and has good application ability for drug delivery application in the human body.

[0003] The multiple-input multiple-output (MIMO) molecule communication system is a molecule communication system using multiple transmitters and receivers to send and receive. Since the multiple-input multiple-output molecule communication system can improve the data transmission rate, reduce interference and increase the communication range and reduce the power consumption of the system, it is a very promising nanonetwork communication technology. In the multiple-input multiple-output molecule communication network, the molecule release quantity of each user, i.e. each sender nanomachine, is limited. When the distances of each link are different, under the condition that the molecule release quantity has upper and lower bound constraints, how to set the optimal molecule release quantity for the corresponding sender nanomachine of each link. When there are two senders, the optimal molecule release quantity can be searched by simple exhaustive search. However, when there are more senders, the time required by the exhaustive algorithm increases geometrically. How to obtain the optimal molecule release quantity of each sender in as little time as possible is a very challenging work. SUMMARY

[0004] The purpose of the present application is to solve the above problems, and a molecule release quantity optimization method of a multiple-input multiple-output molecule communication system is proposed. The molecule release quantity optimization of the sender nanomachine in the multiple-input multiple-output molecule communication system can be completed with less time and lower computational complexity, which greatly reduces the average bit error rate.

[0005] To achieve the above object, the technical scheme adopted by the present application is:

[0006] The present application provides a method for optimizing the number of molecules released by a multiple-input multiple-output molecular communication system, which comprises a plurality of sender nanomachines and a receiver nanomachine.

[0007] S1, obtain simulation data and establish a time-varying channel impulse response function of the multiple-input multiple-output molecular communication system, wherein the time-varying channel impulse response function is The formula is as follows:

[0008]

[0009] Wherein,

[0010] In the formula, is the initial distance between the ith sender nanomachine Tx i and the receiver nanomachine Rx1, i = 1 ~ M, represents an initial distance vector formed by each link, is the volume of the receiver nanomachine Rx1, t is the time, τ is the relative time at time t, D A is the molecular diffusion coefficient, is the diffusion coefficient of the receiver nanomachine, is the diffusion coefficient of the ith sender nanomachine Tx i , erfc is the complementary error function, and b1, b2 and b3 are fitting parameters.

[0011] S2, minimize the error between the predicted value of the time-varying channel impulse response function and the simulation data by using a nonlinear fitting method, and obtain a fitted time-varying channel impulse response function.

[0012] S3, derive the average bit error rate of the multiple-input multiple-output molecular communication system in the mth time slot according to the fitted time-varying channel impulse response function, and establish a multivariate optimization objective function for the number of molecules released by each sender nanomachine in the current time slot, wherein:

[0013] The average bit error rate Pe avg [m] of the multiple-input multiple-output molecular communication system in the mth time slot is as follows:

[0014]

[0015] The multivariate optimization objective function is as follows:

[0016]

[0017] In the formula, Indicates link Tx i →The bit error rate of Rx1 in the m-th time slot, Tx represents the i-th sending nanomachine. i The number of molecules released, where N represents the vector of the number of molecules released by each sending nanomachine. T represents transpose, ψ represents the upper bound of the number of molecules released, and Ψ represents the upper bound of the number of molecules released.

[0018] S4. Build and train a deep neural network;

[0019] S5. Input the initial distance vector formed by each preset link into the trained deep neural network to obtain the optimal molecular release vector for each sending nanomachine.

[0020] Preferably, the i-th sending nanomachine Tx i Initial distance between the receiver nanomachine Rx1 and the receiver nanomachine The calculation is as follows:

[0021]

[0022] In the formula, (a i ,b i ,c i ) represents the i-th sending nanomachine Tx i The coordinates (a1,b1,c1) represent the coordinates of the receiver nanomachine Rx1, and r is the radius of the receiver nanomachine Rx1.

[0023] Preferably, the process of acquiring simulation data is as follows:

[0024] S11. A Brownian motion simulator is used to simulate the random movement of each nanomachine in a multi-input multi-output molecular communication system.

[0025] S12. During the simulation, the simulation duration is divided into several independent time slots Δt, and the duration of each time slot Δt is T. s At the beginning of each time slot Δt, if the transmitted bit is '1', each sending nanomachine will transmit a preset number of molecules; if the transmitted bit is '0', each sending nanomachine will not transmit molecules.

[0026] S13. During the simulation, the receiving nanomachine counts the number of molecules received at each time slot Δt to obtain the number of molecules received from each sending nanomachine in each time slot Δt, thus forming simulation data.

[0027] Preferably, link Txi →Bit error rate of Rx1 in the m-th time slot The calculation process is as follows:

[0028] S31, Assume the i-th sending nanomachine Tx i If molecules are released at the beginning of the m-th time slot and propagate via free diffusion, the probability of them being received by the receiving nanomachine Rx1 in the n-th time slot can be calculated based on the fitted time-varying channel impulse response function.

[0029]

[0030] In the formula, T s For the duration of each time slot, the i-th sending nanomachine Tx i At t=(m-1)T s The molecules are released at time, and the receiving nanomachine Rx1 is at t = (nm)T s +τ counts the number of molecules received;

[0031] S32, will and Represented as a binomial distribution:

[0032]

[0033]

[0034] In the formula, This indicates that the receiver nanomachine Rx1 receives the data from the sender nanomachine Tx in the m-th time slot. i The number of molecules released, This represents the inter-symbol interference of the receiver's nanomachine Rx1 in the m-th time slot. Tx represents the i-th sending nanomachine. i The bits waiting to be transmitted at the beginning of the m-th time slot, for the i-th sender nanomachine Tx i The bits waiting to be sent are encoded by releasing different numbers of molecules, and this is done by releasing them at the beginning of the time slot. One molecule is used to transmit bit '1', and zero molecules are released to transmit bit '0'. Tx represents the i-th sending nanomachine. i The probability that the molecules released in the m-th time slot are received by the receiving nanomachine Rx1 in the m-th time slot;

[0035] S33. Use the Poisson distribution to express the binomial distribution form in step S32 as a Poisson distribution form:

[0036]

[0037]

[0038] In the formula, Tx represents the i-th sending nanomachine. i The probability of transmitting bit '1' in the nth time slot;

[0039] S34. Calculate the probability distribution of the number of molecules received by the receiver nanomachine Rx1 in the m-th time slot.

[0040]

[0041] S35. The MAP decision threshold that minimizes the average BER is derived using the maximum a posteriori probability estimation algorithm, as follows:

[0042] S351. Under assumptions H0 and H1, the probability distribution of the number of molecules received by the receiver nanomachine Rx1 in the m-th time slot is expressed as follows:

[0043]

[0044]

[0045] in,

[0046]

[0047]

[0048] In the formula, This represents the Poisson distribution coefficient under assumption H0. This represents the Poisson distribution coefficient under assumption H1, where H0 represents the Mth sender nanomachine Tx. M In the case of transmitting bit '0' in the m-th time slot, let H1 represent the M-th sending nanomachine Tx. M The case where bit '1' is sent in the m-th time slot;

[0049] S352, Assumption For link Tx i →The decision threshold of Rx1 in the m-th time slot, then link Tx i →Rx1

[0050] Decision bits for transmitting bits in the m-th time slot Represented as:

[0051]

[0052] Likelihood ratio test of the number of molecules received in the m-th time slot Represented as:

[0053]

[0054] in:

[0055]

[0056]

[0057] In the formula, Let H0 be the probability mass function. Let Tx be the probability mass function under hypothesis H1, and let Tx be the ith sender nanomachine. i The probability Pr(H0) of sending bit '0' in the m-th time slot is: The i-th sender nanomachine Tx i The probability Pr(H1) of sending bit '1' in the m-th time slot is:

[0058] S353. Obtain link Tx according to step S352. i → MAP decision threshold of Rx1 in the m-th time slot Represented as:

[0059]

[0060] S36. Derive link Tx based on MAP decision threshold. i →Bit error rate of Rx1 in the m-th time slot

[0061]

[0062] in,

[0063]

[0064]

[0065] Preferably, the deep neural network includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence, wherein the number of neurons in the first hidden layer, the second hidden layer, and the third hidden layer are 8, 6, and 6, respectively.

[0066] Preferably, the activation function of each neuron in each hidden layer is a rectified linear unit activation function, and if the input of the activation function is a positive number, the input is directly output; otherwise, the output is 0.

[0067] Preferably, the training process of the deep neural network is as follows:

[0068] S41. Obtain the first dataset, which includes initial distance vectors for different link lengths;

[0069] S42. Initialize the distance vector of the k-th link length in the first dataset. As the k-th input to a deep neural network, the corresponding output of the deep neural network is:

[0070]

[0071] S43, Adopt an extension strategy to Expand to V nearest neighbor solutions, as shown in the following formula:

[0072]

[0073] in,

[0074]

[0075] In the formula, Indicates Let be an M-dimensional Euclidean space centered at , and in this space, any M-dimensional vector with vector Let R be the distance between them, and ||·|| denote the Euclidean norm;

[0076] S44, Calculate d k The corresponding v-th neighboring solution average bit error rate

[0077] S45. Select the optimal solution from the average bit error rates of the current V nearest neighbor solutions.

[0078]

[0079] S46. Constructing Dataset Units And store it in the queue of the second dataset;

[0080] S47. Select δ of the latest dataset units in the second dataset to train the deep neural network once, set k = k + 1, and return to execute step S42 until the first dataset is traversed.

[0081] S48. Every L / δ training iterations, the actual output and the expected output of the deep neural network are compared, and the difference between the actual output and the expected output is recorded. After L training iterations, the deep neural network with the smallest difference is taken as the trained deep neural network, where L is the preset number of training iterations.

[0082] Preferably, the sending nanomachine is a point transmitter and the receiving nanomachine is a passive spherical receiver.

[0083] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0084] This invention studies a multiple-input multiple-output (MIMO) molecular communication system composed of multiple transmitting nanomachines and one receiving nanomachine. For the time-varying channel in the MIMO system, simulation data is used to fit a proposed time-varying channel impulse response function with fitting parameters to model the time-varying channel. The average bit error rate (BER) of the MIMO system is derived using the fitted BER. An optimization problem is established, with lower and upper bounds on the number of molecules released by all transmitting nanomachines, aiming to minimize the average BER of the system. Finally, a deep neural network is designed to optimize the number of molecules released for different link lengths in the MIMO system. The deep neural network is trained using an optimized training algorithm, enabling it to perform the optimization task effectively, thereby optimizing the number of molecules released by each transmitting nanomachine in the MIMO system. Experimental results show that, compared with other traditional optimization algorithms such as genetic algorithms, the trained deep neural network has a faster convergence speed and better optimization effect. Compared with exhaustive search, the deep neural network has a very short running time, but the optimization effect is close to or equal to that of exhaustive search. It uses less time and lower computational complexity to optimize the number of molecules released by the sending nanomachine in a multi-input multi-output molecular communication system, resulting in the lowest average bit error rate. It is suitable for drug design and drug implantation. Attached Figure Description

[0085] Figure 1 This is a flowchart of the method for optimizing the number of molecules released in the multi-input multi-output molecular communication system of the present invention;

[0086] Figure 2 This is a topological diagram of the multi-input multi-output molecular communication system of the present invention in Cartesian coordinates.

[0087] Figure 3 This is a schematic diagram illustrating the variation of the time-varying channel impulse response function of the present invention with simulation time;

[0088] Figure 4 This is a schematic diagram of fitting the time-varying channel impulse response function of the present invention;

[0089] Figure 5 This is a schematic diagram showing the relationship between the fitting parameters of this invention and d0 under different r and h.

[0090] Figure 6This is a schematic diagram illustrating the process by which the deep neural network of this invention optimizes the number of molecules released by each sending nanomachine.

[0091] Figure 7 This diagram illustrates the relationship between the unoptimized average bit error rate (BER) of a multiple-input multiple-output molecular communication system under random molecule release conditions and the optimal BER using the method of this invention, and the upper bound of the number of molecules released by the sending nanomachine.

[0092] Figure 8 This is a schematic diagram illustrating the relationship between the optimal number of molecules released in this invention and different constraint conditions;

[0093] Figure 9 This is a schematic diagram showing the variation of the optimal molecular release quantity corresponding to each sending nanomachine of the present invention under different constraint conditions.

[0094] Figure 10 This is a schematic diagram showing the relationship between the optimal number of molecules released by each sending nanomachine of the present invention and different link lengths.

[0095] Figure 11 This is a comparison chart of the average bit error rates of the present invention, genetic algorithm, and exhaustive search. Detailed Implementation

[0096] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0097] It should be noted that when a component is referred to as being "connected" to another component, it can be directly connected to the other component or there may be an intervening component. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein in the specification of this application is for the purpose of describing particular embodiments only and is not intended to limit the application.

[0098] To overcome the shortcomings of existing technologies and optimize the number of molecules released by the sender nanomachines in multi-input multi-output (MIMO) molecular communication systems with less time and computational complexity, and to make the optimization result as close as possible to the optimal solution, a deep neural network (DNN) is used to optimize the number of molecules released by the sender in MIMO molecular communication. To enable the deep neural network to perform the optimization task, a deep neural network training algorithm specifically designed for optimizing the number of molecules released by the sender nanomachines in molecular communication was developed. The deep neural network obtained through this training algorithm can complete the optimization task of the number of molecules released in molecular communication, thereby minimizing the average bit error rate of the MIMO molecular communication system.

[0099] like Figures 1-11 As shown, a method for optimizing the number of molecules released in a multiple-input multiple-output (MIMO) molecular communication system is presented. The MIMO molecular communication system includes M sending nanomachines and one receiving nanomachine. The method for optimizing the number of molecules released in the MIMO molecular communication system includes the following steps:

[0100] S1. Obtain simulation data and establish the time-varying channel impulse response function of the multi-input multi-output molecular communication system. The formula is as follows:

[0101]

[0102] in,

[0103] In the formula, For the i-th sending nanomachine Tx i The initial distance between the receiver nanomachine Rx1 and the receiver nanomachine, i = 1 to M, This represents the initial distance vector formed by each link. Let D be the volume of the receiving nanomachine Rx1, t be time, τ be the relative time to time t, and D be the volume of the receiving nanomachine Rx1. A The molecular diffusion coefficient is... The diffusion coefficient of the receiving nanomachine is . For the i-th sending nanomachine Tx i The diffusion coefficient is given by , erfc is the complement error function, and b1, b2, and b3 are fitting parameters. Here, b1 is used to study how inter-link interference affects the peak amplitude, while b2 and b3 are used to study the effect of the random mobility of the nanomachines on the peak amplitude.

[0104] In one embodiment, the i-th sending nanomachine Tx i Initial distance between the receiver nanomachine Rx1 and the receiver nanomachine The calculation is as follows:

[0105]

[0106] In the formula, (a i ,b i ,c i ) represents the i-th sending nanomachine Tx i The coordinates (a1,b1,c1) represent the coordinates of the receiver nanomachine Rx1, and r is the radius of the receiver nanomachine Rx1.

[0107] Specifically, this embodiment includes three sending nanomachines Tx1, Tx2, and Tx3, and one receiving nanomachine Rx1. It should be noted that this application is also applicable to the case of multiple receiving nanomachines, i.e., it can be considered as communication between each sending nanomachine and any receiving nanomachine. Figure 2 For the topology of a multiple-input multiple-output (MIMO) molecular communication system in Cartesian coordinates, the sending nanomachines Tx1, Tx2, and Tx3 are placed along the z-axis, with their Cartesian coordinates being (0,0,(2r+h)), (0,0,0), and (0,0,-(2r+h)), respectively. The receiving nanomachines Rx1, Rx2, and Rx3 are all spheres of radius r, with their centers parallel to the y-axis along the line connecting Tx1, Tx2, and Tx3, respectively. Therefore, their Cartesian coordinates are (0,(d0+r),(2r+h)), (0,(d0+r),0), and (0,(d0+r),-(2r+h)), respectively. For the receiving nanomachine Rx1, the initial distances between Tx1, Tx2, Tx3, and Rx1 are defined as follows: According to the distance calculation formula...

[0108]

[0109]

[0110]

[0111] In one embodiment, the process of acquiring simulation data is as follows:

[0112] S11. A Brownian motion simulator is used to simulate the random movement of each nanomachine in a multi-input multi-output molecular communication system.

[0113] S12. During the simulation, the simulation duration is divided into several independent time slots Δt, and the duration of each time slot Δt is T. s At the beginning of each time slot Δt, if the transmitted bit is '1', each sending nanomachine will transmit a preset number of molecules; if the transmitted bit is '0', each sending nanomachine will not transmit molecules.

[0114] S13. During the simulation, the receiving nanomachine counts the number of molecules received at each time slot Δt to obtain the number of molecules received from each sending nanomachine in each time slot Δt, thus forming simulation data.

[0115] A Brownian motion simulator was chosen to simulate the random movement of the nanomachines. In this simulation, the simulation duration was divided into many independent time slots Δt, and it was assumed that the duration of each time slot was equal to T. s At the beginning of each time slot, if the bit transmitted in the current time slot is '1', then Tx1, Tx2, and Tx3 transmit 5000 molecules; if the bit transmitted is '0', then Tx1, Tx2, and Tx3 do not transmit molecules. The receiving nanomachine Rx1 counts the number of molecules received every Δt throughout the simulation experiment. After sufficient experimentation, the number of molecules received by Rx1 from Tx1, Tx2, and Tx3 at each Δt is obtained.

[0116] Figure 3 The variation of the channel impulse response over simulation time is shown. Reference and These represent the number of molecules received at each time slot Δt during the entire simulation time for Tx1, Tx2, and Tx3, respectively. In this simulation experiment, the molecular diffusion coefficient D... A =5×10 -9 m 2 / s, the diffusion coefficients of the sending nanomachine and the receiving nanomachine are respectively and The topology parameters d0, h, and r are 15 μm, 1 μm, and 3 μm, respectively. d0 is the initial distance between Tx1 and Rx1 in the y-axis direction, and h is the initial distance between Rx1 and Rx2 in the z-axis direction. Figure 3 Discover, Each time slot has a unique corresponding peak value. However, due to the initial distance between Tx1, Tx2, Tx3 and Rx1... The peak amplitude times and peak values ​​for Tx1, Tx2, and Tx3 are not identical; they show shifts and reductions. Furthermore, within the second time slot, simulation results reveal the impact of the random mobility of the nanomachines on the channel impulse response.

[0117] In one embodiment, the transmitting nanomachine is a point transmitter, and the receiving nanomachine is a passive spherical receiver. For example, the receiving nanomachine is a passive spherical receiver with a radius of r.

[0118] S2. The error between the predicted value and the simulation data of the time-varying channel impulse response function is minimized by using a nonlinear fitting method to obtain a well-fitted time-varying channel impulse response function.

[0119] To obtain the optimal fitting parameters, nonlinear fitting is used to minimize the error between the predicted value and the simulated data of the time-varying channel impulse response function. The time-varying channel impulse response function after fitting is shown below. It was used to derive the average bit error rate of a multiple-input multiple-output molecular communication system.

[0120] Figure 4 This demonstrates how to use simulation data to fit the time-varying channel impulse response function. The detailed process. From the beginning to the end of the simulation experiment, and Both increase accordingly with the increase of simulation time, and at the same time It rose rapidly at the beginning of the simulation experiment and The fluctuations were small throughout the experiment. Table 1 lists the results obtained using nonlinear fitting. The corresponding fitting parameters are b1, b2, and b3.

[0121] Table 1

[0122]

[0123] S3. Based on the fitted time-varying channel impulse response function, derive the average bit error rate of the multi-input multi-output molecular communication system in the m-th time slot, and establish a multivariate optimization objective function for the number of nanomachine molecules released by each sender in the current time slot, where:

[0124] The average bit error rate Pe of a multiple-input multiple-output molecular communication system in the m-th time slot avg [m], the formula is as follows:

[0125]

[0126] The objective function for multivariate optimization is shown in the following formula:

[0127]

[0128] In the formula, Indicates link Tx i →The bit error rate of Rx1 in the m-th time slot, Tx represents the i-th sending nanomachine. i The number of molecules released, where N represents the vector of the number of molecules released by each sending nanomachine. T represents the transpose, ψ represents the upper bound of the number of molecules released, and Ψ represents the upper bound of the number of molecules released. The average bit error rate of the multi-input multi-output molecular communication system is optimized by establishing a constrained multivariate optimization objective function to optimize the number of molecules released by the sending nanomachine.

[0129] In one embodiment, link Tx i →Bit error rate of Rx1 in the m-th time slot The calculation process is as follows:

[0130] S31, Assume the i-th sending nanomachine Tx i If molecules are released at the beginning of the m-th time slot and propagate via free diffusion, the probability of them being received by the receiving nanomachine Rx1 in the n-th time slot can be calculated based on the fitted time-varying channel impulse response function.

[0131]

[0132] In the formula, T s For the duration of each time slot, the i-th sending nanomachine Tx i At t=(m-1)T s The molecules are released at time, and the receiving nanomachine Rx1 is at t = (nm)T s +τ represents the number of molecules received.

[0133] Using the well-fitted time-varying channel impulse response function and Let's derive the bit error rates for links Tx1→Rx1, Tx2→Rx1, and Tx3→Rx1. Assume Tx... i Molecules are released at the beginning of the m-th time slot, and then these molecules propagate through free diffusion, and are finally received by Rx1 in the n-th time slot.

[0134] In formula (5), Tx i At t=(m-1)T s Release molecules, Rx1 at t=(nm)T s +τ counts the number of received molecules. For Tx1, Tx2, and Tx3, different numbers of molecules are released to encode the bits waiting to be sent, for example, by releasing N molecules at the beginning of the time slot. Txi One molecule is used to transmit bit '1', and zero molecules are released to transmit bit '0'. It can be set according to actual needs. The molecules received by Rx1 in the m-th time slot mainly come from the molecules released by Tx1, Tx2 and Tx3 in the m-th time slot, as well as the molecules released by Tx1, Tx2 and Tx3 in the first (m-1) time slots. The molecules released by Tx1, Tx2 and Tx3 in the first (m-1) time slots cause inter-symbol interference (ISI).

[0135] S32, will and Represented as a binomial distribution:

[0136]

[0137]

[0138] In the formula, This indicates that the receiver nanomachine Rx1 receives the data from the sender nanomachine Tx in the m-th time slot. i The number of molecules released, This represents the inter-symbol interference of the receiver's nanomachine Rx1 in the m-th time slot. Tx represents the i-th sending nanomachine. i The bits waiting to be transmitted at the beginning of the m-th time slot, for the i-th sender nanomachine Tx i The bits waiting to be sent are encoded by releasing different numbers of molecules, and this is done by releasing them at the beginning of the time slot. One molecule is used to transmit bit '1', and zero molecules are released to transmit bit '0'. Tx represents the i-th sending nanomachine. i The probability that a molecule released in the m-th time slot is received by the receiving nanomachine Rx1 in the m-th time slot.

[0139] use and Let Rx1 represent the number of molecules received by Rx1 in the m-th time slot from Tx1, Tx2, and Tx3 released in the same time slot, using... Let represent the inter-symbol interference of Rx1 in the m-th time slot.

[0140] S33. Use the Poisson distribution to express the binomial distribution form in step S32 as a Poisson distribution form:

[0141]

[0142]

[0143] In the formula, Tx represents the i-th sending nanomachine.i The probability of transmitting bit '1' in the nth time slot;

[0144] S34. Calculate the probability distribution of the number of molecules received by the receiver nanomachine Rx1 in the m-th time slot.

[0145]

[0146] S35. The MAP decision threshold that minimizes the average BER is derived using the maximum a posteriori probability estimation algorithm, as follows:

[0147] S351. Using formulas (9) and (10), under assumptions H0 and H1, the probability distribution of the number of molecules received by the receiver nanomachine Rx1 in the m-th time slot is expressed as follows:

[0148]

[0149]

[0150] in,

[0151]

[0152]

[0153] In the formula, This represents the Poisson distribution coefficient under assumption H0. This represents the Poisson distribution coefficient under assumption H1, where H0 represents the Mth sender nanomachine Tx. M In the case of transmitting bit '0' in the m-th time slot, let H1 represent the M-th sending nanomachine Tx. M The case where bit '1' is sent in the m-th time slot;

[0154] S352, Assumption For link Tx i →The decision threshold of Rx1 in the m-th time slot, then link Tx i →Decision bit of Rx1 transmitting bits in the m-th time slot Represented as:

[0155]

[0156] Likelihood ratio test of the number of molecules received in the m-th time slot Represented as:

[0157]

[0158] in:

[0159]

[0160]

[0161] In the formula, Let H0 be the probability mass function. Let Tx be the probability mass function under hypothesis H1, and let Tx be the ith sender nanomachine. i The probability Pr(H0) of sending bit '0' in the m-th time slot is: The i-th sender nanomachine Tx i The probability Pr(H1) of sending bit '1' in the m-th time slot is:

[0162] S353. Obtain the link Tx according to step S352 (substitute formulas (17) and (18) into formula (16)). i → MAP decision threshold of Rx1 in the m-th time slot Represented as:

[0163]

[0164] S36. Derive link Tx based on MAP decision threshold. i →Bit error rate of Rx1 in the m-th time slot

[0165]

[0166] in,

[0167]

[0168]

[0169] S4. Build a deep neural network and train it.

[0170] In one embodiment, the deep neural network includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence, wherein the number of neurons in the first hidden layer, the second hidden layer, and the third hidden layer are 8, 6, and 6, respectively.

[0171] In one embodiment, the activation function of each neuron in each hidden layer is a rectified linear unit activation function, and if the input of the activation function is a positive number, the input is directly output; otherwise, the output is 0.

[0172] A deep neural network (DNN) is constructed for a multiple-input multiple-output (MIMO) molecular communication system. The input to this DNN is the length of different links in the system, and the output is the optimal number of molecules released by the sending nanomachine corresponding to each link. In the initialization phase, a DNN model with an input layer, three hidden layers, and an output layer is built to obtain the optimal number of molecules released by the sending nanomachine. For each hidden layer, the number of neurons is set to 8, 6, and 6 respectively. The activation function for each neuron in the hidden layer is the rectified linear unit activation function (RELU). If the input to the activation function is positive, the input is directly output; otherwise, 0 is output. The deep neural network is represented as follows:

[0173]

[0174] In the formula, This represents the output of a deep neural network. Tx represents the i-th sending nanomachine of the deep neural network output. i The optimal number of molecules to release, f w,b (d) Defines a deep neural network with hidden layer weights w and bias b. The input of this deep neural network is d. For each neuron in the hidden layer, first, a weighted sum is calculated based on the current input, then the bias b is added, and finally the accumulated result (weighted sum w) is calculated. sum The input is fed into the corresponding activation function to obtain the output of the corresponding neuron. Taking the neurons in the first hidden layer as an example, the weighted sum w of each neuron in the current hidden layer... sum The calculation is as follows:

[0175]

[0176] In the formula, w i Tx represents the i-th sending nanomachine. i Initial distance between the receiver nanomachine Rx1 and the receiver nanomachine The corresponding weights, where b is the bias.

[0177] For example, in a multiple-input multiple-output (MIMO) molecular communication system comprising three sending nanomachines and one receiving nanomachine, the input to the neuron is... and The corresponding weights are w1, w2, and w3, respectively. Therefore, the weighted sum of the neuron's inputs is:

[0178]

[0179] In this context, weights w1, w2, and w3 determine the degree of influence of the neuron's input on its output, and b is defined as the bias. Subsequently, the weighted sum w... sumIt is fed into the ReLU activation function to prepare for subsequent output. and Let f represent the optimal number of molecules released corresponding to Tx1, Tx2, and Tx3 output by the deep neural network. For simplicity, a parameterized function f is used. w,b (d) represents the deep neural network. and These represent the input and output of the deep neural network, respectively.

[0180] In one embodiment, the training process of the deep neural network is as follows:

[0181] S41. Obtain the first dataset, which includes initial distance vectors for different link lengths;

[0182] S42. Initialize the distance vector of the k-th link length in the first dataset. As the k-th input to a deep neural network, the corresponding output of the deep neural network is: Represented as:

[0183]

[0184] S43, Adopt an extension strategy to Expand to V nearest neighbor solutions, as shown in the following formula:

[0185]

[0186] in,

[0187]

[0188] In the formula, Indicates Let be an M-dimensional Euclidean space centered at , and in this space, any M-dimensional vector with vector Let R be the distance between them, and ||·|| denote the Euclidean norm;

[0189] S44, Calculate d k The corresponding v-th neighboring solution average bit error rate

[0190] S45. Select the optimal solution from the average bit error rates of the current V nearest neighbor solutions.

[0191]

[0192] S46. Constructing Dataset Units And store it in the queue of the second dataset;

[0193] S47. Select δ of the latest dataset units in the second dataset to train the deep neural network once, set k = k + 1, and return to execute step S42 until the first dataset is traversed.

[0194] S48. Every L / δ training iterations, the actual output and the expected output of the deep neural network are compared, and the difference between the actual output and the expected output is recorded. After L training iterations, the deep neural network with the smallest difference is taken as the trained deep neural network, where L is the preset number of training iterations.

[0195] S5. Input the initial distance vector formed by each preset link into the trained deep neural network to obtain the optimal molecular release vector for each sending nanomachine.

[0196] As iterations proceed, the output of the deep neural network gets closer and closer to the desired output because each data point in the training dataset becomes increasingly similar to the ideal output. Expanding and filtering the deep neural network's output yields outputs that better meet expectations. Adding these outputs to a second dataset and using them as training samples further enhances the deep neural network's ability to obtain the optimal output corresponding to the input. In other words, each training sample re-added to the second dataset is closer to the optimal molecule release number for its corresponding link length. Once sufficiently accurate training samples are collected, the deep neural network is retrained. As the training dataset is updated, the weights and biases in the deep neural network are also updated, and more accurate training samples improve the accuracy of the deep neural network's output. With increasing training iterations, the deep neural network's ability to optimize the molecule release number of the sending nanomachine in a multi-input multi-output molecular communication system becomes increasingly stronger. Finally, the deep neural network training is complete, resulting in a trained deep neural network. By setting the initial distance vector values ​​for each link and inputting these values ​​into the trained deep neural network, the optimal molecule release number for different link distances can be obtained.

[0197] exist Figure 5 (a) and Figure 5 (b) shows in detail and The relationship between the fitting parameter b1 and d0 under different r and h. Figure 5 In (a), The fitting parameter b1 in the model continuously decreases as d0 increases. This differs from... Figure 5 The changes of b1 in (a) Figure 5 (b) The fitting parameter b1 increases accordingly as d0 increases. Figure 5(b) It also reveals that r has a more significant impact on the fitting parameter b1. Figure 5 (c) and Figure 6 In (d), h is fixed for observation. The relationship between fitting parameters b2, b3 and d0 was analyzed, and the results showed that b2 and b3 did not fluctuate significantly as d0 increased.

[0198] Figure 7 This diagram illustrates the detailed process of optimizing the number of molecules released by each sending nanomachine in a multi-input multi-output molecular communication system based on a deep neural network. It shows each step from deep neural network initialization, training, to updating, where the input to the deep neural network is the length of each link, and the output is the optimal number of molecules released by the corresponding sending nanomachine for each link. The results demonstrate that by expanding and filtering the output of the deep neural network to repeatedly update the dataset, the process ensures that the dataset used to train the deep neural network closely approximates the input and the corresponding expected output. After repeated dataset updates and training, the output of the deep neural network reaches the optimal solution.

[0199] exist Figure 8 In this study, the relationship between the unoptimized average bit error rate (ABER) and the optimal ABER of a multi-input multi-output (MIMO) molecular communication system and the upper limit of the number of molecules released by the sender nanomachines was investigated. For the unoptimized ABER (where the number of molecules released is random), corresponding to Pe... avg The analysis assumes that the number of molecules released by Tx1, Tx2, and Tx3 is the same and equal to the upper limit of the number of molecules released. The results show that the unoptimized average bit error rate of the system decreases as the upper limit of the number of molecules released increases. This is because after increasing the upper limit of the number of molecules released, Tx1, Tx2, and Tx3 will send more molecules to Rx1, and the probability of molecules released by Tx1, Tx2, and Tx3 being captured by Rx1 will increase, thereby reducing the average bit error rate of the system. For the optimal average bit error rate of the system (the method of this application, corresponding to the optimal Pe), ... avg The analysis utilizes a deep neural network-based optimization algorithm to determine the optimal number of molecules released by each sending nanomachine, further calculating the optimal average bit error rate (BER). Results show that, under different upper bounds on the number of molecules released, the molecule release number obtained by the deep neural network-based optimization algorithm significantly reduces the system's BER. The study also explored the impact of different time slot lengths on the unoptimized BER and the optimal BER. The results show that when T... s At 2s, both the unoptimized average bit error rate and the optimal average bit error rate are significantly better than T. s =1s.

[0200] exist Figure 8(a) shows the optimal molecular release numbers for Tx1, Tx2, and Tx3 under different lower bounds of molecular release numbers. and The changes. Figure 8 In (a), and The molecular mass was set to 15 μm, 16.7 μm, and 21 μm respectively. When the lower bound of the number of molecules released, ψ = 5000, and The results are 6886, 13209, and 15000 respectively. It increases with the increase of the lower bound ψ of the number of molecules released, and and There was no significant change. Furthermore, when ψ = {7000, 8000, 9000}, The lower bound ψ equals the number of molecules released, because Tx1 is closest to Rx1. Figure 8 In (b), the optimal molecular release quantities corresponding to Tx1, Tx2, and Tx3 were investigated. and The relationship between Tx3 and the upper bound Ψ for different molecule release numbers shows that for different values ​​of Ψ, the optimal number of molecules released corresponds to Tx3. Far more than and and It is always equal to Ψ, for example, when Ψ = 15000. and They are equal to 8293, 10958 and 15000 respectively.

[0201] Unlike in Figure 9 In addition to simply increasing the lower or upper bound of the number of molecules released, Figure 10 This study investigated the variation of the optimal molecular release numbers corresponding to Tx1, Tx2, and Tx3 under different upper and lower bound constraints. First, the lower and upper bounds of the molecular release number were set to 7000 and 15000, respectively. and These values ​​are 7000, 13260, and 15000 respectively, because Tx1 and Tx3 are the two transmitters closest and farthest from Rx1, respectively. Next, the lower and upper bounds of the number of molecules released are simultaneously increased to observe... and The changes in ψ = 8000 and Ψ = 16000. and They are still equal to ψ and Ψ respectively, while It increased from 13260 to 14401.

[0202] exist Figure 11This paper elucidates the relationship between the optimal molecule release numbers corresponding to Tx1, Tx2, and Tx3 and different link lengths. First, the lower bound ψ and upper bound Ψ for the molecule release number are set to 5000 and 15000, respectively. and When the thickness is equal to 15.5μm, 18μm and 25μm respectively, and The results are 8386, 11209, and 15000 respectively. Much larger and This is because the distance between Tx3 and Rx1 is much greater than the distances between Tx1 and Tx2 and Rx1. To study the effect of link length variation on... and The effect of this on d0 = [15.5μm, 18μm, 25μm] T and d0 = [17μm, 18μm, 25μm] T The optimal molecular release amounts were compared at d0 = [17 μm, 18 μm, 25 μm]. T hour, and The numbers are 10067, 11209, and 15000 respectively. It must be greater than d0 = [15.5μm, 18μm, 25μm] T Below

[0203] exist ​ In this paper, the performance of the deep neural network-based optimization algorithm (i.e., the method described in this application) in terms of bit error rate (BER) is compared with that of the genetic algorithm (GA) and exhaustive search. The lower bound ψ of the number of molecules released is fixed, while only the upper bound Ψ of the number of molecules released is varied to compare the differences in BER performance among the three methods. The results show that the average BER calculated by the number of molecules released by the deep neural network-based optimization algorithm is very close to the average BER obtained by exhaustive search. The genetic algorithm, however, performs worse than the deep neural network-based optimization algorithm in terms of BER performance. This is mainly because the genetic algorithm has unguided mutations in each generation, thus failing to guarantee the optimality of the solution.

[0204] Table 2

[0205]

[0206] According to Table 2, by comparing the runtime required for different optimization algorithms to optimize the number of molecules released by each sender in a multi-input multi-output (MIMO) molecular communication network, it can be found that compared to genetic algorithms and exhaustive search, the trained deep neural network takes the least time to determine the optimal number of molecules released by each sender nanomachine. Furthermore, the average bit error rate (BER) of the MIMO network corresponding to the optimal BER output by the deep neural network is better than that of the genetic algorithm and infinitely close to that of exhaustive search. Observing the optimal BER output by the deep neural network, it can be found that in the MIMO system, the deep neural network selects to allocate more molecules to the sender nanomachines in longer links. More importantly, the deep neural network can find the optimal molecule release allocation scheme while minimizing the system's average BER. The minimum average BER obtained by the deep neural network is lower than that obtained by the genetic algorithm and the same as that obtained by exhaustive search, fully demonstrating the advantages of the trained deep neural network in optimizing the number of molecules released by each sender nanomachine in MIMO networks.

[0207] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0208] The embodiments described above are merely specific and detailed examples of the embodiments described in this application, and should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the appended claims.

Claims

1. A method for optimizing the number of molecules released in a multiple-input multiple-output (MIMO) molecular communication system, wherein the MIMO molecular communication system comprises M transmitting nanomachines and one receiving nanomachine, characterized in that: The method for optimizing the number of molecules released in the multiple-input multiple-output molecular communication system includes the following steps: S1. Obtain simulation data and establish the time-varying channel impulse response function of the multi-input multi-output molecular communication system. The formula is as follows: in, , , In the formula, For the i-th sending nanomachine Tx i The initial distance between the receiver nanomachine Rx1 and the receiver nanomachine Rx1, i=1~M, d= [ , ,…, ,…, ] T , representing the initial distance vector formed by each link. The volume of the receiving nanomachine Rx1, For a moment, Let D be the relative time at time t. A The molecular diffusion coefficient is... The diffusion coefficient of the receiving nanomachine is . For the i-th sending nanomachine Tx i diffusion coefficient, Here, b1, b2, and b3 are the complementary error functions, and b1, b2, and b3 are the fitting parameters. S2. Use a nonlinear fitting method to minimize the error between the predicted value and the simulation data of the time-varying channel impulse response function, and obtain a well-fitted time-varying channel impulse response function. S3. Based on the fitted time-varying channel impulse response function, derive the average bit error rate of the multi-input multi-output molecular communication system in the m-th time slot, and establish a multivariate optimization objective function for the number of nanomachine molecules released by each sender in the current time slot, where: The average bit error rate of a multiple-input multiple-output molecular communication system in the m-th time slot The formula is as follows: The objective function for multivariate optimization is shown in the following formula: In the formula, Indicates link Tx i The bit error rate of Rx1 in the m-th time slot, Tx represents the i-th sending nanomachine. i The number of molecules released, where N represents the vector of the number of molecules released by each sending nanomachine. T represents transpose. This represents a lower bound on the number of molecules released. This represents the upper bound of the number of molecules released; S4. Build and train a deep neural network; S5. Input the initial distance vector formed by each preset link into the trained deep neural network to obtain the optimal molecular release vector for each sending nanomachine.

2. The method for optimizing the number of molecules released in a multi-input multi-output molecular communication system as described in claim 1, characterized in that: The i-th sending nanomachine Tx i Initial distance between the receiver nanomachine Rx1 and the receiver nanomachine The calculation is as follows: In the formula, (a i , b i , c i ) represents the i-th sending nanomachine Tx i The coordinates (a1, b1, c1) represent the coordinates of the receiver nanomachine Rx1, and r is the radius of the receiver nanomachine Rx1.

3. The method for optimizing the number of molecules released in a multiple-input multiple-output molecular communication system as described in claim 1, characterized in that: The process of acquiring the simulation data is as follows: S11. A Brownian motion simulator is used to simulate the random movement of each nanomachine in a multi-input multi-output molecular communication system. S12. During the simulation, the simulation duration is divided into several independent time slots. Each time slot The duration is T s In each time slot Initially, if the transmitted bit is '1', each sending nanomachine will transmit a preset number of molecules; if the transmitted bit is '0', each sending nanomachine will not transmit molecules. S13, The receiver nanomachines, during the simulation process, at intervals of time slots... The number of received molecules is counted to obtain the data for each time slot. The number of molecules received from the nanomachines of each sender is used to form simulation data.

4. The method for optimizing the number of molecules released in a multiple-input multiple-output molecular communication system as described in claim 1, characterized in that: The link Tx i The bit error rate of Rx1 in the m-th time slot The calculation process is as follows: S31, Assume the i-th sending nanomachine Tx i If molecules are released at the beginning of the m-th time slot and propagate via free diffusion, the probability of them being received by the receiving nanomachine Rx1 in the n-th time slot can be calculated based on the fitted time-varying channel impulse response function. : In the formula, T s For the duration of each time slot, the i-th sending nanomachine Tx i At t = (m-1)T s The molecules are released at time, and the receiving nanomachine Rx1 is at t = (nm)T s +τ counts the number of molecules received; S32, will and Represented as a binomial distribution: In the formula, This indicates that the receiver nanomachine Rx1 receives the data from the sender nanomachine Tx in the m-th time slot. i The number of molecules released, This represents the inter-symbol interference of the receiver's nanomachine Rx1 in the m-th time slot. Tx represents the i-th sending nanomachine. i The bits waiting to be transmitted at the beginning of the m-th time slot, for the i-th sender nanomachine Tx i The bits waiting to be sent are encoded by releasing different numbers of molecules, and this is done by releasing them at the beginning of the time slot. One molecule is used to transmit bit '1', and zero molecules are released to transmit bit '0'. Tx represents the i-th sending nanomachine. i The probability that the molecules released in the m-th time slot are received by the receiving nanomachine Rx1 in the m-th time slot; S33. Use the Poisson distribution to express the binomial distribution form in step S32 as a Poisson distribution form: In the formula, Tx represents the i-th sending nanomachine. i The probability of transmitting bit '1' in the nth time slot; S34. Calculate the probability distribution of the number of molecules received by the receiver nanomachine Rx1 in the m-th time slot. : S35. The MAP decision threshold that minimizes the average BER is derived using the maximum a posteriori probability estimation algorithm, as follows: S351. Under assumptions H0 and H1, the probability distribution of the number of molecules received by the receiver nanomachine Rx1 in the m-th time slot is expressed as follows: in, In the formula, This represents the Poisson distribution coefficient under assumption H0. This represents the Poisson distribution coefficient under assumption H1, where H0 represents the Mth sender nanomachine Tx. M In the case of transmitting bit '0' in the m-th time slot, let H1 represent the M-th sending nanomachine Tx. M The case where bit '1' is sent in the m-th time slot; S352, Assumption For link Tx i If Rx1 is the decision threshold in the m-th time slot, then link Tx i Rx1 Decision bits for transmitting bits in the m-th time slot Represented as: Likelihood ratio test of the number of molecules received in the m-th time slot Represented as: in: In the formula, Let H0 be the probability mass function. Let Tx be the probability mass function under hypothesis H1, and let Tx be the ith sender nanomachine. i The probability of sending bit '0' in the m-th time slot for The i-th sending nanomachine Tx i The probability of sending bit '1' in the m-th time slot for ; S353. Obtain link Tx according to step S352. i Rx1 in the m-th time slot MAP decision threshold , is represented as: S36. Derive link Tx based on MAP decision threshold. i The bit error rate of Rx1 in the m-th time slot : in, 5. The method for optimizing the number of molecules released in a multiple-input multiple-output molecular communication system as described in claim 1, characterized in that: The deep neural network includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer connected in sequence, with the number of neurons in the first hidden layer, the second hidden layer, and the third hidden layer being 8, 6, and 6, respectively.

6. The method for optimizing the number of molecules released in a multiple-input multiple-output molecular communication system as described in claim 5, characterized in that: The activation function of each neuron in each hidden layer is a rectified linear unit activation function, and if the input of the activation function is a positive number, the input is directly output; otherwise, the output is 0.

7. The method for optimizing the number of molecules released in a multiple-input multiple-output molecular communication system as described in claim 1, characterized in that: The training process of the deep neural network is as follows: S41. Obtain the first dataset, which includes initial distance vectors for different link lengths; S42. Initialize the distance vector of the k-th link length in the first dataset. As the k-th input to a deep neural network, the corresponding output of the deep neural network is: : S43, Adopt an extension strategy to Expand to V nearest neighbor solutions, as shown in the following formula: in, In the formula, Indicates Let be an M-dimensional Euclidean space centered at , and in this space, any M-dimensional vector with vector The distance between them is R. Denotes the Euclidean norm; S44, Calculation The corresponding number Neighboring solutions average bit error rate ; S45. Select the optimal solution from the average bit error rates of the current V nearest neighbor solutions. : S46. Constructing Dataset Units And store it in the queue of the second dataset; S47. Select δ of the latest dataset units in the second dataset to train the deep neural network once, set k=k+1, return to execute step S42, and continue until the first dataset is traversed. S48. Every L / δ training iterations, the actual output and the expected output of the deep neural network are compared, and the difference between the actual output and the expected output is recorded. After L training iterations, the deep neural network with the smallest difference is taken as the trained deep neural network, where L is the preset number of training iterations.

8. The method for optimizing the number of molecules released in a multiple-input multiple-output molecular communication system as described in claim 1, characterized in that: The transmitting nanomachine is a point transmitter, and the receiving nanomachine is a passive spherical receiver.

Images