An Optimization Method for Release Rate Allocation in Mobile Multi-User Molecular Communication

Abstract

The present invention discloses an optimization method for release rate allocation in mobile multi-user molecular communication. First, a mathematical model of the average bit error rate of the mobile multi-user molecular communication system is constructed. Then, the release rate allocation deep neural network is iteratively trained. In each iteration, the initial distances between each sender nanomachine and the receiver nanomachine are first initialized to form an initial distance vector, which is input into the release rate allocation deep neural network to generate a corresponding release rate allocation scheme. Then, the release rate allocation scheme is extended, and the allocation scheme with the minimum average bit error rate and its corresponding initial distance vector are used as a training sample and put into the training dataset. The above process is repeated to generate a preset number of training samples, and the generated training samples are used to train the release rate allocation deep neural network. Finally, the trained release rate allocation deep neural network is used to obtain the optimal release rate allocation optimization scheme. The present invention has a faster convergence speed and better optimization effect.

Description

Technical Field

[0001] This application belongs to the technical field of molecular communication, and particularly relates to an optimization method for the release rate allocation of mobile multi-user molecular communication. Background Art

[0002] With the development and progress of science and technology and the further improvement of human communication needs, molecular communication has begun to occupy a place in the field of communication and has become an important communication means in aspects such as biology, environmental protection, and microscopic communication. Due to its good biocompatibility, it can meet people's communication needs in the biological environment, thus realizing important applications in fields such as biological research, medical rescue, and environmental protection.

[0003] One of the most important applications of molecular communication in the biological and medical fields is drug delivery in the human body, which lays the foundation for the networking of bio-nanomaterials. In order to improve the performance of molecular communication, enhance the efficiency of the communication system, and further meet the effect of molecular communication in drug delivery applications in the human body, a multi-user communication system structure can be adopted. A multi-user molecular communication system can satisfy the communication of multiple users with the receiver simultaneously, meet the communication requirements of multiple links, and has good application capabilities for drug delivery applications in the human body.

[0004] In a multi-user mobile molecular communication network, the total release rate of each user, that is, each sender nano-machine, is limited. When the distances of each link are different, how to allocate the release rate for the sender nano-machines of each link needs to be optimized. When there are two users, the optimal release rate allocation scheme can be searched through simple exhaustive search. However, when there are more users, the time required by the exhaustive algorithm increases geometrically. How to obtain the optimal release rate allocation scheme in as little time as possible has become a very challenging task. Summary of the Invention

[0005] The purpose of this application is to provide an optimization method for the release rate allocation of mobile multi-user molecular communication, so as to complete the optimization problem of the release rate allocation of multi-user mobile molecular communication with less time and computational complexity, and make the optimization result as close to the optimal solution as possible.

[0006] To solve the above technical problems, the present invention adopts the following technical solutions:

[0007] An optimization method for the release rate allocation of mobile multi-user molecular communication includes:

[0008] Constructing a mathematical model of the average bit error rate of a mobile multi-user molecular communication system;

[0009] The iterative training releases the rate allocation deep neural network. In each iteration, first, the initial distances between each sender nanomachine and the receiver nanomachine are initialized to form an initial distance vector, which is input into the rate allocation deep neural network to generate a corresponding rate allocation scheme. Then, an extended rate allocation scheme is obtained through extension. The average bit error rate mathematical model is used to calculate the average bit error rate corresponding to each allocation scheme in the extended rate allocation scheme. The allocation scheme with the minimum average bit error rate and its corresponding initial distance vector are used as a training sample and put into the training dataset. This process is repeated to generate a preset number of training samples, and the generated training samples are used to train the rate allocation deep neural network;

[0010] Input the initial distance vector in the mobile molecular communication network into the trained rate allocation deep neural network to obtain the optimal rate allocation optimization scheme.

[0011] Furthermore, the average bit error rate mathematical model is expressed as:

[0012]

[0013] where Pe aver [j] represents the average bit error rate in the j-th time slot, M represents the number of sender nanomachines, represents the bit error rate corresponding to the i-th sender nanomachine Tx i in the j-th time slot;

[0014]

[0015] where p represents the probability that Tx i sends bit 1 at the beginning of time slot j, represents the probability that Tx i sends bit 1 at the beginning of the j-th time slot and Rx receives and decodes it as 0 at the end of the j-th time slot, represents the probability that Tx i sends bit 0 at the beginning of the j-th time slot and Rx receives and decodes it as 1 at the end of the j-th time slot.

[0016] Furthermore, the process of first initializing the initial distances between each sender nanomachine and the receiver nanomachine to form an initial distance vector in each iteration, inputting it into the rate allocation deep neural network to generate an initial rate allocation scheme, then obtaining an extended rate allocation scheme through extension, using the average bit error rate mathematical model to calculate the average bit error rate corresponding to each allocation scheme in the extended rate allocation scheme, using the allocation scheme with the minimum average bit error rate and its corresponding initial distance vector as a training sample and putting it into the training dataset, repeating to generate a preset number of training samples, and using the generated training samples to train the rate allocation deep neural network includes:

[0017] Step 2.1: Input the initial distance vector into the release rate allocation deep neural network to obtain the initial release rate allocation scheme;

[0018] Step 2.2: Expand the initial release rate allocation scheme to obtain the expanded release rate allocation scheme;

[0019] Step 2.3: Use the average bit error rate mathematical model to calculate the average bit error rate corresponding to each allocation scheme in the expanded release rate allocation scheme;

[0020] Step 2.4: Take the allocation scheme with the minimum average bit error rate and its corresponding initial distance vector as a training sample and put it into the training data set;

[0021] Step 2.5: Repeat Steps 2.1 - 2.4 to generate a preset number of training samples, and use the generated training samples to train the release rate allocation deep neural network.

[0022] Furthermore, the expanded initial release rate allocation scheme includes adding perturbations, local search, or / and random generation.

[0023] An optimization method for release rate allocation in mobile multi - user molecular communication proposed in this application optimizes the release rate allocation in mobile molecular communication with the help of a deep neural network (Deep Neural Network, DNN). In order to enable the deep neural network to complete the optimization task, a training algorithm for the deep neural network for optimizing the release rate allocation in mobile molecular communication is designed. The neural network obtained through this algorithm can complete the optimization task of release rate allocation in mobile molecular communication, so as to minimize the average bit error rate of the communication system. This application has the following beneficial effects:

[0024] 1. A multi - user mobile molecular communication network model consisting of multiple sender nanomachines and one receiver nanomachine is studied;

[0025] 2. The average bit error rate of the multi - user mobile molecular communication network system is derived;

[0026] 3. The release rate allocation schemes of each sending nanomachine under different initial distance values of the links in the system are optimized using DNN, and a training algorithm for DNN is designed;

[0027] 4. Numerical results show that compared with other traditional optimization algorithms including the bisection method and genetic algorithm, DNN has a faster convergence speed and better optimization effect. Compared with exhaustive search, DNN has a very short running time, but the optimization effect is close to or equal to that of exhaustive search.

[0028] The present invention provides a guiding direction for optimizing the release rate allocation of mobile multi - user molecular communication for drug delivery using DNN. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flowchart of the optimization method for the release rate allocation of mobile multi - user molecular communication in this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0030] In order to make the objectives, technical solutions and advantages of this application more clear and understandable, the following further details this application in conjunction with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not used to limit this application.

[0031] The present invention studies a multi - user mobile molecular communication network model composed of multiple sender nanomachines and one receiver nanomachine. When the sum of the release rates of all sender nanomachines in the multi - user mobile molecular communication network is constant, the optimal scheme of the release rate allocation of each sender nanomachine when the average BEP (Bit Error Probability) of the network is minimized is obtained using DNN. First, the average bit error rate of the multi - user mobile molecular communication network is deduced; second, an optimization problem of minimizing the average bit error rate of the system under the constraint that the sum of the release rates of all sender nanomachines in the multi - user mobile molecular communication network is constant is established; finally, a training algorithm is designed to train DNN, and the trained DNN can complete the optimization task, thereby realizing the optimization of the release rate allocation of the multi - user mobile molecular communication network.

[0032] In one embodiment, as Figure 1 shown, an optimization method for the release rate allocation of mobile multi - user molecular communication is proposed, including:

[0033] Step S1, constructing a mathematical model of the average bit error rate of the mobile multi - user molecular communication system.

[0034] In this embodiment, M sender nanomachines Tx 1 , Tx 2 , …, Tx M , and one receiver nanomachine Rx are distributed in a three - dimensional fluid environment, where the sender nanomachines are point - like transmitters and the receiver nanomachine is a passive spherical receiver, and their initial distance is Communicating on the premise of time synchronization, the communication process in the j - th time slot is as follows:

[0035] First, each sending nanomachine releases information molecules at the beginning of the j-th time slot; second, the receiving nanomachine counts the number of information molecules received from each link at the end of the j-th time slot; finally, the receiving nanomachine decodes the received information molecules. For the link Tx i →Rx, Tx i releases information molecules into the channel and propagates through free diffusion. Finally, some of the information molecules reach the receiving node Rx. During the propagation process, the collisions between information molecules are ignored. Tx i The probability that an information molecule sent at t = 0 is received by the node Rx within t time is:

[0036]

[0037] where D p,eff = D p + D Rx D p and D Rx represent the diffusion coefficients of the information molecule and Rx respectively, V Rx represents the volume of Rx, τ s represents the time relative to time t, that is, Rx receives information molecules within [t, t + τ s .

[0038] The mean μ s (t, τ h )(t, τ s ) and variance are:

[0039]

[0040]

[0041] where v = D p,eff τ s + D tot t, represents the diffusion coefficient of Tx i .

[0042] is used to represent the bit sent by Tx i at the beginning of time slot j. is the release rate of Tx i . For the link Tx i →Rx, the number of information molecules received by Rx comes from the number of molecules released by Tx i in the current time slot and from Tx iThe number of molecules released in the previous time slot is the inter-symbol interference (ISI). and can be expressed as:

[0043]

[0044]

[0045] where m represents the time slot variable, belonging to 1 to j - 1. T s represents the duration of each time slot.

[0046] can be expressed in the form of a normal distribution, with its mean and variance Then there is:

[0047]

[0048] Based on (4)-(6), and are:

[0049]

[0050]

[0051] Similarly, ISI can also be expressed in the form of a normal distribution:

[0052]

[0053] Its mean and variance The calculation formulas are:

[0054]

[0055]

[0056] where p is the probability that Tx i sends bit 1 at the beginning of time slot j.

[0057] For the link Tx i →Rx, there is also noise interference in the information molecules sent from Tx i and received by Rx. In addition, there is also a part of the information molecules from other transmitting nanomachines. This part is the multiuser interference (MUI). Considering these interference factors, the total amount of information molecules received by Rx in the link Tx i →Rx is:

[0058]

[0059] Among them, and respectively represent the MUI from other links Tx q →Rx (1 ≤ q ≤ M, q ≠ i) and the noise interference in Tx i →Rx.

[0060] The MUI in the link Tx i →Rx is:

[0061]

[0062] It can also be written in the form of a normal distribution:

[0063]

[0064] When calculating the MUI in the link Tx i →Rx, it is necessary to consider the q in the link Tx and →Rx (1 ≤ q ≤ M, q ≠ i). Next, obtain the mean value variance which is:

[0065]

[0066]

[0067] The noise in the link Tx i →Rx can be expressed as:

[0068]

[0069] Among them, the mean value is generally 0, and the variance is positively correlated with the number of information molecules received by Rx. Finally, the normal distribution form of (12) is obtained:

[0070]

[0071] Therefore, the mean value and variance in equation (18) are expressed as:

[0072]

[0073]

[0074] H 0 is used in combination with H 1 to represent Tx i For the two cases of transmitting bit 0 and bit 1 at the start of the j-th time slot, according to (18), we have:

[0075]

[0076]

[0077] Among them, in combination with the calculation formula is

[0078]

[0079]

[0080]

[0081]

[0082] Tx i The probability that Tx transmits bit 0 at the beginning of the j-th time slot and Rx receives and decodes it as 1 at the end of the j-th time slot is:

[0083]

[0084] Among them, represents the decoding result of Rx for the information from Tx i at the end of the j-th time slot. is the decision threshold for Rx to decode the information from Tx i and can be calculated as follows:

[0085]

[0086] Among them, round represents the rounding operation because the threshold should be an integer. A, B, and C are

[0087]

[0088]

[0089]

[0090] Tx i The probability that Tx transmits bit 1 at the beginning of the j-th time slot and Rx receives and decodes it as 0 at the end of the j-th time slot is:

[0091]

[0092] Then the link Txi The bit error rate of →R is:

[0093]

[0094] Considering that there are M links (Tx i →R, 1 ≤ i ≤ M) in the multi-user mobile molecular communication system, the formula for calculating the average bit error rate of the system is:

[0095]

[0096] Using to represent the initial distance vector composed of the initial distances between each transmitting nanomachine Tx i (1 ≤ i ≤ M) and Rx. to represent the release rate vector composed of the release rates of each transmitting nanomachine Tx i (1 ≤ i ≤ M). According to (31), regarding Pe aver [j] as a function of d and N, then Pe aver [j] can be written as Pe aver [j](d, N).

[0097] When the initial distance d of each link is given, how to obtain the optimal release rate of each transmitting nanomachine Tx i (1 ≤ i ≤ M) under the release rate constraint conditions. Therefore, the following constrained optimization problem is established:

[0098]

[0099]

[0100] where N S is the total release rate of all transmitting nanomachines.

[0101] Step S2: Iteratively train the release rate allocation deep neural network. Each iteration first initializes the initial distance vector composed of the initial distances between each transmitting nanomachine and the receiving nanomachine, inputs it into the release rate allocation deep neural network to generate an initial release rate allocation scheme, then obtains an extended release rate allocation scheme through extension, calculates the average bit error rate corresponding to each allocation scheme in the extended release rate allocation scheme using the average bit error rate mathematical model, and takes the allocation scheme with the minimum average bit error rate and its corresponding initial distance vector as a training sample and puts it into the training dataset. Repeat to generate a preset number of training samples, and use the generated training samples to train the release rate allocation deep neural network.

[0102] Specifically, the training algorithm for the release rate allocation deep neural network (DNN) in mobile multi-user molecular communication is as follows:

[0103] 2.1, Input the initial distance vector into the release rate allocation deep neural network to obtain an initial release rate allocation scheme.

[0104] Input the initial distance vector into the deep neural network DNN to obtain an initial release rate allocation scheme where each component of d l is randomly generated within the range of the initial distance magnitudes of each link in the molecular communication system.

[0105] 2.2, Expand the initial release rate allocation scheme to obtain an expanded release rate allocation scheme.

[0106] Expand by using methods such as adding perturbations, local search, or random generation to obtain a set of W new release rate allocation schemes which is also called the expanded release rate allocation scheme, and is the k-th element of, where Here is the new release rate for link Tx i →Rx. The parameter W is obtained through simulation experiments, and in this embodiment, the value of W is selected to be 6.

[0107] Since the output of the DNN before the DNN is trained is not the desired output, in this embodiment, this output is expanded by using methods such as adding perturbations, local search, or random generation and the one that satisfies minimizing Pe aver (d,N) is found among these expansions. In a specific embodiment, a total of W expansion results are jointly generated, with 2 based on adding perturbations, 2 through local search, and 2 through random generation. It can also be expanded by any one or two of these methods.

[0108] Among them, the method of adding perturbations is as follows:

[0109]

[0110] where is the perturbation to Tx based on i and is a random number between 0 and 1.

[0111] The purpose of adding perturbations is to generate a changed value based on so as to obtain The new release rate generated by the change. N s It represents the sum of the release rates of the sender nanomachines of each link in the molecular communication system.

[0112] The method of local search is as follows:

[0113]

[0114] Among them, w is a positive integer used as the step size of local search. is a random number between 0 and 1. Its purpose is to generate a new release rate that is slightly different from the original one.

[0115] The method of random generation is as follows:

[0116]

[0117] Among them, is a random number between 0 and 1. The purpose of this operation is to generate a completely new release rate allocation scheme.

[0118] Considering the possibility of , when , this embodiment sets:

[0119]

[0120] Among them, represents the new value of

[0121] If , then for each it is reset to ensure The setting method is as follows:

[0122]

[0123] Among them, is the new value of

[0124] 2.3. Calculate the average bit error rate corresponding to each allocation scheme in the extended release rate allocation scheme using the average bit error rate mathematical model.

[0125] Calculate

[0126] 2.4. Put the allocation scheme with the minimum average bit error rate and its corresponding initial distance vector into the training dataset as a training sample.

[0127] According to the calculation results in step 2.3, to meet the minimization of Pe in (32) aver (d,N) principle in select from Let d l and Add it into the dataset. Thus, a piece of data in the dataset is generated.

[0128] 2.5, repeat steps 2.1 - 2.4 to generate a preset number of training samples, and use the generated training samples to train the release rate allocation deep neural network.

[0129] In a specific embodiment, first set a preset value L, that is, repeat steps 2.1 - 2.4 a total of L times, and thus L pieces of data are generated.

[0130] Then repeat step 2.5. If after step 2.5 has been carried out a times, there is data in the dataset Set the size of the training samples to Q. If aL ≤ Q, select aL pieces of data as training samples to train the DNN. Otherwise, if aL > Q, select Q pieces of data as training samples to train the DNN.

[0131] It should be noted that when steps 2.1 - 2.5 are completely executed once, it is considered to complete one iteration. After one iteration, re - execute steps 2.1 - 2.5 to perform another iteration. As the iteration progresses, the output obtained by the DNN will be closer and closer to the expected output because each piece of data in the training samples used to train the DNN is closer to the ideal output.

[0132] Step S3: Input the initial distance vector in the mobile molecular communication network into the trained release rate allocation deep neural network to obtain an optimal release rate allocation optimization scheme. By expanding and screening the output of the DNN, a more satisfactory output can be obtained. Adding such an output to the dataset and using it as a training sample to train the DNN can enable the DNN to be more capable of obtaining the optimal output corresponding to the input. As the DNN algorithm runs, the optimization ability of the DNN for the release rate allocation of multi - user mobile molecular communication becomes stronger and stronger. Finally, the DNN training is completed, and thus the trained DNN is obtained.

[0133] Then set the value of the initial distance vector d for each link. In practical applications, the initial distance of each link is generally 15 microns - 30 microns. Input the initial distance vector into the trained DNN, and an optimized release rate allocation scheme under different values of the initial distance vector d of each link can be obtained using this DNN.

[0134] In a multi-user mobile molecular communication network, the optimal release rate allocation scheme selects to allocate more release rates to the sender nanomachines in the longer-distance links. Compared with other traditional optimization algorithms such as the bisection method and genetic algorithm, the DNN has a faster convergence speed and better optimization effect. Compared with the exhaustive search, the running time of the DNN is extremely short, but the optimization effect is close to or equal to that of the exhaustive search.

[0135] By comparing the running durations required for optimizing the release rate allocation of each sender nanomachine in the multi-user mobile molecular communication network using each algorithm under different numbers of sender nanomachines, it can be found that when the number of sender nanomachines, i.e., the number of users, is the same, the DNN takes significantly less time than the bisection method, genetic algorithm, and exhaustive search. In addition, by comparing the optimization results obtained by using each algorithm to optimize the release rate allocation of each sender nanomachine in the multi-user mobile molecular communication network under different numbers of sender nanomachines, it can be found that in the multi-user mobile molecular communication network, the optimal release rate allocation scheme selects to allocate more release rates to the sender nanomachines in the longer-distance links. More importantly, the optimal release rate allocation scheme can be found through the DNN on the premise of minimizing the average error rate of the system. The minimum average error rate of the system obtained by using the DNN is lower than that obtained by the bisection method and genetic algorithm, and is the same as the minimum average error rate of the system obtained by the exhaustive search, fully demonstrating the advantages of the DNN in optimizing the release rate allocation of each sender nanomachine in the multi-user mobile molecular communication network.

[0136] This application is also verified through experiments. Table 1 is a table of the running durations required for optimizing the release rate allocation of each sender nanomachine in the multi-user mobile molecular communication network using each algorithm under different numbers of sender nanomachines.

[0137] Running duration Exhaustive search DNN Bisection method Genetic algorithm M＝3 0.445s 0.478s 20.38s 9.558s M＝4 2.911s 0.957s 32.14s 10.97s M＝5 59.57s 1.595s 44.27s 12.69s M＝6 1206s 2.186s 546.41s 15.30s

[0138] Table 1

[0139] As can be seen from Table 1, when M = 3, the running duration required for the exhaustive search is less than 0.5 s, but as the number of sender nanomachines increases, its required running duration increases geometrically. The running durations required for the bisection method and genetic algorithm increase linearly. Among them, the technical solution of this application using the DNN takes the least time. When M = 6, the DNN only takes about 2.2 s to complete the optimization of the release rate allocation of each sender nanomachine in the communication network. When the M value is the same, the DNN takes significantly less time than the bisection method and genetic algorithm. The parameter settings in Table 1 are as follows T S = 100 ms, j = 5, (σ Noise ) 2 = 1000, N S = 2×10 4 ×M.

[0140] Table 2 shows the optimization result tables obtained by optimizing the release rate allocation of each sender nanomachine in a multi-user mobile molecular communication network using each algorithm with different numbers of sender nanomachines.

[0141]

[0142] Table 2

[0143] Table 2 shows the release rate allocation schemes obtained by each algorithm when M takes different values. It can be seen from Table 2 that the optimization results obtained by the DNN technical solution of this application are the same as those of the exhaustive search. Comparing the results of the exhaustive search and DNN under different M values, in a multi-user mobile molecular communication network, the optimal release rate allocation scheme is to allocate more release rates to the sender nanomachines in the longer links. The parameter settings in Table 2 are as follows T S = 100 ms, j = 5, (σ Noise ) 2 = 1000, N S = 2 × 10 4 × M.

[0144] The above-described embodiments merely represent several implementation manners of this application. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the invention patent. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several modifications and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of this application patent shall be subject to the appended claims.

Claims

1. An optimization method for release rate allocation in mobile multi-user molecular communication, characterized in that, the optimization method for release rate allocation in mobile multi-user molecular communication includes: constructing a mathematical model of the average bit error rate of a mobile multi-user molecular communication system; iteratively training a deep neural network for release rate allocation. In each iteration, first initialize the initial distances between each sending nanomachine and the receiving nanomachine to form an initial distance vector, input it into the deep neural network for release rate allocation to generate a corresponding release rate allocation scheme, then obtain an extended release rate allocation scheme through extension. Calculate the average bit error rate corresponding to each allocation scheme in the extended release rate allocation scheme using the mathematical model of the average bit error rate. Take the allocation scheme with the minimum average bit error rate and its corresponding initial distance vector as a training sample and put it into the training dataset. Repeat to generate a preset number of training samples, and use the generated training samples to train the deep neural network for release rate allocation; input the initial distance vector in the mobile molecular communication network into the trained deep neural network for release rate allocation to obtain an optimal release rate allocation optimization scheme; wherein, the obtaining of the extended release rate allocation scheme through extension includes adding perturbations, local search, or / and random generation.

2. The optimization method for release rate allocation in mobile multi-user molecular communication according to claim 1, characterized in that, the mathematical model of the average bit error rate is expressed as: Among them, Pe aver [j] represents the average bit error rate in the j-th time slot, M represents the number of sender nanomachines, represents the i-th sender nanomachine Tx i corresponding bit error rate in the j-th time slot; where p represents Tx i The probability of transmitting bit 1 at the start of time slot j, represents Tx i The probability that bit 1 is transmitted at the beginning of the j-th time slot and Rx receives and decodes it as 0 at the end of the j-th time slot, represents Tx i The probability that bit 0 is transmitted at the beginning of the j-th time slot and Rx receives and decodes it as 1 at the end of the j-th time slot.

Images