CN112511474A - Intelligent equipment vibration communication method based on convolutional neural network and transfer learning - Google Patents

Intelligent equipment vibration communication method based on convolutional neural network and transfer learning Download PDF

Info

Publication number
CN112511474A
CN112511474A CN202011351885.1A CN202011351885A CN112511474A CN 112511474 A CN112511474 A CN 112511474A CN 202011351885 A CN202011351885 A CN 202011351885A CN 112511474 A CN112511474 A CN 112511474A
Authority
CN
China
Prior art keywords
vibration
layer
intelligent equipment
neural network
bit
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202011351885.1A
Other languages
Chinese (zh)
Other versions
CN112511474B (en
Inventor
王勇
赵广荣
沈益冉
张越
王天一
辛显楠
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Harbin Engineering University
Original Assignee
Harbin Engineering University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Harbin Engineering University filed Critical Harbin Engineering University
Priority to CN202011351885.1A priority Critical patent/CN112511474B/en
Publication of CN112511474A publication Critical patent/CN112511474A/en
Application granted granted Critical
Publication of CN112511474B publication Critical patent/CN112511474B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/38—Synchronous or start-stop systems, e.g. for Baudot code
    • H04L25/40—Transmitting circuits; Receiving circuits
    • H04L25/49—Transmitting circuits; Receiving circuits using code conversion at the transmitter; using predistortion; using insertion of idle bits for obtaining a desired frequency spectrum; using three or more amplitude levels ; Baseband coding techniques specific to data transmission systems
    • H04L25/4906—Transmitting circuits; Receiving circuits using code conversion at the transmitter; using predistortion; using insertion of idle bits for obtaining a desired frequency spectrum; using three or more amplitude levels ; Baseband coding techniques specific to data transmission systems using binary codes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B13/00—Transmission systems characterised by the medium used for transmission, not provided for in groups H04B3/00 - H04B11/00
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L25/00—Baseband systems
    • H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
    • H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
    • H04L25/03006—Arrangements for removing intersymbol interference
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biophysics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Power Engineering (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Telephonic Communication Services (AREA)

Abstract

The invention discloses an intelligent equipment vibration communication method based on a convolutional neural network and transfer learning. The invention relates to the technical field of vibration communication of the Internet of things, which carries out bit stream packet modulation and transmission on a vibration signal at a transmitting end of intelligent equipment; determining a vibration starting point by adopting a beacon bit detection algorithm for a receiving end of the intelligent equipment; extracting principal component characteristics of the signals of the triaxial accelerometer to remove signal noise; carrying out convolutional neural network decoding on the acceleration signal after the principal component analysis feature extraction to obtain a symbol label corresponding to a bit group; when the communication environment changes, transfer learning is carried out, and the vibration signal identification accuracy is improved. The invention adopts the mode of combining bit stream block coding with convolutional neural network decoding without concerning the influence of intersymbol interference in the block on the transmission accuracy rate, and can accelerate the vibration communication rate. The present invention is generally applicable to a variety of commercial intelligent devices, both amplitude modulation and frequency modulation techniques.

Description

Intelligent equipment vibration communication method based on convolutional neural network and transfer learning
Technical Field
The invention relates to the technical field of vibration communication of the Internet of things, in particular to an intelligent equipment vibration communication method based on a convolutional neural network and transfer learning.
Background
The intelligent Internet of things equipment is developing vigorously, and relates to the fields of intelligent home furnishing, smart cities, smart medical treatment and the like, so that the life style of people is greatly changed and enriched, and the life quality and the service level of people are effectively improved. Information interaction and communication are basic functions of operation and use of intelligent equipment, and equipment communication ensuring safety is of great importance. Currently, common communication modes of intelligent devices include WiFi, bluetooth, infrared, ZigBee and the like, which have respective advantages, but from the perspective of information security, they are all vulnerable to wireless eavesdropping.
And the communication safety under the daily environment is better guaranteed by the intelligent equipment vibration communication technology. The vibration communication technology uses a built-in vibration motor of the intelligent device as a vibration transmitter and an accelerometer or a microphone as a vibration receiver. The propagation of vibration only needs a rigid solid as a transmission medium, so that the technology can be generally applied to various intelligent devices, and the existing vibration communication technology can be roughly divided into three types: (1) techniques based on switching control, which represent 1 and 0 symbols by switching the on and off state of the vibration, each symbol carrying only one bit of information; (2) an amplitude modulation based protocol, which quantifies the vibration to n levels, each symbol carrying
Figure BDA0002801538040000011
Information of one bit(ii) a (3) Frequency modulation based protocols, which require finer control of the frequency of the vibration, can enable vibration communication rates to reach orders of magnitude higher. However, most of the above technologies have the problems of low transmission rate, low transmission accuracy, limited use conditions and the like, and cannot be widely applied to daily intelligent equipment of people.
With the rapid development of hardware and software of various embedded intelligent devices, the computing power of the embedded intelligent devices is also greatly improved. This makes it possible to implement applications that deeply learn on smart devices. Neural networks combine neurons with "non-linear" connections to form complex network structures that can fit a variety of complex functional maps. The convolutional neural network, one of the deep learning methods, has better performance in various fields.
In the conventional vibration communication technology based on switch control, the state of vibration is related to the transmitted binary value, i.e. vibration on indicates 1 and vibration off indicates 0. The speed of switching states determines the transmission rate. As shown in fig. 1, the vibration motor cannot stop the vibration of the motor immediately after receiving a stop command due to inertia. The vibration is started at a time stamp of 100ms, and the vibration amplitude reaches a maximum value after 30ms, at which time a command to stop the motor is executed, after which the motor takes about 70ms to cool sufficiently. Thus, as the bit rate increases, vibrations are likely to propagate to adjacent inter-symbols and cause significant inter-symbol interference. This limits the transmission rate of switch control based technologies. In addition, vibration techniques based on amplitude modulation and frequency modulation cannot be universally supported by current smart phones. Compared with the traditional vibration communication technology based on switch control, the method has the advantages that the influence of intersymbol interference in the group on the transmission accuracy rate is not needed to be concerned in the mode of combining bit stream block coding with convolutional neural network decoding, and the vibration communication speed can be accelerated. The present invention is generally applicable to a variety of commercial intelligent devices, both amplitude modulation and frequency modulation techniques.
Disclosure of Invention
In order to achieve higher vibration signal transmission rate, the invention provides the following technical scheme:
a vibration communication method of intelligent equipment based on a convolutional neural network and transfer learning is disclosed, wherein the intelligent equipment comprises a sending end and a receiving end, and the method comprises the following steps:
step 1: carrying out bit stream packet modulation transmission on a vibration signal on a transmitting end of the intelligent equipment;
step 2: determining a vibration starting point of a receiving end of the intelligent equipment, adopting a beacon bit detection algorithm, and acquiring an original triaxial accelerometer signal;
and step 3: extracting principal component characteristics of a signal of a triaxial accelerometer at a receiving end, and removing signal noise;
and 4, step 4: carrying out convolutional neural network decoding on the acceleration signal after the principal component analysis feature extraction to obtain a symbol label corresponding to a bit group;
and 5: when the communication environment changes, transfer learning is carried out, and the vibration signal identification accuracy is improved.
Preferably, the step 1 specifically comprises:
step 1.1: grouping bit stream data to form a bit group, carrying out bit stream grouping operation, dividing the bit stream data into bit groups with fixed sizes, taking each 4 bits of data as one bit group, and selecting the shortest single symbol length to be transmitted by a transmitting end of the intelligent equipment to improve the transmission rate;
step 1.2: carrying out bit group vibration coding, and converting a bit group 1001 into a vibration signal by a sending end of the intelligent device for transmission, wherein the length of a single symbol is 15ms, when 1 is sent, the single symbol comprises an idle state of 10ms and a vibration state of 5ms, and when 0 is sent, the 15ms is all the idle state;
step 1.3: adding a cooling time of 40ms at the end of each 4-bit group to enable the vibration motor to be decelerated to a quiet state; the sending end of the intelligent device places a special beacon bit at the beginning of bit group transmission, so that the receiving device receives the starting point of the whole vibration message, the beacon bit enables the vibration motor to continuously vibrate for 30ms, the motor reaches the maximum vibration amplitude, and then the idle state is switched to 70ms, so that the vibration motor is fully cooled and quiet.
Preferably, the step 2 specifically comprises:
adopting a beacon bit detection algorithm to carry out vibration decoding on the starting point of data transmission found by the receiving end of the intelligent equipment; the beacon bit detection algorithm uses the square sum gradient of the amplitude of the triaxial accelerometer to find a starting point, the length of an adopted time window is 100ms and the step length is 5ms according to the absolute value of the beacon bit and the gradient of two 4-bit groups containing a cooling period, and the base line value BS of an idle period in the sliding window is obtained by calculating the average value of the gradients of the first half part and the second half part of the sliding windowidleAnd base line value BS of period of vibrationvib；
According to BSidleAnd BSvibTo determine whether the middle of the time window is the starting point of vibration, when the absolute value of the gradient of the middle point of the time window is larger than BSidleThen the point in the time window is marked as a vibration candidate point at this time.
Preferably, the life cycle of the vibration should also be taken into account, i.e. the BS is judgedvibWhether or not it is greater than BSidleThat is, it is determined whether the increase of the current gradient value is continued for a certain period of time and has higher energy, and if the above condition is satisfied, the midpoint in the current time window is the starting point of the vibration.
Preferably, the step 3 specifically comprises:
step 3.1: extracting its features by principal component analysis, S ∈ RM×NIs an accelerometer segment of 100ms, M is 3 to represent three axes of the accelerometer, N represents N data points collected by each axis, N is 20 when the sampling rate is 200Hz, and a projection matrix phi epsilon R is found by a principal component analysis algorithmM×mProjecting the raw data matrix to a representation at a low latitude, Fv∈RNThe projection values are dispersed as much as possible, namely the variance is maximum, before the projection matrix is calculated, the original data must be subjected to zero mean value processing, namely decentralization, and the mean value is subtracted from each row of the data matrix to obtain a centered matrix as follows:
Figure BDA0002801538040000031
wherein i is the column index of the S matrix;
step 3.2: and (3) decomposing the EVD by adopting a matrix decomposition method characteristic value to calculate the characteristic vector and the characteristic value of the matrix, and expressing the characteristic vector and the characteristic value of the matrix by the following formula:
Figure BDA0002801538040000032
wherein, the diagonal line EorderedIs the descending order arrangement of the eigenvalue, the column vector of phi contains the eigenvector corresponding to the eigenvalue;
step 3.3: projecting the original data matrix to a feature space F, wherein F belongs to Rm×NThe first row of (1) corresponds to FvMaximum eigenvalue E oforderedThe maximum information of the original signal is reserved; two rows in F are discarded and contain the least information of the original signal.
Preferably, the step 4 specifically includes:
step 4.1: and decoding the data vector after the principal component analysis processing by adopting a convolutional neural network. The input vector of the input layer of the network is a one-dimensional vector subjected to principal component analysis feature extraction and standardization, when data is transmitted at 40bit/s, a bit group with the information content of 4 bits is transmitted every 100ms, the length N of the input vector is 20, and the shape of the input data is 20 multiplied by 1;
step 4.2: the convolutional layer is used for extracting the characteristics of the vibration sequence, the kernel size of the four convolutional layers is set to be 5, the step length is set to be 1, the number of channels is set to be 8, 16, 32 and 64 respectively, the ReLU is used as an activation function to enable the network to have sparsity, and the target characteristics in the vibration signal are located by the model and the data noise is reduced;
step 4.3: the pooling layer is a maximum pooling Max Pooling layer behind the convolution layer, and the size of the filter is 4;
step 4.4: a full link layer, wherein two full link layers are stacked behind the feature extraction module, the node number of the full link layer is 128 and 16 respectively, so as to integrate the local features extracted by the convolution layer, learn the optimal representation of the vibration of each bit group category, and use Relu as the activation function of the full link layer;
step 4.5: output layer, using softmax classifier, maps feature representations to 16 different symbols from 0000 to 1111.
Preferably, the step 5 specifically comprises: when the vibration communication environment changes, the convolutional neural network is subjected to transfer learning, the last convolutional layer, the full-connection layer and the output layer in the network model are selected, training is continued on the former three convolutional layers, the last convolutional layer, the full-connection layer and the output layer in the network model are transferred, the detail characteristics of vibration signals are extracted by the convolutional layers, the overall trend of the vibration signals is extracted by the fourth convolutional layer, and the last convolutional layer, the full-connection layer and the output layer in the network model are selected.
The invention has the following beneficial effects:
compared with the traditional vibration communication technology based on switch control, the method has the advantages that the influence of intersymbol interference in a group on the transmission accuracy rate is not needed to be concerned in the mode of combining bit stream block coding with convolutional neural network decoding. The present invention is generally applicable to a variety of commercial intelligent devices, both amplitude modulation and frequency modulation techniques.
The invention is designed based on a switch type vibration communication technology and combined with a convolutional neural network. At a vibration transmitting end, the smart phone converts a binary bit stream into bit groups, each bit group corresponds to a specific symbol label, then a vibration motor of the smart phone is controlled to rapidly switch a vibration state to transmit a vibration signal, and cooling time is added between the bit groups to inhibit symbol interference between the bit groups. According to the technology, an accelerometer of an intelligent watch is used for receiving an acceleration signal at a vibration receiving end, principal component analysis is adopted for preprocessing accelerometer data, a convolutional neural network is used for classifying bit groups, and finally symbols corresponding to the bit groups are restored to binary bit streams. The invention improves the vibration communication speed and accuracy, and ensures the universality of the technology implemented on the daily intelligent equipment.
Drawings
FIG. 1 is a schematic diagram of the inter-symbol interference problem of a conventional switch-controlled vibration-based communication technique;
FIG. 2 is a schematic illustration of bitstream packet encoding;
FIG. 3 is a schematic diagram of beacon bit detection;
FIG. 4 is a schematic diagram of principal component analysis feature extraction;
FIG. 5 is a block diagram of a convolutional neural network decoder model;
FIG. 6 is a flow diagram of an implementation of a convolutional neural network and transfer learning based vibration communication technique;
fig. 7 is an overall functional schematic.
Detailed Description
The present invention will be described in detail with reference to specific examples.
The first embodiment is as follows:
as shown in fig. 2 to 7, the present invention provides an intelligent device vibration communication method based on a convolutional neural network and transfer learning, and an intelligent device vibration communication method based on a convolutional neural network and transfer learning, where the intelligent device includes a sending end and a receiving end, and includes the following steps:
step 1: carrying out bit stream packet modulation transmission on a vibration signal on a transmitting end of the intelligent equipment;
the step 1 specifically comprises the following steps:
step 1.1: bit stream data is grouped to form a bit group, in order to reduce the influence of intersymbol interference on decoding precision, bit stream grouping operation is carried out, the bit stream data is divided into bit groups with fixed sizes, each 4bit data is used as one bit group, and a sending end of the intelligent equipment selects the shortest single symbol length to be transmitted to improve the transmission rate;
step 1.2: carrying out bit group vibration coding, and converting a bit group 1001 into a vibration signal by a sending end of the intelligent device for transmission, wherein the length of a single symbol is 15ms, when 1 is sent, the single symbol comprises an idle state of 10ms and a vibration state of 5ms, and when 0 is sent, the 15ms is all the idle state;
step 1.3: adding a cooling time of 40ms at the end of each 4-bit group to enable the vibration motor to be decelerated to a quiet state; the sending end of the intelligent device places a special beacon bit at the beginning of bit group transmission, so that the receiving device receives the starting point of the whole vibration message, the beacon bit enables the vibration motor to continuously vibrate for 30ms, the motor reaches the maximum vibration amplitude, and then the idle state is switched to 70ms, so that the vibration motor is fully cooled and quiet.
Step 2: determining a vibration starting point of a receiving end of the intelligent equipment, adopting a beacon bit detection algorithm, and acquiring an original triaxial accelerometer signal;
the step 2 specifically comprises the following steps:
adopting a beacon bit detection algorithm to carry out vibration decoding on the starting point of data transmission found by the receiving end of the intelligent equipment; the beacon bit detection algorithm uses the square sum gradient of the amplitude of the triaxial accelerometer to find a starting point, the length of an adopted time window is 100ms and the step length is 5ms according to the absolute value of the beacon bit and the gradient of two 4-bit groups containing a cooling period, and the base line value BS of an idle period in the sliding window is obtained by calculating the average value of the gradients of the first half part and the second half part of the sliding windowidleAnd base line value BS of period of vibrationvib；
According to BSidleAnd BSvibTo determine whether the middle of the time window is the starting point of vibration, when the absolute value of the gradient of the middle point of the time window is larger than BSidleThen the point in the time window is marked as a vibration candidate point at this time.
The life cycle of the vibration should also be taken into account, i.e. determining BSvibWhether or not it is greater than BSidleThat is, it is determined whether the increase of the current gradient value is continued for a certain period of time and has higher energy, and if the above condition is satisfied, the midpoint in the current time window is the starting point of the vibration.
And step 3: because the noise characteristic of the low-cost accelerometer and various placing postures of the receiver are different, the original acceleration data are not better processed directly, the main component of vibration is subjected to feature extraction, high-frequency noise is removed, and a triaxial accelerometer signal is obtained;
the step 3 specifically comprises the following steps:
step 3.1: extracting its features by principal component analysis, S ∈ RM×NIs an accelerometer segment of 100ms, M is 3 to represent three axes of the accelerometer, N represents N data points collected by each axis, N is 20 when the sampling rate is 200Hz, and a projection matrix phi epsilon R is found by a principal component analysis algorithmM×mProjecting the raw data matrix to a representation at a low latitude, Fv∈RNThe projection values are dispersed as much as possible, namely the variance is maximum, before the projection matrix is calculated, the original data must be subjected to zero mean value processing, namely decentralization, and the mean value is subtracted from each row of the data matrix to obtain a centered matrix as follows:
Figure BDA0002801538040000061
wherein i is the column index of the S matrix;
step 3.2: and (3) decomposing the EVD by adopting a matrix decomposition method characteristic value to calculate the characteristic vector and the characteristic value of the matrix, and expressing the characteristic vector and the characteristic value of the matrix by the following formula:
Figure BDA0002801538040000062
wherein, the diagonal line EorderedIs the descending order arrangement of the eigenvalue, the column vector of phi contains the eigenvector corresponding to the eigenvalue;
step 3.3: projecting the original data matrix to a feature space F, wherein F belongs to Rm×NFirst row F ofvCorrespond to EorderedThe maximum eigenvalue in the signal retains the maximum information of the original signal; two rows in F are discarded and contain the least signaled information. They contain minimal information about the signal and are usually associated with high frequency noise and other irregular artifacts, and the triaxial accelerometer signals are extracted by PCA features as shown in fig. 4.
And 4, step 4: carrying out convolutional neural network decoding on the acceleration signal after the principal component analysis feature extraction to obtain a symbol label corresponding to a bit group;
the step 4 specifically comprises the following steps:
step 4.1: decoding the data vector after principal component analysis processing by adopting a convolutional neural network, wherein the input vector of an input layer of the network is a one-dimensional vector subjected to principal component analysis feature extraction and standardization, when data is transmitted at 40 bits/s, a bit group with the information content of 4 bits is transmitted every 100ms, the length N of the input vector is 20, and the shape of the input data is 20 multiplied by 1;
step 4.2: the convolutional layer is used for extracting the characteristics of the vibration sequence, the kernel size of the four convolutional layers is set to be 5, the step length is set to be 1, the number of channels is set to be 8, 16, 32 and 64 respectively, the ReLU is used as an activation function to enable the network to have sparsity, and the target characteristics in the vibration signal are located by the model and the data noise is reduced;
step 4.3: the pooling layer is a maximum pooling Max Pooling layer behind the convolution layer, and the size of the filter is 4; the use of pooling layers reduces the complexity of the model.
Step 4.4: a full link layer, wherein two full link layers are stacked behind the feature extraction module, the node number of the full link layer is 128 and 16 respectively, so as to integrate the local features extracted by the convolution layer, learn the optimal representation of the vibration of each bit group category, and use Relu as the activation function of the full link layer;
step 4.5: output layer, using softmax classifier, maps feature representations to 16 different symbols from 0000 to 1111.
And 5: when the communication environment changes, transfer learning is carried out, and the vibration signal identification accuracy is improved.
The step 5 specifically comprises the following steps: when the use environment of the intelligent device is greatly changed, for example, the communication distance of vibration is changed from 0cm to 40cm, a small amount of newly collected data can be used for transfer learning on the basis of the original model. In the field of computer vision, it is common that the first layers of a convolutional neural network are not particularly strongly correlated with a particular image dataset, while the last layers of the network are closely correlated with the selected dataset and its task goals. Therefore, the first few layers of the network model are often selected to be frozen, and the later layers of the network model are continuously trained by using newly acquired data. I.e. the first few layers of weights of the migration network model. And performing migration learning according to the decoded signals of the triaxial accelerometer, selecting and freezing the last layers of the model by using vibration signals, continuously training the first three layers of convolutional layers, migrating the last layer of convolutional layer and the fully-connected layer in the network model to the output layer, extracting the detailed characteristics of the vibration signals when the convolutional layers extract, extracting the overall trend of the vibration signals after the fourth layer of convolutional layer, and selecting the last layer of convolutional layer and the fully-connected layer in the network model to the output layer. The reason is that the detailed characteristics of the time-varying vibration signals extracted from the first few layers of the convolution layer and the overall trend of the vibration signals extracted from the fourth layer of the convolution layer are the same, and the detailed characteristics are different due to the environmental change, so that the last convolution layer and the later fully-connected layer in the migration network model are selected to be the output layer. Thus, a higher vibration signal identification accuracy can be achieved with less new environmental data.
The above is only a preferred embodiment of the intelligent device vibration communication method based on the convolutional neural network and the transfer learning, and the protection range of the intelligent device vibration communication method based on the convolutional neural network and the transfer learning is not limited to the above embodiments, and all technical solutions belonging to the idea belong to the protection range of the present invention. It should be noted that modifications and variations which do not depart from the gist of the invention will be those skilled in the art to which the invention pertains and which are intended to be within the scope of the invention.

Claims (7)

1. A vibration communication method of intelligent equipment based on a convolutional neural network and transfer learning is provided, wherein the intelligent equipment comprises a sending end and a receiving end, and is characterized in that: the method comprises the following steps:
step 1: carrying out bit stream packet modulation transmission on a vibration signal on a transmitting end of the intelligent equipment;
step 2: determining a vibration starting point of a receiving end of the intelligent equipment, adopting a beacon bit detection algorithm, and acquiring an original triaxial accelerometer signal;
and step 3: extracting principal component characteristics of a signal of a triaxial accelerometer at a receiving end, and removing signal noise;
and 4, step 4: carrying out convolutional neural network decoding on the acceleration signal after the principal component analysis feature extraction to obtain a symbol label corresponding to a bit group;
and 5: when the communication environment changes, transfer learning is carried out, and the vibration signal identification accuracy is improved.
2. The intelligent equipment vibration communication method based on the convolutional neural network and the transfer learning as claimed in claim 1, wherein: the step 1 specifically comprises the following steps:
step 1.1: grouping bit stream data to form a bit group, carrying out bit stream grouping operation, dividing the bit stream data into bit groups with fixed sizes, taking each 4 bits of data as one bit group, and selecting the shortest single symbol length to be transmitted by a transmitting end of the intelligent equipment to improve the transmission rate;
step 1.2: carrying out bit group vibration coding, and converting a bit group 1001 into a vibration signal by a sending end of the intelligent device for transmission, wherein the length of a single symbol is 15ms, when 1 is sent, the single symbol comprises an idle state of 10ms and a vibration state of 5ms, and when 0 is sent, the 15ms is all the idle state;
step 1.3: adding a cooling time of 40ms at the end of each 4-bit group to enable the vibration motor to be decelerated to a quiet state; the sending end of the intelligent device places a special beacon bit at the beginning of bit group transmission, so that the receiving device receives the starting point of the whole vibration message, the beacon bit enables the vibration motor to continuously vibrate for 30ms, the motor reaches the maximum vibration amplitude, and then the idle state is switched to 70ms, so that the vibration motor is fully cooled and quiet.
3. The intelligent equipment vibration communication method based on the convolutional neural network and the transfer learning as claimed in claim 1, wherein: the step 2 specifically comprises the following steps:
adopting a beacon bit detection algorithm to carry out vibration decoding on the starting point of data transmission found by the receiving end of the intelligent equipment; the beacon bit detection algorithm uses the square sum gradient of the amplitude of the triaxial accelerometer to find a starting point, the length of an adopted time window is 100ms and the step length is 5ms according to the absolute value of the beacon bit and the gradient of two 4-bit groups containing a cooling period, and the base line value BS of an idle period in the sliding window is obtained by calculating the average value of the gradients of the first half part and the second half part of the sliding windowidleAnd base line value BS of period of vibrationvib；
According to BSidleAnd BSvibTo determine whether the middle of the time window is the starting point of vibration, when the absolute value of the gradient of the middle point of the time window is larger than BSidleThen the point in the time window is marked as a vibration candidate point at this time.
4. The intelligent equipment vibration communication method based on the convolutional neural network and the transfer learning as claimed in claim 3, wherein: the life cycle of the vibration should also be taken into account, i.e. determining BSvibWhether or not it is greater than BSidleThat is, it is determined whether the increase of the current gradient value is continued for a certain period of time and has higher energy, and if the above condition is satisfied, the midpoint in the current time window is the starting point of the vibration.
5. The intelligent equipment vibration communication method based on the convolutional neural network and the transfer learning as claimed in claim 1, wherein: the step 3 specifically comprises the following steps:
step 3.1: extracting its features by principal component analysis, S ∈ RM×NIs an accelerometer segment of 100ms, M is 3 to represent three axes of the accelerometer, N represents N data points collected by each axis, N is 20 when the sampling rate is 200Hz, and a projection matrix phi epsilon R is found by a principal component analysis algorithmM×mProjecting the raw data matrix to a representation at a low latitude, Fv∈RNThe projection values are dispersed as much as possible, i.e. the variance is maximum, before calculating the projection matrix, the original data must be zero-averaged, i.e. decentralized, and each row of the data matrix isThe mean is subtracted to obtain the centered matrix as follows:
Figure FDA0002801538030000021
wherein i is the column index of the S matrix;
step 3.2: and (3) decomposing the EVD by adopting a matrix decomposition method characteristic value to calculate the characteristic vector and the characteristic value of the matrix, and expressing the characteristic vector and the characteristic value of the matrix by the following formula:
Figure FDA0002801538030000022
wherein, the diagonal line EorderedIs the descending order arrangement of the eigenvalue, the column vector of phi contains the eigenvector corresponding to the eigenvalue;
step 3.3: projecting the original data matrix to a feature space F, wherein F belongs to Rm×NFirst row F ofvCorrespond to EorderedThe maximum eigenvalue in the signal retains the maximum information of the original signal; two rows in F are discarded and contain the least signaled information.
6. The intelligent equipment vibration communication method based on the convolutional neural network and the transfer learning as claimed in claim 1, wherein: the step 4 specifically comprises the following steps:
step 4.1: carrying out convolutional neural network decoding on the acceleration signal subjected to principal component analysis feature extraction, wherein an input vector of an input layer of a network is a one-dimensional vector subjected to principal component analysis feature extraction and standardization, when data is transmitted at 40bit/s, a bit group with the information content of 4 bits is transmitted every 100ms, the length N of the input vector is 20, and the shape of the input data is 20 multiplied by 1;
step 4.2: the convolutional layer is used for extracting the characteristics of the vibration sequence, the kernel size of the four convolutional layers is set to be 5, the step length is set to be 1, the number of channels is set to be 8, 16, 32 and 64 respectively, the ReLU is used as an activation function to enable the network to have sparsity, and the target characteristics in the vibration signal are located by the model and the data noise is reduced;
step 4.3: the pooling layer is a maximum pooling Max Pooling layer behind the convolution layer, and the size of the filter is 4;
step 4.4: a full link layer, wherein two full link layers are stacked behind the feature extraction module, the node number of the full link layer is 128 and 16 respectively, so as to integrate the local features extracted by the convolution layer, learn the optimal representation of the vibration of each bit group category, and use Relu as the activation function of the full link layer;
step 4.5: output layer, using softmax classifier, maps feature representations to 16 different symbols from 0000 to 1111.
7. The intelligent equipment vibration communication method based on the convolutional neural network and the transfer learning as claimed in claim 1, wherein: the step 5 specifically comprises the following steps: when the communication environment changes, transfer learning is carried out, the vibration signal selects and freezes the later layers of the model, the training is continued on the former three layers of the convolutional layers, the last layer of convolutional layer and the full-connection layer in the network model are transferred to the output layer, the detail characteristics of the vibration signal are extracted by the convolutional layers, the overall trend of the vibration signal is extracted by the fourth layer of convolutional layer, and the last layer of convolutional layer and the full-connection layer in the network model are selected and transferred to the output layer.
CN202011351885.1A 2020-11-26 2020-11-26 Intelligent equipment vibration communication method based on convolutional neural network and transfer learning Active CN112511474B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011351885.1A CN112511474B (en) 2020-11-26 2020-11-26 Intelligent equipment vibration communication method based on convolutional neural network and transfer learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011351885.1A CN112511474B (en) 2020-11-26 2020-11-26 Intelligent equipment vibration communication method based on convolutional neural network and transfer learning

Publications (2)

Publication Number Publication Date
CN112511474A true CN112511474A (en) 2021-03-16
CN112511474B CN112511474B (en) 2022-05-17

Family

ID=74966461

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011351885.1A Active CN112511474B (en) 2020-11-26 2020-11-26 Intelligent equipment vibration communication method based on convolutional neural network and transfer learning

Country Status (1)

Country Link
CN (1) CN112511474B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115015390A (en) * 2022-06-08 2022-09-06 华侨大学 MWTLMDS-based curtain wall working modal parameter identification method and system

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130329526A1 (en) * 2009-12-09 2013-12-12 Shlomo Selim Rakib System and method for vibration mediated networks
CN104104779A (en) * 2013-04-08 2014-10-15 国民技术股份有限公司 Information transmission system and method
CN107342962A (en) * 2017-07-03 2017-11-10 北京邮电大学 Deep learning intelligence Analysis On Constellation Map method based on convolutional neural networks
CN107548544A (en) * 2015-07-24 2018-01-05 惠普发展公司有限责任合伙企业 The sensor carried out with mode of vibration communicates
US10332367B1 (en) * 2018-10-17 2019-06-25 Capital One Services, Llc Systems and methods for using haptic vibration for inter device communication
CN111757127A (en) * 2019-03-29 2020-10-09 慕尼黑工业大学 Encoding/decoding apparatus and method for encoding/decoding vibrotactile signal

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130329526A1 (en) * 2009-12-09 2013-12-12 Shlomo Selim Rakib System and method for vibration mediated networks
CN104104779A (en) * 2013-04-08 2014-10-15 国民技术股份有限公司 Information transmission system and method
CN107548544A (en) * 2015-07-24 2018-01-05 惠普发展公司有限责任合伙企业 The sensor carried out with mode of vibration communicates
CN107342962A (en) * 2017-07-03 2017-11-10 北京邮电大学 Deep learning intelligence Analysis On Constellation Map method based on convolutional neural networks
US10332367B1 (en) * 2018-10-17 2019-06-25 Capital One Services, Llc Systems and methods for using haptic vibration for inter device communication
CN111757127A (en) * 2019-03-29 2020-10-09 慕尼黑工业大学 Encoding/decoding apparatus and method for encoding/decoding vibrotactile signal

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
INHWAN HWANG: "Privacy-Aware Communication for Smartphones Using Vibration", 《2012 IEEE INTERNATIONAL CONFERENCE ON EMBEDDED AND REAL-TIME COMPUTING SYSTEMS AND APPLICATIONS》 *
KYUIN LEE: "SYNCVIBE: Fast and Secure Device Pairing through Physical Vibration on Commodity Smartphones", 《2018 IEEE 36TH INTERNATIONAL CONFERENCE ON COMPUTER DESIGN (ICCD)》 *
ZHEJIE SHEN: "Near Field Service Initiation via Vibration Channel", 《2016 12TH INTERNATIONAL CONFERENCE ON MOBILE AD-HOC AND SENSOR NETWORKS (MSN)》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115015390A (en) * 2022-06-08 2022-09-06 华侨大学 MWTLMDS-based curtain wall working modal parameter identification method and system

Also Published As

Publication number Publication date
CN112511474B (en) 2022-05-17

Similar Documents

Publication Publication Date Title
CN109067688B (en) Dual-drive OFDM (orthogonal frequency division multiplexing) receiving method of data model
CN112600772B (en) OFDM channel estimation and signal detection method based on data-driven neural network
Wu et al. A CNN-based end-to-end learning framework toward intelligent communication systems
Wang et al. Distributed learning for automatic modulation classification in edge devices
CN112637093B (en) Signal detection method based on model-driven deep learning
CN111614587B (en) SC-FDE system signal detection method based on self-adaptive integrated deep learning model
KR101979394B1 (en) Adaptive transmission scheme determination apparatus based on MIMO-OFDM System using machine learning model and adaptive transmission method the same
CN110113288A (en) A kind of design and demodulation method of the ofdm demodulator based on machine learning
CN111786923A (en) Channel estimation method for time-frequency double-channel selection of orthogonal frequency division multiplexing system
Perenda et al. Learning the unknown: Improving modulation classification performance in unseen scenarios
Camelo et al. A semi-supervised learning approach towards automatic wireless technology recognition
CN109787926A (en) A kind of digital signal modulation mode recognition methods
CN112511474B (en) Intelligent equipment vibration communication method based on convolutional neural network and transfer learning
Zhang et al. Real-time OFDM signal modulation classification based on deep learning and software-defined radio
CN113708855A (en) OTFS data drive receiving method, system and medium based on deep learning
CN110311876A (en) The implementation method of underwater sound OFDM receiver based on deep neural network
Al-Baidhani et al. Learning for detection: A deep learning wireless communication receiver over Rayleigh fading channels
CN113114603B (en) Information recovery method and device for MIMO-OFDM system
Hua et al. Signal detection in uplink pilot-assisted multi-user MIMO systems with deep learning
Ye et al. Bilinear convolutional auto-encoder based pilot-free end-to-end communication systems
Ilhan et al. Offloading deep learning empowered image segmentation from uav to edge server
Elwekeil et al. Deep learning for environment identification in vehicular networks
CN114826832B (en) Channel estimation method, neural network training method, device and equipment
CN114745234B (en) Deep learning MIMO system signal detection method and system
Bhatt et al. Analysis of the fifth generation NOMA system using LSTM algorithm

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant