CN107743103B - Multi-node access detection and channel estimation method of MMTC (multimedia messaging and control) system based on deep learning - Google Patents
Multi-node access detection and channel estimation method of MMTC (multimedia messaging and control) system based on deep learning Download PDFInfo
- Publication number
- CN107743103B CN107743103B CN201711021155.3A CN201711021155A CN107743103B CN 107743103 B CN107743103 B CN 107743103B CN 201711021155 A CN201711021155 A CN 201711021155A CN 107743103 B CN107743103 B CN 107743103B
- Authority
- CN
- China
- Prior art keywords
- training
- node
- matrix
- channel
- data
- 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.)
- Active
Links
Images
Classifications
-
- 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
- H04L25/0202—Channel estimation
- H04L25/0204—Channel estimation of multiple channels
-
- 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
- H04L25/0202—Channel estimation
- H04L25/024—Channel estimation channel estimation algorithms
- H04L25/0254—Channel estimation channel estimation algorithms using neural network algorithms
Abstract
The invention provides a multi-node access detection and channel estimation method of an MMTC (multimedia messaging and traffic control) system based on deep learning, which comprises the following steps: determining a pilot frequency sequence of each node according to a modulation scheme adopted by the MMTC, and determining channel impulse response of each node; generating input data according to certain node activity, further generating a training set and a verification set for training a deep neural network and a test set for testing model performance, designing DNN and BRNN models for detecting active users and carrying out simulation verification, and solving a linear equation set by using a least square method according to a user activity detection result of the models to carry out channel estimation. According to the channel estimation method provided by the invention, under the conditions of different pilot frequency lengths and different numbers of active users, the accuracy rate of user access detection is higher than that of the traditional method, and the time of node access detection is greatly shortened.
Description
Technical Field
The invention relates to the technical field of channel estimation in a communication system, in particular to a multi-node access detection and channel estimation method of an MMTC (multimedia messaging and control) system based on deep learning.
Background
MMTC (Massive Machine-type Communication) is a hot research problem of iot (internet of things), and is mainly characterized in that a large amount of access nodes only sporadically transmit small data packets at a low data rate. The cellular system used by the traditional voice communication is mainly designed for high data rate and large data packets, the node and the receiving end follow the rule of access reservation when communicating, and the communication characteristics of the MMTC determine that the communication rule can cause the overhead occupied by the control packet header information to be larger than the real packet information to be sent, which is a great waste for the occupation of the wireless resources of the MMTC communication network.
The number of user nodes in the MMTC communication system communicating at the same time is not large, and the data packets transmitted are also small, so the MMTC communication system is a sparse communication system. The core of the CS (Compressive Sensing) theory is to project a sparse or compressible high-dimensional signal to a low-latitude space through a specific matrix transformation, and when performing signal reconstruction, reconstruct an original signal by using a linear or nonlinear recovery algorithm by using the sparsity of the sparse signal or the compressed signal. Currently, commonly used compressed sensing signal recovery algorithms include MP (Matching Pursuit), OMP (Orthogonal Matching Pursuit), CoSaMP (Compressive Sampling Matching Pursuit), and the like. The algorithms basically adopt a cyclic iteration optimization mode to realize signal reconstruction, the reconstruction time is generally limited by the iteration times, and the reconstruction precision is limited by the properties of a measurement matrix.
Disclosure of Invention
Compared with the traditional compressed sensing recovery algorithm, the deep learning method has the advantages that the algorithm of the operation unit is stable, the model speed is high, the network structure can be independently learned and evolved, and the system gain is obviously improved along with the increase of the data training amount. The large data volume of the MMTC communication system meets the requirement of a deep learning method, and the high speed of deep learning meets the real-time requirement of MMTC communication. The speed and the accuracy rate of multi-node access detection and channel joint estimation in the MMTC communication system are improved.
In order to achieve the purpose, the invention adopts the following technical scheme:
a multi-user access detection and channel estimation method of an MMTC system based on deep learning comprises the following steps:
s1: preparing a data set;
determining a pilot frequency sequence of each node according to a modulation scheme adopted by a communication system MMTC, generating a corresponding convolution matrix, and confirming that each node obeys a channel impulse response of standard normal distribution to obtain a channel matrix;
setting node activity, generating input data by channel impact response and convolution matrix of active nodes, and generating a training set and a verification set for training a deep neural network and a test set for testing model accuracy by using the input data;
s2: training a deep learning model;
constructing a deep neural network model DNN and a block threshold neural network model BRNN, training the DNN and the BRNN by using the training set through an algorithm, verifying through the verification set at the end of each training period, adjusting a training method, and storing the DNN and BRNN models after training;
s3: detecting an active node;
testing the accuracy of the active nodes of the stored DNN and BRNN models by using a test set; outputting active node information;
s4: estimating a channel;
solving a linear equation set by using a least square method to carry out channel estimation according to the active node information output in the S3;
the data set preparation of S1 further includes:
(1) according to a constellation point set of a modulation scheme in an MMTC system, determining a pilot sequence of each node, and obtaining a pilot matrix S ═ S of all nodes1,…,sk,…]Wherein s iskFor the pilot sequence of node k, a corresponding convolution matrix is generated for the pilot sequence of each node
Wherein the content of the first and second substances,
obtaining convolution matrix of all nodes
For the convolution matrix
Performing complex number domain to real number domain transformation to obtain convolution matrixWherein SrAnd SiAre respectively as
The real and imaginary parts of (c);
(2) determining a channel impulse response of each node following a standard normal distribution asWherein L represents the number of tap delays of the channel, k represents a node, and the observation result of the pilot sequence received at the receiving end of the channel is:
wherein x represents convolution, n represents additive white gaussian noise, and k represents a node;
(3) setting the impact response of the non-active node to the channel as 0 value to obtain a matrix h corresponding to the access signal, and converting the matrix h from a complex number field to a real number field to obtain a matrix
According to a matrixAnd obtaining a received signal matrix y by the convolution matrix Q, adding noise when receiving signals, and obtaining received signals of mixed noise
Input data for training the model;
(4) marking sequence indexes of active nodes as training labels, and generating a training set, a verification set and a test set of the deep neural network model according to a set proportion by using the input data and the format of the training labels;
the block threshold neural network model is structurally characterized in that a pooling layer is added between a full connection layer and an activation function, the output of the pooling layer enters the activation layer and is spliced and stretched, the splicing and stretching result and the output of the full connection layer are subjected to point multiplication, and the point multiplication is used as the input of the next layer.
Further, the training deep learning model of S2 further includes:
constructing a model;
designing a block threshold structure in a deep neural network by using the block sparse characteristic of nodes in an MMTC communication system, wherein the block threshold structure is used for carrying out block-based threshold limitation after batch normalization on full-connection layer output, and respectively constructing a deep neural network model and a block threshold neural network model;
reconstructing sparse data;
training the deep neural network model and the block threshold neural network model by using the training set through a gradient descent method, verifying a training result by using the verification set after each training period is finished, adjusting the training method according to the result of the verification set until the model is optimal in performance, and storing the trained model.
Further, the block threshold structure in the designed deep neural network is realized by a pooling layer and an activation function layer, output data for batch normalization of the output of the full connection layer is divided into data blocks with set length, the maximum value of the data in the data blocks is smaller than 0, all the data in the data blocks are set to be 0, otherwise, the data in the data blocks are kept unchanged, and the set length is the channel vector length allocated to the nodes.
Further, the deep neural network model is a residual neural network model and comprises a full connection layer, a batch normalization layer, a pooling layer and a cross entropy loss function layer, wherein the batch normalization layer is used for avoiding gradient disappearance and gradient dispersion, the pooling layer is arranged between the batch normalization layer and an activation function, the activation function layer and the output of the batch normalization layer are subjected to point multiplication, the point multiplication is used as input data of the cross entropy loss function layer, and the input data and channel input data have the same block distribution characteristics.
Further, the loss function of the cross entropy loss function layer is:
wherein l represents a set of tags, piIndicating the probability of the correct category and i indicates the index of the active node in the label.
Furthermore, the activation function is a relu function, and performs point-to-point threshold judgment and limitation on the input, and relu is defined as relu (x) max {0, x }.
Furthermore, according to the index value output in the multi-node detection process, the corresponding column in the matrix Q is taken out, and the input data is input
And the matrix Q solves the channel matrix h through a least square method to obtain the mean square error of channel estimation.
According to the technical scheme provided by the invention, the channel detection estimation accuracy rate of the method provided by the invention is higher than that of the traditional method under the conditions of different signal-to-noise ratios, different pilot frequency lengths and different user activation probabilities, and the time for node access detection is greatly shortened.
Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a flowchart of a multi-node access detection and channel estimation method of an MMTC system based on deep learning according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of an MMTC communication system provided by an embodiment of the invention;
FIG. 3 is a block threshold network schematic diagram provided by an embodiment of the present invention;
FIG. 4 is a comparison graph of performance of methods at different user activities according to an embodiment of the present invention;
FIG. 5 is a graph comparing the performance of different methods for different length pilot sequences according to an embodiment of the present invention;
fig. 6 is a time chart of the single sample test run by the methods provided by the embodiments of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention.
As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or coupled. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
For the convenience of understanding the embodiments of the present invention, the following description will be further explained by taking several specific embodiments as examples in conjunction with the drawings, and the embodiments are not to be construed as limiting the embodiments of the present invention.
The invention provides a multi-node access detection and channel estimation method of an MMTC system based on deep learning, wherein FIG. 1 is a flow chart of the method, and as shown in FIG. 1: the method comprises the following steps:
s1: preparing a data set;
determining a pilot frequency sequence of each node according to a modulation scheme adopted by a communication system MMTC, generating a corresponding convolution matrix, and confirming that each node obeys a channel impulse response of standard normal distribution to obtain a channel matrix;
setting node activity, generating input data by channel impact response and convolution matrix of active nodes, and generating a training set and a verification set for training a deep neural network and a test set for testing model accuracy by using the input data;
s2: training a deep learning model;
constructing a deep neural network model DNN and a block threshold neural network model BRNN, training the DNN and the BRNN by using the training set through an algorithm, verifying through the verification set at the end of each training period, adjusting a training method, and storing the DNN and BRNN models after training;
s3: detecting an active node;
testing the accuracy of the active nodes of the stored DNN and BRNN models by using a test set; outputting active node information;
s4: estimating a channel;
and solving a linear equation system by using a least square method according to the active node information output in the S3 to carry out channel estimation.
In a specific embodiment, the data set preparation of S1 further includes:
(1) determining each point according to the constellation point set of the modulation scheme in the MMTC systemObtaining pilot matrix S ═ S of all nodes by pilot sequence of each node1,…,sk,…]Wherein s iskFor the pilot sequence of node k, a corresponding convolution matrix is generated for the pilot sequence of each node
Wherein the content of the first and second substances,
obtaining convolution matrix of all nodes
For the convolution matrix
Performing complex number domain to real number domain transformation to obtain convolution matrix
Wherein SrAnd SiAre respectively as
The real and imaginary parts of (c);
(2) determining a channel impulse response of each node following a standard normal distribution as
Wherein L represents the number of tap delays of the channel, k represents a node, and the observation result of the pilot sequence received at the receiving end of the channel is:
wherein x represents convolution, n represents additive white gaussian noise, and k represents a node;
(3) setting the impact of the non-active node on a channel to be 0 value to obtain a matrix h corresponding to an access signal, and converting the matrix h from a complex number field to a real number field to obtain a matrix
According to a matrix
And obtaining a received signal matrix y by the convolution matrix Q, adding noise when receiving signals, and obtaining received signals of mixed noise
Input data for training the model;
(4) and marking the sequence index of the active node as a training label, and generating a training set, a verification set and a test set of the deep neural network model according to a set proportion by using the input data and the format of the training label.
In a specific embodiment, the training deep learning model of S2 further includes:
constructing a model;
designing a block threshold structure in a deep neural network by using the block sparse characteristic of nodes in an MMTC communication system, wherein the block threshold structure is used for carrying out block-based threshold limitation after batch normalization on full-connection layer output, and respectively constructing a deep neural network model and a block threshold neural network model;
reconstructing sparse data;
training the deep neural network model and the block threshold neural network model by using the training set through a gradient descent method, verifying a training result by using the verification set after each training period is finished, adjusting the training method according to the result of the verification set until the model is optimal in performance, and storing the trained model.
In a specific embodiment, the block threshold structure in the designed deep neural network is implemented by a pooling layer and an activation function layer, and output data for batch normalization of the output of the fully-connected layer is divided into data blocks with set lengths, the maximum value of the data in the data blocks is smaller than 0, all the data in the data blocks are set to be 0, otherwise, the data in the data blocks are kept unchanged, and the set lengths are channel vector lengths allocated to nodes.
In a specific embodiment, the neural network model is a residual neural network model, and includes a full connection layer, a batch normalization layer, a pooling layer, and a cross entropy loss function layer, where the batch normalization layer is used to avoid gradient disappearance and gradient diffusion, the pooling layer is between the batch normalization layer and an activation function, and performs point multiplication on the output of the activation function layer and the batch normalization layer, and the point-multiplied output is used as input data of the cross entropy loss function layer, and the input data and channel input data have the same block distribution characteristics.
In a specific embodiment, the loss function of the cross entropy loss function layer is:
wherein l represents a set of tags, piIndicating the probability of the correct category and i indicates the index of the active node in the label.
In a specific embodiment, the activation function is a relu function, and performs point-to-point threshold determination and limitation on the input, where relu is defined as relu (x) ═ max {0, x }.
In a specific embodiment, according to the index value output in the multi-node detection process, the corresponding column in the matrix Q is taken out, and the data is inputAnd the matrix Q solves the channel matrix h through a least square method to obtain the mean square error of channel estimation.
Example (b):
fig. 2 is a schematic view of a sparse-type system MMTC, and as shown in fig. 2, there are K user nodes in the sparse-type MMTC communication system, and at most n users in the system need to send data to a base station at the same time, that is, the users' data are sent to the base stationMaximum activity probability Pan/K < 1. The active users send respective pilot frequencies, and the base station performs multi-user access detection and channel joint estimation through a compressed sensing algorithm. The base station then uses the estimated channel state information to estimate the data subsequently transmitted by the user.
The embodiment of the invention provides a multi-user sparse access detection and channel estimation method in an MMTC communication system based on deep learning, which comprises the following steps:
the method comprises the following steps: determining an initial pilot sequence s of a user k according to a modulation scheme adopted by an MMTC systemkAnd corresponding channel impulse response hkAnd further obtaining a convolution matrix Q and a channel matrix H.
(1) Determining a constellation point set Λ corresponding to a modulation scheme adopted by a system, for example, Λ of a BPSK (Binary Phase shift Keying) modulation scheme is [1, -1], Λ of a QPSK (Quadrature Phase shift Keying) modulation scheme is [1+ i,1-i, -1+ i, -1-i ], or adopting other modulation schemes.
(2) Repeatedly and randomly selecting N from constellation point set lambdasInitial pilot sequence of k user composed by element
The channel impulse response corresponding to the user node k is generated by standard normal distribution as
Wherein L represents the number of tap delays of the discrete channel, and the observation result of the pilot sequence received at the receiving end is:
wherein x represents convolution and n represents additive white gaussian noise; performing convolution transformation on the expression, and transforming the expression according to the matrix convolution transformation
Wherein
Representing a vector skThe convolution matrix of (2).
(3) Sequentially generating initial pilot sequences of all user nodes so as to obtain convolution matrixes of all user nodesFor the convolution matrix
Performing complex number domain to real number domain transformation to obtain convolution matrix
Step two: preparing a training set and a validation set for training a neural network and a test set for testing model performance, determining a block threshold structure principle of the neural network for sparse signal recovery is shown in fig. 3:
(1) setting the maximum number of nodes K, and setting the activation probability P of the usera>0, obtaining channel impulse responses corresponding to n active nodes from h, distributing channel impulse responses with the length of L and the value of 0 to the inactive nodes, and combining the channel impulse responses of all the nodes into an input signal
Performing conversion from a complex number domain to a real number domain on h to obtain a block sparse vector with the dimension of 2KL multiplied by 1
(2) For the generated convolution matrix Q, the production of batchesAnd is composed of
Generating received data y, and adding noise into the received data according to the set signal-to-noise ratio to obtain noise data
n is white noise subject to a gaussian distribution;
(3) the compressed sensing recovery problem of the sparse signal is further simplified into a position index problem of a nonzero element, an index vector l with a label of a training set as the position of the nonzero element is determined, and the training set, a verification set and a test set are generated according to a certain proportion;
(4) determining a Neural Network model DNN (Deep Neural Network) for solving the multi-label classification problem as a residual Neural Network model, wherein connection layers of data are all full connection layers, adding a batch normalization layer to prevent gradient disappearance and gradient diffusion, and selecting a relu activation function as an activation function layer. A Block-threshold-based Neural Network (BRNN) model is constructed, the main structure is that a pooling layer is added between a full connection layer and an activation function, the output of the pooling layer enters the activation layer and is spliced and stretched, point multiplication is carried out on the output of the pooling layer and the output of the full connection layer, the point multiplication is carried out on the output of the pooling layer and is used as the input of the next layer, and the input data and the channel input have the same Block distribution characteristics.
(5) As a multi-label classification model, the loss function of the last layer of the model is cross entropy loss,
wherein l represents a set of tags, piIndicating the probability of the correct category and i indicates the index of the active node in the label. The accuracy of the final output is given by the proportion of the total number of data correctly predicted for all active users to the total data.
(6) Determining a compressed sensing recovery neural network structure based on block sparse signals, wherein the block structure designed by the invention mainly comprises:
and dividing vectors which are output by the full connection layer and have the same dimension with the transmitted signal into K blocks with the length of L, wherein each block corresponds to the channel response of a user K, and L is the number of tap time delays. And for each block of data, if all the values of the block of data are not positive, all the blocks of data are set to be 0, otherwise, the data are kept unchanged. The output vector of the block structure has the same block structure as the transmitted signal, so as to realize faster and more accurate fitting of data distribution, the block structure is composed of a pooling layer and an activation function layer, and the activation function is a relu activation function: relu (x) max {0, x }.
(7) Determining the evaluation standard of the check model, and taking the maximum probability value of n for the probability value table output by the training neural network, wherein the corresponding index vector is
And performing cross comparison with the label vector l to obtain an accuracy acc of sum/num, wherein sum is the number of all correctly predicted active user indexes, and num is the total number of samples in the label vector.
(8) And training a data set by using a fully-connected neural network in a stochastic gradient descent method, checking a training result by using a verification set at the end of each period to adjust parameter setting, and storing the trained DNN model and the BRNN model.
Step three: multi-user sparse access detection and channel estimation.
(1) Since the channel information of the inactive users is treated as a zero element, and the activity rate P of the usersa1, so h is a sparse vector, thus
This equation can be solved using a compressed perceptual reconstruction signal algorithm, such as GOMP (Group orthogonal Matching Pursuit) or Group Lasso (Group least absolute shrinkage and selection algorithm). The method provided by the embodiment of the invention uses a deep learning method instead of the traditional iterative method in the process of carrying out compressed sensing recovery on the received signal.
(2) Running the stored model on the test set, taking the index of n values with the maximum probability and the label to cross-compare the output classification probability value table, and outputting the most accurateThe rate of certainty is the activation of the user index to all sample values predicted to be correct divided by the total number of samples. Extracting corresponding columns in the matrix Q according to the output index values of the maximum n probability values, and receiving data
And solving the linear equation set by the matrix Q by a least square method to obtain the value of the predicted channel matrix and the signal-to-noise ratio of the channel estimation.
To sum up, the embodiment of the present invention provides a method for performing joint multi-user sparse access detection and channel estimation in an MMTC communication system through deep learning, and fig. 4 is a comparison graph of performance of each method under different user liveness levels provided by the embodiment of the present invention; FIG. 5 is a diagram comparing the performance of the methods under the pilot sequences of different lengths; as can be seen from fig. 4 and 5, the accuracy of the method of the present invention is higher than that of the conventional method under different pilot lengths and different user activation probabilities, fig. 6 is a single sample detection time chart for each method provided by the embodiment of the present invention, and as can be seen from fig. 6, the method of the present invention greatly reduces the user access detection time.
In multi-user sparse access and channel joint estimation based on compressed sensing in an MMTC communication system, the length of a channel vector is usually fixed, so that an access signal presents the characteristic of block sparsity, namely, nonzero elements exist adjacently along the length of the channel vector. According to the block sparse structure of the access signal, the block structure is introduced into the neural network, and the output data can be quickly and accurately fitted with the distribution characteristic of the access signal by carrying out block structure constraint on the output of the upper layer.
Those of ordinary skill in the art will understand that: the figures are merely schematic representations of one embodiment, and the blocks or flow diagrams in the figures are not necessarily required to practice the present invention.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for apparatus or system embodiments, since they are substantially similar to method embodiments, they are described in relative terms, as long as they are described in partial descriptions of method embodiments. The above-described embodiments of the apparatus and system are merely illustrative, and the units described as separate parts may or may not be physically separate, and the parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.
Claims (7)
1. A multi-node access detection and channel estimation method of an MMTC system based on deep learning is characterized by comprising the following steps:
s1: preparing a data set;
determining a pilot frequency sequence of each node according to a modulation scheme adopted by a communication system MMTC, generating a corresponding convolution matrix, and confirming that each node obeys a channel impulse response of standard normal distribution to obtain a channel matrix;
setting node activity, generating input data by channel impact response and convolution matrix of active nodes, and generating a training set and a verification set for training a deep neural network and a test set for testing model accuracy by using the input data;
s2: training a deep learning model;
constructing a deep neural network model DNN and a block threshold neural network model BRNN, training the DNN and the BRNN by using the training set through an algorithm, verifying through the verification set at the end of each training period, adjusting a training method, and storing the DNN and BRNN models after training;
s3: detecting an active node;
testing the accuracy of the stored DNN and BRNN models by using a test set to detect active nodes, and outputting active node information;
s4: estimating a channel;
solving a linear equation set by using a least square method to carry out channel estimation according to the active node information output in the S3;
the data set preparation of S1 further includes:
(1) according to a constellation point set of a modulation scheme in an MMTC system, determining a pilot sequence of each node, and obtaining a pilot matrix S ═ S of all nodes1,…,sk,…]Wherein s iskFor the pilot sequence of node k, a corresponding convolution matrix is generated for the pilot sequence of each node
Wherein the content of the first and second substances,
obtaining convolution matrix of all nodes
For the convolution matrix
Performing complex number domain to real number domain transformation to obtain convolution matrix
Wherein SrAnd SiAre respectively as
The real and imaginary parts of (c);
(2) determining each nodeThe channel impulse response following the standard normal distribution is
Wherein L represents the number of tap delays of the channel, k represents a node, and the observation result of the pilot sequence received at the receiving end of the channel is:
wherein x represents convolution, n represents additive white gaussian noise, and k represents a node;
(3) setting the impact response of the non-active node to the channel as 0 value to obtain a matrix h corresponding to the access signal, and converting the matrix h from a complex number field to a real number field to obtain a matrix
According to a matrix
And obtaining a received signal matrix y by the convolution matrix Q, adding noise when receiving signals, and obtaining received signals of mixed noise
Input data for training the model;
(4) marking sequence indexes of active nodes as training labels, and generating a training set, a verification set and a test set of the deep neural network model according to a set proportion by using the input data and the format of the training labels;
the block threshold neural network model is structurally characterized in that a pooling layer is added between a full connection layer and an activation function, the output of the pooling layer enters the activation layer and is spliced and stretched, the splicing and stretching result and the output of the full connection layer are subjected to point multiplication, and the point multiplication is used as the input of the next layer.
2. The method according to claim 1, wherein the training of S2 is a deep learning model, further comprising:
constructing a model;
designing a block threshold structure in a deep neural network by using the block sparse characteristic of nodes in an MMTC communication system, wherein the block threshold structure is used for carrying out block-based threshold limitation after batch normalization is carried out on full-connection layer output, and respectively constructing a DNN and a BRNN;
reconstructing sparse data;
and training the DNN and the BRNN by using a gradient descent method for the training set, verifying a training result by using the verification set after each training period is finished, adjusting the training method according to the result of the verification set, and storing the trained model.
3. The method of claim 2,
the block threshold structure in the designed deep neural network is realized by a pooling layer and an activation function layer, output data for batch normalization of output of a full connection layer is divided into data blocks with set length, the maximum value of the data in the data blocks is smaller than 0, all the data in the data blocks are set to be 0, otherwise, the data in the data blocks are kept unchanged, and the set length is the channel vector length distributed by nodes.
4. The method of claim 3,
the deep neural network model is a residual neural network model and comprises a full connection layer, a batch normalization layer, a pooling layer and a cross entropy loss function layer, wherein the batch normalization layer is used for avoiding gradient disappearance and gradient dispersion, the pooling layer is arranged between the batch normalization layer and an activation function, the activation function layer and the output of the batch normalization layer are subjected to point multiplication, the input data after the point multiplication is used as the input data of the cross entropy loss function layer, and the input data and the channel input data have the same block distribution characteristics.
5. The method of claim 4, wherein the loss function of the cross-entropy loss function layer is:
wherein l represents a set of tags, piIndicating the probability of the correct category and i indicates the index of the active node in the label.
6. The method of claim 5, wherein the activation function is a relu function, and the threshold decision limit for point-to-point input is performed, and relu is defined as
relu(x)＝max{0,x}。
7. The method of claim 6,
according to the index value output in the multi-node detection process, the corresponding column in the matrix Q is taken out, and the data is input
And the matrix Q solves the channel matrix h through a least square method to obtain the mean square error of channel estimation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711021155.3A CN107743103B (en) | 2017-10-26 | 2017-10-26 | Multi-node access detection and channel estimation method of MMTC (multimedia messaging and control) system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711021155.3A CN107743103B (en) | 2017-10-26 | 2017-10-26 | Multi-node access detection and channel estimation method of MMTC (multimedia messaging and control) system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107743103A CN107743103A (en) | 2018-02-27 |
| CN107743103B true CN107743103B (en) | 2020-02-14 |
Family
ID=61236954
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711021155.3A Active CN107743103B (en) | 2017-10-26 | 2017-10-26 | Multi-node access detection and channel estimation method of MMTC (multimedia messaging and control) system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107743103B (en) |
Families Citing this family (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108540410B (en) * | 2018-03-06 | 2020-11-03 | 东南大学 | Joint active user detection and channel estimation method for scheduling-free transmission system |
| CN108566257B (en) * | 2018-04-27 | 2020-08-18 | 电子科技大学 | Signal recovery method based on back propagation neural network |
| CN108848519B (en) * | 2018-05-25 | 2021-05-18 | 东南大学 | Heterogeneous network user access method based on cross entropy learning |
| US20210367702A1 (en) * | 2018-07-12 | 2021-11-25 | Intel Corporation | Devices and methods for link adaptation |
| CN108833313A (en) * | 2018-07-12 | 2018-11-16 | 北京邮电大学 | A kind of radio channel estimation method and device based on convolutional neural networks |
| CN109617847B (en) * | 2018-11-26 | 2021-04-06 | 东南大学 | OFDM receiving method without cyclic prefix based on model-driven deep learning |
| CN109787715B (en) * | 2018-12-18 | 2021-01-19 | 中国科学院深圳先进技术研究院 | DNN decoding method and decoding communication equipment of SCMA system |
| CN110011775B (en) * | 2019-03-22 | 2022-06-07 | 安徽师范大学 | Method and system for jointly realizing active user detection and channel estimation |
| CN110071881B (en) * | 2019-04-26 | 2020-12-15 | 北京理工大学 | Active user detection and channel estimation method with adaptive overhead |
| CN113748614B (en) * | 2019-04-30 | 2023-06-20 | 华为技术有限公司 | Channel estimation model training method and device |
| CN110661734B (en) * | 2019-09-20 | 2020-07-28 | 西安交通大学 | Channel estimation method, device and readable storage medium based on deep neural network |
| CN111045861B (en) * | 2019-10-22 | 2023-11-07 | 南京海骅信息技术有限公司 | Sensor data recovery method based on deep neural network |
| CN111641570B (en) * | 2020-04-17 | 2021-04-27 | 浙江大学 | Joint equipment detection and channel estimation method based on deep learning |
| CN111786915B (en) * | 2020-05-09 | 2023-03-14 | 广州番禺职业技术学院 | Wireless communication method and system based on deep learning under nonlinear distortion condition |
| CN111614584B (en) * | 2020-05-20 | 2021-11-23 | 西安电子科技大学 | Transform domain adaptive filtering channel estimation method based on neural network |
| CN111865845A (en) * | 2020-06-04 | 2020-10-30 | 国网江西省电力有限公司供电服务管理中心 | MIMO user detection and channel estimation device and method |
| WO2022000365A1 (en) * | 2020-07-01 | 2022-01-06 | Qualcomm Incorporated | Machine learning based downlink channel estimation and prediction |
| CN111835464B (en) * | 2020-07-14 | 2021-06-25 | 浙江大学 | Large-scale cooperation access method irrelevant to source address |
| CN112073106B (en) * | 2020-08-14 | 2022-04-22 | 清华大学 | Millimeter wave beam prediction method and device, electronic device and readable storage medium |
| CN112968853B (en) * | 2021-01-29 | 2022-07-01 | 北京交通大学 | Multi-user access detection and channel estimation method for MMTC (multimedia messaging and control) system |
| CN112929445B (en) * | 2021-02-20 | 2022-06-07 | 山东英信计算机技术有限公司 | Recommendation system-oriented link prediction method, system and medium |
| EP4362404A1 (en) * | 2021-06-22 | 2024-05-01 | Beijing Xiaomi Mobile Software Co., Ltd. | Channel estimation method, channel estimation apparatus, and storage medium |
| CN113766541B (en) * | 2021-09-07 | 2023-10-17 | 北京交通大学 | Active device in MMTC scene and detection method of using channel thereof |
| CN113766669B (en) * | 2021-11-10 | 2021-12-31 | 香港中文大学（深圳） | Large-scale random access method based on deep learning network |
| CN115208944B (en) * | 2022-06-30 | 2023-07-11 | 深圳市大数据研究院 | Method, system, computer device and medium for detecting active users and data |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008076505A2 (en) * | 2006-10-27 | 2008-06-26 | Karthikeyan Chandrashekar | Modeling and simulating wireless mac protocols |
| CN101399803A (en) * | 2007-09-27 | 2009-04-01 | 大唐移动通信设备有限公司 | Multi-user detection method and device for OFDM signal |
| CN102868422A (en) * | 2012-09-07 | 2013-01-09 | 天津理工大学 | MMSE-BDFE (Minimum Mean Square Error-Blind Decision Feedback Equalizer) multi-user detection system based on neural network, and working method of MMSE-BDFE multi-user detection system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR101412006B1 (en) * | 2012-10-25 | 2014-06-26 | 서울대학교산학협력단 | WLAN service method and WLAN system |
-
2017
- 2017-10-26 CN CN201711021155.3A patent/CN107743103B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008076505A2 (en) * | 2006-10-27 | 2008-06-26 | Karthikeyan Chandrashekar | Modeling and simulating wireless mac protocols |
| CN101399803A (en) * | 2007-09-27 | 2009-04-01 | 大唐移动通信设备有限公司 | Multi-user detection method and device for OFDM signal |
| CN102868422A (en) * | 2012-09-07 | 2013-01-09 | 天津理工大学 | MMSE-BDFE (Minimum Mean Square Error-Blind Decision Feedback Equalizer) multi-user detection system based on neural network, and working method of MMSE-BDFE multi-user detection system |
Non-Patent Citations (2)
| Title |
|---|
| Delay-aware massive random access for machine-type communications via hierarchical stochastic learning;Yannan Ruan; Wei Wang; Zhaoyang Zhang; Vincent K. N. Lau;《2017 IEEE International Conference on Communications (ICC)》;20170731;全文 * |
| 基于压缩感知理论的无线多径信道估计方法研究;王妮娜;《CNKI博士学位论文全文库》;20120430;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107743103A (en) | 2018-02-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107743103B (en) | Multi-node access detection and channel estimation method of MMTC (multimedia messaging and control) system based on deep learning | |
| CN111224779B (en) | Physical layer key generation method and device based on codebook, storage medium and terminal | |
| CN109560841A (en) | Extensive mimo system channel estimation methods based on improved distributed compression perception algorithm | |
| CN111884976B (en) | Channel interpolation method based on neural network | |
| Woodruff et al. | An optimal lower bound for distinct elements in the message passing model | |
| CN112910811B (en) | Blind modulation identification method and device under unknown noise level condition based on joint learning | |
| CN108599820A (en) | The extensive mimo system channel estimation methods of matching pursuit algorithm are sampled based on block structure self-adapting compressing | |
| CN106130689A (en) | A kind of non-linear self-feedback chaotic neural network signal blind checking method | |
| CN114268388B (en) | Channel estimation method based on improved GAN network in large-scale MIMO | |
| CN104866558A (en) | Training method of social networking account mapping model, mapping method and system | |
| CN106911443B (en) | Pilot tone optimum design method in compressed sensing based M2M communication system | |
| CN111835464A (en) | Large-scale cooperation access method irrelevant to source address | |
| CN113422695A (en) | Optimization method for improving robustness of topological structure of Internet of things | |
| CN115276855A (en) | ResNet-CBAM-based spectrum sensing method | |
| CN106612158A (en) | Signal blind detection method based on complex sinusoidal chaotic neural network | |
| Wang et al. | Statistical Model on CRAFT | |
| CN110059401B (en) | OFDM system underwater sound channel impulse response reconstruction method | |
| CN110944002A (en) | Physical layer authentication method based on exponential average data enhancement | |
| CN116170066A (en) | Load prediction method for low-orbit satellite Internet of things | |
| CN110365434A (en) | Multi-antenna cooperative frequency spectrum sensing method based on information geometry and differential evolution clustering algorithm | |
| CN114980086A (en) | Model training method, secret key generation method, training equipment, communication party and system | |
| CN111045861B (en) | Sensor data recovery method based on deep neural network | |
| CN109286408B (en) | Method for energy harvesting and energy receiver | |
| CN106911431B (en) | Improved partial edge information transmission method applied to demodulation process of sparse code multiple access system | |
| CN111082973A (en) | Network delay estimation method based on lever sampling |
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 |