CN114363218B - Communication reachable rate detection method based on end-to-end learning - Google Patents
Communication reachable rate detection method based on end-to-end learning Download PDFInfo
- Publication number
- CN114363218B CN114363218B CN202210015129.4A CN202210015129A CN114363218B CN 114363218 B CN114363218 B CN 114363218B CN 202210015129 A CN202210015129 A CN 202210015129A CN 114363218 B CN114363218 B CN 114363218B
- Authority
- CN
- China
- Prior art keywords
- layer
- representing
- sequence
- node
- output
- 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
- 238000004891 communication Methods 0.000 title claims abstract description 18
- 238000001514 detection method Methods 0.000 title claims abstract description 17
- 238000004364 calculation method Methods 0.000 claims abstract description 51
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000013528 artificial neural network Methods 0.000 claims description 25
- 230000006870 function Effects 0.000 claims description 18
- 239000013598 vector Substances 0.000 claims description 18
- 239000011159 matrix material Substances 0.000 claims description 12
- 108010076504 Protein Sorting Signals Proteins 0.000 claims description 11
- 238000013507 mapping Methods 0.000 claims description 11
- 238000000034 method Methods 0.000 claims description 11
- 230000005540 biological transmission Effects 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 238000010801 machine learning Methods 0.000 claims description 7
- 230000003213 activating effect Effects 0.000 claims description 6
- 125000003275 alpha amino acid group Chemical group 0.000 claims description 3
- 150000001875 compounds Chemical class 0.000 claims description 3
- 230000007704 transition Effects 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 abstract description 8
- 238000004422 calculation algorithm Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 5
- 238000005259 measurement Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011897 real-time detection Methods 0.000 description 2
- 238000007476 Maximum Likelihood Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
Classifications
-
- 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
- Error Detection And Correction (AREA)
- Mobile Radio Communication Systems (AREA)
- Communication Control (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention discloses a communication reachable rate detection method based on end-to-end learning, which comprises the following steps: 1. training by using a neural network model to calculate a Log Likelihood Ratio (LLR); 2. the result 3 is precisely calculated by a gradient descent algorithm, and the information reachable rate based on bit decoding, namely Generalized Mutual Information (GMI), is calculated according to the log likelihood ratio. The invention can greatly improve the calculation efficiency while ensuring the accuracy of the calculation result, thereby improving the calculation instantaneity of the information reachable rate.
Description
Technical Field
The invention relates to the technical field of communication, in particular to a generalized mutual information calculation method for calculating a log-likelihood ratio LLR based on a neural network model, which relates to end-to-end learning and machine learning.
Background
In recent years, machine learning has been widely used in the field of communications, and from encoding to channel modeling to decoding to modulation and demodulation, machine learning has been applied to a certain extent and achieved with good results. The training structure can tend to be optimal by training some parameters in the neural network through a machine learning method, so that the calculation result is more accurate.
With the advancement of a series of technological innovations, the development of communication technology has made an important contribution to the increase in capacity of the internet. However, the capacity of the communication core network cannot meet the increasing traffic demand of people in the information age, and in order to make the information transmission rate approach to shannon limit, information theory analysis of the communication system is particularly important.
The reachable information rate (achievable informationrates, AIR) is a performance evaluation index based on information theory information measurement, is defined as the maximum information quantity which can be reliably transmitted in a given channel, can be used for evaluating the maximum reachable transmission rate of the channel, and has important practical significance in practical engineering application because the generalized mutual information is used as the reachable information rate based on a bit decoder code modulation system. In the prior art, because a plurality of factors, such as actual distribution of channels, channel dimensions, mutual interference among bits and the like, need to be considered in the calculation of the generalized mutual information in the prior art, the actual application of the generalized mutual information in the prior art is less, the calculation steps of the generalized mutual information in the actual application are complex, the time is long, and absolute accuracy cannot be achieved, so that the method has more application of the mutual information, namely the other reachable information rate without considering the relation among the bits, but has more practical significance in the actual engineering application because of more factors considered by the generalized mutual information in comparison. Therefore, measuring the generalized mutual information of the channel is of great importance for understanding the channel transmission information capability and evaluating the channel performance.
Disclosure of Invention
The invention aims to solve the defects in the prior art, and provides a communication reachable rate detection method based on end-to-end learning, so that the calculation result is ensured to be accurate, the calculation efficiency is greatly improved, and the calculation instantaneity of generalized mutual information is improved.
In order to achieve the aim of the invention, the invention adopts the following technical scheme:
the invention discloses a communication reachable rate detection method based on end-to-end learning, which is characterized by comprising the following steps:
step 1, defining a transmission signal sequence of a transmitting end as s= { s 1 ,s 2 ,…,s i ,…,s n }，s i Representing the ith transmission signal, defining the sequence of the reception signal at the receiving end as Represents the i-th received signal, i.e. [1, n ]]N represents the sequence length;
defining a received signal sequenceEach received signal is mapped into a bit sequence with length m, and a sequence formed by a kth bit mapped by the received signal sequence is defined as B k ＝{B k,1 ,B k,2 ,…,B k,i ,…,B 1,n }, wherein B is k,i Representing the received signal +.>The k-th bit of the mapping, k.epsilon.1, m]，B k,i ∈{0,1}；
Step 2, the received signal sequenceEach received signal of (a) is divided into two parts, real part and imaginary part, and is marked as +.>Wherein (1)>A 2 n-th input value representing an input layer, i.e., layer 0;
let the node bias vector from the input layer to the hidden layer be recorded asWherein (1)>Representing the bias of the 2 n-th node of the input layer;
the number of nodes of an input layer and a hidden layer of the neural network is 2n, and the number of nodes of an output layer of the neural network is m; the number of layers of the hidden layer is H;
let the weight matrix from 2n nodes in the input layer to 2n nodes in the hidden layer 1 be recorded asWherein (1)>Representing weights of the 2 n-th node in the input layer to the 2 n-th node in the hidden layer of the 1 st layer;
the weight matrix from 2n nodes in any h layer hidden layer to 2n nodes in h+1th layer hidden layer is recorded asWherein (1)>Representing weights from the 2 n-th node in the h layer hidden layer to the 2 n-th node in the h+1 layer hidden layer;
let the weight matrix from 2n nodes in the H layer hidden layer to m nodes in the output layer be recorded asWherein (1)>Representing the weight from the 2n node in the H-th layer hidden layer to the m node in the output layer, H E [1, H]；
Let node bias vector in any h-th hidden layer be recorded asWherein (1)>Representing the bias of the 2 n-th node in the h-th hidden layer;
causing the calculation result sequence of any h-th hidden layerIs listed asWherein (1)>Representing the calculation result of the 2n node in the h layer hidden layer;
let the linear equation of the h hidden layer be y (h) ＝ω (h) x T +b (h) Wherein T represents a transpose;
output sequence after activating h layer hidden layerWherein (1)>Represents the 2n output after the h layer hidden layer is activated, and x (h) ＝f(y (h) ) F (·) is the activation function;
output sequence z after activating H-th hidden layer (H) As an input sequence of output layers and by linear equation y=ω in the output layers (H) (x (H) ) T +b (H) The output calculation result sequence is y= { y1, y2, …, yn }, wherein yn represents the calculation result output by the nth node of the output layer; x is x (H) Representing the output sequence after activation of the H-th hidden layer, b (H) Representing node bias vectors in the H-th hidden layer;
step 2.1, assume channel distribution f Y|X (y|x) estimating a log likelihood ratio sequence of the binary mapped bit sequences of the n received signals as using the maximum log approximation method shown in the formula (1)Wherein (1)>Log likelihood ratio sequence representing k-th bit of n received signal maps> Log-likelihood ratio of kth bit representing nth received signal mapping, k e [1, m]；
In the formula (1), the components are as follows,representing the maximum probability of the kth bit decision of the mapping of the n received signals being 1 or 0, f Y|X (y|x) represents the channel transition probability of X for the transmit signal and Y for the receive signal, X, Y representing the transmit and receive signal sequences, respectively;
step 2.2, defining the current iteration number as I, and defining the maximum iteration number as I max Initializing i=1;
randomly initializing all weight matrixes and node offset vectors in the neural network when I=1, and recording the initialized weight matrixes and node offset vectors as a set theta of the I-th iteration I ＝{ω (0) ,ω (1) ,…,ω (H) ；b (1) ,b (2) ,…b (H) }；
Step 2.3, establishing a loss function l (theta) of the ith iteration of the neural network by using the formula (2) I )：
In the formula (2), the amino acid sequence of the compound,representing the calculation result sequence output by the kth node of the output layer after the ith iteration, andwherein (1)>Representing the log-likelihood ratio of the ith received signal output by the kth node after the ith iteration;
step 2.4 gradient θ for the I-th iteration using equation (3) I Updating, namely updating the ownership weight matrix and the bias vector in the network to obtain the gradient theta of the (I+1) th iteration I+1 ；
In the formula (3), alpha is the learning rate in machine learning, and alpha is more than 0;
step 2.5, after assigning I to I+1, judging that I is greater than I max If so, the neural network training is completed, and a trained calculation network model is obtained, otherwise, the step 2.3 is returned to be sequentially executed until the loss function l (theta I ) Stopping training until the training frequency is smaller than the set training standard epsilon, and obtaining a trained calculation network model for calculating a received signal sequenceOptimal log likelihood ratio sequence of each signal mapping bit sequence +.>Wherein,,optimal log likelihood ratio sequence representing output of kth node of output layer +.>Representing the optimal log-likelihood ratio of the ith received signal output by the kth node of the computational network output layer, and recording the neural network parameters after training as theta as a network of computational stagesComplexing parameters;
step 2.6, calculating generalized mutual information G through the formula (4);
in the formula (4), E represents a mathematical expectation,representing receipt of B k Is decided as +.>Conditional probability of->Representation->Is a probability distribution of (c).
Compared with the prior art, the invention has the beneficial effects that:
1. the invention realizes the calculation training of the log likelihood ratio LLR based on the neural network model, has simple network structure and simple algorithm realization, overcomes the defect that the accuracy and the real-time measurement of the channel performance index cannot be considered in the prior art, ensures that the measurement result is relatively real-time by improving the algorithm used by the measuring device, and simultaneously ensures that the measurement result is accurate by the gradient descent of the end-to-end neural network and the minimum value of the loss function, thereby ensuring that the detection method is easy to design and has good practical value.
2. The invention has simple design logic, estimates the approximate log likelihood ratio LLR of the channel by using the maximum likelihood method, stops training when the value of the loss function is smaller than a certain selected value epsilon, and can use the neural network model under the parameter as an LLR calculation model to greatly improve the LLR calculation efficiency.
3. The invention uses log-likelihood ratio LLR and generalized mutual information GMI as indexes for measuring channel performance, and has a certain relation between the two, so the GMI can be calculated by using LLR, thereby simplifying the calculation of the GMI, improving the calculation efficiency of the GMI, and having real-time property by adjusting the update time.
Drawings
FIG. 1 is a schematic diagram of the communication reachable rate detection method based on end-to-end learning of the present invention;
FIG. 2 is a flow chart of the communication achievable rate detection method based on end-to-end learning of the invention;
FIG. 3 is a neural network diagram of the communication achievable rate detection method parameter training based on end-to-end learning of the present invention;
fig. 4 is a neural network computational model diagram of the communication achievable rate detection method based on end-to-end learning of the invention.
Detailed Description
In an actual communication system, an implementation flow chart of an achievable information rate detection method based on end-to-end learning is shown in fig. 2, a neural network model is built and parameters thereof are initialized, the neural network is trained and a value of a loss function is calculated, when the training times exceed the maximum iteration times, the training is stopped, otherwise, a loss function value is calculated, when the training times are smaller than epsilon, the training is stopped, epsilon=2 is taken, a log likelihood ratio LLR of a signal is calculated by using the trained model, finally generalized mutual information GMI of the signal is calculated by using the LLR and a calculation result is output, and simultaneously, a channel condition is detected, when the channel condition is changed, namely, the channel is shown to be bent in an actual condition or weather is changed in wireless communication, the network is retrained, or the network parameters are updated to recalculate the LLR and the GMI by setting fixed update time, so that the calculation result is accurate and accords with the actual condition.
In this embodiment, a schematic structure diagram of a communication reachable rate detection method based on end-to-end learning is shown in fig. 1, which includes four modules, and input parameters are a signal sequence s= { s sent by a sending end 1 ,s 2 ,…，s n Sequence of signals received by the receiverTaking n=32, calculating the value lambda of log likelihood ratio LLR by using neural network model m And the value of LLR calculated by maximum logarithmic approximation +.>As a loss function argument, when the training frequency does not reach the maximum iteration frequency I max When the loss function value is not satisfied with the requirement, i.e. l (theta) is not less than epsilon (epsilon=2), the training mode is entered, and the network parameters are trained through the parameter initialization module and the network training module; when l (theta) is less than epsilon (epsilon=2), the method enters a calculation mode by using the trained network parameter theta, realizes the real-time detection of the GMI through a log likelihood ratio LLR and generalized mutual information GMI calculation stage, and can detect the generalized mutual information of the channel at the moment by connecting interfaces at two ends of the detection device with two ends of the channel; the schematic diagram of the neural network model is shown in fig. 3, that is, the neural network model of the channel log likelihood ratio LLR is trained through a hidden layer, and the method can realize real-time calculation of the channel generalized mutual information GMI.
The method can be extended to multidimensional signals, and the detection device can be extended to a detector for multidimensional signal generalized mutual information by adjusting the number of neurons of a network input layer. Specifically, the method comprises the following steps:
step 1, defining a transmission signal sequence of a transmitting end as s= { s 1 ,s 2 ,…,s i ,…,s n }，s i Representing the ith transmission signal, defining the sequence of the reception signal at the receiving end as Represents the i-th received signal, i.e. [1, n ]]N represents the sequence length.
Defining a received signal sequenceEach received signal of (a) is mapped into a bit sequence of length m, defining a received signalThe k bit of the signal sequence map forms a sequence B k ＝{B k,1 ,B k,2 ,…,B k,i ,…,B 1,n }, wherein B is k,i Representing the received signal +.>The k-th bit of the mapping, k.epsilon.1, m]，B k,i E {0,1}; the transmission and reception procedure is as shown in fig. 1, taking the example that the bit sequence length of each signal map is 8, i.e. m=8.
Step 2, the received signal sequenceEach received signal of (a) is divided into two parts, real part and imaginary part, and is marked as +.>Wherein (1)>A 2 n-th input value representing an input layer;
let the node bias vector from the input layer to the hidden layer be recorded asWherein (1)>Representing the bias of the 2 n-th node of the input layer;
the number of nodes of the input layer and the hidden layer of the neural network is 2n, and the number of nodes of the output layer of the neural network is m; the number of hidden layers is H;
let the weight matrix from 2n nodes in the input layer to 2n nodes in the hidden layer 1 be recorded asWherein (1)>Representing weights of the 2 n-th node in the input layer to the 2 n-th node in the hidden layer of the 1 st layer;
the weight matrix from 2n nodes in any h layer hidden layer to 2n nodes in h+1th layer hidden layer is recorded asWherein (1)>Representing weights of the 2 n-th node in the h-th layer hidden layer to the 2 n-th node in the h+1-th layer hidden layer;
let the weight matrix from 2n nodes in the H layer hidden layer to m nodes in the output layer be recorded asWherein (1)>Representing the weight of the 2n node in the H-th hidden layer to the m node in the output layer, H E [1, H]；
Let node bias vector in any h-th hidden layer be recorded asWherein (1)>Representing the bias of the 2 n-th node in the h-th hidden layer;
making the calculation result sequence of any h-th hidden layer be recorded asWherein (1)>Representing the calculation result of the 2n node in the h layer hidden layer;
let the h layerThe linear equation of the hidden layer is y (h) ＝ω (h) x T +b (h) Wherein T represents a transpose;
output sequence x after activating h layer hidden layer (h) ＝{x 1 (h) ,x 2 (h) ,…,x 2n (h) And } wherein,represents the 2n output after the h layer hidden layer is activated, and x (h) ＝f(y (h) ) F (·) is the activation function;
output sequence z after activating H-th hidden layer (H) As an input sequence of the output layer and by the linear equation y=ω in the output layer (H) (x (H) ) T +b (H) The output calculation result sequence is y= { y1, y2, …, yn }, wherein yn represents the calculation result output by the nth node of the output layer; x is x (H) Representing the output sequence after activation of the H-th hidden layer, b (H) Representing node bias vectors in the H-th hidden layer; as shown in fig. 3, in the example of the neural network, re represents the real part of the signal, im represents the imaginary part of the signal, n=32 is taken as an example, the number of input nodes is 2n=64, one layer of hidden layers is taken as an example, i.e. h=1, the number of hidden layer nodes is 64, the number of output layer nodes is m=8, and the network output result is m log likelihood ratio sequences. The calculation of the first node of the hidden layer is shown in FIG. 4, in whichRepresenting all weights of all nodes of the input layer to the first node of the hidden layer, calculating the result +.>The activation function is exemplified by the ReLU function, i.e. f (x) =max (0, x), the activation function is applied to +.>After that get->As input to the next level node.
Step 2.1, assume channel distribution f Y|X (y|x) estimating a log likelihood ratio sequence of the binary mapped bit sequences of the n received signals as using the maximum log approximation method shown in the formula (1)Wherein (1)>Log likelihood ratio sequence representing k-th bit of n received signal maps> Representing log-likelihood ratios of the kth bit representing the nth received signal map, k e [1, m]；
In the formula (1), the components are as follows,representing the maximum probability of the kth bit decision of the mapping of the n received signals being 1 or 0, f Y|X (y|x) represents the channel transition probability of X for the transmit signal and Y for the receive signal, X, Y representing the transmit and receive signal sequences, respectively;
step 2.2, defining the current iteration number as I, and defining the maximum iteration number as I max Initializing i=1;
randomly initializing all weight matrixes and node offset vectors in the neural network when I=1, and recording the initialized weight matrixes and node offset vectors as a set theta of the I-th iteration I ＝{ω (0) ,ω (1) ,…,ω (H) ；b (1) ,b (2) ,…b (H) }；
Step 2.3, establishing a loss function l (theta) of the ith iteration of the neural network by using the formula (2) I )：
In the formula (2), the amino acid sequence of the compound,representing the calculation result sequence output by the kth node of the output layer after the ith iteration and marking asWherein (1)>Representing the log-likelihood ratio of the ith received signal output by the kth node after the ith iteration;
step 2.4 gradient θ for the I-th iteration using equation (3) I Updating, namely updating the ownership weight matrix and the bias vector in the network to obtain the gradient theta of the (I+1) th iteration I+1 ；
In the formula (3), alpha is the learning rate in machine learning, and alpha is more than 0; taking alpha=0.1 as an example, the training speed and the calculating speed of the network are improved while the calculating precision is ensured;
step 2.5, after assigning I to I+1, judging that I is greater than I max If so, the neural network training is completed, and a trained calculation network model is obtained, otherwise, the step 2.3 is returned to be sequentially executed until the loss function l (theta I ) Stopping training until the training frequency is smaller than the set training standard epsilon, and obtaining a trained calculation network model for calculating a received signal sequenceMapping the optimal log likelihood ratio sequence of the bit sequence to each signal, and recording the calculation result asWherein (1)>Representing an optimal log likelihood ratio sequence output by a kth node of an output layerWherein->Representing the optimal log likelihood ratio of the ith received signal output by the kth node of the network output layer, and marking the neural network parameter after training as theta as the network parameter of the calculation stage; as shown in fig. 1, in the training mode, the trained network parameter θ is transmitted into the network in the calculation mode, and the log-likelihood ratio of the signal can be quickly calculated by using the network.
Step 2.6, calculating generalized mutual information G through the formula (4);
in the formula (4), E represents a mathematical expectation, m and B k Are all defined in the above-mentioned step 1,for the definition in step 2.5 above, < > where->Representing receipt of B k Is decided as +.>Conditional probability of->Representation->Is a probability distribution of (c). As shown in fig. 1 and fig. 2, the detector includes a parameter initialization module, a network training module, an LLR calculation module and a generalized mutual information GMI calculation module, which implement real-time detection of GMI, can implement AIR detection of multidimensional signals by adjusting the number of network input nodes, can adjust calculation time and calculation accuracy by modifying a training end mark, and can improve detection instantaneity and ensure calculation accuracy by determining a proper end mark.
Claims (1)
1. A communication reachable rate detection method based on end-to-end learning is characterized by comprising the following steps:
step 1, defining a transmission signal sequence of a transmitting end as s= { s 1 ,s 2 ,…,s i ,…,s n }，s i Representing the ith transmission signal, defining the sequence of the reception signal at the receiving end asRepresents the i-th received signal, i.e. [1, n ]]N represents the sequence length;
defining a received signal sequenceEach received signal is mapped into a bit sequence with length m, and a sequence formed by a kth bit mapped by the received signal sequence is defined as B k ＝{B k,1 ,B k,2 ,…,B k,i ,…,B 1,n }, wherein B is k,i Representing the received signal +.>The k-th bit of the mapping, k.epsilon.1, m]，B k,i ∈{0,1}；
Step 2, the received signal sequence is processedColumn ofEach received signal of (a) is divided into two parts, real part and imaginary part, and is marked as +.>Wherein (1)>A 2 n-th input value representing an input layer, i.e., layer 0;
let the node bias vector from the input layer to the hidden layer be recorded asWherein (1)>Representing the bias of the 2 n-th node of the input layer;
the number of nodes of an input layer and a hidden layer of the neural network is 2n, and the number of nodes of an output layer of the neural network is m; the number of layers of the hidden layer is H;
let the weight matrix from 2n nodes in the input layer to 2n nodes in the hidden layer 1 be recorded asWherein (1)>Representing weights of the 2 n-th node in the input layer to the 2 n-th node in the hidden layer of the 1 st layer;
the weight matrix from 2n nodes in any h layer hidden layer to 2n nodes in h+1th layer hidden layer is recorded asWherein the method comprises the steps of，/>Representing weights from the 2 n-th node in the h layer hidden layer to the 2 n-th node in the h+1 layer hidden layer;
let the weight matrix from 2n nodes in the H layer hidden layer to m nodes in the output layer be recorded asWherein (1)>Representing the weight from the 2n node in the H-th layer hidden layer to the m node in the output layer, H E [1, H]；
Let node bias vector in any h-th hidden layer be recorded asWherein (1)>Representing the bias of the 2 n-th node in the h-th hidden layer;
making the calculation result sequence of any h-th hidden layer be recorded asWherein (1)>Representing the calculation result of the 2n node in the h layer hidden layer;
let the linear equation of the h hidden layer be y (h) ＝ω (h) x T +b (h) Wherein T represents a transpose;
output sequence x after activating h layer hidden layer (h) ＝{x 1 (h) ,x 2 (h) ,…,x 2n (h) And } wherein,represents the 2n output after the h layer hidden layer is activated, and x (h) ＝f(y (h) ) F (·) is the activation function;
output sequence z after activating H-th hidden layer (H) As an input sequence of output layers and by linear equation y=ω in the output layers (H) (x (H) ) T +b (H) The output calculation result sequence is y= { y 1 ,y 2 ,…,y n -wherein y n Representing the calculation result output by the nth node of the output layer; x is x (H) Representing the output sequence after activation of the H-th hidden layer, b (H) Representing node bias vectors in the H-th hidden layer;
step 2.1, assume channel distribution f Y|X (y|x) estimating a log likelihood ratio sequence of the binary mapped bit sequences of the n received signals as using the maximum log approximation method shown in the formula (1)Wherein (1)>Log likelihood ratio sequence representing k-th bit of n received signal maps>Log-likelihood ratio of kth bit representing nth received signal mapping, k e [1, m]；
In the formula (1), the components are as follows,k bit decisions representing n received signal mappings are 1 or 0, respectivelyMaximum probability, f Y|X (y|x) represents the channel transition probability of X for the transmit signal and Y for the receive signal, X, Y representing the transmit and receive signal sequences, respectively;
step 2.2, defining the current iteration number as I, and defining the maximum iteration number as I max Initializing i=1;
randomly initializing all weight matrixes and node offset vectors in the neural network when I=1, and recording the initialized weight matrixes and node offset vectors as a set theta of the I-th iteration I ＝{ω (0) ,ω (1) ,…,ω (H) ；b (1) ,b (2) ,…b (H) }；
Step 2.3, establishing a loss function l (theta) of the ith iteration of the neural network by using the formula (2) I )：
In the formula (2), the amino acid sequence of the compound,representing the calculation result sequence output by the kth node of the output layer after the ith iteration, andwherein (1)>Representing the log-likelihood ratio of the ith received signal output by the kth node after the ith iteration;
step 2.4 gradient θ for the I-th iteration using equation (3) I Updating, namely updating the ownership weight matrix and the bias vector in the network to obtain the gradient theta of the (I+1) th iteration I+1 ；
In the formula (3), alpha is the learning rate in machine learning, and alpha is more than 0;
step 2.5, after assigning I to I+1, judging that I is greater than I max If so, the neural network training is completed, and a trained calculation network model is obtained, otherwise, the step 2.3 is returned to be sequentially executed until the loss function l (theta I ) Stopping training until the training frequency is smaller than the set training standard epsilon, and obtaining a trained calculation network model for calculating a received signal sequenceOptimal log likelihood ratio sequence of each signal mapping bit sequence +.>Wherein (1)>Optimal log likelihood ratio sequence representing output of kth node of output layer +.>Representing the optimal log likelihood ratio of the ith received signal output by the kth node of the network output layer, marking the neural network parameter after training as theta, and taking the neural network parameter as the network parameter of the calculation stage;
step 2.6, calculating generalized mutual information G through the formula (4);
in the formula (4), E represents a mathematical expectation,representing receipt of B k Is decided as +.>Conditional probability of->Representation->Is a probability distribution of (c).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210015129.4A CN114363218B (en) | 2022-01-07 | 2022-01-07 | Communication reachable rate detection method based on end-to-end learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210015129.4A CN114363218B (en) | 2022-01-07 | 2022-01-07 | Communication reachable rate detection method based on end-to-end learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114363218A CN114363218A (en) | 2022-04-15 |
| CN114363218B true CN114363218B (en) | 2023-07-28 |
Family
ID=81108189
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210015129.4A Active CN114363218B (en) | 2022-01-07 | 2022-01-07 | Communication reachable rate detection method based on end-to-end learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114363218B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108540267A (en) * | 2018-04-13 | 2018-09-14 | 北京邮电大学 | A kind of multi-user data information detecting method and device based on deep learning |
| WO2021041862A1 (en) * | 2019-08-30 | 2021-03-04 | Idac Holdings, Inc. | Deep learning aided mmwave mimo blind detection schemes |
| CN113839744A (en) * | 2021-09-22 | 2021-12-24 | 重庆大学 | Blind detection method of generalized wireless optical MIMO system based on deep learning |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102265546B (en) * | 2008-12-23 | 2015-07-29 | 爱立信电话股份有限公司 | Channel quality based on the radio communication channel receiving data is determined |
| CN104009822B (en) * | 2014-05-14 | 2017-07-11 | 上海交通大学 | Based on new demodulation modification method of the imperfect channel estimation containing arrowband interference |
| US9749089B2 (en) * | 2015-11-04 | 2017-08-29 | Mitsubishi Electric Research Laboratories, Inc. | Fast log-likelihood ratio (LLR) computation for decoding high-order and high-dimensional modulation schemes |
| CN110741553B (en) * | 2017-06-22 | 2023-11-03 | 瑞典爱立信有限公司 | Neural network for forward error correction decoding |
| CN109241392A (en) * | 2017-07-04 | 2019-01-18 | 北京搜狗科技发展有限公司 | Recognition methods, device, system and the storage medium of target word |
| EP3553953A1 (en) * | 2018-04-13 | 2019-10-16 | Université De Reims Champagne-Ardenne | Approximation of log-likelihood ratios for soft decision decoding in the presence of impulse noise channels |
| US12107679B2 (en) * | 2019-04-29 | 2024-10-01 | Nokia Technologies Oy | Iterative detection in a communication system |
| CN111181607B (en) * | 2020-01-09 | 2021-04-27 | 杭州电子科技大学 | Physical layer coding optimization antenna selection method based on soft message selection forwarding |
-
2022
- 2022-01-07 CN CN202210015129.4A patent/CN114363218B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108540267A (en) * | 2018-04-13 | 2018-09-14 | 北京邮电大学 | A kind of multi-user data information detecting method and device based on deep learning |
| WO2021041862A1 (en) * | 2019-08-30 | 2021-03-04 | Idac Holdings, Inc. | Deep learning aided mmwave mimo blind detection schemes |
| CN113839744A (en) * | 2021-09-22 | 2021-12-24 | 重庆大学 | Blind detection method of generalized wireless optical MIMO system based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114363218A (en) | 2022-04-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ye et al. | Channel agnostic end-to-end learning based communication systems with conditional GAN | |
| CN108566257B (en) | Signal recovery method based on back propagation neural network | |
| CN110365612A (en) | A kind of deep learning Beam Domain channel estimation methods based on approximate Message Passing Algorithm | |
| CN112468230B (en) | Wireless ultraviolet light scattering channel estimation method based on deep learning | |
| CN112926265A (en) | Atmospheric porous probe measurement calibration method based on genetic algorithm optimization neural network | |
| CN113381828B (en) | Sparse code multiple access random channel modeling method based on condition generation countermeasure network | |
| CN110233810B (en) | MSK signal demodulation method based on deep learning under mixed noise | |
| CN110336594A (en) | A kind of deep learning signal detecting method based on conjugate gradient decent | |
| CN110300075A (en) | A kind of radio channel estimation method | |
| CN112422208B (en) | Signal detection method based on antagonistic learning under unknown channel model | |
| CN114584230A (en) | Predictive channel modeling method based on countermeasure network and long-short term memory network | |
| CN114325245B (en) | Power transmission line fault line selection and positioning method based on traveling wave data deep learning | |
| CN114363218B (en) | Communication reachable rate detection method based on end-to-end learning | |
| CN110474798B (en) | Method for predicting future signal of wireless communication by using echo state network | |
| US11489560B2 (en) | Method of parameter estimation for a multi-input multi-output system | |
| CN114614920B (en) | Signal detection method based on data and model combined driving of learning factor graph | |
| CN116760491A (en) | Signal-to-noise ratio estimation method based on deep learning | |
| CN116362328A (en) | Federal learning heterogeneous model aggregation method based on fairness characteristic representation | |
| CN116527180A (en) | SCMA method based on CWGAN-GP satellite-ground link channel modeling | |
| CN113489545B (en) | Light space pulse position modulation step-by-step classification detection method based on K-means clustering | |
| CN113890633B (en) | Underwater acoustic communication system self-adaptive selection method based on deep neural network | |
| CN114070415A (en) | Optical fiber nonlinear equalization method and system | |
| CN112821971A (en) | Time-varying channel signal detection method based on countermeasure learning | |
| CN113052081A (en) | Modulation mode identification method based on Cauchy Score constellation diagram | |
| Li et al. | Stochastic channel modeling for deep neural network-aided sparse code multiple access communications |
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 |