Visible light spatial modulation signal detection method based on vector machine

By using a vector machine-based signal detection method, the problem of difficulty in obtaining channel state information is solved, achieving efficient signal detection under low signal-to-noise ratio, avoiding complex channel estimation, and improving detection performance.

CN118971955APending Publication Date: 2024-11-15SOUTHEAST UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410837507.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-26
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

Existing visible light communication signal detection methods require perfect channel state information. Imperfect channel state information will reduce algorithm performance, and channel estimation is complex and it is difficult to obtain the statistical characteristics of the signal, which increases the design difficulty.

Method used

A signal detection method based on vector machines is adopted. A signal detection classification model is established during the training phase, and the one-to-one support vector machine model is transformed into a convex quadratic programming model to obtain weights. Signal detection is performed during the detection phase to avoid channel estimation. The decision region is divided in a data-driven manner.

Benefits of technology

It achieves near-optimal signal detection performance at low signal-to-noise ratios, requires no channel estimation, and provides detailed decision region division.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure BDA0004913578910000015
    Figure BDA0004913578910000015
  • Figure BDA0004913578910000025
    Figure BDA0004913578910000025
  • Figure BDA0004913578910000027
    Figure BDA0004913578910000027
Patent Text Reader

Abstract

The invention discloses a signal detection method for visible light spatial modulation based on a vector machine, and relates to the technical field of visible light communication signal detection, signal detection is divided into a training stage and a detection stage, in the training stage, a signal detection classification model is established, a one-to-one support vector machine model is established for the classification model, and the one-to-one support vector machine model is established for the detection stage. Converting the one-to-one support vector machine model into a convex quadratic programming model by adopting an interval maximization method, solving the convex quadratic programming model, and obtaining the weight of the one-to-one support vector machine model; in the detection stage, the one-to-one support vector machine model weights are used for completing signal detection. Compared with a traditional signal detection method, the method has the advantages that channel state information is obtained based on data driving without channel estimation, the judgment area is divided carefully, and the performance is close to the optimal detection under the low signal-to-noise ratio.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of visible light communication signal detection, and in particular to a visible light spatial modulation signal detection method based on a vector machine. Background Art

[0002] At present, the spatial modulation visible light communication signal detection method adopts a design method based on the statistical characteristics of the received signal, and designs corresponding detection methods for different statistical characteristics. The algorithm designed by this method requires perfect channel state information. Imperfect channel state information will reduce the performance of the algorithm. Therefore, the receiver requires a complex channel estimation algorithm. At the same time, the pilot sequence to complete the channel estimation requires a higher received signal-to-noise ratio. In addition, when designing the algorithm, it is necessary to know the statistical characteristics of the received signal, but the statistical characteristics of the actual signal are difficult to obtain, which also increases the difficulty of algorithm design. Summary of the invention

[0003] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art and provide a signal detection method for visible light spatial modulation based on a vector machine. Compared with the traditional signal detection method, the present invention is based on data-driven, does not require channel estimation to obtain channel state information, and divides the decision area in detail, and the performance is close to optimal detection under low signal-to-noise ratio.

[0004] The present invention adopts the following technical solutions to solve the above technical problems:

[0005] A signal detection method for visible light spatial modulation based on a vector machine proposed in the present invention includes a training phase and a detection phase of the vector machine, wherein:

[0006] The vector machine training phase includes:

[0007] Establish a signal detection classification model; establish a one-to-one support vector machine model for the signal detection classification model, then use the margin maximization method to transform the one-to-one support vector machine model into a convex quadratic programming model, solve the convex quadratic programming model, and obtain the one-to-one support vector machine model weight;

[0008] The detection phase includes:

[0009] Signal detection is completed using the one-to-one support vector machine model weights trained in the training phase.

[0010] As a further optimization scheme of the signal detection method for visible light spatial modulation based on a vector machine described in the present invention, the training stage is specifically as follows:

[0011] Step A-1: ​​Establish a signal detection classification model, including: Send signal set Receive signal set and class number set in,

[0012] Send signal set It is expressed as:

[0013]

[0014] Among them, x i is the i-th transmitted spatial modulation symbol, denoted by x i =s m a n ;s m is the mth pulse amplitude modulation symbol, s m =2mP t (M+1),m=1,2,...,M, where P t is the average transmission power, M is the order of the pulse amplitude modulation symbol; a n Select the vector for the nth LED, whose dimension is N t , N t is the number of all emitting LEDs, a n Medium t elements are 1, and the rest are 0, n t is the number of activated LEDs; Indicates rounding down. Indicates that from N t Pick n different elements from t The number of combinations of elements, assuming that the LED selection vector has 2 L , using L bits for encoding; the number of spatial modulation symbols sent is K, and K = 2 L M;

[0015] Receive signal set for:

[0016]

[0017] Among them, r i represents the received spatial modulation symbol under the i-th noise-free channel, whose dimension is N r , N r is the number of receiving photodetectors; H is the visible light channel matrix, whose dimension is N r ×N t ;

[0018] Class Number Set for:

[0019] C={1,2,...,...,K},

[0020] Among them, 1, 2, ..., ..., K represent categories. The categories and the transmitted spatial modulation symbols satisfy a one-to-one correspondence. Category i corresponds to the i-th transmitted spatial modulation symbol x i .

[0021] As a further optimization scheme of a signal detection method for visible light spatial modulation based on a vector machine described in the present invention, a one-to-one support vector machine model includes a multi-classification data set, a one-to-one two-classification data set and a two-classification support vector machine, wherein:

[0022] The multi-classification dataset T is:

[0023] T={(y1,c1),(y2,c2),...,(y i ,c i ),...,(y N ,c N )},

[0024] Among them, (y i ,c i ) is the sample point of the multi-classification dataset, y i represents the received spatial modulation symbol under the i-th noisy channel, and its dimension is N t ,y i =Hx i +n, n is the additive noise vector; N is the number of training data in the multi-classification dataset, N = KP, P is x i The number of repetitions;

[0025] The one-to-one binary classification data set is:

[0026]

[0027] in, represents a binary classification data set consisting of categories k1 and k2, Indicates when When c i =+1; Indicates when When c i =-1, The number of elements in is 2P; and It is a single-class dataset split from a multi-class dataset T, satisfying and k1 and k2 satisfy k1∈{1,...,K-1}, k2∈{2,...,K}, k1<k2.

[0028] As a further optimization scheme of the signal detection method of visible light spatial modulation based on vector machine described in the present invention,

[0029] against Using the margin maximization method, a binary classification support vector machine optimization model is established for:

[0030]

[0031] in, and are the support vector machine model parameters, N r dimensional weight vector, is the deviation; · represents the dot product between two vectors, and P is an integer.

[0032] As a further optimization scheme of the signal detection method of visible light spatial modulation based on vector machine described in the present invention, solving The specific method is as follows:

[0033] Step A-3-1: Transformed into a dual model for:

[0034]

[0035] where α1,α2,...,α 2P is the Lagrange multiplier, and The relationship between the solutions is: there exists α j >0 makes the original problem The solution is and Where j≠i;

[0036] Step A-3-2: Place the model Transformed into a dual model including kernel function for:

[0037]

[0038] Among them, K(·),·) is the kernel function, and K(y i ,y j )=φ(y i )·φ(y j ), φ(·) is the mapping function; and The relationship between the solutions is: there exists α j >0 makes the original problem The solution is and Where j≠i;

[0039] Step A-3-3: Solving the model Thus we get and

[0040] As a further optimization scheme of the signal detection method for visible light spatial modulation based on vector machine described in the present invention, the kernel function K(·,·) is a Gaussian kernel function, where j≠i, σ 2 is the Gaussian kernel function parameter, set to 1.

[0041] As a further optimization scheme of the signal detection method of visible light spatial modulation based on vector machine described in the present invention, the specific process of the detection stage is as follows:

[0042] Step B-1: Calculate the classification label is the classification decision function, and its expression is as follows:

[0043]

[0044] y is y i Any one of;

[0045] Step B-2: Constructing a multiset Statistics all The most frequent As detection categories, namely:

[0046]

[0047] Among them, num(k) is the statistical multiset The number of categories k in ;

[0048] Step B-3: Obtain the transmitted spatial modulation symbol estimate and

[0049] Compared with the prior art, the present invention adopts the above technical solution and has the following technical effects:

[0050] The present invention divides signal detection into a training phase and a detection phase. In the training phase, a signal detection classification model is established, a one-to-one support vector machine model is established for the classification model, and then an interval maximization method is used to transform the one-to-one support vector machine model into a convex quadratic programming model, the convex quadratic programming model is solved, and the weight of the one-to-one support vector machine model is obtained; in the detection phase, the one-to-one support vector machine model weight is used to complete signal detection; compared with the traditional signal detection method, the method of the present invention is based on data-driven, does not require channel estimation to obtain channel state information, and divides the decision area in detail, and the performance is close to optimal detection under low signal-to-noise ratio. DETAILED DESCRIPTION

[0051] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention is described in detail below.

[0052] The present invention provides a visible light spatial modulation signal detection method based on a one-to-one support vector machine. The method can realize signal detection without channel estimation in an optical communication system based on a data-driven method for spatial modulation. Signal detection is divided into two stages: training and detection. In the training stage, a signal detection classification model is established, and a one-to-one support vector machine model is established for the classification model. Then, the one-to-one support vector machine model is converted into a convex quadratic programming model by using an interval maximization method, and the convex quadratic programming model is solved to obtain the one-to-one support vector machine model weight; in the detection stage, the one-to-one support vector machine model weight is used to complete signal detection. Compared with the traditional signal detection method, the method of the present invention is based on data-driven, does not require channel estimation to obtain channel state information, and divides the decision area in detail, and the performance is close to optimal detection under low signal-to-noise ratio.

[0053] In a spatial modulation visible light communication system, there are N t Transmitter LEDs and N r The receiving photodetectors activate only n t The spatial modulation visible light communication system transmits bits according to bits are grouped into groups, each group contains two subgroups. The first subgroup contains The LED selection bits contain the number and index of activated LEDs, which are called spatial information. The LED selection bits generate the LED selection vector a n , whose dimension is N t , used to describe LED index information, where n t The first element is 1, and the rest are 0; the second subgroup contains log2M bits, which are data bits, and M is the pulse amplitude modulation order. The data bits are converted into transmission symbols after pulse amplitude modulation, which is represented by s m, the i-th transmitted spatial modulation symbol can be expressed as x i =s m a n .

[0054] The spatial modulation symbols sent pass through the visible light channel H, whose dimension is N r ×N t , after adding noise, the received signal is obtained. When performing signal detection, it is necessary to transform the signal detection problem into a classification problem, and then use a one-to-one support vector machine model to model and solve it. Therefore, signal detection is divided into two stages: training and detection. The training stage can be divided into three steps: (1) establishing a classification model; (2) establishing a one-to-one support vector machine model; (3) solving the optimization problem. The purpose of the training stage is to determine the optimal support vector machine model parameters by processing a set of known data to achieve accurate signal classification. The detection stage is relatively simple, that is, using the trained one-to-one support vector machine model to classify the new signal to achieve signal detection.

[0055] First, the training phase is introduced.

[0056] Step 1: Build a classification model

[0057] Each time data is sent, the combination of the activated LED index group and the pulse amplitude modulation symbol is limited, so the transmission signals of different combinations can be numbered, each combination corresponds to a category. L In the combination, Using L bits, the pulse amplitude modulation symbol has M amplitudes, using log2M bits. Therefore, there are K = 2 L Class M is represented by L+log2M bits. Therefore, the transmitted signal set is represented as Assuming the channel matrix is ​​known, without considering the noise, the received signal set is r i is the received spatial modulation symbol under the i-th noiseless channel, there are K categories in total, and the category number set is c is the category, and the category and the transmitted spatial modulation symbol satisfy a one-to-one correspondence, such as category c=i corresponds to the i-th transmitted spatial modulation symbol x i .

[0058] Step 2: Build a one-to-one support vector machine model

[0059] After the classification model is established, a multi-classification support vector machine is established to solve the classification model. The signal in the training signal-to-noise ratio SNR trainAfter passing through the additive Gaussian channel, a received signal is generated. Therefore, a multi-class dataset T is established as T = {(y1, c1), (y2, c2),..., (y i , c i ),..., (y N , c N )}, where (y i , c i ) is a sample point of the multi-class dataset, y i = Hx i + n, which is the received spatial modulation symbol under the i-th noisy channel, n is the additive Gaussian noise, and c i is the corresponding class number.

[0060] Next, a one-against-one support vector machine model is established. First, each class of data is split to generate K sets. The k-th set is T k = {(y i , c i ) | (y i , c i ) ∈ T, c i = k}, and each time two sets of k1 and k2 are selected and The two sets are merged, and the class numbers are modified to +1 and -1 to obtain indicating that when , c i = +1; indicating that when , c i = -1, The number of elements in is 2P. k1 and k2 satisfy k1 ∈ {1,..., K - 1}, k2 ∈ {2,..., K}, k1 < k2. Thus, a pair of one-against-one binary classification datasets For this pair of one-against-one binary classification datasets A binary classification support vector machine optimization model is established

[0061]

[0062] Subject to:

[0063] Where and are the support vector machine model parameters, is the N r -dimensional weight vector, is the bias, and · represents the dot product. For a pair of one-against-one binary classification datasets, the binary classification support vector machine optimization model is finally obtained, and the support vector machine model parameters and

[0064] Step 3: Solve the optimization problem

[0065] In order to obtain the parameters of the binary support vector machine model, it is necessary to solve the optimization model The optimization model can be transformed into the following dual model

[0066]

[0067] Limited by:

[0068] α i ≥0,i=1,2,...,2P

[0069] Among them, α1, α2, ..., α 2P is the Lagrange multiplier, and The relationship between the solutions is: there exists α j > 0, so that the solution to the original problem is and

[0070] Transform the dual problem to obtain the dual model with kernel function

[0071]

[0072] Among them, K(·,·) is the kernel function, and K(y i ,y j )=φ(y i )·φ(y j ), φ(·) is the mapping function; and The relationship between the solutions is: there exists α j > 0, so that the solution to the original problem is and

[0073] Solving the dual model with kernel function using interior point method We get α1, α2, ..., α 2P , according to the corresponding relationship and Finally get and

[0074] Dual model for all kernel functions where k1∈{1,...,K-1},k2∈{2,...,K},k1<k2 for k1 and k2 Solve and get K(K-1) / 2 pairs and

[0075] Next comes the detection phase.

[0076] In the detection phase, for any signal-to-noise ratio y under the current channel H i , use the solved K(K-1) / 2 pairs of parameters to complete signal detection. Calculate K(K-1) / 2 classification labels is the classification decision function, and its expression is as follows:

[0077]

[0078] Among them, x is The calculated scalar value. Get K(K-1) / 2 Then, construct a multi-set Statistics all The most frequent As the detection category,

[0079]

[0080] Where num(k) represents the statistical multiset According to the correspondence between the transmitted spatial modulation symbol and the category, the transmitted spatial modulation symbol estimation value is obtained. and

[0081] The above description is only a specific implementation of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a technician familiar with the technical field within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A signal detection method for visible light spatial modulation based on vector machine, characterized in that: It includes the training phase and the detection phase of the vector machine, in which: The vector machine training phase includes: Establish a signal detection classification model; establish a one-to-one support vector machine model for the signal detection classification model, then use the margin maximization method to transform the one-to-one support vector machine model into a convex quadratic programming model, solve the convex quadratic programming model, and obtain the one-to-one support vector machine model weight; The detection phase includes: Signal detection is completed using the one-to-one support vector machine model weights trained in the training phase.

2. The signal detection method of visible light spatial modulation based on vector machine according to claim 1, characterized in that: The training phases are as follows: Step A-1: ​​Establish a signal detection classification model, including: Send signal set Receive signal set and class number set in, Send signal set It is expressed as: Among them, x i is the i-th transmitted spatial modulation symbol, denoted by x i =s m a n ;s m is the mth pulse amplitude modulation symbol, s m =2mP t / (M+1),m=1,2,...,M,where P t is the average transmission power, M is the order of the pulse amplitude modulation symbol; a n Select the vector for the nth LED, whose dimension is N t , N t is the number of all emitting LEDs, a n Medium t elements are 1, and the rest are 0, n t is the number of activated LEDs; Indicates rounding down. Indicates that from N t Pick n different elements from t The number of combinations of elements, assuming that the LED selection vector has 2 L , using L bits for encoding; the number of spatial modulation symbols sent is K, and K = 2 L M; Receive signal set for: Among them, r i represents the received spatial modulation symbol under the i-th noise-free channel, whose dimension is N r , N r is the number of receiving photodetectors; H is the visible light channel matrix, whose dimension is N r ×N t ; Class Number Set for: C={1,2,...,...,K}, Among them, 1, 2, ..., ..., K represent categories. The categories and the transmitted spatial modulation symbols satisfy a one-to-one correspondence. Category i corresponds to the i-th transmitted spatial modulation symbol x i .

3. The signal detection method of visible light spatial modulation based on vector machine according to claim 2, characterized in that: The one-to-one support vector machine model includes a multi-classification dataset, a one-to-one binary classification dataset, and a binary classification support vector machine, where: The multi-classification dataset T is: T={(y1,c1),(y2,c2),...,(y i ,c i ),...,(y N ,c N )}, Among them, (y i ,c i ) is the sample point of the multi-classification dataset, y i represents the received spatial modulation symbol under the i-th noisy channel, and its dimension is N t ,y i =Hx i +n, n is the additive noise vector; N is the number of training data in the multi-classification dataset, N = KP, P is x i The number of repetitions; The one-to-one binary classification data set is: in, represents a binary classification data set consisting of categories k1 and k2, Indicates when When c i =+1; Indicates when When c i =-1, The number of elements in is 2P; and It is a single-class dataset split from a multi-class dataset T, satisfying and k1 and k2 satisfy k1∈{1,...,K-1}, k2∈{2,...,K}, k1<k2.

4. The signal detection method of visible light spatial modulation based on vector machine according to claim 3 is characterized in that: against Using the margin maximization method, a binary support vector machine optimization model is established for: in, and are the support vector machine model parameters, N r dimensional weight vector, is the deviation; · represents the dot product between two vectors, and P is an integer.

5. The signal detection method of visible light spatial modulation based on vector machine according to claim 4 is characterized in that: Solution The specific method is as follows: Step A-3-1: Transformed into a dual model for: where α1,α2,...,α 2P is the Lagrange multiplier, and The relationship between the solutions is: there exists α j >0 makes the original problem The solution is and Where j≠i; Step A-3-2: Place the model Transformed into a dual model including kernel function for: Among them, K(·,·) is the kernel function, and K(y i ,y j )=φ(y i )·φ(y j ), φ(·) is the mapping function; and The relationship between the solutions is: there exists α j >0 makes the original problem The solution is and Where j≠i; Step A-3-3: Solving the model Thus we get and 6. The signal detection method of visible light spatial modulation based on vector machine according to claim 5, characterized in that: The kernel function K(·,·) is a Gaussian kernel function. where j≠i, σ 2 is the Gaussian kernel function parameter, set to 1.

7. The signal detection method of visible light spatial modulation based on vector machine according to claim 5, characterized in that: The specific process of the detection phase is as follows: Step B-1: Calculate the classification label is the classification decision function, and its expression is as follows: y is y i Any one of; Step B-2: Constructing a multiset Count the most frequent occurrences among all D As detection categories, namely: , Among them, num(k) is the statistical multiset The number of categories k in ; Step B-3: Obtain the transmitted spatial modulation symbol estimate and