A serial separation method for aliased unknown wireless signals
The serial separation method for aliased unknown wireless signals iteratively decomposes signals based on their characteristics and noise levels, addressing recognition failures and enhancing accuracy in signal decomposition.
Patent Information
- Application Number
- CN202111237401.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-10-22
- Publication Date
- 2024-12-13
- Estimated Expiration
- 2041-10-22
AI Technical Summary
Existing wireless signal decomposition methods, such as NMF and NNSC, struggle with accurately separating aliased unknown wireless signals, particularly when signal components have significant magnitude differences, leading to recognition failures and inaccurate decompositions.
A serial separation method for aliased unknown wireless signals, involving iterative updates of sparse coding matrices and spectrum feature dictionaries, decomposes signals one by one and layer by layer, accounting for signal characteristics and noise levels to enhance accuracy.
The method improves signal decomposition accuracy by ensuring fair representation of all signal components, even when large signals do not overshadow small ones, and effectively handles both noise-free and noisy scenarios.
Smart Images
Figure CN114118211B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of wireless communication technology, and in particular to a serial separation method for aliased unknown wireless signals. Background Art
[0002] The core of data analysis-based problems is to find an appropriate representation of the data. Such useful representations can be to clearly separate the underlying structure of the data, or to reduce the dimension of the data to further reduce the complexity. Such methods for analyzing data already exist, such as PCA (Principal Component Analysis), ICA (Independent Component Analysis), MF (Matrix Factorization) and NMF (Non-negative Matrix Factorization). ICA performs blind source separation in an unsupervised manner, however, it cannot obtain the actual number of components and cannot perform quantitative analysis on each component. MF can effectively handle data separation problems, but it is not suitable for situations where each component is non-negative in actual problems.
[0003] Compared with the above methods, NMF proposed by Lee and Seung in Nature in 1999 decomposes a non-negative matrix into the product of two non-negative matrices, namely the coding matrix and the basis matrix. The components after NMF decomposition are all non-negative values, which is more suitable for practical scenarios, that is, decomposing the whole into the sum of parts. Therefore, the application of NMF in many fields has attracted widespread attention and research, such as wireless signal separation, noise suppression, image recognition, energy solution, etc. In addition, the non-negativity of NMF also leads to the requirement of coding sparsity. It uses fewer active components to encode data, making the coding more in line with the actual situation. However, the sparsity of the code under the NMF method is uncertain, so more existing solutions are to add the requirement of coding sparsity to NMF to control the sparsity of the code, namely NNSC (Non-Negative Sparse Coding).
[0004] In terms of noise suppression, the noise uses the source signal model and updates the signal base matrix based on NNSC to suppress the influence of noise. In terms of energy solution, sparse coding is applied to the energy solution task to obtain effective results of energy usage of each household appliance. In terms of wireless signal decomposition, NNSC is used to decompose the superimposed signal to further complete the monitoring of the electromagnetic environment, which is of great significance to the safety of the electromagnetic environment and the use of the spectrum. If the signal to be decomposed is a superposition of base signals, the coding matrix obtained by the NNSC method can clearly quantify each base signal. Generally speaking, the signal components of the superimposed decomposition matrix will not be completely identical to the base signal, and the base signal is only a representation of each signal, so it will cause the result of the coding matrix to be biased. Especially when the magnitude difference of the component signals in the signal to be decomposed is large, the result of the signal component with a small proportion is generally inaccurate, that is, it is submerged.
[0005] That is to say, if the magnitudes of the various parts of the signal to be decomposed are very different, the larger part is more likely to be recognized, and the smaller part may be submerged, resulting in recognition failure; and in actual scenarios, the components that make up the decomposed signal are somewhat different from the signal base matrix. In order to solve the above problems, it is urgent to propose a serial wireless signal decomposition method, which peels off the signal components in the superimposed signal one by one and layer by layer, thereby improving the accuracy of signal decomposition.
[0006] In summary, it is very necessary to study the serial wireless signal decomposition method from the perspective of the signal base matrix and the small signals in the decomposed signal. Summary of the invention
[0007] The purpose of the present invention is to solve the above-mentioned defects in the prior art and provide a serial separation method for aliased unknown wireless signals.
[0008] The purpose of the present invention can be achieved by adopting the following technical solutions:
[0009] A serial separation method for aliased unknown wireless signals, the serial separation method comprising the following steps:
[0010] S1. Initialize the sparse coding matrix A of the superimposed signal at the receiving end, where all elements in the sparse coding matrix A are non-negative values, and initialize the total power P of the separated wireless signal to 0;
[0011] S2. The power spectrum density of the M wireless signals in the frequency domain is used as the spectrum feature dictionary matrix D of the M wireless signals. s and noise characteristic matrix D n ;
[0012] S3, the receiving end receives an aliased unknown signal X emitted by multiple unknown transmitting sources, and obtains a sparse coding matrix A of the aliased unknown wireless signal and a spectrum feature dictionary matrix D of the aliased unknown wireless signal through iterative updating;
[0013] S4. The receiving end separates the signal components contained in the aliased signal one by one in series until the total power constraint is met.
[0014] Furthermore, the process of step S2 is as follows:
[0015] S201, initialize M independent wireless signals S=[S1, ..., S m , ..., S M ],in is the power spectral density of the mth signal in the frequency domain, W is the number of samples of the signal spectrum, each column of S contains the sampling points obtained within the bandwidth B of the signal receiving end, and each signal S is obtained m The power spectral density of
[0016] S202, training each signal S m Get the corresponding dictionary matrix vector D s_m And construct M wireless signal spectrum feature dictionary matrices D s , that is, the signal basis matrix, D s The mth column of is the power spectral density of the mth signal and And according to the formula Get the unitized basis matrix Record the initialization of Ds ||·|| represents the L2 norm of the vector, Represents the field of real numbers.
[0017] Furthermore, in step S3, the unknown signal is superimposed on the receiving end without noise. The processing process is as follows:
[0018] S30A1, assuming that the wireless signal spectrum feature dictionary matrix D in a noise-free scenario is D s , the unit basis matrix of D According to the formula Update the encoding matrix A of the aliased unknown signal, where ⊙ and Respectively represent the multiplication and division of each element in the matrix, ← represents the calculation result of the formula on the right side of the arrow as the iteration result of the variable on the left side of the arrow, the sparsity parameter λ>0, [·] T Indicates transposing the matrix in brackets;
[0019] S30A2, according to the formula Update the wireless signal spectrum feature dictionary matrix D, where is obtained by column standardization of D, is a square matrix with all elements set to 1 and of appropriate size;
[0020] S30A3, according to A and The iterative result of the noise-free superposition unknown signal is Return to steps S30A1 and S30A2 to perform cyclic updates until the coding matrix of step S30A1 converges to obtain the result of the coding matrix.
[0021] Furthermore, in step S3, for the unknown signal containing noise superimposed The processing process is as follows:
[0022] S30B1, sparse coding matrix of superimposed signal at receiving end Among them A s and A n Initialize the wireless signal and noise sparse coding matrices under noisy conditions respectively, and their elements are all non-negative;
[0023] S30B2, the spectrum feature dictionary of wireless signals under noisy conditions is expanded to D = [D s , D n ], D n is the initialization noise characteristic matrix obtained by the noise power spectrum density within bandwidth B received by the receiver in the no-signal scenario, D s is the dictionary matrix of M wireless signal spectrum features under noisy conditions, then the unitized basis matrix of D in Yes D s The unit matrix of Yes D n The unit basis matrix of ;
[0024] S30B3, update A according to the following update rules s and A n :
[0025]
[0026]
[0027] Among them l s >0 and l n >0 are the sparsity parameters of signal and noise respectively;
[0028] S30B4, update the wireless signal spectrum feature dictionary matrix D according to the following update rules s as follows
[0029]
[0030] in is the matrix obtained by column normalization of D, is a square matrix with all elements set to 1 and of appropriate size;
[0031] S30B4, the decomposition result of the receiving end signal part is Return to steps S30B2 and S30B3 to perform cyclic updates until the coding matrix of step S30B2 converges to obtain the result of the coding matrix.
[0032] Furthermore, the process of step S4 is as follows:
[0033] S401, the receiving end performs serial separation on the aliased signals one by one, and calculates the sequence number m0 corresponding to the signal with the maximum sparse coefficient, which is Among them A m represents the sparse coefficient of the mth column of A, i.e. the mth signal, and then updates the received signal X at the receiving end.
[0034]
[0035] in, is the sparse coefficient corresponding to the m0th signal, is the power spectrum density of the m0th wireless signal in the spectrum feature dictionary matrix, is the power spectrum density of the m0th wireless signal in the basis matrix of the spectrum feature dictionary matrix; if there are negative elements in the updated matrix X, correct them to 0;
[0036] S402, Definition and P are the sparse coefficients and the separated signal power after serial separation in step S401, respectively. and P,
[0037]
[0038]
[0039] Until P>ε∑(X), ε is the power fitting coefficient, ε∈(0,1).
[0040] Compared with the prior art, the present invention has the following advantages and effects:
[0041] 1) Compared with the existing wireless signal decomposition algorithm, the serial wireless signal decomposition algorithm proposed in the present invention decomposes the wireless superposition signal based on the characteristics of each signal.
[0042] 2) For the decomposition of noise-free superposition signals, the serial wireless signal decomposition algorithm proposed in the present invention decomposes the signals to be decomposed layer by layer and one by one based on the characteristics of each signal, thus overcoming the problem of large signals drowning out small signals and improving the fairness of the entire decomposition algorithm.
[0043] 3) For the decomposition of the noisy superposition signal, the present invention first obtains the dictionary matrix of the noise according to the noise-free scene, and then uses the proposed algorithm to perform serial signal decomposition on the wireless superposition signal. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 It is a technical block diagram of a serial separation method for aliased unknown wireless signals in an embodiment of the present invention;
[0045] Figure 2 It is a flow chart of a serial separation method for aliased unknown wireless signals in a noise-free scenario according to an embodiment of the present invention;
[0046] Figure 3 is a flow chart of a serial separation method for aliased unknown wireless signals in a noisy scenario according to an embodiment of the present invention;
[0047] Figure 4 is a schematic diagram of frequency domain sampling of eight independent signals in an embodiment of the present invention, including Bluetooth, GPS, FDDLTE, TD-LTE, WiMAX, WLAN (11a), WLAN (11b) and WLAN (11g);
[0048] Figure 5 is a schematic diagram of the spectrum of the target signal and the estimated signal in different decomposition algorithms in a noise-free scenario in an embodiment of the present invention, wherein: Figure 5 (a) is a schematic diagram of the spectrum in the NNSC algorithm; Figure 5 (b) is a schematic diagram of the spectrum in the SWSD algorithm;
[0049] Figure 6 is a schematic diagram of energy ratios of various components of the target signal and various components of the estimated signal obtained using different algorithms in a noise-free scenario in an embodiment of the present invention, wherein: Figure 6 (a) is a schematic diagram of the energy ratio of each component of the target signal; Figure 6 (b) is a schematic diagram of the energy ratio of each component of the estimated signal under the NNSC algorithm; Figure 6 (c) is a schematic diagram of the energy ratio of each component of the estimated signal under the SWSD algorithm;
[0050] Figure 7 is a schematic diagram of the energy ratio of each component of the target signal and each component of the estimated signal obtained by using different algorithms in a noisy scene (SNR=20dB) in an embodiment of the present invention, where: Figure 7(a) is a schematic diagram of the energy ratio of each component of the target signal; Figure 7 (b) is a schematic diagram of the energy ratio of each component of the estimated signal under the NNSC algorithm; Figure 7 (c) is a schematic diagram of the energy ratio of each component of the estimated signal under the SWSD algorithm. DETAILED DESCRIPTION
[0051] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0052] Example
[0053] This embodiment combines Figures 1 to 7 , specifically describes a serial separation method for aliased unknown wireless signals. Assume that there are M independent signals, S = [S1, ..., S m , ..., S M ],in is the mth signal and is a 1-dimensional column vector, and W is the number of samples of the signal spectrum. m Each column of contains the sampling points obtained within the bandwidth B (if analyzing time domain signals, the sampling points are obtained within the time period T). First, we need to train each signal S m The features of the signal are then used to decompose the superimposed signal. To obtain the features of each signal, a large amount of sample data needs to be trained to obtain its features. D is the dictionary matrix composed of the above features, that is, the signal basis matrix. Assume that the mth column of D is the spectral feature of the mth signal, and
[0054] like is the superposition signal to be decomposed, then X≈DA, where It corresponds to the mth row of the weight matrix A.
[0055] The signal to be decomposed is composed of multiple unknown signals superimposed, and their strengths and characteristics are different. Obviously, compared with other signals, signals with large energy or obvious characteristics are easier to detect. If the signal to be decomposed is detected in layers, the accuracy of signal decomposition can be improved. Therefore, this embodiment proposes a SWSD (successive wireless signals disaggregation) algorithm, which improves the performance of existing methods.
[0056] As mentioned above, the SWSD-based algorithm can decompose the superposition signal in different scenarios, namely, the noise-free superposition signal and the noisy superposition signal.
[0057] (1) Noise-free superposition signal
[0058] Since the superposition signal is noise-free, the dictionary matrix does not need the characteristics of the noise. The process of the noise-free SWSD algorithm is as follows:
[0059] T1, initialize the encoding matrix Total power P=0.
[0060] T2. Train each signal S m Get the corresponding dictionary matrix D m , and follow Normalize the basis matrix. Record the initialization of D
[0061] T3. Obtain the values of the encoding matrix A and the dictionary matrix D according to the following steps:
[0062] T301. Randomly initialize the encoding matrix A, whose elements are non-negative.
[0063] T302, update the coding matrix according to the following rules Among them ⊙ and They represent the multiplication and division of each element in the matrix, and the sparsity parameter λ>0.
[0064] T303, update the dictionary matrix according to the following rules
[0065]
[0066] in It is obtained by column standardization of D, which is defined here is a square matrix whose elements are all 1 and of the appropriate size.
[0067] T304, the superposition signal is decomposed into Return to step T302 until convergence to obtain the result of the encoding matrix.
[0068] T4. Calculation Then update X, the formula is as follows:
[0069]
[0070] If there are negative elements in the updated X, correct them to 0.
[0071] T5. Update according to the following rules and P
[0072]
[0073] Return to step T3 until This means that the estimated total power is ε times higher than the target power, with ε∈(0,1).
[0074] (2) Noisy superposition signal
[0075] In a noisy scenario, the superposition signal is composed of AWG (Additive White Noise) and multiple signals. Then it is necessary to train the noise to obtain its characteristics, which will be an important part of the dictionary matrix. Obviously, the weight matrix can be divided into two parts, namely the signal part and the noise part. Regarding the decomposition of noisy signals,
[0076]
[0077] The subscripts s and n represent signal and noise respectively. The process of the serial wireless signal decomposition algorithm in a noisy scenario is as follows:
[0078] R1, initialize the encoding matrix Total power P=0.
[0079] R2, dictionary matrix D of the signal part s is the dictionary matrix D in the noise-free scene. In order to reduce the influence of noise, D is obtained by decomposing the noisy superposition signal in the noise-free scene. n , the converged result is D n Initialization.
[0080] R3. Obtain the values of the encoding matrix A and the dictionary matrix D according to the following steps:
[0081] R301, randomly initialize the encoding matrix A s and A n , and all its elements are non-negative.
[0082] R302, A s and A n The update rules are as follows:
[0083]
[0084]
[0085] Among them l s >0 and l n >0 are the sparsity parameters of signal and noise respectively.
[0086] R303, dictionary matrix D of the signal part s The update rule is as follows
[0087]
[0088] in It is obtained by column standardization of D, which is defined here is a square matrix whose elements are all 1 and of the appropriate size.
[0089] R304, the decomposition result of the pure signal part is: Return to step R302 until convergence.
[0090] R4. Calculation Then update X,
[0091]
[0092] If there are negative elements in the updated X, correct them to 0.
[0093] R5. Update according to the following rules and P
[0094]
[0095]
[0096] Return to step R3 until This means that the estimated total power is ε times higher than the target power, with ε∈(0,1).
[0097] Figure 4 In this embodiment, a frequency domain sampling diagram of eight independent signals is used, including Bluetooth, GPS, FDDLTE, TD-LTE, WiMAX, WLAN (11a), WLAN (11b) and WLAN (11g) to construct a wireless signal spectrum feature dictionary matrix D at the receiving end. s The receiving end receives signals from five unknown transmitting sources and performs serial separation on the received aliased signals. In a noise-free scenario, the spectrum diagram of the target signal and the estimated signal in different decomposition algorithms is shown in the figure below. Figure 5 As shown, Figure 5 (a) is a schematic diagram of the spectrum in the NNSC algorithm; Figure 5 (b) is a schematic diagram of the spectrum in the SWSD algorithm. In a noise-free scenario, the energy ratios of the target signal components and the estimated signal components obtained using different algorithms are shown in the figure below: Figure 6 As shown, Figure 6 (a) is a schematic diagram of the energy ratio of each component of the target signal; Figure 6 (b) is a schematic diagram of the energy ratio of each component of the estimated signal under the NNSC algorithm; Figure 6 (c) is a schematic diagram of the energy ratio of each component of the estimated signal under the SWSD algorithm; Figure 6 It can be seen that the signal energy ratio separated by the serial separation method of wireless signals proposed in this patent, namely the SWSD algorithm, is closest to the target signal and has the best performance.
[0098] In a noisy scenario (SNR = 20dB), the energy ratios of the target signal components and the estimated signal components obtained using different algorithms are shown in the following figure. Figure 7 As shown. Among them, Figure 7 (a) is a schematic diagram of the energy ratio of each component of the target signal; Figure 7 (b) is a schematic diagram of the energy ratio of each component of the estimated signal under the NNSC algorithm; Figure 7 (c) is a schematic diagram of the energy ratio of each component of the estimated signal under the SWSD algorithm. By comparison, it can be seen that the signal energy ratio separated by the serial separation method of wireless signals proposed in this patent, namely the SWSD algorithm, is closest to the target signal and has the best performance.
[0099] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.
Claims
1. A serial separation method for aliased unknown wireless signals, characterized in that: The serial separation method comprises the following steps: S1. Initialize the sparse coding matrix A of the superimposed signal at the receiving end, where all elements in the sparse coding matrix A are non-negative values, and initialize the total power P of the separated wireless signal to 0; S2. The power spectrum density of the M wireless signals in the frequency domain is used as the spectrum feature dictionary matrix D of the M wireless signals. s and noise characteristic matrix D n ; S3, the receiving end receives an aliased unknown signal X emitted by multiple unknown transmitting sources, and obtains a sparse coding matrix A of the aliased unknown wireless signal and a spectrum feature dictionary matrix D of the aliased unknown wireless signal through iterative updating; S4. The receiving end separates the signal components contained in the aliased signal one by one in series until the total power constraint is met.
2. The serial separation method for aliased unknown wireless signals according to claim 1, characterized in that: The process of step S2 is as follows: S201, initialize M independent wireless signals S=[S1, ..., S m , ..., S M ],in is the power spectral density of the mth signal in the frequency domain, W is the number of samples of the signal spectrum, each column of S contains the sampling points obtained within the bandwidth B of the signal receiving end, and each signal S is obtained m The power spectral density of S202, training each signal S m Get the corresponding dictionary matrix vector D s_m And construct M wireless signal spectrum feature dictionary matrices D s , that is, the signal basis matrix, D s The mth column of is the power spectral density of the mth signal and And according to the formula Get the unitized basis matrix Record D s Initialization ||·|| represents the L2 norm of the vector, Represents the field of real numbers.
3. The serial separation method for aliased unknown wireless signals according to claim 1, characterized in that: In step S3, the unknown signal is superimposed on the receiving end without noise. The processing process is as follows: S30A1, assuming that the wireless signal spectrum feature dictionary matrix D in a noise-free scenario is D s , the unit basis matrix of D According to the formula Update the encoding matrix A of the aliased unknown signal, where ⊙ and Respectively represent the multiplication and division of each element in the matrix, ← represents the calculation result of the formula on the right side of the arrow as the iteration result of the variable on the left side of the arrow, the sparsity parameter λ>0, [·] T Indicates transposing the matrix in brackets; S30A2, according to the formula Update the wireless signal spectrum feature dictionary matrix D, where is obtained by column standardization of D, is a square matrix with all elements set to 1 and of appropriate size; S30A3, according to A and The iterative result of the noise-free superposition unknown signal is Return to steps S30A1 and S30A2 to perform cyclic updates until the coding matrix of step S30A1 converges to obtain the result of the coding matrix.
4. The serial separation method for aliased unknown wireless signals according to claim 3, characterized in that: In step S3, for the unknown signal containing noise superimposed The processing process is as follows: S30B1, sparse coding matrix of superimposed signal at receiving end Among them A s and A n Initialize the wireless signal and noise sparse coding matrices under noisy conditions respectively, and their elements are all non-negative; S30B2, the spectrum feature dictionary of wireless signals under noisy conditions is expanded to D = [D s , D n ], D n is the initialization noise characteristic matrix obtained by the noise power spectrum density within bandwidth B received by the receiver in the no-signal scenario, D s is the dictionary matrix of M wireless signal spectrum features under noisy conditions, then the unitized basis matrix of D in Yes D s The unit matrix of Yes D n The unit basis matrix of ; S30B3, update A according to the following update rules s and A n : Among them l s >0 and l n >0 are the sparsity parameters of signal and noise respectively; S30B4, update the wireless signal spectrum feature dictionary matrix D according to the following update rules s as follows in is the matrix obtained by column normalization of D, is a square matrix with all elements set to 1 and of appropriate size; S30B4, the decomposition result of the receiving end signal part is Return to steps S30B2 and S30B3 to perform cyclic updates until the coding matrix of step S30B2 converges to obtain the result of the coding matrix.
5. The serial separation method for aliased unknown wireless signals according to claim 4, characterized in that: The process of step S4 is as follows: S401, the receiving end performs serial separation on the aliased signals one by one, and calculates the sequence number m0 corresponding to the signal with the maximum sparse coefficient, which is Among them A m represents the sparse coefficient of the mth column of A, i.e. the mth signal, and then updates the received signal X at the receiving end. in, is the sparse coefficient corresponding to the m0th signal, is the power spectrum density of the m0th wireless signal in the spectrum feature dictionary matrix, is the power spectrum density of the m0th wireless signal in the basis matrix of the spectrum feature dictionary matrix; if there are negative elements in the updated matrix X, correct them to 0; S402, Definition and P are the sparse coefficients and the separated signal power after serial separation in step S401, respectively. and P, Until P>ε∑(X), ε is the power fitting coefficient, ε∈(0,1).