A method for spatial electromagnetic spectrum identification based on parameterized quantum circuits

Abstract

The application discloses a space electromagnetic spectrum identification method based on a parameterized quantum circuit, space electromagnetic spectrum data collection, electromagnetic spectrum time-frequency image sample set construction based on a discrete short-time Fourier transform, parameter quantumization circuit design for the space electromagnetic spectrum time-frequency image, and electromagnetic spectrum data processing based on the parameter quantumization circuit.The space electromagnetic spectrum identification method based on the parameterized quantum circuit has stronger feature extraction capability compared with other neural network models under space conditions, and can quickly identify electromagnetic spectrum under a complex electromagnetic environment.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic spectrum identification technology, specifically to a spatial electromagnetic spectrum identification method based on parameterized quantum circuits. Background Technology

[0002] Situational awareness is a key development direction for countries in the space domain, and electromagnetic situational awareness is a crucial component. Whether it's detecting the space electromagnetic environment or conducting radio communication between satellites, both rely on electromagnetic spectrum identification. Therefore, accurately identifying whether electromagnetic signals captured in complex electromagnetic environments are indeed satellite signals is paramount. However, in the space environment, electromagnetic spectrum identification depends solely on the satellite itself; ground personnel cannot promptly obtain information about the electromagnetic spectrum in space and issue commands to the satellite. Thus, there is an urgent need for mission satellites equipped with modules to identify the characteristics of the space electromagnetic spectrum, enabling them to distinguish the electromagnetic spectrum features of satellites in complex electromagnetic environments and execute subsequent missions. Summary of the Invention

[0003] This invention provides a space electromagnetic spectrum identification method based on parameterized quantum circuits, which is used to accurately identify the electromagnetic spectrum characteristics of space satellites from a complex electromagnetic spectrum environment under space conditions, thereby gaining an advantage in electromagnetic situational awareness.

[0004] To achieve the above objectives, the present invention adopts the following solution:

[0005] Step 1: Acquisition of spatial electromagnetic spectrum data;

[0006] Step 2: Constructing an electromagnetic spectrum time-frequency image sample set based on discrete short-time Fourier transform;

[0007] Step 3, Parametric Quantization Circuit Design for Space Electromagnetic Spectrum Time-Frequency Diagram;

[0008] Step 4: Electromagnetic spectrum data processing based on parameter quantization circuits.

[0009] Furthermore, the specific steps in step 2 to generate the electromagnetic spectrum time-frequency image dataset based on discrete short-time Fourier transform are as follows:

[0010] The electromagnetic spectrum is converted into a time-frequency image sample set through discrete short-time Fourier transform, i.e.:

[0011]

[0012] Where s(k) is the discrete signal to be identified, g(t) is a short-time window function, and * denotes conjugate;

[0013] This sample set is used as a time-frequency image dataset of the spatial electromagnetic spectrum, and it is divided into a training set and a test set in a certain proportion to train and test a quantum convolutional neural network for recognizing the spatial electromagnetic spectrum.

[0014] Furthermore, the algorithm flow for constructing the parameter quantized circuit for identifying the spatial electromagnetic spectrum in step 3 is as follows:

[0015] S1 converts the information of time-frequency image samples into quantum states that can be recognized by a quantum neural network through quantum encoding, and uses it as the input layer of the quantum convolutional neural network;

[0016] The S2 quantum convolution kernel uses a 2×2 window. After input information is processed by the Hadamard gate feature mapping, the four qubits are mapped to the quantum state |ψ1>.

[0017] S3|ψ1> evolves through the quantized circuit U(θ) into |ψ2>=U(θ)|ψ1>, and measurements are then performed on its qubits. Take the expected value

[0018] S4 moves the convolution window, scans the entire time-frequency image sample, and obtains an eigenvalue matrix composed of expected values.

[0019] Furthermore, in step 3, quantum encoding refers to the process of converting the input information of a quantum neural network into a quantum state. The mapping performed by quantum computing is (x1, x2, ..., x...). m ,0)→(x1',x2',...,x m ',y), using |ψ> 1,...,m This represents a quantum state encoded with classical data. Quantum state |ψ> 1,...,m Let be a superposition of 2m computational basis vectors in an m-dimensional Hilbert space, where If the basis vectors are {|0,0,…,0>,...,|1,1,…,1>}, then in

[0020] Furthermore, in step 3, the quantum state is the state of a qubit, that is, the physical state of the quantum system described by a two-dimensional vector. The quantum state is represented by |ψ>, which can change over time and is usually normalized in quantum computing.

[0021] Furthermore, in step 3, the quantized circuit U(θ) is a learnable parameterized quantum circuit that essentially performs a unitary transformation. The parameterized quantum circuit can be represented as a Pauli matrix {σ...} x ,σ y ,σ z A linear combination of the tensor product of} and the identity matrix I, if there are k layers in the network, and Uk (Θ) represents the unitary matrix of the k-th layer network, then the entire quantum circuit can be represented as:

[0022] U(Θ)=U k (Θ k )U k-1 (Θ k )...U 1 (Θ1).

[0023] Furthermore, in step 3, the measurement on the qubit transforms the quantum state output by the quantum circuit into information that a computer can recognize, which is a crucial step in the quantum neural network. In quantum computing, projective measurements are generally used, described by the observable Λ. The spectral decomposition form of the observable Λ is:

[0024]

[0025] Where, λ i For the characteristic value of the observable Λ, P i For the eigenvalue λ i The projection onto the characteristic space. The probability of obtaining result i after measuring the quantum state |ψ> is:

[0026] p i =p(λ=λi)=<ψ|P i |ψ>

[0027] After measurement, the quantum state becomes:

[0028]

[0029] Furthermore, the specific steps of step 4, electromagnetic spectrum data processing based on parameter quantization circuits, are as follows:

[0030] The electromagnetic spectrum data processed by the parameterized quantum circuit forms a feature matrix; the feature matrix is ​​input into the pooling layer, and the pooling method is max-pooling; the pooled electromagnetic spectrum data enters a fully connected layer, and the loss is calculated using a loss function. The loss is fed back to the parameterized quantum circuit and the fully connected layer through the optimizer to update the parameters, and the output layer uses the Softmax activation function.

[0031] After training the quantum convolutional neural network with a time-spectrum image training set generated by the short-time Fourier transform of the spatial electromagnetic spectrum, the training dataset is compared and analyzed with the test dataset to identify the spatial electromagnetic spectrum.

[0032] The present invention has the following advantages:

[0033] Due to the superposition and entanglement properties of quantum states, quantum neural networks can perform parallel computations, resulting in a faster computation speed compared to traditional neural networks. Furthermore, n qubits can store 2^n qubits. n Bit information has a larger information storage space.

[0034] Due to the structural advantages of its quantum neural network, the electromagnetic spectrum identification method occupies less memory compared to traditional neural networks, making it better suited to the limited hardware conditions caused by complex spatial environments. Attached Figure Description

[0035] Figure 1 A flowchart illustrating the workflow of a spatial electromagnetic spectrum identification method based on parameterized quantum circuits;

[0036] Figure 2 Here is a flowchart of the quantum convolutional neural network algorithm;

[0037] Figure 3 This is a diagram of a quantum convolutional neural network structure. Detailed Implementation

[0038] Specific steps of a spatial electromagnetic spectrum identification method based on quantum neurons:

[0039] Step 1: Space Electromagnetic Spectrum Data Acquisition

[0040] Raw electromagnetic spectrum data is collected by the ground terminal of a space satellite and sampled to obtain raw samples of the space electromagnetic spectrum.

[0041] Step 2: Construction of an electromagnetic spectrum time-frequency image sample set based on discrete short-time Fourier transform

[0042] The electromagnetic spectrum is converted into a time-frequency image sample set through discrete short-time Fourier transform, i.e.:

[0043]

[0044] Where s(k) is the discrete signal to be identified, g(t) is a short-time window function, and * denotes conjugate;

[0045] This sample set is used as a time-frequency image dataset of the spatial electromagnetic spectrum, and it is divided into a training set and a test set in a certain proportion to train and test a quantum convolutional neural network for recognizing the spatial electromagnetic spectrum.

[0046] Step 3: Parametric Quantization Circuit Design for Electromagnetic Spectrum Time-Frequency Diagram

[0047] Due to the superposition and entanglement properties of quantum states, quantum neural networks can perform parallel computations, resulting in a faster computation speed compared to traditional neural networks. Furthermore, n qubits can store 2^n qubits.n Bit information has a larger information storage space;

[0048] The quantum convolutional neural network used is primarily a parameterized quantum circuit.

[0049] The parameterized quantum circuit algorithm flow is as follows:

[0050] S1 encodes the information of the time-frequency image samples of the electromagnetic spectrum into quantum states that can be recognized by the parameterized quantum circuit, and uses them as the input layer of the quantum convolutional neural network.

[0051] The S2 quantum convolution kernel uses a 2×2 window. After input information is processed by the Hadamard gate feature mapping, the four qubits are mapped to the quantum state |ψ1>.

[0052] S3|ψ1> evolves through the quantized circuit U(θ) into |ψ2>=U(θ)|ψ1>, and measurements are then performed on its qubits. Take the expected value

[0053] S4 moves the convolution window and scans the time-frequency image samples of the entire spatial electromagnetic spectrum to obtain the eigenvalue matrix composed of the expected values.

[0054] Step 4: Electromagnetic spectrum data processing based on parametric quantization circuits

[0055] The electromagnetic spectrum data processed by the parameterized quantum circuit forms a feature matrix. This feature matrix is ​​then input into a pooling layer, where pooling can be represented by the following formula:

[0056]

[0057] in, It is the j-th output feature image of the l-th layer of the network after pooling operation. It is the input image of the j-th pooling layer in the l-th layer of the network. It is a pooling operation method, and the pooling method used is max-pooling;

[0058] The pooled electromagnetic spectrum data enters a fully connected layer, where the loss is calculated using a loss function. This loss is then fed back to the parameterized quantum circuit and the fully connected layer via an optimizer to update the parameters. The output layer uses the Softmax activation function, which can be expressed by the following formula:

[0059]

[0060] Where x is the input data;

[0061] The quantum convolutional neural network was trained using a time-spectrum image training set generated by short-time Fourier transform of the spatial electromagnetic spectrum. The training dataset was then compared and analyzed with the test dataset to obtain the recognition probability of the spatial electromagnetic spectrum.

[0062] The trained quantum convolutional neural network ultimately achieves the goal of identifying the electromagnetic spectrum in space.

Claims

1. A method for identifying spatial electromagnetic spectrum based on parameterized quantum circuit, characterized in that: (1) spatial electromagnetic spectrum data acquisition Collecting electromagnetic spectrum raw data through satellite ground terminal, and sampling the raw data to obtain original samples of spatial electromagnetic spectrum; (2) construction of electromagnetic spectrum time-frequency image sample set based on discrete short-time Fourier transform Converting spatial electromagnetic spectrum into time-frequency image sample set through discrete short-time Fourier transform, that is: taking the sample set as the time-frequency image data set of spatial electromagnetic spectrum, and dividing it into training set and test set at a certain ratio to train and test the quantum convolutional neural network for identifying spatial electromagnetic spectrum; (3) parameterized quantum circuit design for electromagnetic spectrum time-frequency image Using quantum convolutional neural network as the main body of the parameterized quantum circuit; the parameterized quantum circuit algorithm process is: S1 converting the information of the time-frequency image sample of spatial electromagnetic spectrum into quantum state that can be recognized by the parameterized quantum circuit through quantum encoding, as the input layer of the quantum convolutional neural network; S4 moving the convolution window to scan the time-frequency image sample of spatial electromagnetic spectrum to obtain the characteristic value matrix composed of expected values; (4) electromagnetic spectrum data processing based on parameterized quantum circuit The electromagnetic spectrum data processed by the parameterized quantum circuit forms a feature matrix, and the feature matrix is input into the pooling layer, and the pooling can be represented by the following formula: The electromagnetic spectrum data after pooling enters a fully connected layer, then a loss is calculated by a loss function, the parameters of the parameterized quantum circuit and the fully connected layer are updated through an optimizer, and the output layer uses a Softmax activation function, which can be expressed by the following formula: Wherein, x is the input data; the time-frequency image training set generated by the short-time Fourier transform of the spatial electromagnetic spectrum is used to train the quantum convolutional neural network, the output results of the training data set and the test data set are compared and analyzed, and the identification probability of the spatial electromagnetic spectrum is obtained; the trained quantum convolutional neural network finally achieves the purpose of identifying spatial electromagnetic spectrum. The quantum encoding is to convert the input information into quantum state, and the mapping performed by quantum computing is: After measurement, the quantum state becomes: ​ ​ ​ ； wherein is a discrete signal to be identified, is a window function of short time length, denotes the conjugate; ​ ​ ​ ​ ​ The S2 quantum convolution kernel adopts a window with a size of 2*2, and after inputting information, the four quantum bits are mapped to a quantum state through Hadamard gate feature mapping ; S3 Through the quantization circuit Evolution, into And make a measurement on the qubit , Take the expected value ; ​ ​ ​ wherein, is the jth output feature map of the ith layer in the network, is the jth output feature map of the ith layer in the network, is the jth input feature map of the ith pooling layer in the network, is the jth input feature map of the ith pooling layer in the network, is the pooling operation, and the used pooling operation is max-pooling. ​ ； ​ ​ ​ 2. A parametric quantum circuit based space electromagnetic spectrum identification method as claimed in claim 1, wherein: ​ ； with denotes a quantum state encoding classical data, the quantum state is a superposition of 2 m computational basis vectors in an m-dimensional Hilbert space, where , the basis vectors are then where .

3. A parametric quantum circuit based space electromagnetic spectrum identification method as claimed in claim 1, wherein: The quantum state is the state of a qubit, i.e. the physical state of a quantum system described by a two-dimensional vector, the quantum state is denoted by which can vary over time and is usually normalized in quantum computing.

4. A parametric quantum circuit based space electromagnetic spectrum identification method as claimed in claim 1, wherein: The quantized circuit A parameterized quantum circuit that is learnable, which essentially performs a unitary transformation, can be represented as a linear combination of tensor products of Pauli matrices and the identity matrix I, if there are t layers of the network, and where U t represents the unitary matrix represented by the tth layer of the network, then the entire quantum circuit can be represented as: 。 5. A parametric quantum circuit based space electromagnetic spectrum identification method as claimed in claim 1, wherein: The measurement on the quantum bit is the process of changing the quantum state output by the quantum circuit into information that can be recognized by the computer, and is a key step of the quantum neural network. In quantum computing, a projection measurement is generally used, and the observable The description, the observable The spectral decomposition form is: ； where is an observable whose eigenvalues, are the eigenvalues of the projection onto the eigenspace of the eigenvalue of the observable , the probability of obtaining the result i after measuring the observable ； ​ 。

Images