Search method and apparatus for synchronization signal, storage medium, and electronic device

Abstract

The application discloses a synchronization signal searching method and device, a storage medium and an electronic device, and relates to the field of communication. The method comprises the following steps: acquiring a signal set to be searched for a primary synchronization signal; generating target energy information of the signal set, wherein the target energy information is used for indicating the maximum energy of the signals in different frequency bands in the signal set in different time periods; detecting target synchronization information of the signal set by using a signal searching model and the target energy information, wherein the target synchronization information is used for indicating the signal position of the primary synchronization signal in the signal set, the signal searching model is obtained by training an initial searching model by using an energy information sample marked with the signal position of the primary synchronization signal; and searching for the primary synchronization signal in the signal set according to the target synchronization information. By using the technical scheme, the problem of low searching efficiency of the primary synchronization signal is solved, and the technical effect of improving the searching efficiency of the primary synchronization signal is achieved.

Description

Technical Field

[0001] This application relates to the field of communications, and more specifically, to a method and apparatus for searching synchronization signals, a storage medium, and an electronic device. Background Technology

[0002] The Primary Synchronization Signal (PSS) is a crucial component of the physical layer synchronization process, used to help user equipment (UE, such as a mobile phone) achieve initial synchronization with the base station in time and frequency. However, in high-bandwidth communication systems, especially when the observation window is limited, searching for the PSS within the time and frequency range can consume a significant amount of time, making the localization process very slow.

[0003] No effective solution has yet been proposed to address the low search efficiency of the master synchronization signal in related technologies. Summary of the Invention

[0004] This application provides a method and apparatus for searching synchronization signals, a storage medium, and an electronic device to at least solve the problem of low search efficiency of main synchronization signals in related technologies.

[0005] According to one embodiment of this application, a method for searching a synchronization signal is provided, comprising: acquiring a signal set of a primary synchronization signal to be searched; generating target energy information of the signal set, wherein the target energy information is used to indicate the maximum energy of signals in different frequency bands in the signal set at different time periods; detecting target synchronization information of the signal set through a signal search model and the target energy information, wherein the target synchronization information is used to indicate the signal position of the primary synchronization signal in the signal set, and the signal search model is obtained by training an initial search model using energy information samples labeled with the signal position of the primary synchronization signal; and searching for the primary synchronization signal in the signal set according to the target synchronization information.

[0006] Optionally, generating target energy information for the signal set includes: dividing the signal set into multiple signal subsets in the time domain according to time order; determining the frequency domain responses of multiple frequency components of the multiple signal subsets to obtain multiple target frequency domain responses, wherein the multiple target frequency domain responses correspond one-to-one with the multiple signal subsets and one-to-one with the multiple frequency components; detecting the maximum value of the square of the absolute value of each target frequency domain response to obtain multiple target elements, wherein the target elements are used to indicate the maximum energy of the signal in the target frequency band in the signal set during the target time period, the target frequency band is the frequency band of the frequency component corresponding to the target element, and the target time period is the time period corresponding to the target element; inserting the multiple target elements into the initial energy matrix according to the target format to obtain the target energy matrix, wherein the target energy information includes the target energy matrix.

[0007] Optionally, the frequency domain responses of multiple frequency components of multiple signal subsets are determined to obtain multiple target frequency domain responses, including obtaining the target frequency domain response of the k-th frequency component of the m-th signal subset using the following formula:

[0008] ;

[0009] in, The target frequency domain response of the k-th frequency component of the m-th signal subset. The target transform parameters are used to indicate the number of multiple frequency components. L represents the time length of each signal subset. For the m-th signal subset, For window functions.

[0010] Optionally, detecting target synchronization information of a signal set using a signal search model and target energy information includes: generating a feature matrix that meets the input requirements of a second sub-model using a first sub-model and a target energy matrix, wherein multiple target elements in the target energy matrix are used to indicate the maximum energy of signals in different frequency bands in the signal set at different times, and the target energy information includes the target energy matrix; concatenating an initial classification vector in the first row of the feature matrix to obtain an initial classification matrix; inputting the initial classification matrix into the second sub-model, wherein the second sub-model includes a multi-layer self-attention transform encoder; receiving the target classification matrix output by the second sub-model; extracting the first row vector from the target classification matrix to obtain a target classification vector; inputting the target classification vector into a third sub-model and a fourth sub-model, wherein the signal search model includes the first sub-model, the second sub-model, the third sub-model, and the fourth sub-model; receiving the first sub-information output by the third sub-model and the second sub-information output by the fourth sub-model, wherein the first sub-information is used to indicate whether a primary synchronization signal exists in the signal set, and the second sub-information is used to indicate the time position and frequency position of the primary synchronization signal in the signal set, and the target synchronization information includes the first sub-information and the second sub-information.

[0011] Optionally, generating an initial classification matrix that meets the input requirements of the second sub-model using the first sub-model and the target energy matrix includes: converting the target energy matrix into an information matrix of target dimension using the first sub-model, wherein multiple row vectors in the information matrix correspond one-to-one with multiple target elements; generating position codes for the multiple row vectors based on the frequency index and time index of the multiple target elements corresponding to the multiple row vectors, thereby obtaining a position matrix, wherein the time index is used to indicate the time period corresponding to the target element, and the frequency index is used to indicate the frequency band corresponding to the target element; and adding the information matrix and the position matrix to obtain a feature matrix.

[0012] Optionally, the target energy matrix is ​​converted into an information matrix of the target dimension through the first sub-model, including: performing a convolution operation on the target energy matrix with a one-dimensional convolution kernel along the frequency direction to obtain a local enhancement tensor; expanding the local enhancement tensor into a one-dimensional sequence to obtain an intermediate sequence; and mapping each vector in the intermediate sequence to a vector of the reference dimension through a projection matrix to obtain an information matrix, wherein the column dimension in the target dimension is equal to the reference dimension.

[0013] Optionally, a position matrix is ​​obtained by generating position codes for multiple row vectors based on the frequency index and time index of multiple target elements corresponding to multiple row vectors, including: generating time dimension codes for multiple row vectors based on the time index to obtain a time dimension matrix, and generating frequency dimension codes for multiple row vectors based on the frequency index to obtain a frequency dimension matrix; and concatenating the time dimension matrix and the frequency dimension matrix to obtain the position matrix.

[0014] According to another embodiment of this application, a synchronization signal search apparatus is also provided, comprising: an acquisition module for acquiring a signal set of a primary synchronization signal to be searched; a generation module for generating target energy information of the signal set, wherein the target energy information is used to indicate the maximum energy of signals in different frequency bands in the signal set at different time periods; a detection module for detecting target synchronization information of the signal set through a signal search model and the target energy information, wherein the target synchronization information is used to indicate the signal position of the primary synchronization signal in the signal set, and the signal search model is obtained by training an initial search model using energy information samples labeled with the signal position of the primary synchronization signal; and a search module for searching for the primary synchronization signal in the signal set according to the target synchronization information.

[0015] According to another aspect of the embodiments of this application, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer program, and the computer program is configured to execute the above-described method for searching for synchronization signals when it is run.

[0016] According to another aspect of the embodiments of this application, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the above-mentioned method for searching for synchronization signals through the computer program.

[0017] In this embodiment, after obtaining the signal set of the primary synchronization signal to be searched, target energy information is generated to indicate the maximum energy of signals in different frequency bands within the signal set at different time periods. The target synchronization information, indicating the signal position of the primary synchronization signal within the signal set, is detected using a signal search model and the target energy information. This achieves the determination of the signal position of the primary synchronization signal within the signal set based on the energy distribution characteristics exhibited by the signal set. Then, by combining the target synchronization information—that is, the determined signal position of the primary synchronization signal within the signal set—the search for the primary synchronization signal is performed within the signal set. Compared to a blind search across the entire time domain, this reduces the search time for the primary synchronization signal. Therefore, the above technical solution solves the problem of low search efficiency for primary synchronization signals in related technologies, achieving the technical effect of improving the search efficiency of primary synchronization signals. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the hardware environment for a method of searching for a synchronization signal according to an embodiment of this application;

[0021] Figure 2 This is a flowchart of a method for searching a synchronization signal according to an embodiment of this application;

[0022] Figure 3 This is a hardware system architecture diagram of a synchronization block undersampling search method according to an embodiment of this application;

[0023] Figure 4 This is a schematic diagram of a synchronization signal search method according to an embodiment of this application;

[0024] Figure 5 This is a structural block diagram of a synchronization signal search device according to an embodiment of this application. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] The methods and embodiments provided in this application can be executed on a computer terminal, device terminal, or similar computing device. Taking running on a computer terminal as an example, Figure 1 This is a schematic diagram of the hardware environment for a synchronization signal search method according to an embodiment of this application. For example... Figure 1 As shown, a computer terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 (which may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.) and a memory 104 for storing data are also shown. In one exemplary embodiment, the computer terminal may further include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the computer terminal described above. For example, the computer terminal may also include components that are more complex than those described above. Figure 1 The more or fewer components shown, or having the same Figure 1 Equivalent functions or ratios shown Figure 1 The functions shown have more different configurations.

[0028] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the synchronization signal search method in this embodiment of the invention. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, thereby implementing the above-described method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to a computer terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0029] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by a communication provider for the computer terminal. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0030] This embodiment provides a method for searching for synchronization signals, applied to the aforementioned computer terminal. Figure 2 This is a flowchart of a synchronization signal search method according to an embodiment of this application, such as... Figure 2 As shown, the method may include, but is not limited to, the following steps:

[0031] Step S202: Obtain the set of signals to be searched for the master synchronization signal;

[0032] Step S204: Generate target energy information for the signal set, wherein the target energy information is used to indicate the maximum energy of signals in different frequency bands in the signal set at different time periods;

[0033] Step S206: Detect the target synchronization information of the signal set through the signal search model and target energy information. The target synchronization information is used to indicate the signal position of the main synchronization signal in the signal set. The signal search model is obtained by training the initial search model using energy information samples labeled with the signal position of the main synchronization signal.

[0034] Step S208: Search for the primary synchronization signal in the signal set based on the target synchronization information.

[0035] After obtaining the signal set of the primary synchronization signal to be searched through the above steps, target energy information is generated to indicate the maximum energy of signals in different frequency bands within the signal set at different times. This target synchronization information, indicating the signal position of the primary synchronization signal within the signal set, is then detected using a signal search model and the target energy information. This allows for the determination of the primary synchronization signal's position within the signal set based on its energy distribution characteristics. Combining this with the target synchronization information—that is, the determined signal position—the search for the primary synchronization signal within the signal set is performed. Compared to a blind search across the entire time domain, this reduces the search time for the primary synchronization signal. Therefore, the above technical solution solves the problem of low search efficiency for primary synchronization signals in related technologies, achieving the technical effect of improving the search efficiency of primary synchronization signals.

[0036] Optionally, in this embodiment, this application can be applied, but is not limited to, to the initial access process of the terminal side (including various terminal devices or terminal function modules in base station equipment, etc.) of various wireless communication systems (such as the initial access process of the terminal side of high-bandwidth, high-frequency wireless communication systems such as 5G NR or 4G LTE), especially in scenarios such as fast cell search during user equipment (UE) power-on, cell reselection, or handover. In these scenarios, the signal receiver needs to lock the primary synchronization signal from the radio frequency signal within a very short time (such as a few milliseconds) to complete time synchronization and frequency synchronization.

[0037] Figure 3 This is a hardware system architecture diagram of a synchronization block undersampling search method according to an embodiment of this application. For example... Figure 3 As shown, the signal set can be downsampled after digital-to-analog converter, and FFT and sampling control can be performed on the signal set to obtain model input. Then, the possible location of the main synchronization signal can be determined based on model inference, and then fine scanning can be performed.

[0038] In the embodiment provided in step S202, the signal set may, but is not limited to, refer to a discrete digital sampled data sequence with clear time boundaries and length limitations that is divided from the continuously received time-domain data stream by the receiving end before the detection of the main synchronization signal. The signal set may, but is not limited to, consist of in-phase (I) and quadrature (Q) components in complex form, representing the baseband waveform data obtained after down-conversion, filtering and analog-to-digital conversion of the original radio frequency signal within a specific observation window. This set may, but is not limited to, include time-domain segments that may exist in the main synchronization signal to be detected, as well as mixed information of background noise, interference signals and other channel components, and is the original input data source for subsequent time-frequency analysis and pattern recognition.

[0039] Optionally, in this embodiment, the signal set for obtaining the master synchronization signal to be searched can be achieved, but is not limited to, through various hardware and software collaborative methods. For example, after the radio frequency front-end receives the analog signal, it is digitized by the analog-to-digital converter, and the baseband processor copies the continuous sampled data blocks to the memory buffer through the high-speed cache to form a signal set. Alternatively, a sliding window mechanism can be used to extract multiple continuous time-domain sampling sequences from the circular buffer at fixed or overlapping step sizes, or for a specific bandwidth configuration, the baseband signal block can be directly extracted and output through the digital downconverter, thereby flexibly adapting to the resource constraints and real-time requirements of different processing units.

[0040] In the embodiment provided in step S204, after obtaining the signal set of the master synchronization signal to be searched, the characteristics of the energy distribution of the obtained signal set can be detected, but are not limited to, generating the target energy information of the signal set.

[0041] Optionally, in this embodiment, the target energy information may be, but is not limited to, a preprocessed two-dimensional time-frequency matrix, whose row dimension corresponds to the discretized time step (representing different observation periods) and column dimension corresponds to the discretized frequency bin (representing different frequency bands). Each element value in the time-frequency matrix represents the signal energy intensity of the frequency component in the original signal set within the corresponding time period and frequency band. The target energy information transforms the complex complex time-domain waveform into an intuitive energy distribution heatmap, retaining the main energy distribution characteristics of the signal on the time-frequency plane while discarding high-frequency details such as phase that are not necessary for rapid positioning.

[0042] Optionally, in this embodiment, generating target energy information prepares the way for transforming the computationally complex time-domain waveform matching problem into a much less computationally intensive two-dimensional pattern recognition problem. Directly inputting the original high-sampling-rate signal into the deep learning model would result in huge memory consumption and computational latency, failing to meet the millisecond-level fast synchronization requirements. However, by performing FFT mapping on the signal and executing energy accumulation or pooling operations, undersampling dimensionality reduction can be actively performed. This significantly reduces the data dimensionality to adapt to the input requirements of the Transformer model, while utilizing the unique periodic structure and energy concentration characteristics of the PSS signal in the frequency domain to retain key features sufficient to distinguish between signal and noise. Thus, while ensuring detection performance, extremely fast coarse localization can be achieved, providing an efficient time window for subsequent fine-grained searches.

[0043] As an optional implementation, the target energy information of the signal set can be generated in the following manner, but is not limited to: dividing the signal set into multiple signal subsets in the time domain according to time order; determining the frequency domain responses of multiple frequency components of the multiple signal subsets to obtain multiple target frequency domain responses, wherein the multiple target frequency domain responses correspond one-to-one with the multiple signal subsets and one-to-one with the multiple frequency components; detecting the maximum value of the square of the absolute value of each target frequency domain response to obtain multiple target elements, wherein the target elements are used to indicate the maximum energy of the signal in the target frequency band in the signal set during the target time period, the target frequency band is the frequency band of the frequency component corresponding to the target element, and the target time period is the time period corresponding to the target element; constructing a target energy matrix according to the target format and the multiple target elements, wherein the target energy information includes the target energy matrix.

[0044] Optionally, in this embodiment, the continuous signal can be divided into multiple subsets (time slices), each subset can be frequency-domain transformed to obtain a frequency domain response, and key features can be extracted by calculating the maximum value of the square of the absolute value of the frequency domain response (i.e., the energy spectral density) to obtain target elements representing the energy of a specific frequency band in a specific time period; then, a target energy matrix can be constructed based on these multiple target elements according to the specified target format.

[0045] Optionally, in this embodiment, the signal set may be divided into multiple signal subsets in the time domain according to time order, but is not limited to, and the time span / time length corresponding to each signal subset may be, but is not limited to, equal.

[0046] As an optional implementation, the frequency domain responses of multiple frequency components of multiple signal subsets can be determined, but are not limited to, in the following manner, to obtain multiple target frequency domain responses: the target frequency domain response of the k-th frequency component of the m-th signal subset is obtained using the following formula:

[0047] ;

[0048] in, The target frequency domain response of the k-th frequency component of the m-th signal subset. The target transform parameters are used to indicate the number of multiple frequency components. L represents the time length of each signal subset. For the m-th signal subset, For window functions.

[0049] Optionally, in this embodiment, a windowed discrete Fourier transform can be performed on each signal subset to obtain the frequency domain response of multiple frequency components of multiple signal subsets, that is, to obtain multiple target frequency domain responses.

[0050] Optionally, in this embodiment, the k-th frequency component may, but is not limited to, refer to a specific frequency component corresponding to the discrete frequency index k when performing a frequency domain transformation (such as a discrete Fourier transform) on a subset of signals. The k-th frequency component may, but is not limited to, represent a frequency in the original time-domain signal with frequency k. (in, Sampling rate, The sine or cosine oscillation component (i.e., the target transformation parameter mentioned above) is the number of transformation points. This frequency component carries the amplitude and phase information of the signal at a specific frequency in the frequency domain response. It is the basic unit that constitutes the signal spectrum structure. By extracting the energy of this frequency component (such as the square of the absolute value), the intensity of the signal in the corresponding frequency band can be quantified, thereby locating the energy distribution of the signal in the two-dimensional time-frequency energy matrix.

[0051] Optionally, in this embodiment, the target transform parameter may refer to, but is not limited to, the number of transform points or the discretization granularity of the frequency domain resolution set during the discrete Fourier transform process. The target transform parameter may determine, but is not limited to, how many equally spaced frequency bins (i.e., the number of the k-th frequency components) the signal spectrum is divided into. Optionally, the target transform parameter may define, but is not limited to, the index range of the frequency domain response, and indirectly determine the frequency resolution. This refers to the actual bandwidth represented by each frequency component. By setting the target transformation parameters, the number of columns (frequency dimension) of the output target energy matrix can be controlled while ensuring coverage of the frequency band required by the main synchronization signal, thereby balancing spectral resolution and computational complexity in the feature extraction stage.

[0052] Optionally, in this embodiment, L can be, but is not limited to, the hop size.

[0053] Optionally, in this embodiment, m can be, but is not limited to, a time step index, and k can be a frequency Bin index.

[0054] Optionally, in this embodiment, using a windowed Discrete Fourier Transform can suppress spectral leakage. When a finite-length time-domain signal subset is truncated, the discontinuity at the signal edges will generate high-frequency oscillations and energy diffusion in the frequency domain, leading to interference between adjacent frequency components. By applying window functions (such as Hanning windows, Hamming windows, etc.), the amplitude attenuation of the signal subset at both ends of the time domain can be smoothed, significantly reducing the sidelobe level, thereby ensuring that the extracted maximum energy more accurately reflects the true spectral characteristics of the signal, avoiding pseudo-peak interference, and improving the accuracy of subsequent deep learning models in identifying the location of the master synchronization signal.

[0055] Optionally, in this embodiment, after obtaining multiple target frequency domain responses, the maximum value of the square of the absolute value of each target frequency response can be detected, that is, the maximum energy of the signal in the k-th frequency band during the m-th time period can be detected to obtain each target element.

[0056] Optionally, in this embodiment, the target energy matrix can be constructed according to, but is not limited to, the target format and target elements. The target format can be, but is not limited to, a two-dimensional matrix or a two-dimensional tensor, where the row dimension corresponds to the discretized frequency Bin index, the column dimension corresponds to the discretized time step index, and each element value in the matrix directly stores the maximum energy value of the corresponding specific frequency band in a specific time period.

[0057] Optionally, in this embodiment, the target energy matrix can be constructed using all target elements, but is not limited to; or the size of the target energy matrix can be further reduced: multiple target elements corresponding to the same frequency band can be divided into several element sets, and the largest target element can be selected as a reference element. This operation can be performed on all frequency bands to obtain multiple reference elements, which are then used to construct the target energy matrix using the reference elements and a specified format. The number of target elements included in each element set can be determined based on the model's processing capability and the accuracy requirements for searching the master synchronization signal. When the model's processing capability is poor, the number of target elements included in each element set can be increased; when the accuracy requirements for searching the master synchronization signal are high, the number of target elements included in each element set can be decreased.

[0058] Figure 4 This is a schematic diagram of a synchronization signal search method according to an embodiment of this application. Figure 4 As shown, it is possible, but not limited to, after receiving the original input signal (i.e., the aforementioned signal set), to use... The Fast Fourier Transform (FFT) of a point maps the signal to the frequency domain. For the m-th short-time observation window, its frequency domain response is expressed as:

[0059] ;

[0060] in, Frequency Bin Index, Frequency Resolution m is the time step index, L is the step size, and w[n] is the window function.

[0061] Furthermore, the target energy matrix S can be obtained. , , where g is greater than or equal to 1 and g is an integer, thus obtaining an undersampled two-dimensional feature tensor S.

[0062] Through the above methods, we achieved dimensionality reduction and noise suppression of signal features while preserving the key energy distribution characteristics of the PSS signal. By calculating the maximum energy value of the frequency domain response within each time-frequency unit, we can effectively suppress random fluctuations in background Gaussian noise and highlight the significant energy peak characteristics of the PSS signal in the time-frequency plane. This significantly reduces the data dimensionality (undersampling) to reduce the computational complexity of subsequent Transformer models, while ensuring that the feature matrix of the input model still contains high signal-to-noise ratio information sufficient to distinguish between signal and noise. This lays a high-quality input foundation for subsequent deep learning models to quickly and accurately identify PSS locations.

[0063] In the embodiment provided in step S206, the signal search model may, but is not limited to, be obtained by training the initial search model using energy information samples labeled with the location of the primary synchronization signal. The initial search model may, but is not limited to, include multiple sub-models, which may, but are not limited to, be jointly trained, or may, but are not limited to, being trained based on multi-task learning, or may be trained using other methods; this application does not limit this.

[0064] Optionally, in this embodiment, the energy distribution characteristics of the signal location of the master synchronization signal can be learned by using a signal search model, and then the signal location of the master synchronization signal of the real-time acquired signal set can be quickly determined based on the signal search model and the target energy information of the real-time acquired signal set.

[0065] As an optional implementation, target synchronization information of a signal set can be detected using a signal search model and target energy information in the following manner, but not limited to: generating a feature matrix that meets the input requirements of a second sub-model using a first sub-model and a target energy matrix, wherein multiple reference elements in the target energy matrix are used to indicate the maximum energy of signals in different frequency bands in the signal set at different times, and the target energy information includes the target energy matrix; concatenating an initial classification vector in the first row of the feature matrix to obtain an initial classification matrix; inputting the initial classification matrix into the second sub-model, wherein the second sub-model includes a multi-layer self-attention transform encoder; receiving the target classification matrix output by the second sub-model; extracting the first row vector from the target classification matrix to obtain a target classification vector; inputting the target classification vector into a third sub-model and a fourth sub-model, wherein the signal search model includes the first sub-model, the second sub-model, the third sub-model, and the fourth sub-model; receiving the first sub-information output by the third sub-model and the second sub-information output by the fourth sub-model, wherein the first sub-information is used to indicate whether a primary synchronization signal exists in the signal set, and the second sub-information is used to indicate the time position and frequency position of the primary synchronization signal in the signal set, and the target synchronization information includes the first sub-information and the second sub-information.

[0066] Optionally, in this embodiment, the multiple reference elements may be, but are not limited to, completely identical to the multiple target elements, or the multiple reference elements may be, but are not limited to, some of the multiple target elements.

[0067] Optionally, in this embodiment, the target energy matrix may be converted into a feature matrix first through a first sub-model, and an initial classification vector is concatenated in its first row to form an initial classification matrix. Then, the initial classification matrix is ​​input into a second sub-model including a multi-layer self-attention transform encoder for processing. After obtaining the target classification matrix, the first row vector is extracted as the target classification vector. Finally, the target classification vector is input into the third and fourth sub-models in parallel, and the first sub-information (classification result) used to indicate the existence of the main synchronization signal and the second sub-information (regression result) used to indicate the time and frequency position of the main synchronization signal are received respectively.

[0068] As an optional implementation, an initial classification matrix that meets the input requirements of the second sub-model can be generated using the first sub-model and the target energy matrix in the following manner, but not limited to: converting the target energy matrix into an information matrix of the target dimension using the first sub-model, wherein multiple row vectors in the information matrix correspond one-to-one with multiple reference elements; generating position codes for the multiple row vectors based on the frequency index and time index of the multiple reference elements corresponding to the multiple row vectors, thereby obtaining a position matrix, wherein the time index is used to indicate the temporal order of the time period corresponding to the reference element within the time periods corresponding to the multiple reference elements, and the frequency index is used to indicate the frequency band arrangement order of the frequency band corresponding to the reference element within the frequency bands corresponding to the multiple reference elements; and adding the information matrix and the position matrix to obtain the feature matrix.

[0069] Optionally, in this embodiment, the target energy matrix can be mapped to an information matrix of the target dimension through the first sub-model, such that each row vector in the information matrix corresponds to a reference element. Then, using the frequency index and time index associated with the reference element corresponding to the row vector, codes representing frequency domain and time domain position information are generated for these row vectors respectively. These codes are combined into a position matrix. Finally, the information matrix and the position matrix are added element by element to obtain a feature matrix that includes both energy features and spatiotemporal position information, so as to solve the problem that the self-attention mechanism cannot directly perceive the relative position of elements in a two-dimensional sequence.

[0070] As an optional implementation, the target energy matrix can be converted into an information matrix of the target dimension through the first sub-model in the following manner, but not limited to: performing a convolution operation on the target energy matrix using a one-dimensional convolution kernel along the frequency direction to obtain a local enhancement tensor; unfolding the local enhancement tensor into a one-dimensional sequence to obtain an intermediate sequence; and mapping each vector in the intermediate sequence to a vector of the reference dimension through a projection matrix to obtain an information matrix, wherein the column dimension in the target dimension is equal to the reference dimension.

[0071] Optionally, in this embodiment, the first sub-model can perform convolution operations on the target energy matrix using a one-dimensional convolution kernel along the frequency direction to extract local features and obtain a local enhancement tensor. After the first sub-model flattens the tensor into a one-dimensional sequence, it uses a projection matrix to map each vector in the sequence into a vector with a unified reference dimension, thereby generating an information matrix with a column dimension equal to the reference dimension. This completes the conversion from a two-dimensional energy map to the one-dimensional feature sequence and dimension required by the Transformer.

[0072] Optionally, in this embodiment, the target dimension may refer to, but is not limited to, the dimension of the information matrix, including row dimension and column dimension. The reference dimension may refer to, but is not limited to, the dimension of the vector, that is, the number of elements included in the vector. It may be required, but is not limited to, that the column dimension in the target dimension is equal to the reference dimension of the vector, that is, that is, the number of elements included in the vector is limited to be equal to a specified number.

[0073] like Figure 4 As shown, Conv2D feature extraction can be performed on S, but is not limited to this. It can be, but is not limited to, first utilizing asymmetric convolution kernels. The feature sequence is obtained by mapping S (i.e., the target energy matrix) with a one-dimensional convolution kernel along the frequency direction. (i.e., information matrix):

[0074] (i.e., the first sub-model)

[0075] in This represents a one-dimensional receptive field convolution with a frequency domain window length of P_f. N is the number of patches, and D is the hidden layer dimension.

[0076] That is, it is possible, but not limited to, using asymmetric convolution kernels first. Convolve S using an asymmetric kernel that captures local correlations in the frequency dimension. Then flatten the output from the convolution, and finally multiply the flattened vector by the projection matrix. This allows different numbers of convolutional feature channels to be mapped to a unified Transformer hidden layer dimension D. This gives each token a D-dimensional (i.e., reference dimension) vector representation, suitable for subsequent self-attention computation.

[0077] Optionally, in this embodiment, the various model parameters in the first sub-model (e.g., including...) Parameters and The parameters can be, but are not limited to, determined through training (e.g., supervised training).

[0078] As an optional implementation, a position matrix can be obtained by generating position codes for multiple row vectors based on the frequency index and time index of multiple reference elements corresponding to multiple row vectors in the following manner: generating time dimension codes for multiple row vectors based on the time index to obtain a time dimension matrix, and generating frequency dimension codes for multiple row vectors based on the frequency index to obtain a frequency dimension matrix; concatenating the time dimension matrix and the frequency dimension matrix to obtain the position matrix.

[0079] Optionally, in this embodiment, the time index can be, but is not limited to, used to indicate the relative order of reference elements in the time dimension, that is, the order position of the signal segment corresponding to the reference element in the sequence of time segments corresponding to all reference elements, rather than a specific absolute timestamp; for example, if a long signal is divided (the division formed by the entire method, not just the division formed by dividing the signal set according to time order) into 100 consecutive time subsets and numbered sequentially, then the time index corresponding to the 50th time subset is 50, indicating that it is the 50th time segment in the sequence. The time index can be, but is not limited to, t as described above.

[0080] Optionally, in this embodiment, the frequency index can be, but is not limited to, used to indicate the relative arrangement order of reference elements in the frequency dimension. That is, the discrete ranking of the frequency components corresponding to the reference element in a preset frequency grid from low to high (or from high to low) reflects the relative topological relationship of frequency bins. For example, if the signal spectrum is divided into 128 equally wide frequency bins, the frequency index corresponding to the lowest frequency band is 0, and the frequency index corresponding to the highest frequency band is 127. Regardless of whether the actual frequency is 1kHz or 10MHz, its relative ranking index in the sequence remains unchanged. The frequency index can be, but is not limited to, k as described above.

[0081] Optionally, in this embodiment, the time-dimensional matrix is ​​obtained by generating time-dimensional codes for multiple row vectors based on the time index. This can be achieved, but is not limited to, using a positional encoding formula defined by a sine or cosine function, for each row vector corresponding to a time index position. (in, Given the dimension index i, calculate its corresponding sine or cosine value (e.g., ...). or ), where D is the hidden layer dimension, thus generating a vector containing time position information for each row vector, and stacking the time position encoded vectors corresponding to all row vectors to form a time dimension matrix.

[0082] Optionally, in this embodiment, generating a frequency-dimensional matrix by generating frequency-dimensional codes for multiple row vectors based on the frequency index can, but is not limited to, using a positional coding mechanism based on sine or cosine functions, for each row vector corresponding to the frequency index position. (in, Given the dimension index i, calculate its corresponding sine or cosine value (e.g., ...). or ), where D is the hidden layer dimension, in order to capture the relative positional relationship in the frequency domain, thereby generating a vector containing frequency position information for each row vector, and stacking the frequency position encoded vectors corresponding to all row vectors to form a frequency dimension matrix.

[0083] Optionally, in this embodiment, when using a sine or cosine function position encoding mechanism to encode time position information and frequency position information, different functions can be used for position information of different dimensions, but are not limited to. That is, when using a sine function position encoding mechanism to encode time position information, a cosine function position encoding mechanism is used to encode frequency position information; and when using a sine function position encoding mechanism to encode frequency position information, a cosine function position encoding mechanism is used to encode time position information. By limiting the time position and frequency position to be encoded using sine and cosine functions respectively (i.e., one using sine and the other using cosine), the inherent relationship between sine and cosine functions is utilized. Phase orthogonality ensures that the position signals in the time and frequency dimensions maintain maximum discriminative power and linear independence in the feature space when constructing two-dimensional spatiotemporal location features. This orthogonal design effectively avoids feature redundancy or gradient update direction overlap that may be caused by using the same function. This allows the Transformer's self-attention mechanism to capture the relative positional deviations on the time and frequency axes in the time-frequency plot more clearly and independently, thereby improving the model's estimation accuracy of the coarse position of the master synchronization signal in an extremely undersampled coarse feature matrix.

[0084] Optionally, in this embodiment, after obtaining the time dimension matrix and the frequency dimension matrix, the two can be spliced ​​together to form a position matrix, but not limited to this.

[0085] like Figure 4 As shown, the features obtained through Conv2D feature extraction are... After (i.e., the information matrix), position encoding can be performed, but is not limited to, defining an independent sinusoidal position encoding matrix (i.e., the time-dimensional matrix) in the time dimension. Independent sinusoidal position coding matrices in the frequency dimension (i.e., frequency dimension matrices) :

[0086] ;

[0087] ;

[0088] It is possible, but not limited to, to combine the two to form a (position matrix). .

[0089] For example, to simplify the explanation, it is possible, but not limited to, assuming D=4, meaning each feature vector has only 4 values. It is also possible, but not limited to, setting the first two dimensions for time encoding (sine) and the last two dimensions for frequency encoding (cosine). Similarly, to simplify the explanation, it is possible, but not limited to, assuming the information matrix output by the first sub-model... This includes two vectors (i.e., the aforementioned multiple row vectors) (representing two different time-frequency blocks):

[0090] Vector A: Corresponds to time index t=0, frequency index k=0. Vector B: Corresponds to time index t=1, frequency index k=0.

[0091] Then for vector A, its time position Frequency position ;

[0092] Then, for time encoding: calculate the first two dimensions; when i=0, When i=0.5, The time component [0,0] is obtained. Frequency encoding is performed, and the last two dimensions are calculated. When i=0, When i=0.5, The frequency component [1,1] is obtained. [0,0] and [1,1] are concatenated to form the position vector [0,0,1,1].

[0093] Correspondingly, for vector B, its time position Frequency position ;

[0094] Then, for time encoding: calculate the first two dimensions; when i=0, When i=0.5, The time component is obtained as [0.84, 0.01]. Frequency encoding is performed, and the last two dimensions are calculated. When i=0, When i=0.5, The frequency component [1,1] is obtained. [0.84,0.01] and [1,1] are concatenated to form the position vector [0.84,0.01,1,1].

[0095] Then the time dimension matrix can be obtained. and frequency dimension matrix The final position matrix is ​​obtained. .

[0096] Optionally, in this embodiment, after obtaining the position matrix, the information matrix and the position matrix can be added together, but are not limited to this: that is, the elements at the same position in the information matrix and the position matrix are added together to obtain the feature matrix. That is, the feature matrix... .

[0097] For example, suppose the information matrix composed of vectors A and B mentioned above is... Then the feature matrix corresponding to this information matrix is .

[0098] Optionally, in this embodiment, after obtaining the feature matrix, an initial classification vector may be concatenated into the first row of the feature matrix, but is not limited to this. Optionally, the initial classification vector may be, but is not limited to, a learnable classification vector, which may be, but is not limited to, randomly generated, or may be, but is not limited to, pre-defined.

[0099] like Figure 4 As shown, the positional encoding operation steps may include, but are not limited to, obtaining the feature matrix. Then, the learnable classification vector (i.e., the initial classification vector) is concatenated at the beginning. The initial classification matrix is ​​obtained. .

[0100] Optionally, in this embodiment, the initial classification matrix can be input into a second sub-model including a multi-layer self-attention transform encoder, but is not limited to. By inputting the initial classification matrix into the second sub-model, the powerful global modeling capability of the multi-layer self-attention transform encoder is utilized to perform deep semantic extraction and nonlinear interactive fusion on the time-frequency feature sequence enhanced by position encoding. Through the self-attention mechanism, each position in the sequence can dynamically pay attention to all other positions, thereby capturing long-distance time-frequency dependencies and suppressing noise interference in the extremely undersampled coarse feature map. Finally, feature vectors containing global key information (such as CLS tokens) are aggregated to provide a highly robust abstract feature representation for subsequent accurate estimation of the coarse time, frequency position, and existence of the master synchronization signal.

[0101] like Figure 4 As shown, Transformer Encoders can be applied to the initial classification matrix, but are not limited to this. That is, the initial classification matrix can be controlled to pass through... The Transformer encoder for layer l:

[0102] ;

[0103] Attention distribution is calculated using scaled dot product: ;

[0104] Combining residual connectivity and layer normalization:

[0105] (i.e., for) (Perform multi-head self-attention computation)

[0106] (i.e., for) Perform nonlinear transformation of multilayer perceptron (feedforward network);

[0107] Extraction is possible, but not limited to... The CLS Token feature vector of the last layer in the layer That is, extraction The first row vector of Z, output from the last layer in the layer, yields the target classification vector. .

[0108] Optionally, in this embodiment, the various model parameters in the second sub-model (e.g., including...) (etc.) can be, but is not limited to, determined in advance through early model training.

[0109] Optionally, in this embodiment, after obtaining the target classification vector, the target classification vector can be input into the multi-task head for inference, but is not limited to that, the target classification vector can be input into the third sub-model and the fourth sub-model respectively.

[0110] Optionally, in this embodiment, the model output of the third sub-model after receiving the target classification vector may be, but is not limited to, the first sub-information, which may be, but is not limited to, used to indicate whether a master synchronization signal exists in the signal set.

[0111] Optionally, in this embodiment, the model output of the fourth sub-model after receiving the target classification vector may, but is not limited to, the second sub-information. The second sub-information may, but is not limited to, indicate the time position and frequency position of the primary synchronization signal in the signal set. Specifically, the time position and frequency position indicated by the second sub-information may, but is not limited to, be a relative position or an absolute position.

[0112] like Figure 4 As shown, it is possible, but not limited to, after processing by Transformer Encoders, to obtain Input a multi-task header. The output should be categorized. (i.e., the third sub-model) (where, That is, the Sigmoid activation function. The time and frequency outputs are used to infer whether a primary synchronization signal exists in the signal set. The time and frequency outputs are used to determine the relative offset of the possible primary synchronization signal within the signal set, i.e., the relative offset of the PSS signal within the current observation window. Specifically, the time and frequency outputs provide a normalized relative time deviation. (This represents the relative time position of the main synchronization signal within the window) and a relative center frequency position. (This indicates the offset of the signal center frequency relative to the center of the current analysis frequency band.) (i.e., the fourth sub-model).

[0113] Optionally, in this embodiment, the model parameters in the third and fourth sub-models (e.g., including...) (etc.) can be, but is not limited to, determined by prior model training.

[0114] In the embodiment provided in step S208, the primary synchronization signal can be searched in the signal set by combining the determined primary synchronization signal with the signal position in the signal set.

[0115] Optionally, in this embodiment, after obtaining Afterwards, it can be done, but is not limited to, through calculation. Determine the absolute time coordinate of the primary synchronization signal within the signal set by calculating... Determine the absolute frequency coordinates of the primary synchronization signal within the signal set (where round(x) is the floor function), where... The total time length of the signal set. For frequency resolution, Let S be the number of bins in the frequency direction of the target energy matrix S.

[0116] Based on the above, the complex time-domain matching problem can be transformed into a pattern recognition problem on an extremely low-resolution time-frequency map. Utilizing the perceptual capabilities of the Transformer, the coarse time and frequency position of the PSS can be estimated within an extremely undersampled two-dimensional coarse feature matrix. Time-frequency localization can be performed very quickly.

[0117] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software and necessary general-purpose hardware platforms, and of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application.

[0118] Figure 5 This is a structural block diagram of a synchronization signal search device according to an embodiment of this application; as shown... Figure 5 As shown, the synchronization signal search device includes:

[0119] The acquisition module 502 is used to acquire the set of signals to be searched for the master synchronization signal;

[0120] The generation module 504 is used to generate target energy information of the signal set, wherein the target energy information is used to indicate the maximum energy of signals in different frequency bands in the signal set at different time periods;

[0121] The detection module 506 is used to detect the target synchronization information of the signal set through the signal search model and the target energy information. The target synchronization information is used to indicate the signal position of the main synchronization signal in the signal set. The signal search model is obtained by training the initial search model using energy information samples labeled with the signal position of the main synchronization signal.

[0122] Search module 508 is used to search for the main synchronization signal in the signal set based on the target synchronization information.

[0123] Through the above embodiments, after obtaining the signal set of the primary synchronization signal to be searched, target energy information is generated to indicate the maximum energy of signals in different frequency bands within the signal set at different time periods. The target synchronization information, indicating the signal position of the primary synchronization signal within the signal set, is detected using a signal search model and the target energy information. This achieves the goal of determining the signal position of the primary synchronization signal within the signal set by combining the energy distribution characteristics exhibited by the signal set. Then, by combining the target synchronization information, i.e., the determined signal position of the primary synchronization signal within the signal set, the search for the primary synchronization signal is performed within the signal set. Compared to a blind search across the entire time domain, this reduces the search time for the primary synchronization signal. Therefore, the above technical solution solves the problem of low search efficiency for primary synchronization signals in related technologies, achieving the technical effect of improving the search efficiency of primary synchronization signals.

[0124] Optionally, the generation module includes: a segmentation unit for segmenting the signal set into multiple signal subsets in the time domain according to time order; a determination unit for determining the frequency domain responses of multiple frequency components of the multiple signal subsets to obtain multiple target frequency domain responses, wherein the multiple target frequency domain responses correspond one-to-one with the multiple signal subsets and one-to-one with the multiple frequency components; a detection unit for detecting the maximum value of the square of the absolute value of each target frequency domain response to obtain multiple target elements, wherein the target elements are used to indicate the maximum energy of the signal in the target frequency band in the signal set during the target time period, the target frequency band is the frequency band of the frequency component corresponding to the target element, and the target time period is the time period corresponding to the target element; and a construction unit for constructing a target energy matrix according to the target format and the multiple target elements, wherein the target energy information includes the target energy matrix.

[0125] Optionally, the determining unit is also used to: obtain the target frequency domain response of the k-th frequency component of the m-th signal subset using the following formula:

[0126] ;

[0127] in, The target frequency domain response of the k-th frequency component of the m-th signal subset. The target transform parameters are used to indicate the number of multiple frequency components. L represents the time length of each signal subset. For the m-th signal subset, For window functions.

[0128] Optionally, the detection module includes: a generation unit, used to generate a feature matrix that meets the input requirements of the second sub-model using a first sub-model and a target energy matrix, wherein multiple reference elements in the target energy matrix are used to indicate the maximum energy of signals in different frequency bands in the signal set at different time periods, and the target energy information includes the target energy matrix; a splicing unit, used to splice an initial classification vector into the first row of the feature matrix to obtain an initial classification matrix; a first input unit, used to input the initial classification matrix into the second sub-model, wherein the second sub-model includes a multi-layer self-attention transform encoder; a first receiving unit, used to receive the target classification matrix output by the second sub-model; and an extraction unit. The signal search model includes a first sub-model, a second sub-model, a third sub-model, and a fourth sub-model. The second receiving unit is used to receive the first sub-information output by the third sub-model and the second sub-information output by the fourth sub-model. The first sub-information indicates whether a primary synchronization signal exists in the signal set, and the second sub-information indicates the time and frequency position of the primary synchronization signal in the signal set. The target synchronization information includes both the first and second sub-information.

[0129] Optionally, the generation unit is further configured to: convert the target energy matrix into an information matrix of target dimension through the first sub-model, wherein multiple row vectors in the information matrix correspond one-to-one with multiple reference elements; generate position codes for the multiple row vectors based on the frequency index and time index of the multiple reference elements corresponding to the multiple row vectors, thereby obtaining a position matrix, wherein the time index is used to indicate the temporal order of the time period corresponding to the reference element within the time periods corresponding to the multiple reference elements, and the frequency index is used to indicate the frequency band arrangement order of the frequency band corresponding to the reference element within the frequency bands corresponding to the multiple reference elements; and add the information matrix and the position matrix to obtain a feature matrix.

[0130] Optionally, the generation unit is also used to: perform a convolution operation on the target energy matrix using a one-dimensional convolution kernel along the frequency direction to obtain a local enhancement tensor; expand the local enhancement tensor into a one-dimensional sequence to obtain an intermediate sequence; and map each vector in the intermediate sequence to a vector of the reference dimension through a projection matrix to obtain an information matrix, wherein the column dimension in the target dimension is equal to the reference dimension.

[0131] Optionally, the generating unit is also used to: generate time-dimensional codes for multiple row vectors based on time indices to obtain a time-dimensional matrix, and generate frequency-dimensional codes for multiple row vectors based on frequency indices to obtain a frequency-dimensional matrix; and concatenate the time-dimensional matrix and the frequency-dimensional matrix to obtain a position matrix.

[0132] For a description of the features in the embodiment corresponding to the synchronization signal search device, please refer to the relevant description of the embodiment corresponding to the synchronization signal search method, which will not be repeated here.

[0133] Embodiments of this application also provide a storage medium including a stored program, wherein the program, when executed, performs the search method for any of the above-mentioned synchronization signals.

[0134] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0135] Embodiments of this application also provide an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above-described embodiments of the method for searching for synchronization signals.

[0136] Optionally, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0137] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program: acquiring a signal set of the primary synchronization signal to be searched; generating target energy information of the signal set, wherein the target energy information is used to indicate the maximum energy of signals in different frequency bands in the signal set at different times; detecting target synchronization information of the signal set through a signal search model and the target energy information, wherein the target synchronization information is used to indicate the signal position of the primary synchronization signal in the signal set, and the signal search model is obtained by training an initial search model using energy information samples labeled with the signal position of the primary synchronization signal; and searching for the primary synchronization signal in the signal set according to the target synchronization information.

[0138] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described methods for searching synchronization signals.

[0139] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in any of the above-described methods for searching synchronization signals.

[0140] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0141] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0142] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for searching a synchronization signal, characterized in that, include: Obtain the set of signals to be searched for as the primary synchronization signal; Generate target energy information for the signal set, wherein the target energy information is used to indicate the maximum energy of signals in different frequency bands in the signal set at different time periods; The target synchronization information of the signal set is detected by a signal search model and the target energy information, wherein the target synchronization information is used to indicate the signal position of the main synchronization signal in the signal set, and the signal search model is obtained by training an initial search model using energy information samples labeled with the signal position of the main synchronization signal. The primary synchronization signal is searched in the signal set based on the target synchronization information.

2. The method according to claim 1, characterized in that, The target energy information for generating the signal set includes: The signal set is divided into multiple signal subsets in the time domain according to time order; Determine the frequency domain response of multiple frequency components of the multiple signal subsets to obtain multiple target frequency domain responses, wherein the multiple target frequency domain responses correspond one-to-one with the multiple signal subsets, and the multiple target frequency domain responses correspond one-to-one with the multiple frequency components; The maximum value of the square of the absolute value of each target frequency domain response is detected to obtain multiple target elements, wherein the target elements are used to indicate the maximum energy of the signal in the target frequency band in the signal set during the target time period, the target frequency band is the frequency band of the frequency component corresponding to the target element, and the target time period is the time period corresponding to the target element; A target energy matrix is ​​constructed according to the target format and the plurality of target elements, wherein the target energy information includes the target energy matrix.

3. The method according to claim 2, characterized in that, The step of determining the frequency domain response of multiple frequency components of the multiple signal subsets to obtain multiple target frequency domain responses includes: The target frequency domain response of the k-th frequency component of the m-th subset of signals is obtained by the following formula: ； in, The target frequency domain response is the k-th frequency component of the m-th subset of the signals. The target transformation parameter indicates the number of the plurality of frequency components. L is the time length of each of the aforementioned signal subsets. For the m-th subset of the signals, For window functions.

4. The method according to claim 1, characterized in that, The step of detecting target synchronization information of the signal set through a signal search model and the target energy information includes: A feature matrix that meets the input requirements of the second sub-model is generated by using the first sub-model and the target energy matrix. The multiple reference elements in the target energy matrix are used to indicate the maximum energy of signals in different frequency bands in the signal set at different time periods. The target energy information includes the target energy matrix. The initial classification vector is concatenated into the first row of the feature matrix to obtain the initial classification matrix; The initial classification matrix is ​​input into the second sub-model, wherein the second sub-model includes a multi-layer self-attention transform encoder; Receive the target classification matrix output by the second sub-model; Extract the first row vector from the target classification matrix to obtain the target classification vector; The target classification vector is input into the third sub-model, and the target classification vector is input into the fourth sub-model. The signal search model includes the first sub-model, the second sub-model, the third sub-model, and the fourth sub-model. The system receives first sub-information output by the third sub-model and second sub-information output by the fourth sub-model. The first sub-information indicates whether a primary synchronization signal exists in the signal set, and the second sub-information indicates the time and frequency position of the primary synchronization signal in the signal set. The target synchronization information includes the first sub-information and the second sub-information.

5. The method according to claim 4, characterized in that, The step of generating an initial classification matrix that meets the input requirements of the second sub-model using the first sub-model and the target energy matrix includes: The target energy matrix is ​​converted into an information matrix of target dimension through the first sub-model, wherein multiple row vectors in the information matrix correspond one-to-one with multiple reference elements; A positional code is generated for the multiple row vectors based on the frequency index and time index of the multiple reference elements corresponding to the multiple row vectors, resulting in a positional matrix. The time index is used to indicate the chronological order of the time period corresponding to the reference element in the time periods corresponding to the multiple reference elements, and the frequency index is used to indicate the frequency band arrangement order of the frequency band corresponding to the reference element in the frequency bands corresponding to the multiple reference elements. The feature matrix is ​​obtained by adding the information matrix and the position matrix.

6. The method according to claim 5, characterized in that, The step of converting the target energy matrix into an information matrix of target dimension through the first sub-model includes: The target energy matrix is ​​convolved using a one-dimensional convolution kernel along the frequency direction to obtain a local enhancement tensor. Expand the local enhancement tensor into a one-dimensional sequence to obtain the intermediate sequence; The information matrix is ​​obtained by mapping each vector in the intermediate sequence to a vector of the reference dimension using a projection matrix, wherein the column dimension of the target dimension is equal to the reference dimension.

7. The method according to claim 5, characterized in that, The step of generating position codes for the plurality of row vectors based on the frequency index and time index of the plurality of reference elements corresponding to the plurality of row vectors, to obtain a position matrix, includes: A time-dimensional matrix is ​​obtained by generating time-dimensional codes for the plurality of row vectors based on the time index, and a frequency-dimensional matrix is ​​obtained by generating frequency-dimensional codes for the plurality of row vectors based on the frequency index. The position matrix is ​​obtained by concatenating the time dimension matrix and the frequency dimension matrix.

8. A synchronization signal search device, characterized in that, include: The acquisition module is used to acquire the set of signals to be searched for as the master synchronization signal; A generation module is used to generate target energy information for the signal set, wherein the target energy information is used to indicate the maximum energy of signals in different frequency bands in the signal set at different time periods; The detection module is used to detect the target synchronization information of the signal set through the signal search model and the target energy information, wherein the target synchronization information is used to indicate the signal position of the main synchronization signal in the signal set, and the signal search model is obtained by training an initial search model using energy information samples labeled with the signal position of the main synchronization signal; The search module is used to search for the primary synchronization signal in the signal set based on the target synchronization information.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein the program, when executed, performs the method of any one of claims 1 to 7.

10. An electronic device comprising a memory and a processor, characterized in that, The memory stores a computer program, and the processor is configured to execute the method of any one of claims 1 to 7 through the computer program.

Images