CNN-based time-frequency synchronization blind detection method and system

Abstract

The invention relates to the technical field of communication, discloses a CNN-based time-frequency synchronization blind detection method and system, and aims at a complex channel environment and high-precision synchronization requirements of a 5G / 6G system to generate a real training data set by simulating multipath fading and noise so as to ensure data diversity and reliability. PSS signal features are automatically extracted by using CNN, complexity of manual feature design is avoided, and detection accuracy and robustness are significantly improved; compared with a traditional algorithm, the method is low in calculation complexity and short in detection time delay, PSS signal blind detection can be rapidly completed, and downlink time-frequency synchronization of the base station and the UE is achieved; according to the method, the high requirement of a 5G / 6G system on a synchronous detection algorithm is effectively met, and powerful support is provided for efficient operation and digital transformation of a network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication, in particular to a CNN-based time-frequency synchronization blind detection method and system. BACKGROUND

[0002] 5G is a new generation of broadband mobile communication technology with high speed, high reliability and low latency, and mass connectivity. Its three major application scenarios include enhanced mobile broadband (eMBB), ultra-high reliable low latency communication (uRLLC) and massive machine type communication (mMTC). 6G is the upgrade version of 5G, which will build a fully connected world integrating terrestrial wireless and satellite communication. Compared with 5G, 6G will have a data transmission rate of 50 times that of 5G, a latency of one tenth of 5G, and will be much better than 5G in terms of peak rate, latency, traffic density, connection number density, mobility, spectrum efficiency, positioning capability, etc. At present, various industries are in the key period of digital transformation, and 5G is the key technology to promote this transformation. As the next generation of network communication technology standard, 6G is not only an extension of 5G, but also a core engine for the deep integration of digital economy and real economy, and a core driving force for new productivity, which will better empower more industries. The 6G white paper released at the Global 6G Technology Conference points out that the information interaction of intelligent agents such as unmanned vehicles and drones based on the transformation of productivity and production relations has become a new business scenario for 6G, which puts forward higher requirements for network latency, reliability and other indicators.

[0003] To communicate, first synchronize. The primary synchronization signal (PSS) is the first physical signal received by the user equipment (UE) when accessing the 5G / 6G NR system. At this time, the UE has no prior information about the communication system, has not obtained the time-frequency synchronization information of the system, and the internal reference frequency is not accurate enough. Therefore, the UE needs to detect the PSS in a blind way to achieve downlink frequency synchronization and symbol-level time synchronization, and to establish a downlink synchronization communication link. This is the most complex part of the cell search process. However, the NR system uses a downlink transmission scheme based on CP-OFDM technology, which is very sensitive to carrier frequency offset. Factors such as Doppler shift, local oscillator frequency deviation and multipath effect can destroy the orthogonality between subcarriers, and thus affect the overall performance of the system. In addition, the high sampling rate characteristics of the 5G / 6G NR system make the system have more stringent requirements for carrier frequency offset and timing error. Whether the primary synchronization signal can be quickly and reliably detected directly affects the establishment of the downlink synchronization communication link, and is the key to the successful access of the user.

[0004] Currently, the main synchronization signal detection algorithm of 5G / 6G system is mainly based on the correlation of the local primary synchronization sequence and the received signal. Common algorithms include traditional cross-correlation algorithm, differential correlation algorithm, segmented correlation algorithm and maximum likelihood algorithm based on cyclic prefix. The traditional cross-correlation algorithm is the most commonly used PSS timing synchronization technology, which has the advantages of simple principle and strong correlation. However, the algorithm has high computational complexity, and is prone to frequency offset accumulation, resulting in poor anti-frequency offset performance, especially in the case of large frequency offset, the performance will decrease sharply. The differential correlation algorithm performs differential processing on the primary synchronization signal before cross-correlation, which can weaken the influence of frequency offset and has certain anti-frequency offset ability. However, the processing process will introduce additional interference noise terms, making the algorithm very sensitive to noise, difficult to adapt to complex interference environment, and the complexity of the algorithm is much higher than that of the traditional cross-correlation algorithm. The segmented correlation algorithm shortens the length of the effective correlation window by segmentation, to a certain extent, reduces the cumulative influence of frequency offset, and improves the anti-frequency offset ability. However, too many segments will cause the peak correlation to decrease, thereby reducing the anti-noise ability of the algorithm, and the computational complexity of the algorithm is also high. The maximum likelihood algorithm based on cyclic prefix can obtain good performance, but the computational complexity is very large, resulting in long algorithm delay.

[0005] Morelli and M. Moretti proposed a PSS detection algorithm that relies on the maximum likelihood estimation criterion and uses a suitable reduced-rank representation of the channel frequency response to consider multipath distortion. Although this algorithm can achieve good performance, the computational complexity is very large, resulting in a large delay, which cannot meet the requirements of fast synchronization of NR system downlink. A. Omri and M. Shaqfeh et al. proposed a detection algorithm based on autocorrelation and cross-correlation. The autocorrelation algorithm has certain anti-carrier frequency offset (CFO) ability, but the anti-noise ability is poor, and it cannot complete synchronization at low signal-to-noise ratio. The cross-correlation algorithm can realize synchronization at low signal-to-noise ratio, but when the frequency offset increases, the performance will decrease sharply. L. Xue and S. Qiu et al. proposed an improved time-domain data-aided OFDM / OQAM system carrier frequency offset and time offset joint estimation algorithm. Although this algorithm can improve the estimation accuracy, the introduction of iterative link greatly increases the computational complexity. Y. H. You and H. K. Song proposed a joint PSS and IFO detection algorithm, which reduces part of the complexity, but the performance also decreases accordingly. Dr. Wang Dan of Chongqing University of Posts and Telecommunications proposed a primary synchronization signal timing synchronization algorithm with anti-frequency offset and anti-noise. This algorithm can achieve good detection effect in the case of large frequency offset, but the computational complexity is very large, and the real-time performance is poor, which is difficult to meet the requirements of fast synchronization of 5G / 6G NR system downlink.

[0006] In conclusion, in the new era of higher requirements for network delay, reliability and other indicators, it is of great significance to study a primary synchronization signal fast detection algorithm with high reliability, low complexity and low delay.

[0007] The above information is presented as background information only to assist with an understanding of the present application. No determination or admission is made as to any portion of the above information as having predictive value in relation to the adequacy or existence of prior art with respect to the present application. SUMMARY

[0008] The application provides a CNN-based time-frequency synchronization blind detection method and system to solve the problems in the prior art.

[0009] To achieve the above-mentioned purpose, the application provides the following technical solutions:

[0010] In a first aspect, the application provides a CNN-based time-frequency synchronization blind detection method, which comprises:

[0011] S101, generating a PSS transmission signal data set transmitted by a base station based on the M sequence characteristics of the primary synchronization signal (PSS);

[0012] S102, performing channel processing on the PSS transmission signal data set based on multipath fading and noise characteristics to generate a PSS reception signal data set received by a user equipment (UE) after channel transmission;

[0013] S103, dividing the PSS reception signal data set into a training set, a validation set and a test set;

[0014] S104, building a PSS detection model based on a convolutional neural network (CNN);

[0015] S105, training and validating the PSS detection model using the training set and the validation set to obtain a trained PSS detection model;

[0016] S106, inputting the test set into the trained PSS detection model to obtain a detection result.

[0017] Further, in the CNN-based time-frequency synchronization blind detection method, S102 comprises:

[0018] S1021, mapping the PSS transmission signal data set to a center subcarrier to complete frequency domain mapping;

[0019] S1022, transforming the frequency domain PSS signal into a time domain PSS signal based on inverse fast Fourier transform (IFFT) and adding a cyclic prefix;

[0020] S1023, based on the wireless channel transmission characteristics, adding frequency offset, multipath channel fading effect and noise to the time domain PSS signal to generate a PSS received signal data set received by a user equipment (UE) after channel transmission.

[0021] Further, in the CNN-based time-frequency synchronization blind detection method, in the S105 and S106, the processing process of the PSS detection model includes:

[0022] The convolutional pooling module sequentially uses three one-dimensional convolution kernels for CNN convolution on the PSS received signal after channel transmission, then uses a BatchNorm1d module and a ReLU module for batch normalization and ReLU activation on the convolution result after each convolution, and then uses a MaxPool1d module for pooling operation.

[0023] The PSS feature data output after three-layer convolutional pooling is further extracted by using two linear regression modules, and finally a linear regression module can extract the PSS type to complete the blind detection of the PSS signal.

[0024] Further, in the CNN-based time-frequency synchronization blind detection method, the sizes of the three one-dimensional convolution kernels used in sequence are 7, 5 and 3, respectively.

[0025] The size of the pooling window of the MaxPool1d module is 2*2.

[0026] Further, in the CNN-based time-frequency synchronization blind detection method, the number of neurons of the two linear regression modules used for further feature extraction is 512 and 128, respectively.

[0027] The number of neurons of the linear regression module used for extracting the PSS type is 3.

[0028] Further, in the CNN-based time-frequency synchronization blind detection method, the S103 includes:

[0029] S1031, according to a preset division ratio, randomly extract part of the data in the PSS received signal data set as a training set, and the data amount of the training set accounts for X%;

[0030] S1032, from the remaining data, extract part of the data according to a preset ratio as a validation set, and the data amount of the validation set accounts for Y%;

[0031] S1033. The remaining data that was not selected into the training set and validation set is used as the test set, and the data volume of the test set accounts for Z%; where X%+Y%+Z%=100%, X%>Y%>Z%, and in the process of partitioning, it is ensured that the data in the training set, validation set and test set are representative in terms of channel characteristics, noise level and multipath effect.

[0032] Furthermore, in the CNN-based time-frequency synchronous blind detection method, before step S103, the method further includes:

[0033] S102.5. Perform time delay processing on each signal sample in the PSS received signal dataset to simulate the signal transmission delay between the base station and the UE at different distances, and generate multiple delayed versions of the signal sample;

[0034] S102.6 Perform frequency offset processing on each signal sample in the PSS received signal dataset to simulate the signal frequency change under different Doppler effects and generate multiple frequency offset versions of the signal sample.

[0035] S102.7 Perform signal-to-noise ratio (SNR) adjustment processing on each signal sample in the PSS received signal dataset. By adding Gaussian white noise of different intensities, simulate the signal transmission environment under different SNR conditions and generate multiple signal samples with different SNR levels.

[0036] S102.8 Perform amplitude adjustment processing on each signal sample in the PSS received signal dataset to simulate the attenuation or amplification effect of the signal during transmission and generate multiple signal samples with different amplitudes.

[0037] S102.9 Perform phase rotation processing on each signal sample in the PSS received signal dataset to simulate the phase changes that may occur during signal transmission and generate multiple phase-rotated versions of the signal sample.

[0038] S102.10. Merge all the new signal samples generated after the above processing to form an expanded PSS received signal dataset;

[0039] Correspondingly, S103 includes:

[0040] The expanded PSS received signal dataset is divided into a training set and a validation set, and the original PSS received signal dataset is used as the test set.

[0041] Secondly, the present invention provides a time-frequency synchronous blind detection system based on CNN, the system comprising:

[0042] The first generation module is configured to generate a PSS transmission signal data set transmitted by a base station based on an M sequence characteristic of a primary synchronization signal (PSS);

[0043] The second generation module is configured to perform channel processing on the PSS transmission signal data set based on multipath fading and noise characteristics to generate a PSS reception signal data set received by a user equipment (UE) after channel transmission;

[0044] The data division module is configured to divide the PSS reception signal data set into a training set, a validation set and a test set;

[0045] The model building module is configured to build a PSS detection model based on a convolutional neural network (CNN).

[0046] The model training module is configured to train and validate the PSS detection model using the training set and the validation set to obtain a trained PSS detection model.

[0047] The model detection module is configured to input the test set into the trained PSS detection model to obtain a detection result.

[0048] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the CNN-based time-frequency synchronization blind detection method provided in the first aspect when executing the computer program.

[0049] In a fourth aspect, the present application provides a computer-readable storage medium having computer executable instructions stored thereon, wherein the computer executable instructions are executed by a computer processor to implement the CNN-based time-frequency synchronization blind detection method provided in the first aspect.

[0050] Compared with the prior art, the present application has the following beneficial effects:

[0051] This invention provides a CNN-based blind detection method and system for time-frequency synchronization. By leveraging the powerful feature extraction and nonlinear fitting capabilities of CNNs, it can effectively address the complex channel environment and high-precision synchronization requirements of 5G / 6G systems. Specifically, the method first generates a base station-side transmitted signal dataset based on the M-sequence characteristics of PSS (Pressure Sequence of Signal), and then generates a UE-side received signal dataset by simulating channel characteristics such as multipath fading and noise, ensuring the diversity and authenticity of the training data. Subsequently, the received signal dataset is divided into training, validation, and test sets for model training, validation, and testing, ensuring the model's generalization ability and reliability. In building the PSS detection model, a convolutional neural network is used to extract and classify features of the PSS signal. Compared with traditional algorithms, this can automatically learn complex features in the signal, avoiding the complexity and limitations of manually designing features. Through the training and validation process, the model can adapt to different channel conditions and frequency offsets, significantly improving the accuracy and robustness of PSS detection. Finally, the trained model is tested using a test set, verifying its effectiveness in real-world scenarios. This method can not only quickly and accurately complete the blind detection of PSS signals and realize downlink time-frequency synchronization between the base station and the UE, but also has low computational complexity and latency, meeting the requirements of 5G / 6G systems for high reliability, low complexity and low latency synchronization detection algorithms, and providing strong technical support for the efficient operation and digital transformation of 5G / 6G networks.

[0052] The present invention has other features and advantages, which will be apparent from or will be set forth in detail in the accompanying drawings and the following detailed description, which together serve to explain the particular principles of the invention. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a structural diagram of the SSB block provided in Embodiment 1 of the present invention;

[0055] Figure 2 This is a schematic diagram of the position of SSB in a half-frame (Case B) provided in Embodiment 1 of the present invention;

[0056] Figure 3This is a flowchart illustrating a CNN-based time-frequency synchronous blind detection method provided in Embodiment 1 of the present invention.

[0057] Figure 4 This is a further detailed flowchart of S102 provided in Embodiment 1 of the present invention;

[0058] Figure 5 This is a schematic diagram of the PSS detection model provided in Embodiment 1 of the present invention processing the PSS received signal after transmission through the channel;

[0059] Figure 6 This is a schematic diagram of the structure of a computer device provided in Embodiment 3 of the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Example 1

[0062] The primary synchronization signal (PSS), secondary synchronization signal (SSS), and physical broadcast channel (PBCH) signals of a 5 / 6G base station system constitute the SS / PBCH block, or simply the synchronization signal block (SSB). It is transmitted simultaneously with a narrow beam, covering the entire cell through beam scanning to achieve UE downlink synchronization. The SSB block uses time-division multiplexing, with each SSB block occupying 240 consecutive subcarriers (20 RBs) in the frequency domain and 4 OFDM symbols in the time domain. Figure 1 As shown.

[0063] The PSS is a pseudo-random sequence of length 127 using a frequency-domain BPSK M-sequence, mapped onto the 127 subcarriers in the middle of the SSB block. It is a sequence about... The function is shown in equation (1):

[0064] (1);

[0065] In the formula Defined by equation (2):

[0066] (2);

[0067] The PSS sequence can take three different values, corresponding to the physical layer identifier within the group of Physical Cell Identities (PCIs). Correspondingly.

[0068] The period of the SSB block in the NR system is variable, and can be 5ms, 10ms, 20ms, 40ms, 80ms, or 160ms, which can be flexibly configured in the physical layer frame structure according to different scenario requirements. Within each period, multiple SSBs constituting the SSB Burst are confined to a 5ms half-radio frame and transmitted using a beam scanning method. Based on the different SSB subcarrier spacings, the protocol specifies five candidate SSB block case patterns, defining the slot index of the SSB in the half-frame and the OFDM symbol index value of the SSB in that slot. Taking Case B as an example, when the SSB's SCS = 30kHz, the first OFDM symbol index of the candidate SSB is {4,8,16,20} + 28×n, as shown below. Figure 2 As shown.

[0069] For carrier frequencies less than or equal to 3 GHz, n ∈ {0}, the SSB is transmitted on subframe 0 of a certain half-frame, with a total of 4 candidate positions (Lmax=4); for carrier frequencies within FR1 and greater than 3 GHz, n ∈ {0, 1}, the SS / PBCH block is transmitted on subframes 0 and 1 of a certain half-frame, with a total of 8 candidate positions (Lmax=8).

[0070] Decoding the PSS (Primary Synchronization Signal) allows us to obtain the start position, symbol length, and PCI of the OFDM symbol in the NR system. After decoding the PSS, we can obtain the SSB block timing and subcarrier spacing using the SSB's time-frequency structure. Detecting the Primary Synchronization Signal (PSS) enables downlink frequency synchronization and symbol-level time synchronization between the UE and the base station. This signal can also serve as a reference for frequency generation within the UE, significantly reducing frequency fluctuations between the UE and the network.

[0071] 5 / 6G base station systems employ a downlink transmission scheme based on CP-OFDM technology, which is highly sensitive to carrier frequency offset. The master synchronization signal sent from the base station to the UE, after transmission through the channel, is subject to frequency offset due to factors such as added Gaussian white noise, Doppler shift, and local oscillator frequency deviation at the transmitting and receiving ends. Assuming... If the PSS sequence is sent by the base station, then the PSS sequence received by the UE is... It can be represented as:

[0072] (3);

[0073] In the formula, It is a location Frequency offset at ( The normalized frequency offset relative to the subcarrier spacing is defined as follows: , For frequency offset, (subcarrier spacing) Additive white Gaussian noise is introduced into the channel.

[0074] In traditional cross-correlation synchronization algorithms, the PSS (Pressure Sequence) uses the M-sequence, which has good correlation characteristics. The cross-correlation synchronization algorithm extracts the PSS synchronization point by performing cross-correlation calculations on the received PSS signal and three locally generated PSS sequences at the receiving end, and then searching for the cross-correlation peak. Its expression is:

[0075] (4);

[0076] In the formula, It is receiving signals In position Place and No. The relevant values ​​of a local PSS sequence This is the number of local PSS sequence sampling points. It is the first A local PSS sequence.

[0077] By comparing the peak values ​​of the three cross-correlation algorithms, the synchronization point of the PSS signal can be obtained. The expression for determining the peak value of the three cross-correlation results is shown in Equation 5:

[0078] (5);

[0079] In the formula, The maximum value of the three cross-correlation peaks. The position index corresponding to the maximum value of the three cross-correlation peaks is the synchronization point of the PSS signal. The transmission bandwidth is configured as follows The number of sampling points in the wireless subframe at that time.

[0080] Although the traditional cross-correlation synchronization algorithm is simple in principle, it requires cross-correlation between the received signal and each local PSS sequence. For each received signal, the algorithm needs to perform three cross-correlation operations, which makes the computational load large and the time delay large. In addition, it will accumulate frequency offset when performing cross-correlation operations. It can meet the system synchronization requirements when the frequency offset is small, but the synchronization performance will be significantly reduced when the frequency offset is large.

[0081] To address issues such as reducing algorithm complexity, minimizing algorithm latency, improving the algorithm's resistance to frequency offset, and enhancing the adaptability of PSS blind detection under complex channels, this invention provides a CNN-based time-frequency synchronization blind detection method.

[0082] Please refer to Figure 3 , Figure 3 This is a flowchart illustrating a CNN-based blind detection method for time-frequency synchronization provided in Embodiment 1 of the present invention. This method is applicable to scenarios involving downlink time-frequency synchronization establishment between a user equipment (UE) and a base station in a 5G / 6G NR system. The method is executed by a CNN-based blind detection system, which can be implemented in software and / or hardware. The method specifically includes the following steps:

[0083] S101. Based on the M-sequence characteristics of the primary synchronization signal PSS, generate a dataset of PSS transmission signals sent by the base station.

[0084] It should be noted that the Master Synchronization Signal (PSS) possesses specific M-sequence characteristics, which are inherent properties of the PSS signal. Based on these characteristics, a set of PSS signal data to be actually transmitted at the base station is generated using specific algorithms or rules. This dataset forms the basis for all subsequent operations, representing the original PSS signal transmitted by the base station.

[0085] S102. Based on multipath fading and noise characteristics, channel processing is performed on the PSS transmitted signal dataset to generate the PSS received signal dataset received by the user equipment (UE) after channel transmission.

[0086] It should be noted that in actual wireless communication environments, signals are affected by multipath fading and noise during transmission from the base station to the user equipment (UE). Multipath fading refers to the phenomenon where signals arrive at the receiver via different paths, and the varying path lengths cause changes in signal phase and amplitude, thus affecting signal quality. Noise, on the other hand, is an unavoidable interference factor in the communication environment. By simulating these actual channel characteristics on the PSS transmission signal dataset generated in step S101, the actual PSS signal dataset received by the UE can be obtained, which more closely resembles the real communication scenario.

[0087] S103. Divide the PSS received signal dataset into a training set, a validation set, and a test set;

[0088] It should be noted that, in order to effectively train, validate, and test the subsequently built model, the PSS received signal dataset obtained in step S102 needs to be divided. The training set is used to train the model, allowing it to learn the features and patterns in the data; the validation set is used to evaluate and adjust the model's performance during training to prevent overfitting; and the test set is used to independently evaluate the model's final performance after training is complete, to determine the model's effectiveness in practical applications. This division helps to comprehensively and accurately evaluate the model's performance.

[0089] S104. Build a PSS detection model based on convolutional neural network (CNN);

[0090] It should be noted that Convolutional Neural Networks (CNNs) are a type of deep learning model with strong feature extraction capabilities. Utilizing the structure and characteristics of CNNs, a model specifically designed for detecting PSS signals can be constructed. This model typically includes convolutional layers, pooling layers, and fully connected layers. Through the combination and configuration of these layers, the model can extract features and classify the input PSS signal data, thereby achieving PSS signal detection.

[0091] S105. The PSS detection model is trained and validated using the training set and the validation set to obtain a trained PSS detection model.

[0092] It should be noted that the PSS detection model built in step S104 is trained using the training set obtained in step S103. During training, the model continuously adjusts its parameters to minimize the error between the predicted and actual results, thereby learning the features and patterns in the data. Simultaneously, the model performance is evaluated using a validation set. Based on the evaluation results on the validation set, the model parameters are adjusted and optimized to prevent overfitting, ultimately resulting in a well-trained and high-performance PSS detection model.

[0093] S106. Input the test set into the trained PSS detection model to obtain the detection results.

[0094] It should be noted that after the model training is completed, the trained PSS detection model is finally tested using the test set obtained in step S103. The test set data is input into the model, and the model performs detection on the test set data based on the learned features and patterns, outputting the corresponding detection results. By analyzing these detection results, the performance and effectiveness of the model in practical applications can be comprehensively and objectively evaluated.

[0095] The innovations of this invention are mainly reflected in the following aspects:

[0096] 1. Innovative Algorithm: Based on in-depth research into the characteristics of the primary synchronization signal in 5 / 6G NR systems and existing detection algorithms, this invention innovatively proposes a CNN-based blind detection algorithm for 6G time-frequency synchronization, addressing the shortcomings of existing algorithms in terms of frequency offset resistance and algorithm complexity. This algorithm fully considers the high precision and high reliability requirements of 6G communication systems for time-frequency synchronization. By introducing CNN technology, it effectively improves the algorithm's performance in complex channel environments.

[0097] 2. Comprehensive Algorithm Flow Design: The proposed CNN-based blind detection algorithm for 6G time-frequency synchronization encompasses several key stages. First, based on the characteristics of PSS sequences and the multipath fading and noise characteristics of wireless channels, a PSS dataset is generated and appropriately partitioned to provide sufficient and representative data samples for model training. Second, a PSS detection model is built based on a Convolutional Neural Network (CNN), utilizing CNN's powerful feature extraction capabilities to accurately model PSS signals. Then, the model is trained and validated using training and validation sets to obtain the trained PSS detection model. Finally, the model is tested using a test set, and the PSS detection results are visualized for easy analysis and evaluation by users. The entire algorithm flow design is rigorous and complete, ensuring the algorithm's effectiveness and reliability.

[0098] 3. Introducing Advanced Technology to Break Through Traditional Thinking: This invention introduces Convolutional Neural Networks (CNNs) into the field of blind detection of the Master Synchronization Signal (PSS) in 5 / 6G communication, a significant innovation. Traditional PSS detection algorithms typically rely on manually designed PSS correlation detection algorithms. These algorithms require pre-setting a local PSS sequence and detecting the PSS by calculating the correlation peak between the received PSS signal and the local PSS sequence. However, this traditional method suffers from performance limitations in complex channel environments and has high algorithm design complexity. This invention utilizes a data-driven feature learning method, allowing the model to automatically learn the features of the PSS signal from a large amount of data, replacing manually designed detection algorithms. This breaks free from the constraints of traditional thinking and brings a new technological breakthrough to the field of blind PSS detection.

[0099] 4. Reduced sequence traversal overhead and accelerated detection speed: The CNN-based PSS blind detection algorithm of this invention automatically learns the features of the PSS signal through the model, eliminating the need to traverse and search all possible PSS sequences, thus significantly reducing the PSS sequence traversal overhead. This advantage enables the algorithm to quickly locate the target PSS signal during the detection process, greatly accelerating the detection speed and improving the real-time performance and efficiency of the communication system.

[0100] 5. Enhanced Robustness and Adaptability: The algorithm in this invention fully considers the impact of time-frequency offset and complex channel environments during design and training. Through extensive data training and optimization, the PSS detection algorithm exhibits stronger robustness to time-frequency offset, enabling accurate detection of PSS signals under varying degrees of time-frequency offset. Simultaneously, the algorithm demonstrates better adaptability to blind PSS detection in complex channels, achieving performance breakthroughs in complex scenarios such as low signal-to-noise ratio, large frequency offset, and dynamic channels. Even with significant frequency offset, the algorithm achieves good detection results, providing strong support for the stable operation of 5G / 6G communication systems in complex environments.

[0101] 6. Achieving rapid synchronization and improving system performance and user experience: The CNN-based time-frequency synchronization blind detection algorithm of this invention can quickly and accurately achieve downlink synchronization between the UE and the base station, effectively shortening the synchronization time and improving the overall performance of the communication system. Rapid synchronization enables user equipment to access the network faster, reducing communication interruptions and delays, thereby improving the user experience. At the same time, the high performance of this algorithm lays a solid foundation for the large-scale commercialization and widespread application of 5G / 6G communication systems.

[0102] Please refer to Figure 4 In one embodiment of this example, step S102 can be further refined to include the following steps:

[0103] S1021. Map the PSS transmitted signal dataset to the central subcarrier to complete the frequency domain mapping;

[0104] It's important to note that in wireless communication systems, signals are typically transmitted across multiple subcarriers using frequency division multiplexing (FDM). The PSS (Passive Signal Set) dataset is a set of digital signals containing specific information. To enable its transmission in the wireless spectrum, it needs to be mapped onto specific subcarriers. Selecting the center subcarrier for mapping is a common practice, as it exhibits relative stability and symmetry in the spectrum, reducing the impact of spectral edge effects. Through this mapping operation, the PSS dataset is transformed from its original digital signal form to subcarriers in the frequency domain, preparing it for subsequent frequency-to-time domain transformation and channel transmission.

[0105] S1022. Based on the inverse Fourier transform (IFFT), the frequency domain PSS signal is transformed into a time domain PSS signal, and a cyclic prefix is ​​added.

[0106] It's important to note that Fourier transform and inverse Fourier transform are crucial tools in signal processing. The Fourier transform converts a time-domain signal to a frequency-domain signal, while the inverse Fourier transform (IFFT) does the opposite, converting a frequency-domain signal back to a time-domain signal. In wireless communication, the PSS signal after frequency-domain mapping is a frequency-domain signal. To enable its transmission as a time-domain waveform in a practical wireless channel, it needs to be transformed back to a time-domain signal using IFFT.

[0107] In wireless communication, multipath propagation causes signal delays and spread, leading to inter-symbol interference (ISI). To overcome this problem, this embodiment adds a cyclic prefix to the beginning of each time-domain symbol. The cyclic prefix copies a portion of the signal from the end of the time-domain symbol to the beginning. Thus, at the receiver, even if the signal is time-spread due to multipath propagation, as long as the spread time does not exceed the length of the cyclic prefix, it will not affect the reception of the next symbol, effectively reducing ISI and improving the reliability of signal transmission.

[0108] S1023. Based on the wireless channel transmission characteristics, frequency offset, multipath channel fading effect and noise are added to the time-domain PSS signal to generate a PSS received signal dataset received by the user equipment (UE) after channel transmission.

[0109] It should be noted that in real-world wireless communication environments, due to factors such as the inconsistency of local oscillator frequencies between the transmitter and receiver, or the Doppler effect (when there is relative motion between the transmitter and receiver), a certain deviation (frequency offset) can occur between the received signal frequency and the transmitted signal frequency. This frequency offset affects the signal's phase and frequency characteristics, thus impacting demodulation and detection. To simulate a realistic channel environment, a frequency offset needs to be added to the time-domain PSS signal.

[0110] Furthermore, wireless signals encounter various obstacles during propagation, such as buildings and mountains, causing them to reach the receiver via multiple different paths—this is multipath propagation. Because signals from different paths travel different distances, their arrival times and phases at the receiver also differ. The superposition of these differences leads to random variations in the signal's amplitude and phase; this phenomenon is called multipath fading. Multipath fading severely impacts signal quality. To accurately simulate real-world channels, multipath fading effects need to be added to the time-domain PSS signal.

[0111] Furthermore, various noise sources exist in wireless communication environments, such as thermal noise and interference noise. Noise is superimposed on the signal, interfering with signal transmission and reception, and reducing the signal-to-noise ratio. To make the simulated channel environment more realistic, noise needs to be added to the time-domain PSS signal. By adding frequency offset, multipath channel fading effects, and noise, the time-domain PSS signal becomes a dataset of the actual PSS received signal at the user equipment (UE) side after channel transmission. This dataset more closely resembles the real communication scenario, providing a reliable data foundation for subsequent model training and testing.

[0112] Please refer to Figure 5In one embodiment of this invention, in steps S105 and S106, the PSS detection model processes the received PSS signal after transmission through the channel, mainly in two stages: a convolutional pooling stage and a linear regression stage. Through these two stages, the model can extract features from the received PSS signal and perform blind detection of the PSS signal. That is:

[0113] The convolutional pooling module sequentially uses three one-dimensional convolutional kernels of sizes 7, 5 and 3 to perform CNN convolution on the PSS received signal after channel transmission. After each convolution, the BatchNorm1d module and ReLU module are used to perform batch normalization and ReLU activation on the convolution result. Then, the MaxPool1d module with a pooling window size of 2*2 is used to perform pooling operation.

[0114] The PSS feature data output after three layers of convolutional pooling is further extracted using two linear regression modules with 512 and 128 neurons respectively. Finally, the PSS type can be extracted by a linear regression module with 3 neurons to complete the blind detection of PSS signals.

[0115] It's important to note that a one-dimensional convolutional kernel slides across the one-dimensional signal (here, the PSS received signal) to extract local features through convolution operations. Different kernel sizes have different receptive fields; larger kernels (e.g., 7) can capture a wider range of feature patterns in the signal, while smaller kernels (e.g., 3) are better at capturing local details. By using convolutional kernels of different sizes sequentially, the model can extract features from the PSS received signal at different scales, thus gaining a more comprehensive understanding of the signal's structure and patterns.

[0116] Batch standardization standardizes each batch of data, making the data distribution more stable and reducing internal covariate shift. During neural network training, as the network depth increases, the distribution of input data may change, leading to training difficulties and decreased model performance. Batch standardization adjusts the mean and variance of each batch of data to fixed values ​​(usually 0 and 1), making the input data distribution of each layer more stable, thereby accelerating model convergence and improving training effectiveness.

[0117] After batch standardization, ReLU (Rectified Linear Unit) is used for activation. ReLU is a commonly used activation function, mathematically expressed as f(x) = max(0,x). The ReLU activation function has non-linear characteristics, introducing non-linearity into neural networks, allowing them to learn and represent more complex functional relationships. Furthermore, the ReLU activation function is computationally simple and gradient calculation is convenient, avoiding the vanishing and exploding gradient problems during training, thus improving model training efficiency and performance.

[0118] The main purpose of pooling operations is to reduce the dimensionality of data, decrease computational cost, and retain the key features of the data. MaxPool1d is a max pooling operation that selects the largest value within the pooling window as the output. Pooling operations can effectively reduce redundant information in the data, enhance the robustness of the model, and prevent overfitting. In the PSS detection model, using a 2x2 pooling window can reduce the dimensionality of feature data to a certain extent while retaining important feature information in the signal.

[0119] The linear regression module is a fully connected layer that connects all neurons in the previous layer to all neurons in the current layer. Through the linear regression module, features extracted during the convolutional pooling stage can be further combined and transformed to extract higher-level, more abstract features. Linear regression modules with 512 and 128 neurons can gradually reduce the dimensionality of feature data while increasing the expressive power of features, enabling the model to better capture key features in the PSS signal.

[0120] The linear regression module with 3 neurons corresponds to the three possible PSS types. Through this module, the model maps the extracted features to the classification results of the PSS type. During training, the model learns from the input PSS received signal and its corresponding label (i.e., the true PSS type), adjusting the weights and biases of the linear regression module to accurately predict the PSS signal type. During testing, the model can directly output the predicted PSS type based on the input PSS received signal, thus achieving blind detection of the PSS signal.

[0121] In summary, the PSS detection model processes the received PSS signal after transmission through the channel through the collaborative work of a convolutional pooling module and a linear regression module. The convolutional pooling module extracts features from the received PSS signal at different scales using convolutional kernels of different sizes, batch normalization, ReLU activation, and pooling operations. The linear regression module further combines and transforms the extracted features to ultimately extract the PSS type, completing the blind detection of the PSS signal. This processing effectively utilizes the information in the received PSS signal, improving the accuracy and reliability of PSS signal detection.

[0122] In one embodiment of this example, step S103 can be further refined to include the following steps, mainly focusing on how to scientifically and rationally divide the dataset:

[0123] S1031. According to a preset division ratio, a portion of data is randomly selected from the PSS received signal dataset as a training set, and the data volume of the training set accounts for X% of the total data volume.

[0124] It's important to note that before splitting the dataset, the proportions of the training set, validation set, and test set need to be predetermined. The training set is the dataset used to train the model. The model continuously learns from the data in the training set to adjust its parameters, minimizing the loss function and thus learning the patterns and features in the data.

[0125] Randomly sampling data ensures that the training set covers all scenarios in the PSS received signal dataset, preventing the model from learning partial features due to biased data selection. For example, if the dataset contains signal data under different channel conditions, random sampling ensures that the training set includes samples under various channel conditions, enabling the model to learn signal characteristics under different channel environments.

[0126] X% is typically a relatively large percentage because the training set needs to have enough data for the model to learn effectively. Generally, X% ranges from 60% to 80%, with the specific value depending on the size of the dataset and the complexity of the model. A larger training set provides more information, helping the model learn more accurate features and patterns, but it also increases training time and computational resource consumption.

[0127] S1032. From the remaining data, a portion of the data is extracted as a validation set according to a preset ratio, wherein the data volume of the validation set accounts for Y% of the total data volume.

[0128] It's important to note that after extracting the training set, the remaining data will be used to divide the dataset into validation and test sets. The primary purpose of the validation set is to evaluate and fine-tune the model's performance during training. By evaluating the model's performance on the validation set, problems such as overfitting or underfitting can be identified promptly, allowing for adjustments to the model's parameters to improve its generalization ability.

[0129] Similar to the partitioning of the training set, the validation set also needs to be drawn from the remaining data according to a preset ratio. Determining this preset ratio requires considering the model tuning needs and the size of the dataset. Generally, the value of y% ranges from 10% to 20%.

[0130] The validation set needs to be large enough to accurately evaluate the model's performance. If the validation set is too small, the evaluation results may be inaccurate and fail to truly reflect the model's generalization ability. At the same time, the validation set should not be too large, otherwise it will reduce the amount of data in the training and test sets, affecting the model's training and evaluation effectiveness.

[0131] S1033. The remaining data that was not selected into the training set and validation set is used as the test set, and the data volume of the test set accounts for Z%; where X%+Y%+Z%=100%, X%>Y%>Z%, and in the process of partitioning, it is ensured that the data in the training set, validation set and test set are representative in terms of channel characteristics, noise level and multipath effect.

[0132] It's important to note that the test set is the dataset used for the final evaluation of model performance. After model training and tuning, evaluating the model using the test set allows us to see its performance on unseen data, thus assessing its generalization ability. The test set data should be data that the model has never encountered during training and tuning to ensure the objectivity and accuracy of the evaluation results.

[0133] z% is typically a small percentage, generally between 10% and 20%. The main purpose of the test set is to evaluate the final performance of the model, and it does not require a large amount of data. At the same time, to ensure the reliability of the evaluation results, the amount of data in the test set cannot be too small.

[0134] X%+Y%+Z%=100%, X%>Y%>z%: This is the basic principle of dataset partitioning, ensuring that the entire dataset is completely partitioned and that the proportions of the training, validation, and test sets are reasonable. The training set has the largest proportion to ensure that the model has enough data for learning; the validation set is the next largest, used for model tuning; and the test set has the smallest proportion, used for final evaluation.

[0135] When partitioning the dataset, it's crucial to ensure that the data in the training, validation, and test sets are representative in terms of channel characteristics, noise levels, and multipath effects. This means each set should include PSS received signal data under different channel conditions (e.g., different frequency offsets, different numbers of multipath propagation paths), different noise levels (e.g., different signal-to-noise ratios), and different multipath effects (e.g., different delay spreads). Only in this way can the model achieve good performance in various real-world scenarios, improving its generalization ability and practicality. For example, if the training set only contains high signal-to-noise ratio signal data while the test set contains low signal-to-noise ratio signal data, the model's performance on the test set may be poor because it hasn't learned the signal characteristics under low signal-to-noise ratio conditions.

[0136] In one embodiment of this example, before step S103, the method further includes:

[0137] S102.5. Perform time delay processing on each signal sample in the PSS received signal dataset to simulate the signal transmission delay between the base station and the UE at different distances, and generate multiple delayed versions of the signal sample;

[0138] It should be noted that this simulation addresses signal transmission delays between base stations and user equipment (UEs) at varying distances. In real-world communication scenarios, the time required for signal transmission varies depending on the distance between the base station and the UE, affecting the reception characteristics of the PSS signal. By processing the time delay, data that more closely resembles real-world scenarios can be generated, enhancing the model's adaptability to signals at different distances. This allows the model to learn the characteristic changes of the PSS signal at different distances, improving the model's detection accuracy in real-world deployments.

[0139] S102.6 Perform frequency offset processing on each signal sample in the PSS received signal dataset to simulate the signal frequency change under different Doppler effects and generate multiple frequency offset versions of the signal sample.

[0140] It should be noted that this simulation addresses signal frequency variations under different Doppler effects. When there is relative motion between the base station and the UE, the Doppler effect occurs, causing changes in the frequency of the received signal. By using frequency offset processing, signal samples containing different Doppler effects can be generated, enhancing the model's adaptability to the Doppler effect. This allows the model to identify and process PSS signals under different Doppler effects, improving the model's detection performance in high-speed moving scenarios.

[0141] S102.7 Perform signal-to-noise ratio (SNR) adjustment processing on each signal sample in the PSS received signal dataset. By adding Gaussian white noise of different intensities, simulate the signal transmission environment under different SNR conditions and generate multiple signal samples with different SNR levels.

[0142] It should be noted that this simulation addresses signal transmission environments under different signal-to-noise ratio (SNR) conditions. In actual communication, signals are subject to various noise interferences, and SNR is a crucial indicator of signal quality. By adjusting the SNR, signal samples under different noise levels can be generated, enhancing the model's adaptability to different noise environments. This allows the model to accurately detect PSS signals under varying noise conditions, improving its robustness and practicality.

[0143] S102.8 Perform amplitude adjustment processing on each signal sample in the PSS received signal dataset to simulate the attenuation or amplification effect of the signal during transmission and generate multiple signal samples with different amplitudes.

[0144] It's important to note the attenuation or amplification effects of analog signals during transmission. Signal amplitude changes due to factors such as path loss and antenna gain. Amplitude adjustment processing can generate signal samples of varying amplitudes, enhancing the model's adaptability to these amplitude variations. This allows the model to identify and process PSS signals of different amplitudes, improving its detection accuracy in practical applications.

[0145] S102.9 Perform phase rotation processing on each signal sample in the PSS received signal dataset to simulate the phase changes that may occur during signal transmission and generate multiple phase-rotated versions of the signal sample.

[0146] It is important to note that analog signals may undergo phase changes during transmission. Signals are affected by factors such as channel phase distortion and multipath propagation, leading to phase shifts. Phase rotation processing can generate signal samples with different phases, enhancing the model's adaptability to phase changes. This allows the model to identify and process PSS signals with different phases, improving its detection performance in complex channel environments.

[0147] S102.10. Merge all the new signal samples generated after the above processing to form an expanded PSS received signal dataset;

[0148] It should be noted that all the new signal samples generated after the above processing are merged to form an expanded PSS received signal dataset. Expanding the dataset increases the diversity and quantity of data, providing richer samples for model training and helping to improve the model's generalization ability and performance.

[0149] Expanding the dataset can prevent model overfitting and improve the model's adaptability and detection accuracy in different scenarios.

[0150] Correspondingly, on the other hand, S103 can be further refined to include the following steps:

[0151] The expanded PSS received signal dataset is divided into a training set and a validation set, and the original PSS received signal dataset is used as the test set.

[0152] It should be noted that the expanded training and validation sets contain more signal samples from different scenarios, enabling the model to learn more comprehensive and accurate features and improve the training effect of the model.

[0153] Using the original PSS received signal dataset as the test set ensures that the test set data is independent of the training and validation sets and more closely reflects real-world application scenarios. After model training and tuning, using the test set for final evaluation allows us to obtain the model's performance on unseen data, thus more accurately assessing the model's generalization ability and practicality. This partitioning method fully utilizes the expanded dataset to improve model training effectiveness, while testing on the original dataset ensures that the evaluation results are objective and reliable.

[0154] Although this invention uses terms such as CNN, PSS, and training set frequently, the possibility of using other terms is not excluded. These terms are used merely for the convenience of describing and explaining the essence of this invention; interpreting them as any additional limitation would contradict the spirit of this invention.

[0155] Example 2

[0156] This invention provides a time-frequency synchronized blind detection system based on CNN, the system comprising:

[0157] The first generation module is used to generate a dataset of PSS transmission signals sent by the base station based on the M-sequence characteristics of the primary synchronization signal PSS.

[0158] The second generation module is used to perform channel processing on the PSS transmitted signal dataset based on multipath fading and noise characteristics, and generate the PSS received signal dataset received by the user equipment (UE) side after channel transmission.

[0159] The data partitioning module is used to divide the PSS received signal dataset into a training set, a validation set, and a test set.

[0160] The model building module is used to build a PSS detection model based on a convolutional neural network (CNN).

[0161] The model training module is used to train and validate the PSS detection model using the training set and the validation set to obtain a trained PSS detection model.

[0162] The model detection module is used to input the test set into the trained PSS detection model to obtain the detection results.

[0163] Optionally, the second generation module is specifically used for:

[0164] The PSS transmitted signal dataset is mapped to the central subcarrier to complete the frequency domain mapping.

[0165] The frequency domain PSS signal is transformed into a time domain PSS signal based on the inverse Fourier transform (IFFT), and a cyclic prefix is ​​added.

[0166] Based on the characteristics of wireless channel transmission, frequency offset, multipath channel fading effect and noise are added to the time-domain PSS signal to generate a dataset of PSS received signals received by the user equipment (UE) after channel transmission.

[0167] Optionally, in S105 and S106, the processing procedure of the PSS detection model includes:

[0168] The convolutional pooling module sequentially uses three one-dimensional convolutional kernels to perform CNN convolution on the PSS received signal after channel transmission. After each convolution, the BatchNorm1d module and ReLU module are used to perform batch normalization and ReLU activation on the convolution result, and then the MaxPool1d module is used to perform pooling operation.

[0169] The PSS feature data output after three layers of convolutional pooling is further extracted using two linear regression modules, and finally the PSS type can be extracted by a single linear regression module to complete the blind detection of the PSS signal.

[0170] Optionally, the sizes of the three one-dimensional convolution kernels used in sequence are 7, 5, and 3, respectively;

[0171] The pooling window size of the MaxPool1d module is 2*2.

[0172] Optionally, the two linear regression modules used for further feature extraction have 512 and 128 neurons, respectively;

[0173] The linear regression module used to extract PSS types has 3 neurons.

[0174] Optionally, the data partitioning module is specifically used for:

[0175] S1031. According to a preset division ratio, a portion of data is randomly selected from the PSS received signal dataset as a training set, and the data volume of the training set accounts for X% of the total data volume.

[0176] S1032. From the remaining data, a portion of the data is extracted as a validation set according to a preset ratio, wherein the data volume of the validation set accounts for Y% of the total data volume.

[0177] S1033. The remaining data that was not selected into the training set and validation set is used as the test set, and the data volume of the test set accounts for Z%; where X%+Y%+Z%=100%, X%>Y%>Z%, and in the process of partitioning, it is ensured that the data in the training set, validation set and test set are representative in terms of channel characteristics, noise level and multipath effect.

[0178] Optionally, the system further includes a data expansion module for:

[0179] Time delay processing is performed on each signal sample in the PSS received signal dataset to simulate the signal transmission delay between base stations and UEs at different distances, generating multiple delayed versions of signal samples;

[0180] Each signal sample in the PSS received signal dataset is subjected to frequency shift processing to simulate signal frequency changes under different Doppler effects, generating multiple frequency-shifted versions of the signal sample.

[0181] Each signal sample in the PSS received signal dataset is subjected to signal-to-noise ratio (SNR) adjustment processing. By adding Gaussian white noise of different intensities, the signal transmission environment under different SNR conditions is simulated, generating multiple signal samples with different SNR levels.

[0182] Amplitude adjustment processing is performed on each signal sample in the PSS received signal dataset to simulate the attenuation or amplification effect of the signal during transmission, generating multiple signal samples with different amplitudes.

[0183] Each signal sample in the PSS received signal dataset is subjected to phase rotation processing to simulate the phase changes that may occur during signal transmission, generating multiple phase-rotated versions of the signal sample.

[0184] All the new signal samples generated after the above processing are merged to form an expanded PSS received signal dataset;

[0185] Correspondingly, the data partitioning module is specifically used for:

[0186] The expanded PSS received signal dataset is divided into a training set and a validation set, and the original PSS received signal dataset is used as the test set.

[0187] The above system can execute the methods provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the methods.

[0188] Example 3

[0189] Figure 6 This is a schematic diagram of the structure of a computer device provided in Embodiment 3 of the present invention. Figure 6 A block diagram of an exemplary computer device 12 suitable for implementing embodiments of the present invention is shown. Figure 6 The computer device 12 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0190] like Figure 6 As shown, the computer device 12 is represented in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and bus 18 connecting different system components (including system memory 28 and processing unit 16).

[0191] Bus 18 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0192] Computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0193] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (…). Figure 6 Not shown; usually referred to as a "hard drive"). Although Figure 6Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. Memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0194] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 42 typically perform the functions and / or methods described in the embodiments of the present invention.

[0195] Computer device 12 can also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), and with one or more devices that enable a user to interact with the computer device 12, and / or with any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via input / output (I / O) interface 22. Furthermore, computer device 12 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 20. As shown, network adapter 20 communicates with other modules of computer device 12 via bus 18. It should be understood that, although... Figure 6 As not shown, it can be used in conjunction with computer device 12 with other hardware and / or software modules, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0196] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing the CNN-based time-frequency synchronous blind detection method provided in the embodiments of the present invention.

[0197] Example 4

[0198] Embodiment 4 of the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the CNN-based time-frequency synchronous blind detection method provided in all embodiments of the present invention.

[0199] Any combination of one or more computer-readable media may be used. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device.

[0200] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0201] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0202] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0203] Finally, it should be noted that although the above embodiments have been described in the description and drawings of this invention, this should not limit the scope of patent protection of this invention. Any technical solutions that are based on the essential concept of this invention, utilize the content described in the description and drawings of this invention to make equivalent structural or procedural substitutions or modifications, as well as the direct or indirect application of the technical solutions of the above embodiments to other related technical fields, are all included within the scope of patent protection of this invention.

Claims

1. A time-frequency synchronized blind detection method based on CNN, characterized in that, The method includes: S101. Based on the M-sequence characteristics of the primary synchronization signal PSS, generate a dataset of PSS transmission signals sent by the base station. S102. Based on multipath fading and noise characteristics, channel processing is performed on the PSS transmitted signal dataset to generate the PSS received signal dataset received by the user equipment (UE) after channel transmission. S103. Divide the PSS received signal dataset into a training set, a validation set, and a test set; S104. Build a PSS detection model based on convolutional neural network (CNN); S105. The PSS detection model is trained and validated using the training set and the validation set to obtain a trained PSS detection model. S106. Input the test set into the trained PSS detection model to obtain the detection results.

2. The CNN-based time-frequency synchronous blind detection method according to claim 1, characterized in that, S102 includes: S1021. Map the PSS transmitted signal dataset to the central subcarrier to complete the frequency domain mapping; S1022. Based on the inverse Fourier transform (IFFT), the frequency domain PSS signal is transformed into a time domain PSS signal, and a cyclic prefix is ​​added. S1023. Based on the wireless channel transmission characteristics, frequency offset, multipath channel fading effect and noise are added to the time-domain PSS signal to generate a PSS received signal dataset received by the user equipment (UE) after channel transmission.

3. The CNN-based time-frequency synchronized blind detection method according to claim 1, characterized in that, In S105 and S106, the processing procedure of the PSS detection model includes: The convolutional pooling module sequentially uses three one-dimensional convolutional kernels to perform CNN convolution on the PSS received signal after channel transmission. After each convolution, the BatchNorm1d module and ReLU module are used to perform batch normalization and ReLU activation on the convolution result, and then the MaxPool1d module is used to perform pooling operation. The PSS feature data output after three layers of convolutional pooling is further extracted using two linear regression modules, and finally the PSS type can be extracted by a single linear regression module to complete the blind detection of the PSS signal.

4. The CNN-based time-frequency synchronous blind detection method according to claim 3, characterized in that, The sizes of the three one-dimensional convolutional kernels used in sequence are 7, 5, and 3, respectively; The pooling window size of the MaxPool1d module is 2*2.

5. The CNN-based time-frequency synchronous blind detection method according to claim 4, characterized in that, The two linear regression modules used for further feature extraction had 512 and 128 neurons, respectively. The linear regression module used to extract PSS types has 3 neurons.

6. The CNN-based time-frequency synchronized blind detection method according to claim 1, characterized in that, S103 includes: S1031. According to a preset division ratio, a portion of data is randomly selected from the PSS received signal dataset as a training set, and the data volume of the training set accounts for X% of the total data volume. S1032. From the remaining data, a portion of the data is extracted as a validation set according to a preset ratio, wherein the data volume of the validation set accounts for Y% of the total data volume. S1033. The remaining data that was not selected into the training set and validation set is used as the test set, and the data volume of the test set accounts for Z%; where X%+Y%+Z%=100%, X%>Y%>Z%, and in the process of partitioning, it is ensured that the data in the training set, validation set and test set are representative in terms of channel characteristics, noise level and multipath effect.

7. The CNN-based time-frequency synchronized blind detection method according to claim 1, characterized in that, Prior to S103, the method further includes: S102.

5. Perform time delay processing on each signal sample in the PSS received signal dataset to simulate the signal transmission delay between the base station and the UE at different distances, and generate multiple delayed versions of the signal sample; S102.6 Perform frequency offset processing on each signal sample in the PSS received signal dataset to simulate the signal frequency change under different Doppler effects and generate multiple frequency offset versions of the signal sample. S102.7 Perform signal-to-noise ratio (SNR) adjustment processing on each signal sample in the PSS received signal dataset. By adding Gaussian white noise of different intensities, simulate the signal transmission environment under different SNR conditions and generate multiple signal samples with different SNR levels. S102.8 Perform amplitude adjustment processing on each signal sample in the PSS received signal dataset to simulate the attenuation or amplification effect of the signal during transmission and generate multiple signal samples with different amplitudes. S102.9 Perform phase rotation processing on each signal sample in the PSS received signal dataset to simulate the phase changes that may occur during signal transmission and generate multiple phase-rotated versions of the signal sample. S102.

10. Merge all the new signal samples generated after the above processing to form an expanded PSS received signal dataset; Correspondingly, S103 includes: The expanded PSS received signal dataset is divided into a training set and a validation set, and the original PSS received signal dataset is used as the test set.

8. A time-frequency synchronized blind detection system based on CNN, characterized in that, The system includes: The first generation module is used to generate a dataset of PSS transmission signals sent by the base station based on the M-sequence characteristics of the primary synchronization signal PSS. The second generation module is used to perform channel processing on the PSS transmitted signal dataset based on multipath fading and noise characteristics, and generate the PSS received signal dataset received by the user equipment (UE) side after channel transmission. The data partitioning module is used to divide the PSS received signal dataset into a training set, a validation set, and a test set. The model building module is used to build a PSS detection model based on a convolutional neural network (CNN). The model training module is used to train and validate the PSS detection model using the training set and the validation set to obtain a trained PSS detection model. The model detection module is used to input the test set into the trained PSS detection model to obtain the detection results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the CNN-based time-frequency synchronous blind detection method as described in any one of claims 1-7.

10. A computer-readable storage medium having computer-executable instructions stored thereon, characterized in that, The computer-executable instructions are executed by a computer processor to implement the CNN-based time-frequency synchronous blind detection method as described in any one of claims 1-7.

Images