Detection method, system, electronic device, storage medium and program

Abstract

The application discloses a detection method, a system, an electronic device, a storage medium and a program. The detection method comprises the following steps: sending predefined physical random access channel configuration information to a user end; receiving a physical random access channel sequence generated by the user end according to the physical random access channel configuration information; performing user detection by using the physical random access channel sequence to obtain a user detection probability; and determining that a user is detected if the user detection probability is greater than a preset detection threshold value, or determining that a user is not detected. The technical scheme provided by the application can solve the technical problem that a base station is falsely detected due to interference factors in the prior art.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a detection method, system, electronic device, storage medium, and program. Background Technology

[0002] In communication technology, communication between the user terminal and the base station currently uses technologies such as Long Term Evolution (LTE) and 5G New Radio (NR), all of which require random access technology.

[0003] In random access technology, the user terminal and the base station establish communication through the Physical Random Access Channel (PRACH). The user terminal needs to send a preamble to the base station first. The base station detects the preamble and sends a random access response to the user terminal after detecting the preamble.

[0004] However, due to interference factors such as multipath effect, noise interference, frequency offset, clock offset, non-ideal filtering, hardware nonlinearity and adjacent channel interference, the physical random access channel may experience false detections at the base station. Summary of the Invention

[0005] This application provides a detection method, system, electronic device, storage medium, and program, which aims to effectively solve the technical problem of false detections of base stations caused by interference factors in random access technology.

[0006] According to a first aspect of this application, this application provides a detection method, comprising: sending predefined physical random access channel configuration information to a user terminal; receiving a physical random access channel sequence generated by the user terminal according to the physical random access channel configuration information; performing user detection using the physical random access channel sequence to obtain a user detection probability; if the user detection probability is greater than a preset detection threshold, determining that a user has been detected, otherwise determining that a user has not been detected.

[0007] In one possible design, the user detection using the physical random access channel sequence includes: standardizing the physical random access channel sequence; and using a pre-trained detection model to perform user detection on the standardized physical random access channel sequence to obtain the user detection probability for each detection window.

[0008] In one possible design, the pre-training method of the detection model includes: acquiring a dataset; standardizing the dataset; determining the network structure of an initial detection model, wherein the network structure of the detection model includes: a variable one-dimensional convolutional layer, a first max-pooling layer, a fixed one-dimensional convolutional layer, a second max-pooling layer, a fixed fully connected layer, and a variable fully connected layer; and training the initial detection model using a preset training strategy and the standardized dataset to obtain the detection model.

[0009] In one possible design, the dataset includes: segmentation index data obtained by segmenting the cyclic shift index of the Zadov-Chu sequence according to a predetermined rule, wherein different segmentation index data correspond to different preset values, and the length of the segmentation index data is the same as the length of the detection window.

[0010] In one possible design, the formula for standardizing the dataset includes:

[0011] ; in, Indicates the first i The raw data of each user's detection window; It is the mean of the raw data in this detection window; It is the variance of the raw data of this detection window; It is the standardized data of the original data; i This represents the user index, with values ​​ranging from 1 to maxUserNumber, which is the maximum number of users that can be supported on a single root sequence under a specific physical random access channel configuration.

[0012] In one possible design, the preset training strategy includes an adaptive momentum optimization algorithm, and when the initial detection model is trained using the adaptive momentum optimization algorithm, the training strategy includes a pre-training phase and an optimization phase for the initial detection model, wherein the pre-training phase is constrained by a first learning rate, the optimization phase is constrained by a second learning rate, and the first learning rate is greater than the second learning rate.

[0013] In one possible design, the pre-training method of the detection model further includes: performing a performance evaluation on the detection model to obtain an evaluation value; if the evaluation value is less than a preset evaluation threshold, the detection model passes the evaluation and is saved; if the evaluation value is greater than or equal to the preset evaluation threshold, the detection model is optimized until the evaluation value of the detection model is less than the preset evaluation threshold.

[0014] In one possible design, the formula for evaluating the performance of the detection model includes: ; in, and The preset weighting coefficients, To validate the dataset; FNM is the false negative metric; FPM is the false positive metric. The calculation formula for the missed detection index includes: ; The calculation formula for the false detection index includes: ; in, y This represents the probability of detecting a user in each detection window of the detection model, with a value ranging from zero to one.

[0015] In one possible design, the pre-training method of the detection model further includes: establishing a joint loss function using the missed detection index and the false detection index; and using the loss function to constrain the missed detection rate and false detection rate of the detection model.

[0016] In one possible design, the formula for the joint loss function includes: ; in, denoted as the loss value, where x is the true label value of the dataset.

[0017] In one possible design, the expanded formula for the joint loss function based on the true label values ​​satisfies the following condition:

[0018] Where batchSize represents the number of samples input to the initial detection model at one time during the training process of the detection model; maxUserNumber represents the maximum number of users that a related sequence can support for reuse.

[0019] In one possible design, the detection method further includes: calculating the loss value corresponding to the detection window and backpropagating the calculation result to update the parameters of the detection model.

[0020] In one possible design, the detection method further includes: after saving the detection model, evaluating the detection model using the validation dataset within a preset period to obtain a periodic evaluation value; if the periodic evaluation value is greater than or equal to the preset evaluation threshold, optimizing the detection model until the periodic evaluation value of the detection model is less than the preset evaluation threshold.

[0021] In one possible design, the step of optimizing the detection model includes: acquiring the dataset and acquiring historical datasets of successful or failed random access to user terminals; and using the union of the dataset and the historical datasets to continue training the detection model to optimize it.

[0022] For example, after the base station sends the PRACH configuration, it uses a pre-trained model to detect the PRACH signals fed back by the user. When the gNB detects a user within a PRACH detection window using its detection model, it sends an MSG2 message to the corresponding user. If the user is a real user, the base station will receive an MSG3 response; if the user is a dummy user, the base station will not receive an MSG3 response. The gNB marks the window corresponding to the user who receives the MSG3 response as a real user, and the window corresponding to the user who does not receive the MSG3 response as a dummy user. The historical dataset includes the PRACH configuration sent by the base station, the MSG2 messages sent by the base station, real user labels, and dummy user labels.

[0023] In one possible design, the step of optimizing the detection model further includes: evaluating the optimized detection model to obtain an optimized evaluation value; if the optimized evaluation value is less than the preset evaluation threshold, the detection model passes the evaluation and is saved; if the optimized evaluation value is greater than or equal to the preset evaluation threshold, the detection model is further optimized until the optimized evaluation value of the detection model is less than the preset evaluation threshold.

[0024] In one possible design, the pre-training method for the detection model further includes: training different detection models using different datasets, wherein the different datasets have different physical random access channel configuration data; and storing the different detection models trained using the different datasets, wherein each detection model has a unique identifier.

[0025] In one possible design, the user detection of the standardized physical random access channel sequence using a pre-trained detection model includes: matching the physical random access channel configuration information with the corresponding physical random access channel configuration data, and using a detection model with a unique identifier corresponding to the physical random access channel configuration data to perform user detection on the standardized physical random access channel sequence.

[0026] According to a second aspect of this application, this application also provides a detection method, comprising: receiving physical random access channel configuration information sent by a communication terminal; parsing the physical random access channel configuration information to generate a physical random access channel sequence; and sending the physical random access channel sequence to the communication terminal.

[0027] According to a third aspect of this application, this application also provides a detection system, comprising: an information sending module for sending predefined physical random access channel configuration information to a user terminal; an information receiving module for receiving a physical random access channel sequence generated by the user terminal according to the physical random access channel configuration information; and a user detection module for performing user detection using the physical random access channel sequence; wherein, if the user detection probability is greater than a preset detection threshold, a detected user is determined; otherwise, an undetected user is determined.

[0028] According to a fourth aspect of this application, this application also provides a detection system, comprising: a configuration information receiving module for receiving physical random access channel configuration information sent by a communication terminal; a configuration information parsing module for parsing the physical random access channel configuration information and generating a physical random access channel sequence; and a channel sequence sending module for sending the physical random access channel sequence to the communication terminal.

[0029] According to a fifth aspect of this application, this application also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements any of the detection methods described above.

[0030] According to a sixth aspect of this application, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the detection methods described above.

[0031] According to another aspect of this application, this application also provides a computer program for performing any of the detection methods described above.

[0032] Through one or more embodiments of the above embodiments in this application, at least the following technical effects can be achieved: In the technical solution disclosed in this application, after the terminal sends a random access channel sequence to the base station, the base station can perform user detection after receiving the physical random access channel sequence, and further send an acknowledgment message to the detected user. Through the user's response message, it can further determine whether the detected user is a user terminal, and use the user response message to reduce the probability of false detection by the base station. Attached Figure Description

[0033] The technical solution and other beneficial effects of this application will become apparent from the following detailed description of specific embodiments in conjunction with the accompanying drawings.

[0034] Figure 1 This is a schematic diagram of random access in existing technology; Figure 2 A schematic diagram of the correlation peak generated by the correlation between the first root sequence and the PRACH signal in random access; Figure 3 A schematic diagram of the correlation peak generated by the correlation between the second root sequence and the PRACH signal in random access; Figure 4 A schematic diagram of spurious peaks caused by spectral leakage during inverse Fourier transform in random access; Figure 5 Here is a flowchart of the PRACH detection algorithm in the prior art; Figure 6 A flowchart of the detection method provided in the embodiments of this application; Figure 7 A flowchart of the user detection steps in the detection method provided in the embodiments of this application; Figure 8 A flowchart illustrating the pre-training process of the detection model in the detection method provided in this application embodiment; Figure 9 A flowchart illustrating the pre-training process of the detection model in the detection method provided in this application embodiment; Figure 10 This is a schematic diagram illustrating the data interaction between the base station and the user terminal in the detection method provided in the embodiments of this application; Figure 11 A flowchart of a detection method provided in another embodiment of this application; Figure 12 A framework diagram of the detection system provided in the embodiments of this application; Figure 13 A framework diagram of a detection system provided in another embodiment of this application; Figure 14 A schematic block diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0035] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0036] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the term "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Furthermore, the character " / " in this document, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0037] Random access in LTE and 5G new radio access technologies uses the ZC (Zadoff-Chu) sequence as a preamble. The contention-based random access procedure is as follows: Figure 1 As shown, the user equipment (UE) sends first information to the base station, which is the preamble for the physical random access channel (PRACH). After receiving the first control information, the base station detects the preamble and, upon detection, sends second control information back to the UE, which is the random access response (RAR). After receiving the RAR, the UE sends third control information to the base station, which is a radio resource control (RRC) request. After receiving the third control information, the base station sends fourth control information to the UE, which is a contention resolution message, thus completing the random access procedure and establishing or restoring an RRC connection with the UE.

[0038] However, during the entire random access process, due to interference in the physical random access channel, in a real wireless environment, factors such as multipath effects, noise interference, frequency offset, clock offset, non-ideal filtering, hardware nonlinearity, and adjacent channel interference can cause false detections at the base station, where interference signals are mistaken for user-initiated access requests. When a false detection occurs, the base station will execute the following scheme: 1) The base station needs to handle invalid access requests generated by false detections, which increases the processing load of the base station; 2) This will cause the base station to allocate resources to false detection users, wasting valuable uplink and downlink resources; 3) Processing invalid random access requests increases the signaling overhead of the system, especially the overhead of the first response to the preamble in the second control information, thus reducing system efficiency; 4) Processing invalid random access requests will increase the base station's energy consumption.

[0039] Generally, to reduce the false detection rate of PRACH, the PRACH detection threshold can be increased or the initial power of the user equipment (UE) accessing the PRACH can be increased. However, increasing the PRACH detection threshold may reduce the probability of successful access for cell edge users and increase the false detection rate; increasing the UE access power will lead to increased power consumption of the UE and interference with other channels.

[0040] Specifically, ZC sequences exhibit good autocorrelation and cross-correlation properties. For the ZC root sequence (RS)... Its cyclic shift sequence It is also a ZC sequence, in which This represents the length of the ZC sequence, and CV represents the cyclic shift value. The sequence... and When performing cross-correlation calculations, L RA -CV The location produces a correlation peak. However, for different root sequences... , convert the sequence and its cyclic shift sequence and Performing cross-correlation calculations will not produce correlation peaks. These autocorrelation and cross-correlation properties of the ZC sequence have been effectively utilized in the random access preamble detection process.

[0041] However, various interferences in real wireless environments can affect the cross-correlation results of ZC sequences, leading to spurious peaks during cross-correlation operations between different root sequences and PRACH signals. Furthermore, non-ideal point-count Fast Fourier Transform (FFT) and Inverse Fast Fourier Transform (IFFT) can cause spectral leakage, directly affecting the cross-correlation characteristics between root sequences. This results in spurious peaks when cross-correlation operations are performed between root sequences in the frequency domain and transformed to the time domain. Base stations may misidentify spurious peaks as users, thus causing various negative impacts on the wireless access network system.

[0042] After the user equipment successfully determines the PRACH configuration information, it will select its dedicated root sequence and cyclic shift value and send the corresponding preamble sequence. For the base station, all possible root sequences must be traversed, and cross-correlation operations must be performed with the received PRACH signal. If the received preamble sequence is generated from a different root sequence than that used for detection, potential interference between root sequences will cause glitches in the correlation sequence; if the received preamble sequence is generated from the same root sequence as that used for detection, energy leakage caused by IFFT may also lead to spurious peaks in adjacent detection windows.

[0043] For cases where the preamble sequence is generated differently from the root sequence used for detection, consider the condition that the cyclic shift index of the short format zero-correlation zone configuration (ZCZC) in 3GPP TS 38.211 (3GPP: 3rd generation partnership project; referring to the TS 38.211 standard, TS 38.211 is the physical layer channel and modulation standard in 3GPP) is 3, and assume that the system uses PRACH format B4. In this case, the maximum number of root sequences supported by the system is 3. Each individual root sequence supports multiplexing access for a maximum of 23 users. Assume that a UE (user equipment) with user ID 26 initiates a random access request. This UE selects the second root sequence to generate a PRACH signal. After receiving the PRACH signal, the gNB performs cross-correlation operations on the cyclic root sequence and the received PRACH signal in sequence. When the first root sequence and the received preamble sequence are cross-correlated, due to potential interference in the root sequence, the cross-correlation operation between the first root sequence and the PRACH signal will produce spikes within the detection window. When the spike signal energy exceeds a threshold, a spurious peak is formed, such as... Figure 2 As shown, the presence of these spurious peaks makes it difficult to directly eliminate them using a single threshold judgment method; however, the second root sequence is the root sequence selected by the UE to generate the preamble sequence. Performing a cross-correlation operation between this sequence and the received signal will not produce obvious spurious peaks, but will only detect a larger correlation peak signal within the detection window corresponding to user ID 26, such as... Figure 3 As shown.

[0044] For cases where the preamble sequence is generated from the same root sequence used for detection, considering a ZCZC cyclic shift index of 13, and assuming the system uses PRACH format B4, a preamble format defined in 5G NR, after the base station performs frequency domain correlation using the root sequence and the received signal, the signal is transformed to the time domain and a correlation sequence is obtained through IFFT. Observing the correlation sequence generated between the root sequence corresponding to the actual user and the received signal, it was found that due to potential spectral leakage during the IFFT transformation, the leaked spectral energy increases accordingly with increasing signal-to-noise ratio (SNR). This leaked energy may cause interference between detection windows, leading to false detections in detection windows without signal transmission. Especially under high SNR conditions, the spectral leakage energy may exceed the set detection threshold, potentially causing significant spurious peaks in adjacent detection windows of the actual user. The appearance of these spurious peaks increases the risk of false detections, thus affecting the performance and reliability of the random access channel. Figure 4 The paper details how, under high signal-to-noise ratio conditions, spectral leakage caused by IFFT leads to energy leakage from the (m+1)th detection window containing real users to the mth detection window, resulting in false detections in the mth detection window.

[0045] Threshold-based PRACH detection algorithms, such as Figure 5 As shown, the received time-domain signal is processed through the Lowphy (lower physical layer) and converted to the frequency domain. A preamble sequence is extracted in the frequency domain, and then cross-correlation is performed between the preamble sequence and each root sequence. This cross-correlation is then converted back to the time domain using an IFFT transform. The maximum correlation peak value for each detection window is then found and compared to a preset threshold. If the maximum correlation peak value is greater than the threshold, a user is considered to exist within that detection window; otherwise, no user is considered to exist within that detection window.

[0046] In order to reduce at least one of the negative impacts of high false alarm rates caused by random access, such as resource waste, increased system load, and reduced energy efficiency, embodiments of this application propose a detection method, system, electronic device, storage medium, and program.

[0047] Figure 6 The illustrated detection method provided in this application embodiment can be applied to network device and terminal communication. For example, the network device can be implemented as a base station (not limited to this), such as in 5G communication, where the base station side can be a gNB (gNodeB, next-generation base station). The detection method includes: S101. Send predefined physical random access channel configuration information to the user terminal.

[0048] Accordingly, the UE receives the PRACH configuration information in step S101. The UE can parse this information to determine the parameters of the random access signal.

[0049] Based on the parsed PRACH configuration, the UE can select a suitable Preamble ID and send the corresponding PRACH sequence.

[0050] For example, the gNB sends predefined PRACH configuration information to the UE via the SIB (system information block).

[0051] S102. Receive the physical random access channel sequence generated by the user terminal based on the physical random access channel configuration information.

[0052] S103. Use the physical random access channel sequence to perform user detection and obtain the user detection probability.

[0053] For example, please refer to Figure 7 User detection using physical random access channel sequences includes: S1031. Standardize the physical random access channel sequence.

[0054] S1032. Use the pre-trained detection model to perform user detection on the standardized physical random access channel sequence, and obtain the detection probability of the detected user for each detection window.

[0055] In step S1031, the purpose of the standardization process is to eliminate the power differences that exist when the signals of different users arrive at the base station in the real environment. The standardization process can be performed using the following formula: ; in, Indicates the first i The raw data of each user's detection window; It is the mean of the raw data in this detection window; It is the variance of the raw data of this detection window; It is the standardized version of the original data; i This represents the user index, with values ​​ranging from 1 to maxUserNumber, which is the maximum number of users that can be supported on a single root sequence under a specific physical random access channel configuration.

[0056] The formula for this standardization process represents the first... iPower normalization is performed on the detection windows of each user. Using the above normalization formula, the data from each detection window will be transformed into a normalized form with zero mean and unit standard deviation, thus providing a consistent data input range for model training and enhancing the model's generalization ability to signals with different power levels.

[0057] In step S1032, the detection model is obtained through pre-training and deployed on the base station side. Before deployment to the gNB, the detection model needs to undergo a pre-training phase. In this phase, a neural network model suitable for PRACH signal detection needs to be designed first, and then the model is trained using simulated or laboratory PRACH signal datasets to learn the signal characteristics and optimize the detection algorithm. (See also...) Figure 8 The pre-training process includes dataset collection, data preprocessing, network structure design, training strategy determination, performance evaluation, and model saving.

[0058] For example, please refer to Figure 9 The pre-training methods for detection models include: S1131. Obtain the dataset; S1231. Standardize the dataset; S1331. Determine the network structure of the initial detection model, wherein the network structure of the detection model includes: a variable one-dimensional convolutional layer, a first max pooling layer, a fixed one-dimensional convolutional layer, a second max pooling layer, a fixed fully connected layer, and a variable fully connected layer. S1431. Train the initial detection model using the preset training strategy and the standardized dataset to obtain the detection model.

[0059] For example, in step S1131, the dataset includes: segmentation index data obtained by segmenting the cyclic shift index of the Zadov-Chu sequence according to a predetermined rule, wherein different segmentation index data correspond to different preset values, and the length of the segmentation index data is the same as the length of the detection window.

[0060] The classification of the dataset is based on the number of PRACH detection windows specified in the 3GPP protocol, ensuring that the model can adapt to different PRACH configurations.

[0061] Specifically, the dataset is subdivided according to the ZCZC cyclic shift index defined in the TS38.211 / TS36.211 standards. Different cyclic shift indices correspond to different NCS values, which determine the detection window length of the PRACH signal.

[0062] Since this stage involves offline learning and pre-training of the detection model, the dataset for this stage can be prepared in advance. For a specific cell, its PRACH format and cyclic shift index can generally be determined beforehand. The pre-training stage can extract data applicable to the cell configuration from the pre-prepared dataset. These pre-training datasets include system-level simulation data and experimental data collected in a laboratory environment. This data is used for the initial training of the model, helping the model learn the basic characteristics of the PRACH signal.

[0063] In step S1231, the scheme for standardizing the dataset can be found in the formula for standardizing the dataset in the above embodiments.

[0064] In step S1331, the initial detection model consists of six layers, as detailed below: The first layer is a flexible one-dimensional convolutional layer. One-dimensional convolutional layers in deep learning are mainly used to process sequential data, effectively extracting local features from the sequence. Furthermore, changes in the position of the input features within the sequence do not affect the output, thus enabling the identification of multiple relevant peak features, i.e., multi-user recognition. In this embodiment, this layer dynamically adjusts the size and stride of the convolutional kernel based on the size of the detection window to adapt to different PRACH configurations. The number of convolutional kernels is fixed at 32, and the ReLU activation function is used for non-linear mapping.

[0065] The second layer is the max-pooling layer, which is used for dimensionality reduction and feature extraction in this embodiment. By extracting key peak features, this layer can reduce the size of the relevant sequence, reduce computational complexity and memory usage, and also enhance feature representation, making the model focus more on the peak features of the relevant sequence. This is consistent with the principle of correlation peak detection of PRACH signals.

[0066] The third layer is a fixed-conv1d convolutional layer, independent of the PRACH configuration. It has 64 kernels of size 3 and a stride of 1, and also uses the ReLU activation function. This layer further processes features, enhancing the model's ability to handle interference between detection windows.

[0067] The fourth layer is the second max pooling layer. This layer has a similar function to the second layer, and can further extract and enhance peak features.

[0068] The fifth layer is a fixed fully connected layer (fixed-FNN). This layer maps the features extracted by the convolutional layers to a fixed output space, outputting a feature vector with a fixed dimension of 1×512.

[0069] The sixth layer is a flexible fully connected layer (flexible-FNN). This layer combines and linearly transforms the feature vectors of the fixed output space of the fifth layer, mapping the 512-dimensional feature vectors to the output dimension of the number of users under the specific PRACH configuration. It outputs the predicted probability value of the existence of users in each detection window, with the dimension of [1 × maximum number of users].

[0070] In step S1431, the training strategy may use the Adam optimizer (adaptive moment estimation) to pre-train the initial detection model. The preset training strategy includes: using a preset adaptive momentum optimization algorithm to pre-train and optimize the initial detection model. When pre-training the initial detection model, the first learning rate-constrained adaptive momentum optimization algorithm is used, and when optimizing the initial detection model, the second learning rate-constrained adaptive momentum optimization algorithm is used. The first learning rate is greater than the second learning rate.

[0071] The first learning rate can be used in the pre-training stage, using a higher learning rate to quickly converge and learn general features on the dataset; the second learning rate can be used in the fine-tuning stage to finely adjust the model parameters.

[0072] In some embodiments, the pre-training method for the detection model further includes: The detection model is evaluated to obtain evaluation values; If the evaluation value is less than the preset evaluation threshold, the detection model passes the evaluation and is saved. If the evaluation value is greater than or equal to the preset evaluation threshold, the detection model is optimized until the evaluation value of the detection model is less than the preset evaluation threshold.

[0073] In this embodiment, to fine-tune the model to adapt to the actual wireless environment of gNB deployment, the system will collect real PRACH data generated in actual deployments. Finally, model evaluation will use historical datasets from real-world environments, which are used to verify the model's performance and accuracy in practical applications. Through this meticulous dataset classification and usage strategy, this invention ensures that the PRACH signal detection model has high generalization ability and accuracy in different environments.

[0074] The formulas for evaluating the performance of the detection model include: ; in, and The preset weighting coefficients, To validate the dataset; FNM is the false negative metric; FPM is the false positive metric.

[0075] To balance the false negative rate and false positive rate of the detection model, this embodiment defines a joint loss function related to the false negative rate and false positive rate. This function comprehensively considers both the False Negatives Metric (FNM) and the False Positives Metric (FPM), where the False Negatives Rate (FNR) and False Positives Rate (FPR) are defined as follows: ; Where x represents the actual label value, This indicates the prediction result of the detection model after the judgment, where 0 represents no user and 1 represents a user.

[0076] The formulas for calculating the missed detection index include: ; The formulas for calculating false inspection indicators include: ; in, y This represents the probability of detecting a user in each detection window of the detection model, with a value ranging from zero to one.

[0077] In some embodiments, the pre-training method for the detection model further includes: establishing a joint loss function using false negative and false positive indicators; and using the loss function to constrain the false negative and false positive rates of the detection model.

[0078] For example, FNM represents the missed detection index, i.e., the UE sends a PRACH signal, but the gNB does not detect the PRACH preamble. In this case, the expected model prediction result is close to 1, meaning the probability of detecting a user is higher, and FNM approaches 0; conversely, if the model prediction result is closer to 0, meaning the probability of detecting no user is higher, FNM approaches +∞. FPM represents the false detection index, i.e., the UE does not send a PRACH signal, but the gNB detects the PRACH preamble. In this case, the expected model prediction result is closer to 0, meaning the probability of detecting no user is higher, and FPM approaches 0; conversely, if the model prediction result is closer to 1, meaning the probability of detecting a user is higher, FPM approaches +∞. Then, FNM and FPM are combined into the loss function of the model. The larger the loss function calculated based on the model output value, the less accurate the model's prediction at that point is. In this case, the larger the gradient calculated based on the loss function, the faster the model parameters descent towards the direction of accurate prediction under the action of the Adam optimization algorithm. Joint Loss Function Defined as: ; And in the joint loss function Calculated using the actual label value x, the function can be expanded as follows:

[0079] Here, `batchSize` represents the number of samples input to the initial detection model at one time during the detection model training process; `maxUserNumber` represents the maximum number of users that a relevant sequence can support for reuse. This function can be used to balance the false negative rate and the false positive rate. The formula means that for each detection window, FNM or FPM is calculated once based on whether a PRACH signal is sent in that detection window. When a PRACH signal is sent in the detection window, FNM is calculated; otherwise, FPM is calculated.

[0080] In some embodiments, the detection method further includes: calculating the loss value of the corresponding detection window and backpropagating the calculation result to update the parameters of the detection model.

[0081] Since each PRACH detection window is treated as an independent binary classification problem during model training, the entire model output involves multiple binary classification problems. Therefore, it is necessary to use [a specific method / mechanism] during the forward propagation phase. The loss value is calculated for each detection window so that the model parameters can be updated during the backpropagation phase.

[0082] After saving the detection model, the model is evaluated using the validation dataset within a preset period to obtain the periodic evaluation value. If the periodic evaluation value is greater than or equal to the preset evaluation threshold, the detection model is optimized until the periodic evaluation value of the detection model is less than the preset evaluation threshold.

[0083] Among them, when When the value is less than the preset threshold γ, the model prediction is considered reliable and can be used for user detection. When the evaluation metric exceeds the preset threshold γ, a model fine-tuning process is triggered to optimize model performance.

[0084] In some embodiments, the pre-training method for the detection model further includes: Different detection models are trained using different datasets, where the different datasets have different physical random access channel configuration data; Different detection models trained using different datasets will be stored, with each detection model having a unique identifier.

[0085] In this embodiment, to improve the efficiency and response speed of PRACH signal detection, multiple PRACH detection models are pre-trained. These models are designed for different PRACH configurations, including long and short formats, and different ZCZC cyclic shift index settings. In practical engineering applications, it is usually not necessary to pre-train for every configuration; models can be pre-trained only for the PRACH configurations supported by the system, thereby reducing the complexity of model management and storage space. After model training is complete, each model will be stored in the gNB's storage system. These pre-trained models are optimized for specific PRACH configurations to ensure fast and accurate signal detection in practical applications.

[0086] When saving models, each model is uniquely identified based on its configuration characteristics, such as binding the model ID or name to a specific PRACH configuration. Thus, once the base station determines the corresponding PRACH configuration, the gNB can quickly retrieve the matching pre-trained model using the configuration information. This invention's model management mechanism allows the gNB to retrieve and load corresponding models using simple ID or name indexes, thereby achieving high efficiency and convenience in model retrieval. Furthermore, predefined models are stored only within the gNB, ensuring model security and privacy.

[0087] It should be noted that, with a well-trained detection model, the detection model to be used is selected based on the physical random access channel configuration information sent to the user terminal while executing step S101.

[0088] Therefore, in step S1431, user detection using the pre-trained detection model on the standardized physical random access channel sequence includes: By utilizing physical random access channel configuration information, matching the corresponding physical random access channel configuration data, and using the detection model corresponding to the unique identifier of the physical random access channel configuration data, user detection is performed on the standardized physical random access channel sequence.

[0089] S104. If the user detection probability is greater than the preset detection threshold, the user is determined to be detected; otherwise, the user is determined not to be detected.

[0090] For example, the gNB performs standardized data preprocessing on the received PRACH signal, and then uses the selected AI model to perform user detection on the received PRACH signal. It can output the probability of detecting a user for each detection window, and determine that the window exceeding the probability threshold is a user detected, and the window below the probability threshold is a user not detected.

[0091] Therefore, in the detection method provided in this embodiment, after the terminal sends the random access channel sequence to the base station, the base station can perform user detection after receiving the physical random access channel sequence, and further send an acknowledgment message to the detected user. Through the user's response message, it can be further determined whether the detected user is a user terminal, and the probability of false detection by the base station can be reduced by using the user response message.

[0092] In some embodiments, the step of optimizing the detection model includes: Obtain the dataset, and retrieve the historical dataset when random user access is successful or fails; The detection model is further trained and optimized by using the union of the dataset and the historical dataset.

[0093] For example, after the base station sends the PRACH configuration via the SIB, the UE that needs to access the system randomly will parse the PRACH configuration parameters and send message 1 (MSG1), which is the PRACH Preamble sequence. The base station then uses a pre-trained model to detect the PRACH signal. When the gNB detects a user within a PRACH detection window using its detection model, it sends message 2 (MSG2) to the corresponding user. If the user is a real user, the base station will receive message 3 (MSG3); if the user is a fake user, the base station will not receive an MSG3 response. The gNB marks the window corresponding to the user who receives the MSG3 response as a real user, and the window corresponding to the user who does not receive the MSG3 response as a fake user.

[0094] This data is then collected and stored on the gNB side as a real wireless environment dataset, also known as a historical dataset. The historical dataset includes PRACH configurations sent by the base station, MSG2 messages sent by the base station, real user tags, and fake user tags, and is further divided into fine-tuning datasets. and validation dataset ,in , .

[0095] In some embodiments, the step of optimizing the detection model further includes: The optimized detection model is evaluated to obtain the optimized evaluation value; If the optimized evaluation value is less than the preset evaluation threshold, the detection model passes the evaluation and is saved. If the optimized evaluation value is greater than or equal to the preset evaluation threshold, the detection model will continue to be optimized until the optimized evaluation value of the detection model is less than the preset evaluation threshold.

[0096] For example, the performance metrics Metrics(FNM, FPM) are calculated periodically using the validation dataset according to the performance evaluation method described above. If the calculated performance metrics Metrics(FNM, FPM) are less than a preset threshold (γ), gNB uses the output of the detection model for user detection. This online validation method allows for real-time evaluation of model performance, and the model can be updated promptly when its performance does not meet preset requirements.

[0097] If the performance metrics exceed a threshold (γ), gNB fine-tunes the detection model to optimize its performance. The fine-tuning employs an incremental strategy; when a performance metric exceeds the preset threshold γ, model fine-tuning is initiated. During fine-tuning, let the model before fine-tuning be... The training dataset is The fine-tuned model is Fine-tuned dataset for Compared with the collected real-world incremental data The union of, i.e. By employing the model fine-tuning method described above, this embodiment can effectively enhance the model's false detection performance, while also balancing the issues of missed and false detections in PRACH detection, thereby improving the model's accuracy and reliability in practical applications.

[0098] The fine-tuned model will then be re-evaluated for performance. Evaluation data will be derived from data collected in real-world wireless environments, particularly based on MSG3 response data, using a validation dataset from the real-world dataset. This method allows for the evaluation of the updated model's performance in a real wireless environment, determining whether the model meets preset performance requirements. During model fine-tuning, if the model fails to meet the preset performance requirements, the number of fine-tuning iterations can be increased until the model satisfies them. Once the model meets the performance requirements, gNB will stop fine-tuning and update its AI model library to include the optimized model, subsequently using the optimized model for user detection. This method enables precise model fine-tuning to meet preset performance requirements. By using real-world data and an evaluation mechanism based on MSG3 response, the model evaluation method of this invention can accurately assess model performance and perform model fine-tuning when necessary, thereby ensuring low false negative and low false positive rates in practical applications.

[0099] In summary, the detection method provided in this application has the following advantages: Detection model deployment: This embodiment of the invention deploys a pre-trained detection model in an LTE or NR system to detect PRACH signals during random access procedures, thereby improving the accuracy and efficiency of detection.

[0100] Balancing false positive rate and false negative rate: This embodiment of the invention combines the false positive rate index and the false negative rate index, so that the network can always maintain a balance between the two indexes according to the parameter configuration during the training process, so that the pre-trained model can meet the preset false positive rate and false negative rate index.

[0101] Simplified detection algorithm update: This embodiment of the invention sets online evaluation indicators for false detections and false negatives and collects online data. When the model detection indicators are found to be inconsistent with the preset requirements, the model is adaptively updated using historical data, thereby simplifying the update process of the PRACH detection algorithm.

[0102] Environmental adaptability: Existing PRACH detection methods are highly coupled with the environment, requiring different detection methods to be designed for different scenarios. For example, two detection algorithms might be designed for high-speed scenarios and ordinary scenarios. In contrast, AI-based methods can maintain the structural consistency of the model and adjust the model parameters through fine-tuning, thereby adapting to different usage scenarios and flexibly adapting to various application scenarios.

[0103] Parallel processing capability: In response to the limitation of serial processing of root sequences and received signals in traditional detection methods, the AI ​​detection scheme of this invention can process the correlation detection of multiple root sequences and PRACH signals in parallel, thereby improving the processing speed.

[0104] In summary, the embodiments of this application provide a detection method, such as... Figure 10 As shown, offline model training is required before deploying gNB to obtain the detection model; The gNB then sends the PRACH configuration information to the UE and selects the trained detection model based on the PRACH configuration information.

[0105] The UE then receives and parses the PRACH configuration information. Based on the parsed PRACH configuration, the UE selects a suitable Preamble ID and sends the corresponding PRACH sequence.

[0106] Subsequently, gNB performs standardized data preprocessing on the received PRACH signal, and then uses the selected detection model to perform user detection on the received PRACH signal, outputting the probability of detecting a user in each detection window. Windows that exceed the probability threshold are judged as having detected a user, and windows that are less than the probability threshold are judged as having no detected user.

[0107] Then, verification data collection and performance scheduling are performed. After the base station sends the PRACH configuration via SIB, the UEs that need to access randomly will parse the PRACH configuration parameters and send MSG1, i.e., the PRACH Preamble sequence. Subsequently, the base station uses a pre-trained model to detect the PRACH signal. When the gNB detects a user within a PRACH detection window using the detection model, it will send an MSG2 message to the corresponding user. If the user is a real user, the base station will receive an MSG3 response; if the user is a dummy user, the base station will not receive an MSG3 response. The gNB marks the window corresponding to the user who receives the MSG3 response as a real user, and the window corresponding to the user who does not receive the MSG3 response as a dummy user. Then, this data is collected and stored on the gNB side as a real wireless environment dataset, i.e., a historical dataset, denoted as [database name missing]. Furthermore, it is divided into fine-tuning datasets. and validation dataset ,in , The performance metrics (Metrics(FNM, FPM)) are periodically calculated using the validation dataset according to the performance evaluation method described above. If the calculated performance metrics (Metrics(FNM, FPM)) are less than a preset threshold (γ), gNB uses the output of the detection model for user detection. This online validation method allows for real-time evaluation of model performance, and the model can be updated promptly when its performance fails to meet preset requirements.

[0108] Then, model fine-tuning is performed. If the performance metrics exceed a threshold (γ), gNB fine-tunes the detection model to optimize its detection performance. Model fine-tuning employs an incremental fine-tuning strategy. When the performance metrics exceed the preset threshold γ, model fine-tuning is initiated. During the fine-tuning process, let the model before fine-tuning be... The training dataset is The fine-tuned model is Fine-tuned dataset for Compared with the collected real-world incremental data The union of, i.e. By employing the model fine-tuning method described above, this embodiment can effectively enhance the model's false detection performance, while also balancing the issues of missed and false detections in PRACH detection, thereby improving the model's accuracy and reliability in practical applications.

[0109] Further fine-tuning, model evaluation, and updates are performed: the fine-tuned model will be re-evaluated for performance. Evaluation data is derived from data collected in real-world wireless environments, particularly based on MSG3 response data, using validation datasets from real-world datasets. This method allows for the evaluation of the updated model's performance in a real wireless environment, determining whether the model meets preset performance requirements. During model fine-tuning, if the model fails to meet the preset performance requirements, the number of fine-tuning iterations can be increased until the model satisfies them. Once the model meets the performance requirements, the gNB will stop fine-tuning and update its detection model library to include the optimized model, subsequently using the optimized model for user detection. This method allows for precise model fine-tuning to meet preset performance requirements. By using real-world data and an evaluation mechanism based on MSG3 response, the model evaluation method of this invention can accurately assess model performance and perform model fine-tuning when necessary, thereby ensuring low false negative and low false positive rates in practical applications.

[0110] In this embodiment, the detection method includes the following steps: Detection Model Design and Configuration: Based on the ZCZC configuration specified in the 3GPP TS38.211 / TS36.211 standards, a corresponding AI model was designed for each cyclic shift index (0 to 15). In each configuration, the model's variable one-dimensional convolutional layers and variable fully connected layers are adjusted according to the cyclic shift index, while other layers remain fixed. Each model is uniquely named and numbered and bound one-to-one with the PRACH configuration information.

[0111] Data Collection and Preprocessing: For each configuration of ZCZC cyclic shift index 0 to 15, system-level simulation data or laboratory data were collected and standardized preprocessed to accommodate different detection window lengths and window numbers. 30,000 samples were collected for each configuration, denoted as […]. D t Of which 90% was used as training data, denoted as D t1 10% is used as test data, denoted as D t2 .

[0112] Model training: using the defined loss function The model is pre-trained using the data from the steps above. During pre-training, the initial learning rate is set to α1 = 1e-4, the Adam optimizer is used, the batch size is set to 128, and the training epochs are 1000. Training stops when 1000 epochs are reached or the Metrics(FNM, FPM) on the test set is less than the threshold γ = 1e-4, where β1 and β2 in Metrics(FNM, FPM) are set to 0.3 and 0.7, respectively.

[0113] Model Deployment and Inference: The trained model is stored on the gNB side. The gNB selects the appropriate model for inference based on configuration information and received PRACH signals. For all PRACH signals received by the system, Preamble extraction is performed first. Then, the extracted Preamble is cross-correlated with the root sequence. The correlated signals are then normalized and preprocessed before the corresponding AI model is used for user detection. Subsequently, the gNB sends MSG2 to detected users and collects MSG3 responses from the UE. Users who receive MSG3 are marked as real users, and users who do not receive MSG3 are marked as false detection users. This data from the real wireless environment is saved as a verification dataset. This data is stored in the incremental dataset δ, which is divided into δ1 and δ2, where |δ1|=0.7. |δ|, |δ²|=0.3 |δ|. δ1 is used for model fine-tuning training, and δ2 is used for performance evaluation of the fine-tuned model. During inference, Metrics(FNM, FPM) are periodically calculated using the validation dataset δ, and the obtained metric is compared with the threshold γ=1e-4. If the metric is greater than the threshold, the model update process is triggered; if the metric is less than the threshold, the PRACH detection result of the model is used.

[0114] Model fine-tuning: If the above determines that the model needs to be updated, then fine-tune the model using the incremental dataset D_(t+1)=D_t1∪δ1 defined in step 4, with the fine-tuning learning rate set to α2=5e-5. The fine-tuned model is then evaluated again using Metrics(FNM, FPM) on datasets δ2 and D_t2. Training stops when the metric is less than the threshold γ=1e-4 to ensure that the fine-tuned model meets the performance requirements.

[0115] Therefore, the detection method provided in this embodiment of the invention uses a detection model that can be named RACH-NET. This network structure includes flexible-conv1d (variable one-dimensional convolution), max-pool (maximum pooling), fixed-conv1d (fixed one-dimensional convolution), fixed-FNN (fixed fully connected layer), and flexible-FNN (variable fully connected layer), which can adapt to the detection requirements of different PRACH configurations. It also includes a series of innovative model deployment, inference, data collection, and model fine-tuning steps to ensure performance in various real wireless environments.

[0116] Furthermore, the pre-trained models of RACH-NET are stored on the gNB side, enabling dynamic selection and invocation of the most suitable model for signal detection based on the received PRACH signal characteristics and PRACH configuration information. This deployment strategy improves the response speed and accuracy of the detection process.

[0117] During model inference, RACH-NET uses Metrics (FNM, FPM) to evaluate the reliability of detection results. By comparing the results with a preset threshold γ, the model can automatically determine whether fine-tuning is needed to adapt to changing wireless environments.

[0118] Data collection is a crucial step in RACH-NET training and fine-tuning. System-level simulation data and laboratory-measured data are used to construct diverse datasets, which undergo standardization preprocessing to eliminate differences in signal power among different users and ensure the generalization performance of the model training.

[0119] Model fine-tuning is a crucial component of RACH-NET's adaptive optimization. When a model's evaluation metrics exceed a threshold, gNB uses collected user feedback data from real-world environments to incrementally fine-tune the model. The fine-tuned model is then re-evaluated to ensure its performance meets predetermined requirements.

[0120] This application also provides a detection method applied to a user terminal; please refer to [link / reference]. Figure 11 The detection methods include: S201. Receive physical random access channel configuration information sent by the communication terminal; S202. Parse the physical random access channel configuration information and generate a physical random access channel sequence; S203. Send the physical random access channel sequence to the communication terminal.

[0121] In this embodiment, the UE receives the above-mentioned PRACH configuration information and parses it to determine the parameters of the random access signal.

[0122] Based on the parsed PRACH configuration, the UE selects a suitable Preamble ID and sends the corresponding PRACH sequence.

[0123] After the terminal sends the random access channel sequence to the base station, the base station can perform user detection after receiving the physical random access channel sequence, and further send an acknowledgment message to the detected user. Through the user's response message, it can further determine whether the detected user is a user terminal, and use the user response message to reduce the probability of false detections by the base station.

[0124] This application also provides a detection system applied to a base station side; please refer to [link / reference]. Figure 12The detection system includes: an information sending module 11, an information receiving module 12, and a user detection module 13; the information sending module 11 is used to send predefined physical random access channel configuration information to the user terminal; the information receiving module 12 is used to receive the physical random access channel sequence generated by the user terminal according to the physical random access channel configuration information; the user detection module 13 is used to perform user detection using the physical random access channel sequence; wherein, if the user detection probability is greater than a preset detection threshold, the user is determined to be detected, otherwise the user is determined to be undetected.

[0125] The detection system provided in this embodiment allows the base station to detect users after receiving the physical random access channel sequence from the terminal. The base station then sends an acknowledgment message to the detected user. By using the user's response message, it can further determine whether the detected user is a genuine user and reduce the probability of false detections by the base station.

[0126] In some embodiments, the user detection module 13 includes: a data processing unit for standardizing the physical random access channel sequence; and a detection unit for using a pre-trained detection model to perform user detection on the standardized physical random access channel sequence to obtain the user detection probability for each detection window.

[0127] In some embodiments, the detection system further includes a model training module for training a detection model.

[0128] The model training module includes: a dataset acquisition unit for acquiring the dataset; a standardization unit for standardizing the dataset; a network structure determination unit for determining the network structure of the initial detection model, wherein the network structure of the detection model includes: a variable one-dimensional convolutional layer, a first max pooling layer, a fixed one-dimensional convolutional layer, a second max pooling layer, a fixed fully connected layer, and a variable fully connected layer; and a training unit for training the initial detection model using a preset training strategy and the standardized dataset to obtain the detection model.

[0129] In some embodiments, the dataset includes: segmentation index data obtained by segmenting the cyclic shift index of the Zadov-Chu sequence according to a predetermined rule, wherein different segmentation index data correspond to different preset values, and the length of the segmentation index data is the same as the length of the detection window.

[0130] In some embodiments, the formula for standardizing the dataset includes: ; in, Indicates the first i The raw data of each user's detection window; It is the mean of the raw data in this detection window; It is the variance of the raw data of this detection window; It is the standardized version of the original data; i This represents the user index, with values ​​ranging from 1 to maxUserNumber, which is the maximum number of users that can be supported on a single root sequence under a specific physical random access channel configuration.

[0131] In some embodiments, the preset training strategy includes: pre-training and optimizing the initial detection model using a preset adaptive momentum optimization algorithm, wherein a first learning rate-constrained adaptive momentum optimization algorithm is used when pre-training the initial detection model, and a second learning rate-constrained adaptive momentum optimization algorithm is used when optimizing the initial detection model, and the first learning rate is greater than the second learning rate.

[0132] In some embodiments, the model training module further includes: a performance evaluation unit, used to evaluate the performance of the detection model and obtain an evaluation value; if the evaluation value is less than a preset evaluation threshold, the detection model passes the evaluation and is saved; if the evaluation value is greater than or equal to the preset evaluation threshold, the detection model is optimized until the evaluation value of the detection model is less than the preset evaluation threshold.

[0133] In some embodiments, the formula for evaluating the performance of the detection model includes: ; in, and The preset weighting coefficients, To validate the dataset; FNM is the false negative metric; FPM is the false positive metric. The formulas for calculating the missed detection index include: ; The formulas for calculating false inspection indicators include: ; in, y This represents the probability of detecting a user in each detection window of the detection model, with a value ranging from zero to one.

[0134] In some embodiments, the model training module further includes: a constraint unit, used to establish a joint loss function using the false negative rate and false positive rate of the detection model; and to constrain the false negative rate and false positive rate of the detection model using the loss function.

[0135] The formula for the joint loss function includes: ; in, This represents the loss value, where x is the true label value of the dataset.

[0136] The formula for expanding the loss value satisfies the following conditions:

[0137] Where batchSize represents the number of samples input to the initial detection model at one time during the training process; maxUserNumber represents the maximum number of users that a related sequence can support for reuse.

[0138] In some embodiments, the detection system further includes a reverse training module, used to calculate the loss value of the corresponding detection window and backpropagate the calculation result to update the parameters of the detection model.

[0139] In some embodiments, the detection system further includes: a periodic evaluation module, used to evaluate the detection model within a preset period using a validation dataset after the detection model is saved, and obtain a periodic evaluation value; if the periodic evaluation value is greater than or equal to a preset evaluation threshold, the detection model is optimized until the periodic evaluation value of the detection model is less than the preset evaluation threshold.

[0140] In some embodiments, the performance evaluation unit is specifically used to obtain a dataset and historical datasets when the random access to the user terminal is successful or failed when optimizing the detection model; the detection model is further trained using the union of the dataset and the historical datasets to optimize the detection model.

[0141] In some embodiments, when optimizing the detection model, the performance evaluation unit is further used to evaluate the optimized detection model and obtain an optimized evaluation value; if the optimized evaluation value is less than a preset evaluation threshold, the detection model passes the evaluation and is saved; if the optimized evaluation value is greater than or equal to the preset evaluation threshold, the detection model continues to be optimized until the optimized evaluation value of the detection model is less than the preset evaluation threshold.

[0142] In some embodiments, the model training module further includes: a multi-model training unit for training different detection models using different datasets, wherein the different datasets have different physical random access channel configuration data; and a multi-model storage unit for storing the different detection models trained using different datasets, wherein each detection model has a unique identifier.

[0143] In some embodiments, the detection unit is specifically used to match the corresponding physical random access channel configuration data using physical random access channel configuration information, and to perform user detection on the standardized physical random access channel sequence using a detection model with a unique identifier corresponding to the physical random access channel configuration data.

[0144] This application also provides a detection system applied to a user terminal; please refer to [link / reference]. Figure 13 The detection system includes: a configuration information receiving module 21, a configuration information parsing module 22, and a channel sequence sending module 23; the configuration information receiving module 21 is used to receive physical random access channel configuration information sent by the communication end; the configuration information parsing module 22 is used to parse the physical random access channel configuration information and generate a physical random access channel sequence; the channel sequence sending module 23 is used to send the physical random access channel sequence to the communication end.

[0145] In this embodiment, through data interaction with the base station, after the terminal sends the physical random access channel sequence to the base station, the base station can perform user detection after receiving the physical random access channel sequence, and further send an acknowledgment message to the detected user. Through the user's response message, it can be further determined whether the detected user is a user terminal, and the probability of false detection by the base station can be reduced by using the user response message.

[0146] This application provides an electronic device; please refer to [link / reference]. Figure 14 The electronic device includes a memory 601, a processor 602, and a computer program stored in the memory 601 and executable on the processor 602. When the processor 602 executes the computer program, it implements the detection method described above.

[0147] Furthermore, the electronic device also includes at least one input device 603 and at least one output device 604.

[0148] The aforementioned memory 601, processor 602, input device 603, and output device 604 are connected via bus 605.

[0149] The input device 603 can specifically be a camera, touch panel, physical buttons, or mouse, etc. The output device 604 can specifically be a display screen.

[0150] The memory 601 can be a high-speed random access memory (RAM) or a non-volatile memory, such as a disk storage device. The memory 601 is used to store a set of executable program code, and the processor 602 is coupled to the memory 601.

[0151] Furthermore, this application embodiment also provides a computer-readable storage medium, which may be disposed in the electronic device of the above embodiments, and may be the memory 601 in the foregoing embodiments. The computer-readable storage medium stores a computer program, which, when executed by the processor 602, implements the detection method described in the foregoing method embodiments.

[0152] Furthermore, the storage medium of this computer can also be a USB flash drive, a portable hard drive, a read-only memory (ROM), RAM, a magnetic disk, or an optical disk, or any other medium that can store program code.

[0153] This application also provides a computer program for executing the detection method described in any of the above embodiments.

[0154] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0155] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0156] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0157] If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0158] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0159] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0160] In summary, although the present application has disclosed the preferred embodiments as described above, the above preferred embodiments are not intended to limit the present application. Those skilled in the art can make various modifications and refinements without departing from the spirit and scope of the present application. Therefore, the scope of protection of the present application shall be determined by the scope defined in the claims.

Claims

1. A detection method, characterized in that, include: Send predefined physical random access channel configuration information to the user terminal; Receive the physical random access channel sequence generated by the user terminal based on the physical random access channel configuration information; User detection is performed using the physical random access channel sequence to obtain the user detection probability; If the user detection probability is greater than the preset detection threshold, the user is determined to be detected; otherwise, the user is determined not to be detected.

2. The detection method as described in claim 1, characterized in that, The user detection using the physical random access channel sequence includes: The physical random access channel sequence is standardized. The pre-trained detection model is used to perform user detection on the standardized physical random access channel sequence to obtain the user detection probability for each detection window.

3. The detection method as described in claim 2, characterized in that, The pre-training method for the detection model includes: Obtain the dataset; The dataset is then standardized. The network structure of the initial detection model is determined, wherein the network structure of the detection model includes: a variable one-dimensional convolutional layer, a first max pooling layer, a fixed one-dimensional convolutional layer, a second max pooling layer, a fixed fully connected layer, and a variable fully connected layer; The initial detection model is trained using a preset training strategy and the standardized dataset to obtain the detection model.

4. The detection method as described in claim 3, characterized in that, The dataset includes: segmentation index data obtained by segmenting the cyclic shift index of the Zadov-Chu sequence according to a predetermined rule, wherein different segmentation index data correspond to different preset values, and the length of the segmentation index data is the same as the length of the detection window.

5. The detection method as described in claim 3, characterized in that, The formula for standardizing the dataset includes: ； in, Indicates the first i The raw data of each user's detection window; It is the mean of the raw data in this detection window; It is the variance of the raw data of this detection window; It is the standardized data of the original data; i This represents the user index, with values ​​ranging from 1 to maxUserNumber, which is the maximum number of users that can be supported on a single root sequence under a specific physical random access channel configuration.

6. The detection method as described in claim 3, characterized in that, The preset training strategy includes: pre-training and optimizing the initial detection model using a preset adaptive momentum optimization algorithm, wherein a first learning rate is used to constrain the adaptive momentum optimization algorithm when pre-training the initial detection model, and a second learning rate is used to constrain the adaptive momentum optimization algorithm when optimizing the initial detection model, wherein the first learning rate is greater than the second learning rate.

7. The detection method as described in claim 3, characterized in that, The pre-training method for the detection model also includes: The detection model is evaluated to obtain evaluation values; If the evaluation value is less than the preset evaluation threshold, the detection model passes the evaluation and is saved. If the evaluation value is greater than or equal to the preset evaluation threshold, the detection model is optimized until the evaluation value of the detection model is less than the preset evaluation threshold.

8. The detection method as described in claim 7, characterized in that, The formula for evaluating the performance of the detection model includes: ； in, and The preset weighting coefficients, To validate the dataset; FNM is the false negative metric; FPM is the false positive metric. The calculation formula for the missed detection index includes: ； The calculation formula for the false detection index includes: ； in, y This represents the probability of detecting a user in each detection window of the detection model, with a value ranging from zero to one.

9. The detection method as described in claim 8, characterized in that, The pre-training method for the detection model also includes: A joint loss function is established using the missed detection index and the false detection index; The loss function is used to constrain the false negative rate and false positive rate of the detection model.

10. The detection method as described in claim 9, characterized in that, The formula for the joint loss function includes: ； in, denoted as the loss value, where x is the true label value of the dataset.

11. The detection method as described in claim 10, characterized in that, The formula for calculating the expanded loss value based on the true label value satisfies the following condition: Where batchSize represents the number of samples input to the initial detection model at one time during the training process of the detection model; maxUserNumber represents the maximum number of users that a related sequence can support for reuse.

12. The detection method as described in claim 10, characterized in that, The detection method further includes: The loss value for the corresponding detection window is calculated, and the calculation result is backpropagated to update the parameters of the detection model.

13. The detection method as described in claim 8, characterized in that, The detection method further includes: After the detection model is saved, the detection model is evaluated using the verification dataset within a preset period to obtain a periodic evaluation value. If the periodic evaluation value is greater than or equal to the preset evaluation threshold, the detection model is optimized until the periodic evaluation value of the detection model is less than the preset evaluation threshold.

14. The detection method as described in claim 7, characterized in that, The steps for optimizing the detection model include: Obtain the dataset and retrieve the historical datasets of successful or failed random access to user terminals; The detection model is further trained using the union of the dataset and the historical dataset to optimize it.

15. The detection method as described in claim 14, characterized in that, The step of optimizing the detection model further includes: The optimized detection model is evaluated to obtain an optimized evaluation value; If the optimized evaluation value is less than the preset evaluation threshold, the detection model passes the evaluation and is saved. If the optimized evaluation value is greater than or equal to the preset evaluation threshold, the detection model is further optimized until the optimized evaluation value of the detection model is less than the preset evaluation threshold.

16. The detection method as described in claim 3, characterized in that, The pre-training method for the detection model also includes: Different detection models are trained using different datasets, wherein the different datasets have different physical random access channel configuration data; Different detection models trained using different datasets will be stored, with each detection model having a unique identifier.

17. The detection method as described in claim 16, characterized in that, The user detection using the pre-trained detection model on the standardized physical random access channel sequence includes: Using the physical random access channel configuration information, the corresponding physical random access channel configuration data is matched, and using the detection model corresponding to the unique identifier of the physical random access channel configuration data, user detection is performed on the standardized physical random access channel sequence.

18. A detection method, characterized in that, include: Receive physical random access channel configuration information sent by the communication terminal; Parse the physical random access channel configuration information to generate a physical random access channel sequence; The physical random access channel sequence is sent to the communication terminal.

19. A detection system, characterized in that, include: The information sending module is used to send predefined physical random access channel configuration information to the user terminal; The information receiving module is used to receive the physical random access channel sequence generated by the user terminal according to the physical random access channel configuration information; The user detection module is used to perform user detection using the physical random access channel sequence; wherein, if the user detection probability is greater than a preset detection threshold, the user is determined to be detected, otherwise the user is determined not to be detected.

20. A detection system, characterized in that, include: The configuration information receiving module is used to receive physical random access channel configuration information sent by the communication end; The configuration information parsing module is used to parse the physical random access channel configuration information and generate a physical random access channel sequence; A channel sequence sending module is used to send the physical random access channel sequence to the communication terminal.

21. An electronic device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the method of any one of claims 1 to 18.

22. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1 to 18.

23. A computer program, characterized in that, For performing the method according to any one of claims 1 to 18.

Images