Device for detecting interference of LTE downlink based on air interface parameters and machine learning

Abstract

An LTE downlink interference detection device based on air interface parameters and machine learning includes: a data acquisition module, an interference model training module, an interference detection module, and an interference source localization module. This device establishes a database by sensing parameters reported by the UE and the wireless environment parameters of the base station. It uses the database to train machine learning models under low interference or single interference scenarios to predict the retransmission rate of the base station under low interference or single base station interference conditions, thereby obtaining the interference intensity of the base station and the interference source to achieve interference detection and interference source localization of the base station.

Description

Technical Field

[0001] This invention relates to a technology in the field of wireless communication, specifically an LTE downlink interference detection device based on air interface parameters and machine learning. Background Technology

[0002] Interference issues severely impact communication quality. Due to the dense distribution of base station equipment or unsuitable parameters such as transmit antenna angles, user equipment (UEs) are often subjected to interference from neighboring cells on the same frequency, leading to degraded signal quality and reduced network speed for users at cell edges, thus affecting user experience. Current downlink interference assessments often use the Signal-to-Interference-plus-Noise Ratio (SINR) for description. SINR values ​​are typically measured by UEs and mapped to a Channel State Indicator (CQI), which is then reported via the uplink. The base station obtains the CQI and maps it to the corresponding Modulation and Coding Scheme (MCS), using the MCS information for subsequent downlink scheduling control. However, because downlink SINR is not standardized within 3GPP, different UE vendors use different methods to define it, and these methods are often internal to each application vendor. Therefore, obtaining interference information from the UE side for base station configuration optimization is quite difficult. Summary of the Invention

[0003] To address the aforementioned shortcomings of existing technologies, this invention proposes an LTE downlink interference detection device based on air interface parameters and machine learning. It establishes a database by sensing parameters reported by the UE and the base station's own wireless environment parameters. The database is then used to train machine learning models for low-interference or single-interference scenarios to predict the retransmission rate (PRB) of the base station under low-interference or single-base station interference conditions. This allows for the detection of interference at the base station and the identification of the interference source.

[0004] This invention is achieved through the following technical solution:

[0005] This invention relates to an LTE downlink interference detection device based on air interface parameters and machine learning, comprising: a data acquisition module, an interference model training module, an interference detection module, and an interference source localization module. The data acquisition module acquires relevant parameters from the base station and user side in real time from the base station under test, preprocesses them, and outputs the dataset to the interference model training module. The interference model training module trains the model using a random forest regression algorithm based on the received dataset, and outputs the trained model file to the interference detection module and the interference source localization module respectively. The interference detection module takes the relevant parameters from the base station and user side acquired in real time from the base station under test, inputs them into a first interference detection model, predicts the retransmission rate under interference-free conditions, and obtains the detection result through the interference detection algorithm. The interference source localization module extracts the corresponding base station and user side parameters based on the detection results for medium and strong interference, inputs them into a second interference detection model from the interference model training module, predicts the retransmission rate under high-load interference from a single neighboring base station, and then determines the interference source through the interference source localization algorithm.

[0006] The first interference detection model is the same as the main index for real-time acquisition of relevant parameters on the base station side and user side, and both sub-indexes are under low load.

[0007] The second interference detection model has the same main index as the corresponding base station-side and user-side related parameters, and the sub-index corresponds to the low load and high load of the two base stations, respectively.

[0008] The retransmission rate refers to the number of PRB retransmissions within a measurement time window obtained by periodically collecting the total number of PRB retransmissions and calculating the difference, thus describing the magnitude of downlink interference. Specifically: Where rbRetx_PerCycle represents the number of retransmitted RBs within a period, and rbDl_PerCycle represents the number of transmitted RBs within a period. Related experiments and theoretical derivations demonstrate a negative correlation between the retransmission rate and downlink SINR.

[0009] The base station-side and user-side related parameters include: downlink MCS, CQI, serving cell / neighbor cell reference signal received power (RSRP), serving cell / neighbor cell reference signal received quality (RSRQ), serving cell / neighbor cell PRB occupancy rate, retransmission rate, protocol data unit (PDU) occupancy, and transport block size (TBS) occupancy.

[0010] The preprocessing includes outlier cleaning and index partitioning, wherein:

[0011] Data outlier cleaning includes: determining the numerical range of the measured data, and considering values ​​exceeding the predetermined range as outliers; drawing box plots on historical data to determine the upper and lower edges of the data, and comparing the online data with the upper and lower edges, identifying values ​​exceeding the upper and lower edge ranges as outliers.

[0012] The index partitioning includes: dividing the collected data into segments according to the RSRP size of the serving cell and two neighboring cells to obtain the main index of the data, and dividing the data into segments according to the neighboring cell PRB occupancy rate, i.e. the neighboring cell load, to obtain the secondary index.

[0013] The input parameters for training the random forest regression algorithm are: downlink MCS, CQI, Serving Cell / Neighboring Cell Reference Signal Received Power (RSRP), Serving Cell / Neighboring Cell Reference Signal Received Quality (RSRQ), Serving Cell / Neighboring Cell PRB occupancy rate, Protocol Data Unit (PDU) occupancy, and Transport Block Size (TBS) occupancy; the output parameter is: retransmission rate.

[0014] The model training files are saved with index names, which are the index partitioning results from the preprocessing described above. Multiple trained model files are saved separately according to their index names for the interference detection module to access.

[0015] The phrase "the primary index is the same and the secondary indexes are all low-load" means that the RSRP values ​​of the data to be detected are divided to obtain the corresponding primary indexes, and the primary indexes are the same and the secondary indexes are all low-load among the many trained models.

[0016] The interference detection algorithm includes: a 0-1 interference detection algorithm, an interference intensity detection algorithm, and a threshold selection algorithm, wherein:

[0017] 0-1 interference detection algorithm: Input parameters from the real-time acquired base station and user side parameters of the base station under test are input into the first interference detection model. The retransmission rate under interference-free conditions is predicted using the model and the acquired parameters. The difference between the retransmission rate and the currently measured retransmission rate is calculated. The difference is compared with the detection threshold to obtain the 0-1 interference detection result.

[0018] Interference intensity detection algorithm: The retransmission rate under interference-free conditions is predicted by using the first interference detection model and real-time acquisition of relevant parameters from the base station and user side of the base station under test. The difference is calculated with the currently measured retransmission rate, and the difference is compared with multiple set thresholds to obtain the interference intensity detection result.

[0019] Threshold selection algorithm: For the 0-1 interference detection threshold, using existing historical data within the virtual grid, a CDF diagram of the difference between the predicted retransmission rate and the measured retransmission rate is plotted. The value corresponding to 30% is set as the threshold. For the interference intensity detection threshold, similarly using existing historical data within the virtual grid, a CDF diagram of the difference between the predicted and measured retransmission rates is plotted. Three boundary thresholds a1, a2, and a3 are set for the values ​​corresponding to 30%, 50%, and 70%, respectively. Values ​​less than a1 indicate no interference, a1~a2 indicate weak interference, a2~a3 indicate medium interference, and values ​​greater than a3 indicate strong interference. Threshold selection is calculated based on the established dataset, with an update period of T.

[0020] The second interference detection model, in which the measurement data has the same main index and the secondary index corresponds to the low load and high load of the two base stations respectively, divides the RSRP value of the data to be detected to obtain the corresponding main index. Among the many trained models, it searches for the main index with the same main index and the secondary index corresponding to the high load of neighbor cell one and the low load of neighbor cell two, as well as the main index with the same main index and the secondary index corresponding to the low load of neighbor cell one and the high load of neighbor cell two.

[0021] The second interference detection model is used in the interference source localization algorithm: by using the predicted retransmission rate under single neighboring cell interference, the retransmission rates under two single neighboring cell interference conditions are compared, and the neighboring cell with the larger retransmission rate is the interfering neighboring cell, thus realizing the interference source localization.

[0022] Technical effect

[0023] This invention uses interference detection and interference source localization algorithms, along with base station and UE-side parameters and machine learning models, to simulate wireless environments under different loads in multi-base station scenarios. It evaluates the downlink interference status of the base station by comparing the difference between the predicted retransmission rate and the actual retransmission rate. Compared to existing technologies that only consider using UE parameters to assist in SINR calculation, this invention considers both UE parameters and extracts base station parameters, using the retransmission rate to describe the wireless environment. Attached Figure Description

[0024] Figure 1 This is a schematic diagram of the LTE downlink interference detection device based on air interface parameters and machine learning according to the present invention.

[0025] Figure 2 This is a flowchart of the present invention;

[0026] Figure 3 The image shown is a rendering of an example. Detailed Implementation

[0027] like Figure 1The diagram shows a real-time scenario in this embodiment. The base station is a common commercial communication base station connected to the core network; the terminal is a commercial UE (User Equipment) that communicates with the base station via LTE. The interference scenario considered here involves one serving base station and two interfering neighboring base stations. In actual deployment, the neighboring cell RSRP (Reference Signal Receiving Power) reported by the UE can be sorted, and the neighboring cell with the highest value can be selected as the interfering neighbor for this algorithm. An edge server is set up at the serving base station, where the relevant interference detection program is deployed.

[0028] This embodiment relates to an LTE downlink interference detection device based on air interface parameters and machine learning, which includes: a data acquisition module, an interference model training module, an interference detection module, and an interference source localization module.

[0029] The data acquisition module includes a data acquisition unit, a data outlier cleaning unit, and an index partitioning unit. The data acquisition unit collects parameter information reported by the base station and UE in real time according to a preset list of parameters to be collected, obtaining an unprocessed initial dataset. The data outlier cleaning unit cleans the initial dataset collected by the data acquisition unit, removing outliers to obtain a dataset without outliers. The index partitioning unit, based on the dataset without outliers, and referring to the magnitude of the RSRP value and the neighboring cell load, partitions the data without outliers into multiple datasets named with indexes for model training.

[0030] The interference model training module includes a model training unit and a model storage unit. The model training unit trains interference models based on the dataset output by the data acquisition module and according to different index numbers, using a random forest algorithm to obtain trained models. The models are saved with their index numbers as names. The model storage unit classifies the models trained by the model training unit and stores them according to the main index, resulting in a model library classified by the main index, which facilitates querying by the interference detection and interference source localization modules.

[0031] The interference detection module includes an interference detection unit, an interference intensity detection unit, and a threshold selection unit. Specifically: the interference detection unit calculates the index value corresponding to the base station's test data, retrieves the trained model from the interference model training module to predict the retransmission rate, calculates the difference between the predicted and measured retransmission rates, and compares it with a threshold to obtain a 0-1 interference detection result. The interference intensity detection unit calculates the index value corresponding to the base station's test data, retrieves the trained model from the interference model training module to predict the retransmission rate, calculates the difference between the predicted and measured retransmission rates, and compares it with a threshold to obtain an interference intensity detection result. The threshold selection unit analyzes and calculates the CDF diagram of the difference between the predicted and measured retransmission rates based on historical data recorded in the interference detection and interference intensity detection units, defining three boundary thresholds at 30%, 50%, and 70% for subsequent interference detection. When no historical data is available, the first 100 sets of data are used to calculate the initial threshold.

[0032] The interference source localization module selects data of medium and strong interference based on the output results of the interference detection module, performs the interference source localization algorithm, and obtains the interference source result corresponding to the base station.

[0033] like Figure 2 As shown, this embodiment illustrates the LTE downlink interference detection method based on air interface parameters and machine learning, which specifically includes:

[0034] Step 1: Activate the data acquisition module. Data begins to be output from the base station under test to the acquisition module in real time. The data acquisition module then cleans and indexes the acquired data.

[0035] Step 2: In the interference model construction module, the number of datasets under the same index is judged. When the number exceeds the threshold, the trained model is considered complete.

[0036] Step 3: Perform 0-1 interference detection on the real-time collected data: Obtain the model's input parameters from the base station, and obtain a detection model with the same main index and a secondary index of 00 from the interference model training module. Use the model and the obtained parameters to predict the retransmission rate under interference-free conditions, calculate the difference between this and the currently measured retransmission rate, and compare the difference with the detection threshold to obtain the 0-1 interference detection result.

[0037] Step 4: Determine the detection result of Step 3: If interference occurs, send the data to the interference intensity detection module; otherwise, end the interference detection of the data and wait for the next interference detection cycle.

[0038] Step 5: Detect interference intensity for the input interference data. Obtain the model's input parameters from the interference detection module, and obtain the first interference detection model with the same main index and a secondary index of 00 from the interference model training module. Predict the retransmission rate under interference-free conditions, calculate the difference between this and the currently measured retransmission rate, and compare the difference with multiple set thresholds to obtain the interference intensity detection result.

[0039] Step 6: Determine the result of the interference detection. If the result is medium or strong interference, input the data into the interference source location module; otherwise, end the current interference detection and wait for the next interference detection cycle.

[0040] Step 7: Locate the source of interference in the interference data. Obtain the input parameters of the model from the interference intensity detection module, and obtain a second interference detection model with the same main index and secondary indices of 01 and 10 from the interference model training module. Predict the retransmission rate under single neighboring cell interference conditions. Compare the retransmission rates under two single neighboring cell interference conditions. The neighboring cell with the larger retransmission rate is the interfering neighboring cell, thus locating the interference source.

[0041] In simulated interference scenarios, an OAI (Open Air Interface) open-source soft base station and its associated UE are deployed for service connection and deployment. The data acquisition period in the data acquisition module is set to 1 second, and the frequencies of the serving cell base station and neighboring cell base stations are set to the same to simulate co-channel interference. The specific process for index partitioning and threshold selection is as follows:

[0042] For the index partitioning method, the serving cell RSRP and neighboring cell RSRP are divided into primary index segment numbers according to the rules shown in Table 1, and the neighboring cell load status is divided into secondary index segment numbers according to the rules shown in Table 1. Data with the same index value are stored in the same dataset.

[0043] Table 1. Segment Number Allocation Table

[0044] For the 0-1 interference detection threshold, a CDF plot of the difference between the predicted and measured retransmission rates is drawn using existing historical data within the virtual grid. The value corresponding to 30% is set as the threshold. For the interference intensity detection threshold, a CDF plot of the difference between the predicted and measured retransmission rates is also drawn using existing historical data within the virtual grid. Three boundary thresholds a1, a2, and a3 are set for the values ​​corresponding to 30%, 50%, and 70%, respectively. Values ​​less than a1 indicate no interference, a1-a2 indicate weak interference, a2-a3 indicate medium interference, and values ​​greater than a3 indicate strong interference. The threshold selection is calculated based on the established dataset, with an update cycle of 30 minutes.

[0045] Regarding the threshold for the number of models, which is the threshold in step 2, 100 models were selected as the threshold for the number of models in the dataset in the experiment. When there are more than 100 models, the dataset is considered complete, and the model trained on it can perform interference detection.

[0046] Based on the above parameter selection, relevant experiments were conducted at the OAI base station. Interference detection experiments were performed using three base stations: one serving base station and two interfering base stations. Three UEs were connected to the base stations to play video services with different bitrates to simulate different base station load conditions. Relevant wireless parameters were collected to create a dataset, and interference detection was performed. The results are as follows.

[0047] 1. A total of 104 datasets were collected, containing 36 different primary index numbers. Under each primary index number, there were 8 datasets with one secondary index, 6 rasters with two secondary indexes, 4 with three secondary indexes, and 18 with four secondary indexes. Regarding the time for constructing the virtual raster, since the data was collected offline in the experiment, the raster construction time was 85.66 seconds, excluding the data acquisition time.

[0048] 2. The prediction accuracy was statistically analyzed for each grid cell. Using the random forest prediction method, the maximum prediction error for 104 grid cells was 0.0606, the minimum error was 0.0046, and the average prediction error was 0.0250. The prediction results indicate that the predictions are relatively accurate.

[0049] 3. The collected data was divided into training and testing sets. The training set was used to train the model, and the testing set was used to test interference detection. In the testing set, 28.6% of the data was judged to be interference-free, and 71.4% of the data was judged to have interference. Among the data with interference, the percentage of interference-free data was 28.61%, the percentage of weakly interfered data was 19.45%, the percentage of moderately interfered data was 20.81%, and the percentage of strongly interfered data was 31.13%.

[0050] 4. Interference sources were located for medium and strong interference data. Statistics showed that out of 321 data entries, 126 entries were identified as interference from neighboring cell 1, and 195 entries were identified as interference from neighboring cell 2.

[0051] To verify the compatibility of this interference detection algorithm with the traditional downlink SINR interference detection method, the following experiment was designed to verify the effectiveness of the two methods. Drawing on the definition formula of downlink SINR in OAI-UE and other downlink SINR definition methods, the method of using full-band power measurement was chosen to define downlink SINR.

[0052] Downlink SINR measurement steps: At each measurement point, perform two measurements. The first measurement is the average received power of the frequency band after stabilization when the serving base station is active and there is service; this value is the signal power S. The second measurement is the average received power of the frequency band after stabilization when the interfering base station is active; this value is the interference power I. The SINR value can be obtained from the results of the two measurements, defined as follows: SINR = S / I. Theoretically, the calculation of SINR data should consider noise; however, since neighboring cell interference is much greater than noise interference in this case, the SINR effect caused by noise is not considered in the formula.

[0053] SINR parameters and retransmission rate were measured at 16 measurement points between the two base stations. SINR was selected as the average SINR value within 5 seconds after the interference power and signal power values ​​stabilized, and the retransmission rate was the average value within the corresponding 5-second time period. The measurement data from the 16 measurement points were recorded, and a scatter plot was plotted for linear fitting. The experimental results are shown in Table 2, and the fitted curve is shown in... Figure 3 As shown.

[0054] Table 2. Correspondence between SINR and retransmission rate

[0055] like Figure 3 As shown, there is a mapping relationship between retransmission rate and downlink SINR, that is, there is a correlation between the data size of the two. Using retransmission rate to predict interference can be compatible with SINR to describe interference.

[0056] Compared with existing technologies, this invention shifts the detection location from the UE side to the base station side, and combines UE-side and base station parameters to provide an interference detection algorithm based on retransmission rate. It is compatible with existing SINR detection methods.

[0057] The above-described specific implementations can be partially adjusted by those skilled in the art in different ways without departing from the principles and purpose of the present invention. The scope of protection of the present invention is defined by the claims and is not limited to the above-described specific implementations. All implementation schemes within the scope of the claims are bound by the present invention.

Claims

1. An LTE downlink interference detection device based on air interface parameters and machine learning, characterized in that, include: The system comprises a data acquisition module, an interference model training module, an interference detection module, and an interference source localization module. Specifically: the data acquisition module acquires relevant parameters from the base station and user side in real time from the base station under test, preprocesses them, and outputs the dataset to the interference model training module; the interference model training module trains the model using a random forest regression algorithm based on the received dataset, and outputs the trained model file to the interference detection module and the interference source localization module respectively; the interference detection module takes the relevant parameters from the base station and user side acquired in real time from the base station under test, inputs them into the first interference detection model, predicts the retransmission rate under interference-free conditions, and obtains the detection result through the interference detection algorithm; the interference source localization module extracts the corresponding base station and user side parameters based on the detection results for medium and strong interference, inputs them into the second interference detection model from the interference model training module, predicts the retransmission rate under high-load interference from a single neighboring base station, and then uses the interference source localization algorithm to determine the interference source. The first interference detection model is the same as the main index of the real-time acquisition of relevant parameters on the base station side and user side, and the sub-indexes are both under low load. The second interference detection model is identical to the main index of the corresponding base station-side and user-side related parameters, and the sub-index corresponds to the low load and high load of the two base stations, respectively. The retransmission rate refers to the number of PRB retransmissions within a measurement time window obtained by periodically collecting the total number of PRB retransmissions and calculating the difference, thus describing the magnitude of downlink interference. Specifically: Where: rbRetx_PerCycle is the number of retransmitted RBs within a period, and rbDl_PerCycle is the number of transmitted RBs within a period; through relevant experiments and theoretical derivation, it can be proven that there is a negative correlation between the retransmission rate and the downlink SINR; The base station-side and user-side related parameters include: downlink MCS, CQI, serving cell / neighbor cell reference signal received power, serving cell / neighbor cell reference signal received quality, serving cell / neighbor cell PRB occupancy rate, retransmission rate, protocol data unit occupancy, and transport block size occupancy.

2. The LTE downlink interference detection device based on air interface parameters and machine learning according to claim 1, characterized in that, The preprocessing includes outlier cleaning and index partitioning, wherein: Data outlier cleaning includes: determining the numerical range of the measured data, and considering values ​​that exceed the predetermined range as outliers; drawing box plots on historical data to determine the upper and lower edges of the data, and comparing the online data with the upper and lower edges, and considering values ​​that exceed the upper and lower edge ranges as outliers. The index partitioning includes: dividing the collected data into segments according to the RSRP size of the serving cell and two neighboring cells to obtain the main index of the data, and dividing the data into segments according to the neighboring cell PRB occupancy rate, i.e. the neighboring cell load, to obtain the secondary index.

3. The LTE downlink interference detection device based on air interface parameters and machine learning according to claim 1, characterized in that, The phrase "the primary index is the same and the secondary indexes are all low-load" means that the RSRP values ​​of the data to be detected are divided to obtain the corresponding primary indexes, and the primary indexes are the same and the secondary indexes are all low-load among the many trained models. The interference detection algorithm includes: a 0-1 interference detection algorithm, an interference intensity detection algorithm, and a threshold selection algorithm, wherein: 0-1 interference detection algorithm: Input parameters from the real-time acquired base station and user side parameters of the base station under test are input into the first interference detection model. The retransmission rate under interference-free conditions is predicted using the model and the acquired parameters. The difference between the retransmission rate and the currently measured retransmission rate is calculated. The difference is compared with the detection threshold to obtain the 0-1 interference detection result. Interference intensity detection algorithm: The retransmission rate under interference-free conditions is predicted by using the first interference detection model and real-time acquisition of relevant parameters from the base station and user side of the base station under test. The difference is calculated with the currently measured retransmission rate, and the difference is compared with multiple set thresholds to obtain the interference intensity detection result. Threshold selection algorithm: For the 0-1 interference detection threshold, using existing historical data within the virtual grid, a CDF diagram of the difference between the predicted retransmission rate and the measured retransmission rate is plotted, and the value corresponding to 30% is set as the threshold. For the interference intensity detection threshold, similarly using existing historical data within the virtual grid, a CDF diagram of the difference between the predicted retransmission rate and the measured retransmission rate is plotted, and three boundary thresholds a1, a2, and a3 are set for the values ​​corresponding to 30%, 50%, and 70%, respectively. Values ​​less than a1 indicate no interference, a1~a2 indicate weak interference, a2~a3 indicate medium interference, and values ​​greater than a3 indicate strong interference. The threshold selection is calculated based on the established dataset, with an update period of T. The second interference detection model, in which the main index is the same and the secondary index corresponds to the low load and high load of the two base stations respectively, divides the RSRP value of the data to be detected to obtain the corresponding main index. Among the many trained models, it searches for the main index with the same main index and the secondary index corresponding to the high load of neighbor cell one and the low load of neighbor cell two, as well as the main index with the same main index and the secondary index corresponding to the low load of neighbor cell one and the high load of neighbor cell two. The second interference detection model is used in the interference source localization algorithm: by using the predicted retransmission rate under single neighboring cell interference, the retransmission rates under two single neighboring cell interference conditions are compared, and the neighboring cell with the larger retransmission rate is the interfering neighboring cell, thus realizing the interference source localization.

4. The LTE downlink interference detection device based on air interface parameters and machine learning according to claim 1, characterized in that, The data acquisition module includes a data acquisition unit, a data outlier cleaning unit, and an index partitioning unit. The data acquisition unit collects parameter information reported by the base station and UE in real time according to a preset list of parameters to be collected, obtaining an unprocessed initial dataset. The data outlier cleaning unit cleans the initial dataset collected by the data acquisition unit, removing outliers to obtain a dataset without outliers. The index partitioning unit, based on the dataset without outliers, and referring to the magnitude of the RSRP value and the neighboring cell load, partitions the data without outliers into multiple datasets named with indexes for model training.

5. The LTE downlink interference detection device based on air interface parameters and machine learning according to claim 1, characterized in that, The interference model training module includes a model training unit and a model storage unit. The model training unit trains interference models based on the dataset output by the data acquisition module and according to different index numbers, using a random forest algorithm to obtain trained models. The models are saved with their index numbers as names. The model storage unit classifies the models trained by the model training unit and stores them according to the main index, resulting in a model library classified by the main index, which facilitates querying by the interference detection and interference source localization modules.

6. The LTE downlink interference detection device based on air interface parameters and machine learning according to claim 1, characterized in that, The interference detection module includes an interference detection unit, an interference intensity detection unit, and a threshold selection unit. Specifically: the interference detection unit calculates the index value corresponding to the base station's test data, retrieves the trained model from the interference model training module to predict the retransmission rate, calculates the difference between the predicted and measured retransmission rates, and compares it with a threshold to obtain a 0-1 interference detection result. The interference intensity detection unit calculates the index value corresponding to the base station's test data, retrieves the trained model from the interference model training module to predict the retransmission rate, calculates the difference between the predicted and measured retransmission rates, and compares it with a threshold to obtain an interference intensity detection result. The threshold selection unit analyzes and calculates the CDF diagram of the difference between the predicted and measured retransmission rates based on historical data recorded in the interference detection unit and the interference intensity detection unit, defining three boundary thresholds at 30%, 50%, and 70% for subsequent interference detection.

7. The LTE downlink interference detection device based on air interface parameters and machine learning according to claim 1, characterized in that, The interference source localization module selects data of medium and strong interference based on the output results of the interference detection module, performs the interference source localization algorithm, and obtains the interference source result corresponding to the base station.

8. A method for LTE downlink interference detection based on air interface parameters and machine learning according to any one of claims 1 to 7, characterized in that, Specifically, it includes: Step 1: Turn on the data acquisition module. Data begins to be output from the base station under test to the acquisition module in real time. The data acquisition module cleans and indexes the acquired data. Step 2: In the interference model construction module, the number of datasets under the same index is judged. When the number exceeds the threshold, the trained model is considered complete. Step 3: Perform 0-1 interference detection on the real-time collected data: Obtain the input parameters of the model from the base station, obtain the detection model with the same main index and a secondary index of 00 from the interference model training module; use the model and the obtained parameters to predict the retransmission rate under interference-free conditions, calculate the difference with the currently measured retransmission rate, and compare the difference with the detection threshold to obtain the 0-1 interference detection result. Step 4: Determine the detection result of Step 3: If interference occurs, send the data to the interference intensity detection module; otherwise, end the interference detection of the data and wait for the next interference detection cycle. Step 5: Detect interference intensity for the input interference data; obtain the input parameters of the model from the interference detection module, obtain the first interference detection model with the same main index and a secondary index of 00 from the interference model training module, predict the retransmission rate under interference-free conditions, calculate the difference between the retransmission rate and the currently measured retransmission rate, and compare the difference with multiple set thresholds to obtain the interference intensity detection result. Step 6: Determine the result of the interference detection. If the result is medium or strong interference, input the data into the interference source location module; otherwise, end the current interference detection and wait for the next interference detection cycle. Step 7: Locate the source of interference in the interference data; obtain the input parameters of the model from the interference intensity detection module, obtain the second interference detection model with the same main index and secondary indices of 01 and 10 from the interference model training module, predict the retransmission rate under the single neighboring cell interference condition, compare the retransmission rates under the two single neighboring cell interference conditions, and the neighboring cell with the larger retransmission rate is the interfering neighboring cell, thus locating the interference source.

Images