Mining and prediction method for uplink interference between wireless network users based on local data

Abstract

The present invention proposes a method for mining and predicting uplink interference between wireless network users based on local data, comprising: distinguishing strong-interference users from weak-interference users, and screening out strong-interference users to be predicted; collecting and organizing training data sets for each signal user based on the range of the screened-out strong-interference users to be predicted; training a model to be trained using the training data sets to obtain an interference model for the current signal user; training each signal user, and summarizing the training results of each signal user to obtain an overall interference model for the entire wireless system; and obtaining a predicted UL-SINR value at the serving user base station for a given signal user. The method is simple to implement, closer to actual network scenarios, does not consume additional pilot resources, and consumes relatively little computing resources, thereby achieving real-time, efficient, and high-precision interference prediction.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communications, and in particular relates to a method for mining and predicting uplink interference between wireless network users based on local data. Background Art

[0002] With the increasing development of communication technology, the popularity of mobile devices, and the booming development of traffic-intensive applications, today's wireless networks often adopt the method of increasing the deployment density of network equipment to meet the demand for network capacity. Ultra-dense Network (UDN) greatly improves network capacity by densely deploying access points through near-end transmission and spatial multiplexing. It is one of the key technologies of 5G. However, the density of network nodes leads to higher co-channel interference (CCI) intensity in the network.

[0003] With the centralized processing characteristics of the central unit (CU) of the 5G network architecture, although it is possible to aggregate the wireless resource allocation information generated during network operation and then extract interference information from it, it is not suitable for larger-scale wireless networks.

[0004] The interference matrix plays a vital role in resource allocation. There are two main methods for obtaining the interference matrix: one is to establish the interference matrix based on frequency sweep data, and the other is to establish the interference matrix based on the measurement report message of the mobile phone. The frequency domain information in the frequency sweep data is complete and contains latitude and longitude information, which can accurately reflect the interference situation at the sampling point. The interference matrix generated based on frequency sweep data cannot reflect the interference situation at unknown locations, especially under dense network conditions, where small position changes may bring large interference changes. The measurement cost is too high. The mobile phone measurement report contains the user's actual interference situation, but the interference information only includes a few neighboring cells with strong signals. Therefore, the interference information is incomplete and the established interference matrix has certain errors. In dense networks, the interference situation is even more severe and complex. When in a dense network with a large number of users, the agility and accuracy of the interference matrix establishment have higher requirements. However, the non-real-time and inefficient nature of the current methods, the coarse granularity of information and the rough prediction accuracy cannot adapt to the current network situation.

[0005] In existing cellular networks, the radio resource allocation function is completed by the base station. Each cell basically manages and allocates radio resources independently. To deal with inter-cell interference, the existing network compensates to a certain extent through negotiation and signaling interaction between network units and supplemented by enhanced technologies. For example, by exchanging information through the X2 or Xn interface between base stations, the inter-cell interference coordination (ICIC) or enhanced ICIC (eICIC) technology is used to solve the inter-cell interference problem; or by using coordinated multipoint transmission (Coordinated Multiple Point) CoMP (Coordination of Multiple Points) technology enables different base stations to collaboratively process interference, avoid interference, or convert interference into useful signals, providing users with higher data rates and improving network utilization. However, ICIC and eICIC technologies rely heavily on signaling, so the interference information they can transmit is extremely limited, resulting in poor granularity. Signaling transmission takes time, seriously affecting timeliness. Furthermore, the large number of adjacent cells in a UDN results in considerable signaling exchange overhead, impacting network performance. CoMP technology requires extensive channel measurements, consuming a large number of pilot resources, and consuming significant computing resources for signal processing, making it an unsuitable solution.

[0006] Furthermore, new methods for constructing interference matrices are being explored in academia. Some existing technologies leverage neural network algorithms to mine the vast amounts of data generated during network operation. Without requiring additional hardware or pilot resources, they can precisely and accurately determine the interference relationships and strengths between users and construct an interference matrix. However, this method is time-consuming to mine interference relationships and similarly fails to meet the timeliness requirements of constructing interference models. While some existing algorithms may be faster, mining the interference relationships between signal users and all potential interfering users is not scalable for larger networks.

[0007] Based on the above technical problems existing in the prior art, the present invention provides a method for mining and predicting uplink interference between wireless network users based on local data. Summary of the Invention

[0008] The present invention proposes a method for mining and predicting uplink interference between wireless network users based on local data.

[0009] The present invention adopts the following technical solutions:

[0010] A method for mining and predicting uplink interference between wireless network users based on local data, comprising:

[0011] Step 1: Estimate the interference strength of all adjacent cells based on the path loss model, distinguish between strong-interference users and weak-interference users, and screen out the strong-interference users to be predicted. The inter-user interference strength is characterized by the uplink signal-to-interference-plus-noise ratio (UL-SINR).

[0012] Step 2: Collect and organize training data sets for each signal user (service user) based on the range of strong interference users to be predicted;

[0013] Step 3: Use the training data set to train the model to be trained to obtain the interference model of the current signal user;

[0014] Step 4: Train each signal user, summarize the training results of each signal user (i.e., the interference model of this user), and obtain the total interference model of the entire wireless system;

[0015] Step 5: For a given signal user, the resource allocation vector of the interfering user to be predicted is input into the interference model of the signal user to obtain the predicted value of the UL-SINR of the serving user base station.

[0016] Furthermore, in step 1, the strong interference users to be predicted are screened out and include:

[0017] In the uplink direction, the user's transmission power is determined by the base station through open-loop power control according to the signal path loss, and the given service base station Coordinates and interference user U n Base station Coordinates The interfering user U at (x, y) can be obtained n exist The expected interference power at:

[0018]

[0019] in, is the mathematical expectation, is the coverage range of the base station where the interfering user is located, Only large-scale fading is included and calculated by the path loss model. Since the location of the base station is relatively fixed, it is considered that the location of the base station is known. The signal user will calculate the corresponding Arranged in descending order, the users served by the top 30%-40% of base stations constitute the strong interference user set. Not present The other interfering users in constitute the weak interfering user set

[0020] Further, in step 5, to the service user U mAs well as any interfering user resource allocation vector w to be predicted, the user's UL-SINR at the base station is predicted online using the following formula:

[0021]

[0022] Get service user U m The UL-SINR prediction value received by the base station on this resource block RB

[0023] Furthermore, in step 2, each time a user uses an RB, the training data set stores the resource allocation indicator variable of the interfering user on this RB and the service user U measured by the base station. m The UL-SINR on this RB is taken as a piece of data. Therefore, the data set of this user is:

[0024]

[0025] Furthermore, the model to be trained in step 3 is derived as follows:

[0026] In the uplink direction of the mobile communication network, the service user U m The transmitted uplink signal reaches the home cell The signal receiving power of the base station is:

[0027]

[0028] in, To serve users U m To the home cell The channel gain of

[0029] With service users U m Interfering users U occupying the same wireless resources n Arrival at the community The interference signal power of the base station is:

[0030]

[0031] in, To interfere with user U n To the community The channel gain of

[0032] Thus, serving user U m The signal is in the home cell The UL-SINR at the base station is:

[0033]

[0034] Among them, N m For service usersm The set of interfering users who reuse the same wireless resources, σ 2 is the noise power; further transform equation (6):

[0035]

[0036] in, To serve users U m The received signal-to-noise ratio SNR, γ m,n To serve users U m Interference user U n Signal-to-interference ratio SIR;

[0037] make Then formula (7) is simplified to:

[0038]

[0039] Based on UL-SINR data and with service users U m The set of interfering users N occupying the same wireless resources m Data, the inverse of the signal-to-interference ratio (SIR) between users is obtained through linear regression or other regression algorithms. and the inverse of the serving user's SNR Make predictions and get the signal-to-interference ratio and signal-to-noise ratio.

[0040] Furthermore, in step 1, N m Divided into strong interference user sets and weak interference user set have:

[0041]

[0042] In formula (9), Simplified to the total interference of all weak interfering users Then the above formula (9) becomes:

[0043]

[0044] Furthermore, in step 3, the service user U m The interference model is expressed as:

[0045]

[0046] During training, w,w sec ={w n ∈{0,1}|n∈N' m} is the independent variable of the model function (i.e., the set of resource allocation indicator variables of the interfering user), It is the unknown parameter in the model function, which is obtained by mining the historical data set for training.

[0047] Furthermore, in step 4, the Huber function is selected as the loss metric function:

[0048]

[0049] The Huber function uses a square loss function when the error value is less than the split point δ, and uses a linear function when the error value is greater than the split point δ.

[0050] Compared with the prior art, the present invention has the following advantages:

[0051] 1. The local data-based method for predicting uplink interference between wireless network users does not require physical equipment deployment or complete user measurement reports. Instead, it leverages the massive amount of wireless resource allocation data and wireless measurement data generated during the scheduling process. Through big data analysis and machine learning algorithms, it provides a complete and accurate uplink user interference modeling solution. This method is simple to implement, more closely matches actual network scenarios, consumes no additional pilot resources, and consumes minimal computing resources, achieving real-time, efficient, and high-precision interference prediction.

[0052] 2. The local data-based wireless network inter-user uplink interference mining and prediction method of the present invention uses a nonlinear regression-based model function that considers the causes of wireless network interference and uses a method to express inter-user SIR and user SNR in dB. This makes the algorithm's problem modeling more relevant to the causes of wireless network interference, and the unknown parameters solved (i.e., inter-user SIR and user SNR expressed in dB) are also more suitable for the modeling problem.

[0053] 3. The method for mining and predicting uplink interference between wireless network users based on local data described in the present invention does not require the geographical location information of the user terminal. At the same time, the interference model established is more targeted and closer to the wireless network environment to be applied than the general wireless propagation loss model. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Schematic diagram of a 3GPP dual-strip UDN model in an embodiment of the present invention;

[0055] Figure 2 Schematic diagram of a flow chart of a method for mining and predicting uplink interference between wireless network users based on local data in an embodiment of the present invention;

[0056] Figure 3 In the embodiment of the present invention, Below is a schematic diagram of the interference source identification error of different schemes;

[0057] Figure 4a In the embodiment of the present invention Scatter plot of UL-SINR prediction results of user U0 when ;

[0058] Figure 4b In the embodiment of the present invention, The scatter plot of the UL-SINR prediction results of user U0 is shown below;

[0059] Figure 5 In the embodiment of the present invention, Below is a schematic diagram comparing the UL-SINR prediction errors of various schemes;

[0060] Figure 6 1 is a schematic diagram comparing the average time taken by each user to obtain an interference model in different solutions in an embodiment of the present invention. DETAILED DESCRIPTION

[0061] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below with reference to the accompanying drawings and specific implementation methods. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other unless there is a conflict.

[0062] Example

[0063] like Figure 2 As shown, the method for mining and predicting uplink interference between wireless network users based on local data includes:

[0064] The interference model is trained offline to estimate the interference strength of all potential interfering users and screen out the users with strong interference to be predicted. A training data set is collected and organized for each signal user according to the selected range of strong interfering users. The training data set is used for training to obtain the interference model for the current signal user. After training for each user is completed, the interference model for the entire wireless system is summarized.

[0065] The online prediction of the interference model uses the interference model obtained through offline training, and inputs the resource allocation vector of the interference user to be predicted into the interference model obtained through offline training. In the offline training stage, the training data set is known, and the unknown parameters in the model, namely SIR and SNR, are calculated using a regression algorithm. In the online prediction stage, the resource allocation vector of the interference user to be predicted and the unknown parameters in the model obtained through offline training are known. The resource allocation vector of the interference user to be predicted and the unknown parameters in the model obtained through offline training are substituted into the nonlinear regression model function, and the predicted value of UL-SINR at the serving user base station end is calculated and output.

[0066] Specifically, during the interference model training phase, wireless network interference needs to be analyzed:

[0067] like Figure 1 As shown in the figure, in a small wireless service hotspot area, to meet the huge demand for services and throughput, the operator will deploy a large number of network equipment to form a UDN. There are two rows of rooms on each side of the corridor, each row consists of N r There are q rooms in each room, each of which has a small base station and q users, so there are a total of R = 4N r Small base stations and M=4N r q users;

[0068] In this embodiment, the purpose is to explore the interference relationship between any two users registered with different small base stations in a limited area. This requires an in-depth analysis of the generation mechanism of interference between wireless network cells.

[0069] In the uplink direction of the mobile communication network, the service user U m The transmitted uplink signal reaches the home cell The signal receiving power of the base station is:

[0070]

[0071] in, To serve users U m To the home cell The channel gain, P m The transmission power of the uplink signal serving the user.

[0072] With service users U m Interfering users U occupying the same wireless resources n Arrival at the community The interference signal power of the base station is:

[0073]

[0074] in, To interfere with user U n To the community The channel gain, P n The uplink signal transmission power of the interfering user;

[0075] Thus, serving user U m The signal is in the home cell The UL-SINR at the base station is:

[0076]

[0077] Among them, N m For service users m The set of interfering users who reuse the same wireless resources, σ 2 is the noise power;

[0078] The above formula can be further transformed to obtain:

[0079]

[0080] in, To serve users U m The received signal-to-noise ratio (SNR), γ m,n To serve users U m Interference user U n Signal-to-interference ratio SIR;

[0081] make The above formula is simplified to:

[0082]

[0083] As can be seen from the above formula, the inverse of the UL-SINR data and with service users U m The set of interfering users N occupying the same wireless resources m The inverse of the signal-to-interference ratio (SIR) between users can be obtained by linear regression or other regression algorithms. and the inverse of the serving user's SNR Make predictions;

[0084] Channel gain of interference signal Related to distance, the farther the distance, the n For service users m The less interference is caused, the better the service user U m The smaller the contribution of the total interference experienced, the more important the knowledge of the SIR of this small interfering user is for accurate prediction of U m The SINR has limited effect and is not related to U m The set of interfering users N in the entire network that occupy the same wireless resources m The signaling overhead caused by the data is too large. Therefore, N m Further divided into strong interference user sets and weak interference user set The above formula is further transformed into:

[0085]

[0086] In the above formula Then, some form of simplification is made into the total interference of all weak interfering users The above formula can be further transformed into:

[0087]

[0088] It should be noted that the composition of the interfering user set in different orthogonal resource sharing systems may be different. For example, in an orthogonal frequency division multiplexing (OFDMA) or time division multiplexing (TDMA) system, there is no interference between users in the same cell, so the interfering user set only includes users that are not in the same cell as the serving user; but in a code division multiplexing (CDMA) and non-orthogonal multiplexing (NOMA) system, there may also be interference between users in the same cell, so the interfering user set includes all users in the system except the serving user. However, since the above derivation of the interference relationship is not targeted at a specific orthogonal resource multiplexing method, by constructing different interfering user sets, the inter-user uplink interference modeling scheme obtained in this application based on the above analysis can also be applied to various wireless systems that adopt different orthogonal resource multiplexing methods. Similarly, the above derivation of the interference relationship does not depend on, for example Figure 1 The UDN networking scenario shown is shown, so the inter-user uplink interference modeling and prediction solution proposed in this embodiment is also applicable to any wireless communication networking model.

[0089] During the interference model training phase, an algorithm needs to be selected:

[0090] Regression-based machine learning algorithms can be used to predict the SIR between users and the SNR of the served user, thereby predicting the UL-SINR of the served user under a given RB allocation mode. Based on the analysis of wireless network interference, it can be found that:

[0091]

[0092] in, For all possible service users U m The set of users that cause strong interference, For all weak interference users (forming the set ) of the interference, w n ,w sec ∈{0,1} is the resource allocation indicator variable, indicating the interference user U in the current RB. n Whether it is related to service user U m Reusing the same resource and whether there are weakly interfering users occupying the current RB;

[0093] Due to the unpredictability of small-scale fading in wireless networks, in this embodiment, large-scale fading and path loss are mainly considered. Small-scale fading is regarded as an error term and is not predicted. Adding the error term ∈ to the above formula yields:

[0094]

[0095] The above equation is a linear regression model. However, small-scale fading is multiplicative and cannot be captured by an additive error term. Furthermore, UL-SINR is typically measured in dB, while linear regression uses its natural value as the optimization target, which is inconsistent with actual usage scenarios.

[0096] Therefore, in this embodiment, nonlinear regression is introduced as its model function to solve the above problems:

[0097]

[0098] Among them, ∈' is the error term;

[0099] After taking the logarithms of both sides of the equation, the multiplicative small-scale fading becomes additive, which can be well captured by the error term. In addition, the algorithm optimization target also becomes the dB value of UL-SINR, which is more in line with actual usage scenarios.

[0100] During the interference model training phase, a nonlinear model function needs to be constructed:

[0101] In this embodiment, a nonlinear regression algorithm is selected as the basis. Thus, based on the selected algorithm, the nonlinear model function is:

[0102]

[0103] Among them, w,w sec ={w n ∈{0,1}|n∈N' m} is the independent variable of the model function (i.e., the set of resource allocation indicator variables that interfere with the user), For unknown parameters in the model function, the nonlinear regression algorithm can infer their values by mining historical data sets;

[0104] The unknown parameters directly correspond to the dB values of the SIR between the serving user and the interfering user and the serving user's SNR. This processing can make the physical meaning of the unknown parameters clearer and more specific.

[0105] During the interference model training phase, it is necessary to identify strong interference users:

[0106] Small-scale fading is not considered in this embodiment due to its unpredictability, while large-scale fading is related to distance. In the uplink direction, the user's transmit power is also determined by the base station through open-loop power control according to the signal path loss. Therefore, given the serving base station Coordinates and interference user U n Base station Coordinates The interfering user U at (x, y) can be obtained n exist The expected interference power at:

[0107]

[0108] in, is the coverage range of the base station where the interfering user is located, Only large-scale fading is included and calculated by the path loss model. Since the location of the base station is relatively fixed, it is considered that the location of the base station is known. The signal user will calculate the corresponding Arranged in descending order, the users served by the top-ranked base stations constitute the set of strong interference users. Not present The other interfering users in constitute the weak interfering user set

[0109] It should be noted that although the above formula is based on a specific path loss model, the path loss model does not necessarily approximate the path loss in the real network numerically. The difference between strong and weak interference is determined only by the expected power ranking of each interfering user, and the set of strong interfering users can be reselected based on the subsequent interference model training results.

[0110] During the interference model training phase, offline training of the interference model is required:

[0111] For any user, a training data set is constructed as shown in Table 1, with user U m For example, each time a user uses an RB, the dataset will store the resource allocation indicator variable of the interfering user on this RB and the service user U measured by the base station. m The UL-SINR on this RB is taken as a piece of data. Therefore, the data set of this user can be recorded as:

[0112]

[0113] Table 1 shows the user U m As an example, the structure of the training dataset is shown:

[0114]

[0115] Table 1 User U mAn example of a training dataset

[0116] Model training involves solving all unknown parameters (i.e., the SIR between users and the SNR of the served user), that is, solving the following equation:

[0117]

[0118] It is also necessary to determine the form of the loss metric function ρ(·) before the algorithm can be trained. The most common loss metric function is the squared error function:

[0119]

[0120] That is, the prediction error for each data point is the square of the difference between the predicted value and the label value. The regression problem using squared error is also called least squares regression. The advantage of using least squares regression is that it can make full use of the data, but it has a disadvantage that cannot be ignored: it is easily affected by outliers.

[0121] In actual wireless networks, small-scale fading often fluctuates violently, resulting in a large number of outliers in the data set. Therefore, the squared error function cannot well adapt to the characteristics of the drastically changing wireless network channel environment. Therefore, this embodiment selects the Huber function as the loss metric function:

[0122]

[0123] The Huber function uses a square loss function when the error value is less than the split point δ, and a linear function when it is greater than the split point. This special construction of the Huber function can greatly reduce the impact of outliers on the final training results, making it more suitable for actual wireless networks.

[0124] After determining the loss metric function, the algorithm uses iterative algorithms such as trust region to solve the minimization problem and obtain the unknown parameter value that minimizes the sum of the prediction deviations of each data in the data set as the solution to the problem.

[0125] In the online prediction stage of the interference model:

[0126] For any service user U m As well as any interfering user resource allocation vector w to be predicted, the user's UL-SINR at the base station can be predicted online using the following formula:

[0127]

[0128] Since all γ are obtained in offline training m,n and The predicted value of and Therefore, it is only necessary to substitute the interference user resource allocation vector w to be predicted into the above formula to directly calculate the serving user U m The UL-SINR prediction value received by the base station on this RB

[0129] To further verify the superior technical effect of the local data-based wireless network inter-user uplink interference mining and prediction method described in this embodiment, the performance of the algorithm is described in terms of prediction accuracy and training time. A nonlinear regression algorithm (NLRA) is used as a comparison scheme:

[0130] As shown in Table 2, set the simulation parameters:

[0131]

[0132]

[0133] Table 2 Simulation parameters

[0134] By evaluating the performance of interference source identification, we can more completely reflect whether the interference model obtained by offline training of the proposed solution is accurate.

[0135] For any service user U m The interference source identification error performance is measured using the root mean square error (RMSE):

[0136]

[0137] in, A set of users participating in the interference source identification performance evaluation;

[0138] fixed Changing the number of strong interfering users The interference source identification performance of the two schemes is as follows: Figure 3 As shown in the figure, when n% = 26 / 72 = 36.1%, a good interference source identification effect can be achieved; see the attached Figure 5 When n% = 16 / 72 = 22.2%, good UL-SINR prediction performance can be achieved (RMSE is less than 0.5dB when strong interfering users occupy RBs). n can be adjusted according to the requirements for identification and prediction accuracy. The comparative solution NLRA considers all possible interfering users in the entire network and uses a regression algorithm to mine and predict uplink interference. Although the number of interfering users involved in data set collection and interference model training is significantly reduced, that is, the training complexity is significantly reduced, the RMSE error of the solution described in this embodiment is still within 0.5dB, with extremely high accuracy, showing high application value. As The interference source identification error of the first 16 interfering users is close to that of NLRA, which also shows that the proposed scheme can accurately identify strong interference;

[0139] By evaluating the performance of UL-SINR prediction, we can reflect the accuracy of the online prediction part of the proposed solution;

[0140] The actual data rate does not obey the Shannon formula, but depends on the modulation coding scheme (MCS). Due to the different fading characteristics of each RB, the measured SINR is used to calculate the equivalent SINR value, which is then used to derive the channel quality indicator (CQI). The CQI is then used to select the MCS. The actual data rate is determined by the MCS and the number of RBs used. Therefore, the SINR prediction performance does not need to be completely accurate, as long as it does not affect the SINR selection. For example, Figure 4a and 4b As shown in the figure, each scatter point corresponds to a piece of test data. The gray straight line in the figure represents the dividing line of different MCSs. As can be seen from the figure, when no strong interfering users occupy RBs, the scatter points corresponding to the prediction results form a straight line because different weak interfering users are treated as the same. However, as long as the left endpoint of the line segment is to the right of the rightmost gray line, the accuracy of the MCS selection is not affected.

[0141] For any service user U m The UL-SINR prediction error performance is measured using the root mean square error RMSE:

[0142]

[0143] In the above formula, the ideal UL-SINR value of the i-th test data is It does not include small-scale fading and is calculated only from path loss, inter-user interference, and noise;

[0144] At different numbers of strong interference users Under these conditions, the UL-SINR prediction performance of the two schemes is different, as follows Figure 5 As shown, the average prediction RMSE performance of the solution described in this embodiment in two cases where the strong interfering user occupies the RB and the strong interfering user does not occupy the RB is separately marked. In the case where the strong interfering user occupies the RB, the RMSE of the solution described in this embodiment is close to NLRA, showing a higher prediction accuracy;

[0145] If the training time for interference modeling and prediction is too long, the resulting interference model cannot be guaranteed to be up-to-date and cannot adapt to the rapidly changing characteristics of wireless networks. Therefore, training time is also very important for interference modeling and prediction. The accuracy of the interference model is determined by the spectrum efficiency calculated based on the MCS:

[0146]

[0147] in, and Respectively by and Calculated spectrum efficiency (SE);

[0148] like Figure 5 As shown in the figure, under different numbers of strong interfering users, the average time taken by each user to obtain the interference model is shown. Figure 6 It can be seen from the figure that with the reduction of computational complexity, the time taken by the proposed scheme to reach the same SE RMSE is greatly reduced compared with NLRA. Under the same practicality, the timeliness is greatly enhanced.

[0149] The present invention is not limited to the above-described embodiments. The above-described embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims.

Claims

1. A method for mining and predicting uplink interference between wireless network users based on local data, characterized in that: include: Step 1: Estimate the interference strength of all adjacent cells based on the path loss model, distinguish between strong interference users and weak interference users, and screen out the strong interference users to be predicted. In the uplink direction, the user's transmission power is determined by the base station through open-loop power control according to the signal path loss. Coordinates and interference user U n Base station Coordinates The interfering user U at (x, y) can be obtained n exist The expected interference power at: in, is the mathematical expectation, is the coverage range of the base station where the interfering user is located, Only large-scale fading is included and calculated by the path loss model. Since the location of the base station is relatively fixed, it is considered that the location of the base station is known. The signal user will calculate the corresponding Arranged in descending order, the users served by the top 30%-40% of base stations constitute the strong interference user set. Not present The other interfering users in constitute the weak interfering user set Among them, the inter-user interference intensity is characterized by the uplink signal-to-interference-and-noise ratio UL-SINR; Will interfere with the user set N m Divided into strong interference user sets and weak interference user set have: In formula (9), Simplified to the total interference of all weak interfering users Then the above formula (9) becomes: Step 2: Collect and organize training data sets for each signal user based on the range of strong interference users to be predicted. Each time a user uses an RB, the training data set stores the resource allocation indicator variable of the interfering user on this RB and the service user U measured by the base station. m The UL-SINR on this RB is taken as a piece of data. Therefore, the data set of this user is: Step 3: Use the training data set to train the model to be trained to obtain the interference model of the current signal user: In the uplink direction of the mobile communication network, the service user U m The transmitted uplink signal reaches the home cell The signal receiving power of the base station is: in, To serve users U m To the home cell The channel gain of With service users U m Interfering users U occupying the same wireless resources n Arrival at the community The interference signal power of the base station is: in, To interfere with user U n To the community The channel gain of Thus, serving user U m The signal is in the home cell The UL-SINR at the base station is: Among them, N m For service users m The set of interfering users who reuse the same wireless resources, σ 2 is the noise power; further transform equation (6): in, To serve users U m The received signal-to-noise ratio SNR, γ m,n To serve users U m Interference user U n Signal-to-interference ratio SIR; make Then formula (7) is simplified to: Based on UL-SINR data and with service users U m The set of interfering users N occupying the same wireless resources m Data, the inverse of the signal-to-interference ratio (SIR) between users is obtained through linear regression or other regression algorithms. and the inverse of the serving user's SNR Make predictions and obtain the signal-to-interference ratio and signal-to-noise ratio; Service User U m The interference model is expressed as: During training, w,w sec ={w n ∈{0,1}|n∈N' m } is the independent variable of the model function, It is the unknown parameter in the model function, obtained by mining historical data sets for training; Step 4: Train each signal user, summarize the training results of each signal user, and obtain the total interference model of the entire wireless system: Select the Huber function as the loss metric function: The Huber function uses a square loss function when the error value is less than the split point δ, and a linear function when it is greater than the split point δ; Step 5: For a given signal user, input the interference user resource allocation vector to be predicted into the interference model of the signal user, obtain the predicted value of UL-SINR of the serving user base station, and send the predicted value of UL-SINR to the serving user U. m And any interfering user resource allocation vector w to be predicted, the predicted value of the user's UL-SINR at the base station Use the following formula for online prediction: Get service user U m The UL-SINR prediction value received by the base station on this resource block RB

Images