Prediction of 5G user plane using control plane features and granger causality for feature selection

Abstract

A system for predicting a user plane using control plane features and Granger causality includes a master device, a cloud computing server, and a controller. The controller executes control logic including a predictive application (PA) that obtains sensor data from the sensor, sends control plane data and user plane data to the infrastructure and the cloud computing server, and accesses priori knowledge data stored within a memory of the cloud computing server. Additional control logic performs a Granger causal relationship check on the user plane data and the control plane data and generates a prediction from the fused user plane data and control plane data using a prediction model. The predictive verifier applies to predictions from the predictive model, and the PA enables the master device to make reliable predictions, which allow the master device to adapt to dynamic wireless communication network conditions, even when the best possible model is not fully trained.

Description

Technical Field

[0001] The present disclosure relates to cellular communication systems and, more particularly, to short term variations in cellular network connection quality. Background Art

[0002] Cellular networks are increasingly being used by a variety of different devices, including vehicles. Vehicles often use cellular networks to communicate with other vehicles, infrastructure, etc. In order to optimize performance, manage network traffic, and enhance user experience, machine learning techniques are often used for feature selection. However, traditional machine learning techniques are usually unable to capture the relationship between control plane information and user plane metrics in dynamic environments.

[0003] Therefore, while current systems and methods for optimizing performance, managing network traffic, and enhancing user experience through user plane metric prediction achieve their intended purposes, there is still a need for new and improved systems and methods for predicting user plane metrics using control plane features that are highly adaptable, operate well in dynamic environments, improve cellular communication reliability and user experience, and can be easily applied to new and existing platforms without increasing system complexity. Summary of the invention

[0004] In several aspects of the present disclosure, a system for predicting a user plane using control plane features and Granger causality includes a host device having one or more sensors. The one or more sensors detect telematics information, telecommunication information, host device telemetry information, and host device location information. The system also includes one or more cloud computing servers and one or more controllers. Each of the one or more controllers has a processor, a memory, and one or more input / output (I / O) ports. The I / O port communicates with the one or more sensors and the one or more cloud computing servers. The memory stores program control logic. The processor executes the program control logic. The program control logic includes a prediction application (PA). The PA includes at least first, second, third, fourth, fifth, and sixth control logic parts. The first control logic obtains sensor data from one or more sensors. The second control logic sends control plane data and user plane data to the infrastructure and the cloud computing server through the I / O port. The third control logic accesses prior knowledge data stored in the memory of the cloud computing server. The fourth control logic performs a Granger causality test on the user plane data and the control plane data. The fifth control logic generates a prediction from the fused user plane data and control plane data using a prediction model. The sixth control logic applies a prediction validator to the prediction from the prediction model. PA enables the master device to make reliable predictions even when the best model is not fully trained, which allows the master device to adapt to dynamic wireless communication network conditions.

[0005] In another aspect of the present disclosure, the host device further defines a vehicle that communicates with the cloud computing server via a wireless communication network. The first control logic also includes obtaining telematics information, telecommunication information, vehicle telemetry information, and vehicle location information from one or more sensors.

[0006] In yet another aspect of the present disclosure, the second control logic further includes control logic for sending control plane data radio resource control (RRC) messages to infrastructure (including one or more cellular towers) and one or more cloud computing servers. The second control logic also receives control plane data RRC messages from one or more cellular towers and one or more cloud computing servers, and sends user plane data to one or more cellular towers and one or more cloud computing servers. The second control logic also receives user plane data from one or more cellular towers and one or more cloud computing servers.

[0007] In another aspect of the present disclosure, the third control logic further includes control logic for accessing a priori knowledge data, the a priori knowledge data including: data obtained from sensors of the master device and the attached device, data obtained from a global positioning system (GPS) satellite, data obtained from an infrastructure including one or more cellular towers. The third control logic also compares a control plane data RRC message from the master device with the a priori knowledge data to generate a partial input to the prediction validator.

[0008] In another aspect of the present disclosure, the fourth control logic also includes a control logic for applying a Granger causality test to a control plane data RRC message and user plane data. The Granger causality test outputs a P value that predicts when a change in the control plane data RRC message causes a change in the user plane data. The Granger causality test is applied by at least three different selections based on the P value, including: a threshold P value selection, a fixed number P value selection, and a conditional selection. In the threshold P value selection, all control plane data RRC features with a P value less than a predetermined threshold are used. In the fixed number P value selection, the control plane data RRC features are sorted from small to large, and a predetermined number of features with the smallest P value are selected. In the conditional selection, the control plane data RRC features from the threshold P value selection and from the fixed number P value selection are compared. When it is determined that the P value generated by the fixed number P value selection is less than the P value associated with the threshold P value selection, the fixed number P value selection indicates the Granger causality between the control plane data RRC feature and the user plane data. When it is determined that the P value generated by the threshold P value selection is less than the P value associated with the fixed number P value selection, the threshold P value selection indicates a Granger causal relationship between the control plane data RRC feature and the user plane data. When it is determined that the P values ​​generated by the threshold P value selection and the fixed number P value selection are equal to each other, then the threshold P value selection and the fixed number P value selection equally indicate a Granger causal relationship between the control plane data RRC feature and the user plane data.

[0009] In another aspect of the present disclosure, the fifth control logic also includes a control logic for using the output of the Granger causality test to reduce the amount of data fed into the prediction model from a first amount of data to a second amount of data less than the first amount of data and prevent overfeeding of the prediction model. The fifth control logic also uses one or more of a long short-term memory (LSTM) model, a recurrent neural network (RNN), or an autoregressive integrated moving average model (ARIMA) to generate a multivariate prediction from the user plane data and the control plane data output by the Granger causality test.

[0010] In another aspect of the present disclosure, the fifth control logic also includes control logic for online training the PA by initializing the PA to a zero-feature model as a baseline, sending a control plane message with Granger causality test result information and a predicted root mean square error (RMSE) of the current on-board zero-feature model to a cloud computing server, evaluating the on-board zero-feature model predicted RMSE training speed and prediction, and selectively deploying the new model to the master device.

[0011] In another aspect of the present disclosure, the sixth control logic further includes control logic for receiving data indicating that a triggering event has occurred and initializing a prediction validator using a priori knowledge and the user plane time series u(n) and the control plane time series C(n) as inputs. The sixth control logic also generates a weighted fusion statistical prediction according to the following formula: f =w1·p s +w2·p l , where w1 is the statistical prediction p generated by the statistical prediction model applied to the prior knowledge and user plane time series. s The first weight, w2 is different from the first weight and is applied to the LSTM prediction p l The second weight, and p f It is a fused prediction applied to fused user plane data and control plane data.

[0012] In yet another aspect of the present disclosure, the control logic of the PA further includes: control logic for applying non-uniform upsampling in data collection and time series formulation of user plane data and control plane data.

[0013] In another aspect of the present disclosure, the non-uniform upsampling further includes control logic that reduces bandwidth consumption and processing complexity from a first level to a second level lower than the first level by obtaining user plane data and control plane data only when a trigger event occurs and at one or more additional time points after a predefined time δt after the trigger event. When the value of the control plane time series at time δt after the trigger event is greater than or equal to the value of the control plane time series at a subsequent trigger event, the non-uniform upsampling causes the PA to increase the control plane time series to collect data when the subsequent trigger event occurs, and when the value of the control plane time series at time δt after the trigger event is less than the value of the control plane time series at the subsequent trigger event, causes the PA to upsample the control plane time series.

[0014] In another aspect of the present disclosure, a method for predicting a user plane using control plane features and Granger causality includes: using one or more sensors installed to a master device, using one or more cloud computing servers, and using one or more controllers to detect telematics information, telecommunication information, master device telemetry information, and master device location information. Each of the one or more controllers has a processor, a memory, and one or more input / output (I / O) ports. The I / O port communicates with one or more sensors and one or more cloud computing servers. The memory stores program control logic. The processor executes the program control logic. The program control logic includes a PA having a control logic that obtains sensor data from one or more sensors and sends control plane data and user plane data to an infrastructure and a cloud computing server through an I / O port. The PA control logic also accesses prior knowledge data stored in a memory of a cloud computing server, performs a Granger causality test on the user plane data and the control plane data, and generates a prediction from the fused user plane data and control plane data using a prediction model. The PA control logic also applies a prediction validator to the prediction from the prediction model, and enables the master device to make reliable predictions through the PA even when the best possible model is not fully trained, and the reliable predictions allow the master device to adapt to dynamic wireless communication network conditions.

[0015] In another aspect of the present disclosure, the main device also defines a vehicle that communicates with a cloud computing server via a wireless communication network, and the control logic of the PA also includes control logic for obtaining telematics information, telecommunication information, vehicle telemetry information, and vehicle location information from one or more sensors.

[0016] In yet another aspect of the present disclosure, the method further includes sending control plane data RRC messages to infrastructure (including one or more cellular towers) and one or more cloud computing servers, and receiving control plane data RRC messages from one or more cellular towers and one or more cloud computing servers. The method further includes sending user plane data to one or more cellular towers and one or more cloud computing servers, and receiving user plane data from one or more cellular towers and from one or more cloud computing servers.

[0017] In another aspect of the present disclosure, the method also includes accessing prior knowledge data, the prior knowledge data including: data obtained from sensors of the main device and additional devices, data obtained from GPS satellites, data obtained from infrastructure including one or more cellular towers, and comparing control plane data RRC messages of future main devices with the prior knowledge data to generate partial input to the prediction validator.

[0018] In another aspect of the present disclosure, the method further includes applying a Granger causality test to control plane data RRC messages and user plane data. The Granger causality test outputs a P value that predicts when a change in the control plane data RRC message causes a change in the user plane data. The Granger causality test is applied by at least three different selections based on the P value, including: a threshold P value selection, a fixed number P value selection, and a conditional selection. In the threshold P value selection, all control plane data RRC features with a P value less than a predetermined threshold are used. In the fixed number P value selection, the control plane data RRC features are sorted from small to large, and a predetermined number of features with the smallest P value are selected. In the conditional selection, the control plane data RRC features from the threshold P value selection and from the fixed number P value selection are compared. When it is determined that the P value generated by the fixed number P value selection is less than the P value associated with the threshold P value selection, the fixed number P value selection indicates the Granger causality between the control plane data RRC feature and the user plane data. When it is determined that the P value generated by the threshold P value selection is less than the P value associated with the fixed number P value selection, the threshold P value selection indicates a Granger causal relationship between the control plane data RRC feature and the user plane data. When it is determined that the P values ​​generated by the threshold P value selection and the fixed number P value selection are equal to each other, then the threshold P value selection and the fixed number P value selection equally indicate a Granger causal relationship between the control plane data RRC feature and the user plane data.

[0019] In another aspect of the present disclosure, the method also includes utilizing the output of the Granger causality test to reduce the amount of data fed into the prediction model from a first data amount to a second data amount that is smaller than the first data amount and prevent over-feeding of the prediction model, and utilizing one or more of an LSTM model, RNN, or ARIMA to generate multivariate predictions from user plane data and control plane data output by the Granger causality test.

[0020] In another aspect of the present disclosure, the method further includes online training of the PA by initializing the PA to a zero-feature model as a baseline, sending a control plane message with Granger causality test result information and a predicted RMSE of the current onboard zero-feature model to a cloud computing server. Online training of the PA further includes evaluating the onboard zero-feature model predicted RMSE training speed and prediction, and selectively deploying the new model to the master device.

[0021] In another aspect of the present disclosure, the method further includes receiving data indicating that a triggering event has occurred, initializing a prediction validator using prior knowledge and a user plane time series u(n) and a control plane time series C(n) as input, and generating a weighted fusion statistical prediction according to the following formula: f =w1·p s +w2·p l, where w1 is the statistical prediction p generated by the statistical prediction model applied to the prior knowledge and user plane time series. s The first weight, w2 is different from the first weight and is applied to the LSTM prediction p l The second weight, and p f It is a fused prediction applied to fused user plane data and control plane data.

[0022] In another aspect of the present disclosure, the method further includes applying non-uniform upsampling in data collection and time series formulation of user plane data and control plane data, and reducing bandwidth consumption and processing complexity from a first level to a second level lower than the first level by obtaining user plane data and control plane data only when a trigger event occurs and at one or more additional time points after a predefined time δt after the trigger event. When the value of the control plane time series at time δt after the trigger event is greater than or equal to the value of the control plane time series at a subsequent trigger event, the method causes the PA to increase the control plane time series to collect data when the subsequent trigger event occurs, and when the value of the control plane time series at time δt after the trigger event is less than the value of the control plane time series at the subsequent trigger event, the method causes the PA to upsample the control plane time series.

[0023] In another aspect of the present disclosure, a method for predicting a user plane using control plane features and Granger causality includes detecting telematics information, telecommunication information, master device telemetry information, and master device location information using one or more sensors installed on a host vehicle that communicates with one or more cloud computing servers via a wireless communication network. The method also includes detecting telematics information, telecommunication information, master device telemetry information, and master device location information using one or more controllers, each of the one or more controllers having a processor, a memory, and one or more input / output (I / O) ports, the I / O ports communicating with one or more sensors and one or more cloud computing servers. The memory stores program control logic. The processor executes the program control logic. The program control logic includes a PA. The PA includes control logic for obtaining telematics information, telecommunication information, vehicle telemetry information, and vehicle location information from one or more sensors. The PA also includes control logic for sending control plane data RRC messages to infrastructure (including one or more cellular towers) and one or more cloud computing servers via the I / O port. The PA also includes control logic for receiving control plane data RRC messages from one or more cellular towers and one or more cloud computing servers and sending user plane data to one or more cellular towers and one or more cloud computing servers. The PA also includes control logic for receiving user plane data from one or more cellular towers and one or more cloud computing servers. The PA also includes control logic for accessing prior knowledge data stored in a memory of a cloud computing server, the prior knowledge data including: data obtained from sensors of a main device and an additional device, data obtained from a GPS satellite, and data obtained from an infrastructure including one or more cellular towers. The PA also includes control logic for comparing a control plane data RRC message of a future main device with the prior knowledge data to generate a partial input to a prediction verifier. The PA also includes control logic for applying a Granger causality test to the control plane data RRC message and the user plane data. The Granger causality test outputs a P value that predicts when a change in the control plane data RRC message causes a change in the user plane data. The Granger causality test is applied through at least three different selections based on the P value, including: a threshold P value selection, a fixed number P value selection, and a conditional selection. In the threshold P value selection, all control plane data RRC features with a P value less than a predetermined threshold are used. In the fixed number P value selection, the control plane data RRC features are sorted from small to large, and a predetermined number of features with the smallest P value are selected. In the conditional selection, the control plane data RRC features from the threshold P value selection and from the fixed number P value selection are compared.When it is determined that the P value generated by the fixed number P value selection is less than the P value associated with the threshold P value selection, the fixed number P value selection indicates the Granger causality between the control plane data RRC feature and the user plane data. When it is determined that the P value generated by the threshold P value selection is less than the P value associated with the fixed number P value selection, the threshold P value selection indicates the Granger causality between the control plane data RRC feature and the user plane data. When it is determined that the P values ​​generated by the threshold P value selection and the fixed number P value selection are equal to each other, the threshold P value selection and the fixed number P value selection equally indicate the Granger causality between the control plane data RRC feature and the user plane data. The PA also includes control logic for using the output of the Granger causality test to reduce the amount of data fed into the prediction model from a first amount of data to a second amount of data less than the first amount of data and prevent overfeeding of the prediction model. The PA also includes control logic for generating multivariate predictions from the user plane data and control plane data output by the Granger causality test using one or more of an LSTM model, an RNN, or an ARIMA. The PA also includes control logic for generating predictions from fused user plane data and control plane data using a prediction model, including: online training the PA by initializing the PA to a zero-feature model as a baseline, and sending a control plane message with Granger causality test result information and the prediction RMSE of the current on-board zero-feature model to a cloud computing server. The PA also includes control logic for evaluating the on-board zero-feature model prediction RMSE training speed and prediction and selectively deploying new models to a master device. The PA also includes control logic for applying a prediction validator to predictions from the prediction model, including: receiving data indicating that a trigger event has occurred, initializing the prediction validator using prior knowledge and the user plane time series u(n) and the control plane time series C(n) as inputs, and generating a weighted fused statistical prediction according to the following formula: p. f =w1·p s +w2·p l , where w1 is the statistical prediction p generated by the statistical prediction model applied to the prior knowledge and user plane time series. s The first weight, w2 is different from the first weight and is applied to the LSTM prediction p l The second weight, and p fIt is a fused prediction applied to fused user plane data and control plane data. The PA also includes control logic for applying non-uniform upsampling in data collection and time series formulation of user plane data and control plane data, and reducing bandwidth consumption and processing complexity from a first level to a second level lower than the first level by: obtaining user plane data and control plane data only when a trigger event occurs and at one or more additional time points after a predefined time δt after the trigger event. When the value of the control plane time series at time δt after the trigger event is greater than or equal to the value of the control plane time series at a subsequent trigger event, the method causes the PA to increase the control plane time series to collect data when a subsequent trigger event occurs, and when the value of the control plane time series at time δt after the trigger event is less than the value of the control plane time series at a subsequent trigger event, the method causes the PA to upsample the control plane time series. The PA also includes control logic for enabling the master device to make reliable predictions even when the best possible model is not fully trained, and the reliable predictions allow the master device to adapt to dynamic wireless communication network conditions.

[0024] Further areas of applicability will become apparent from the description provided herein.It should be understood that the description and specific examples are intended for purposes of illustration only and are not intended to limit the scope of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The drawings described herein are for illustration purposes only and are not intended to limit the scope of the present disclosure in any way.

[0026] Figure 1 is a schematic diagram of a system for predicting 5G user plane using control plane features and Granger causality for feature selection according to one aspect of the present disclosure;

[0027] Figure 2 The invention is a diagram illustrating an aspect of the present disclosure. Figure 1 A flowchart of the logic flow of the Granger causality test portion of the system for predicting the 5G user plane using control plane features and Granger causality for feature selection;

[0028] Figure 3A The invention is a diagram illustrating an aspect of the present disclosure. Figure 1 A flowchart of the logical flow of a prediction model for a system for predicting a 5G user plane using control plane features and Granger causality for feature selection;

[0029] Figure 3B The invention is a diagram illustrating an aspect of the present disclosure. Figure 1A graph of exemplary control plane time series data for a system predicting a 5G user plane using control plane features and Granger causality for feature selection;

[0030] Figure 4 It is a description of one aspect of the present disclosure Figure 1 A flowchart of the logic flow of the prediction model portion of the system for predicting the 5G user plane using control plane features and Granger causality for feature selection;

[0031] Figure 5 The invention is a diagram illustrating an aspect of the present disclosure. Figure 1 Schematic diagram of the online training structure of the system for predicting 5G user plane using control plane features and Granger causality for feature selection;

[0032] Figure 6 The invention is a diagram illustrating an aspect of the present disclosure. Figure 1 A flowchart of a logic flow of a prediction validator of a system for predicting a 5G user plane using control plane features and Granger causality for feature selection;

[0033] Fig. 7A According to one aspect of the present disclosure Figure 1 Depiction of the timeline of data collection and time series formulation in a system for predicting 5G user plane using control plane features and Granger causality for feature selection;

[0034] Figure 7B The invention is a diagram illustrating an aspect of the present disclosure. Figure 1 A flowchart of the logical flow of the upsampling process of a system for predicting a 5G user plane using control plane features and Granger causality for feature selection; and

[0035] Figure 8 is a diagram depicting a method according to one aspect of the present disclosure Figure 1 Schematic diagram of a system for predicting 5G user plane using control plane features and Granger causality for feature selection, deploying predictions in a local area and cloud-based model training. DETAILED DESCRIPTION

[0036] The following description is merely exemplary in nature and is not intended to limit the present disclosure, application, or uses.

[0037] refer to Figure 1, a system 10 for predicting a user plane 12 using control plane 14 features and Granger causality for feature selection is shown in schematic form. The system 10 includes one or more devices, such as vehicles 16, each having one or more onboard controllers 18. It should be understood that the device may include any of a variety of wirelessly connected host devices, such as cellular phones, portable computers such as tablet computers, laptop computers, etc., without departing from the scope or intent of the present disclosure. However, for simplicity, clarity, and precision, the following description relates to a non-limiting example in which the device or host device is a vehicle 16.

[0038] The onboard controller 18 is a non-universal electronic control device having a pre-programmed digital computer or processor 20, a non-transitory computer readable medium or memory 22 for storing data such as control logic, software applications, instructions, computer code, data, lookup tables, etc., and a transceiver or input / output (I / O) port 24. Computer readable media include any type of media that can be accessed by a computer, such as read-only memory (ROM), random access memory (RAM), hard drive, compact disk (CD), digital video disk (DVD), or any other type of memory. "Non-transitory" computer readable memory 22 does not include wired, wireless, optical or other communication links that transmit transient electrical signals or other signals. Non-transitory computer readable memory 22 includes media that can permanently store data and media that can store and subsequently rewrite data, such as rewritable optical disks or erasable memory devices. Computer code includes any type of program code, including source code, object code, and executable code. The processor 20 is configured to execute code or instructions. In the vehicle 12 , the controller 18 may be a dedicated Wi-Fi controller or an engine control module, a transmission control module, a body control module, an infotainment control module, etc. The I / O port 24 is configured for wireless communication using a Wi-Fi protocol under IEEE 802.11x, a cellular protocol such as Global System for Mobile Communications (GSM), Code Division Multiple Access (CDMA), Wireless Local Loop (WLL), General Packet Radio Service (GPRS), 1G, 2G, 3G, 4G Long Term Evolution (LTE), 5G, etc.

[0039] The memory 22 may store one or more applications 26. An application 26 is a software program configured to perform a specific function or a set of functions. The application 26 may include one or more computer programs, software components, instruction sets, processes, functions, objects, classes, instances, related data, or a portion thereof suitable for implementation in a suitable computer-readable program code. The application 26 may be stored in the memory 22 of the onboard controller 18 in the vehicle 16, or in an additional or separate memory, such as in the memory 22 of a cloud computing device such as a cloud computing server 28. Examples of the application 26 include audio or video streaming services, games, browsers, social media, and applications that use control plane 14 features and Granger causality for feature selection to predict 5G user plane 12, etc. In the following, for simplicity and to improve clarity, the application that uses control plane 14 features and Granger causality for feature selection to predict 5G user plane 12 is referred to as prediction application (PA) 30 hereinafter.

[0040] In several aspects, the vehicle 16 generates, sends, and receives data while operating. The data may include telematics information, telecommunications information, vehicle telemetry information, location information, data collected from vehicle 16 sensors 31 or systems, etc. The user plane 12 carries and transmits data generated by the vehicle 16, such as data packets processed by protocols such as the Transmission Control Protocol (TCP), the User Datagram Protocol (UDP), the Internet Protocol (IP), etc. In contrast, in the control plane 14, a carrier protocol such as the Radio Resource Control (RRC) protocol handles signaling messages exchanged between one or more vehicles 16, or between end-user devices such as one or more vehicles 16 and base stations or infrastructure such as cellular towers 32.

[0041] Wireless communications, especially under GSM, CDMA, WLL, GPRS, 1G, 2G, 3G, 4G and 5G, are subject to various signaling issues that may alter the quality and reliability of data sent and received thereon. Therefore, optimization of user plane 12 metrics such as bandwidth and throughput is important in managing communication network traffic, optimizing communication network performance, and enhancing user experience.

[0042] In several aspects, the vehicle 16 sends control plane data to infrastructure such as a cellular tower 32 and / or a cloud computing server 28 in the form of an RRC message 34. Likewise, the vehicle 16 receives the control plane data RRC message 34 from the cellular tower 32, the cloud computing server 28, etc. Once the control plane 14 RRC message 34 is received by the cloud computing server 28, the cellular tower 32, etc., the RRC message is added to a database of prior knowledge 36 and compared with the prior knowledge 36 stored in the memory 22 of the cloud computing server 28 and / or the cellular tower 32 and / or in the onboard memory 22 of the vehicle 16. In several aspects, the prior knowledge 36 is a compilation of data obtained from sensors 31 of a host device such as the host vehicle 16, from sensors 31 on additional devices or vehicles 16', and from sensors 31 disposed on or as part of infrastructure such as weather satellites, GPS satellites 37, etc. However, it should be understood that the data from the additional vehicle 16' within the prior knowledge 36 database is filtered or limited in use based on predefined and / or variable metrics. The data from the additional vehicles 16' may be obtained from those additional vehicles 16' at a variety of different locations and at a variety of different time periods. It should be appreciated that in order to make accurate and relevant predictions, the system 10 is best served by utilizing only data from additional vehicles 16' that have been generated by additional vehicles 16' that are physically and / or temporally proximate to the host vehicle 16. Similarly, data from GPS satellites, weather satellites, etc. may be relevant to the host vehicle 16 for a limited time period or within a limited physical distance. Therefore, the data from GPS and weather satellites are similarly filtered.

[0043] The precise definition of "close" may vary depending on the circumstances, but should be understood to mean that the host vehicle 16 and the additional vehicle 16' are within a predetermined physical and / or temporal distance that allows data from the additional vehicle 16' to remain associated with the host vehicle 16. For example, data generated by an additional vehicle 16' traveling at seventy miles per hour on a curve on a highway in changing weather conditions will have a relatively shorter half-life than data from an additional vehicle 16' traveling at fifteen miles per hour on a straight road on a sunny, warm, dry day.

[0044] The vehicle 16 also sends user plane 12 data 38 to infrastructure such as a cellular tower 32 and / or a cloud computing server 28. Likewise, the vehicle 16 receives user plane data 38 messages from the cellular tower 32, the cloud computing server 28, etc. In several aspects, the user plane data 38 and the RRC message 34 are subjected to a Granger causality test 40. The Granger causality test is used to assist in feature selection 42 to generate a user plane 12 prediction for 5G communications. The Granger causality test 40 is used as a predictor, where if X "Granger causes" Y, it is assumed that past values ​​of X contain information that helps predict the value of Y. The Granger causality test 40 outputs a p-value with different sizes or values ​​for different control plane 14 data. In several aspects, when the p-value 44 is less than a predefined significance level, it is inferred that X does have a predictive impact on Y, thereby indicating a Granger causal relationship. Although the predefined significance level can vary depending on the application, in some examples, the significance level is less than 0.05.

[0045] Now turn Figure 2 , Figure 3A and Figure 3B , and continue to refer to Figure 1 , a method 200 for predicting a feature using the Granger causality test 40 is shown in more detail in flowchart form. In one example, the Granger causality test 40 is defined as:

[0046] u(n)=a0+a1u(n-1)+...+amu(nm)+b1c(n-1)+...+bmc(nm)+e1(n) Formula 1

[0047] u(n)=c0+c1u(n-1)+...+cmu(nm)+e2(n) Formula 2

[0048] A feature test is then performed, testing the null hypothesis that b1 = b2 = ... bm = 0. When the null hypothesis is rejected, the control plane feature c(n) Granger induces one or more user plane features u(n), where c(n) is considered as a feature for improving the prediction accuracy of u(n).

[0049] In several examples, at least three different feature selection methods based on p-values ​​of the Granger causality test 40 are utilized. Specifically, a threshold Φ selection method 202, a fixed number selection method 204, and a conditional selection method 206 are used.

[0050] The threshold Φ selection method 202 selects all c(n)s whose p-value 44 is less than the threshold Φ 208 according to the following formula:

[0051] C_1={c(n):p(c(n))<Φ},|C_1|=N1 Formula 3

[0052] Where N1 is the feature number 1 in block 210. It should be understood that the threshold Φ is a predefined value that may vary depending on the application and circumstances, but represents a value selected to ensure a predetermined level of accuracy and reliability for c(n) to be an accurate predictor based on its p-value.

[0053] Conversely, the fixed number selection method 204 calculates all p-values based on the control plane 14 data and sorts the p-values in ascending order 212. The fixed number selection method 204 then selects a predetermined number of features with the smallest P-values. In one example, the fixed number selection method 204 selects three (3) features with the smallest p-values. However, without departing from the scope or intent of the present disclosure, the exact number of features selected can vary. The selection process can be represented as selecting the N2 c(n) with the lowest p-values according to the following formula:

[0054] C_2 = {c(n): rank(p(c(n))) <= N2}, |C_2| = N2 Formula 4

[0055] Where N2 is the feature number 2 in block 214.

[0056] Finally, the conditional selection method 206 compares the values of N1 and N2 in block 216 to determine whether there is Granger causality. More precisely:

[0057] If N2 < N1, then C_3 = C_2 Formula 5

[0058] If N2 > N1, then C_3 = C_1 Formula 6

[0059] Thus, theoretically, subject to the limitation of the number of available features, C_3 is the Granger cause of u(n). That is, when N1 < N2, the fixed number selection method 204 has provided the most accurate result, and when N2 < N1, the threshold Φ selection method 202 has provided the most accurate result. Similarly, when N2 = N1, the threshold Φ selection method 202 and the fixed number selection method 204 are equally likely to produce accurate results.

[0060] Now referring more specifically to Figure 3A and Figure 3B and continuing to refer to Figure 1 and Figure 2 , twenty (20) different control plane 14 time series data points are collected and depicted as a function c(t). The Granger causality test 40 determines the probability that c(n) Granger-causes u(n) in the user plane time series u(n). As previously mentioned, the lower the p-value 44, the higher the probability that c(n) causes u(n) in Granger. Thus, Granger causality can be described as the function P(GC|p) ∝ f(p) in block 46, where f is a decreasing function.

[0061] Now turn Figure 4 And continue to refer to Figure 1 - Figure 3, the system 10 utilizes a prediction model 300 to predict short-term data rates 301. The prediction model 300 begins at box 302, where user plane 12 data [depicted as u(i:i+10)] and control plane 14 data [depicted as C(i:i+10)] are used as inputs. In several aspects, the Granger causality test 40 is used to verify the input data before using it in the prediction model 300 so as to reduce the possibility of over-feeding the prediction model 300. Over-feeding the prediction model 300 may cause the prediction model 300 to generate outputs that are actually meaningless because the prediction model 300 itself cannot determine what input data is actually valuable to the vehicle 16 or wireless communication network function. That is, the Granger causality test 40 reduces the amount of data fed into the prediction model 300 from a first amount of data to a second amount of data that is less than the first amount of data to avoid over-feeding the prediction model 300. In addition to preventing over-feeding (a factor that can reduce the predictive performance of the system 10), the Granger causality test 40 also provides the ability to select "relevant" input variables, thereby improving the predictive accuracy of the model compared to a model without such inputs.

[0062] The user plane 12 and control plane 14 data are normalized 304 and processed by a prediction algorithm 306. The prediction algorithm 306 shown is a long short-term memory (LSTM), however, the LSTM may be replaced by other feature-based time series prediction models, such as a recurrent neural network (RNN), an autoregressive integrated moving average model (ARIMA), etc., without departing from the scope or intent of the present disclosure. The prediction model 300 then performs an inverse normalization 308 and generates a user plane 14 output 310 [depicted as u'(i+11)] which is then compared to a user plane 14 target output 312 [depicted as u(i+11)] to generate a root mean square error (RMSE) 314. In several aspects, the RMSE 314 defines a metric for measuring the prediction error between the predicted output 310 and the target output 312 of the user plane 14, which is the ground truth.

[0063] Now turn Figure 5 And continue to refer to Figure 1-Figure 4, schematically illustrating an online training structure 400 of the PA 30. The PA 30 is initialized to a zero (0) feature model 402 operating on the host vehicle 16 controller 18. The zero feature model 402 is used as a baseline for subsequent comparisons with other features. The host vehicle 16 sends a current control plane 14 message with Granger causality test 40 result information and the predicted RMSE 314 of the current onboard zero feature model 402 to the cloud computing server 28. Within the cloud computing server 28, multiple models are trained in parallel. Based on the comparison of the predicted RMSE 314 and the evaluation of the training speed, the cloud computing server 28 decides whether and when to deploy a new model to the vehicle 16, and thereby replaces 404 the current model with the new model. Training based on the cloud computing server 28 enables the host vehicle 16 to make reliable predictions even when the best possible model is not fully trained. It should be understood that the number of features in a given model is related to the relative speed of the training and deployment process. That is, a single feature model 406 can be trained faster than two (2) feature models 408, etc. In a further exemplary application of the online training structure 400, the results of the single feature model 406 are compared with the two feature model 408 up to the N feature model 410. When the two feature model 408 obtains better empirical results than the single feature model 406, then the two feature model 408 is deployed to the host vehicle 16. Similarly, when the N feature model 410 obtains better results than the N-1 feature model (not specifically shown), then the N feature model 410 is deployed to the host vehicle 16. It should be understood that the training of the N feature model 410 is more computationally complex and therefore slower than the single feature model 406, the two feature model 408, etc. However, the N feature model 410 is more accurate and reliable than the N-1 feature model because the N feature model 410 is optimized according to the Granger causality test 40 of each sub-N feature model 410. Therefore, the N feature model 410 is actually an optimized, fine-tuned feature model of the host vehicle 16.

[0064] Now turn Figure 6 And continue to refer to Figures 1 to 5, a prediction validator 500 is shown in a flowchart form. Over time, the prediction validator 500 utilizes insights gained from field experiments to detect and mitigate the impact of prediction outliers on the predictions of the PA 30. In several aspects, the prediction validator 500 is initialized and generates predictions (m) upon receiving data indicating the occurrence of a trigger event (n). The trigger event (n) can be any of a variety of different types of changes in the user plane 12 and / or control plane 14 conditions. In a non-limiting example, when a handoff event occurs between multiple cellular towers 32, the change in the control plane 14 can directly change or affect the state of the user plane 12. That is, during the handoff between the cellular towers 32, the communication bandwidth between the host vehicle 16 and the cellular tower 32 can be reduced from a first level to a second level that is lower than the first level. Similarly, when line-of-sight communication between the host vehicle 16 and the cellular tower 32, etc., may be hindered by the presence of high-rise buildings, tunnels, mountains, etc., thereby causing changes in the activity of the control plane 14. That is, reference signal received power (RSRP), reference signal received quality (RSRQ), and / or received signal strength indicator (RSSI) may be negatively impacted by line-of-sight obstruction, cellular tower 32 handoffs, and the like. In another example, when the speed of the host vehicle 16 is low, multiple or frequent handoffs between cellular towers 32 may occur within a smaller physical distance, resulting in decreased bandwidth and changes in the user plane 12 experience. In yet another example, when a high-bandwidth demand application (e.g., a video streaming service application) accessed through the host vehicle 16 human machine interface (HMI) or via a user device in electronic communication with the controller 18 is terminated, the amount of bandwidth freed up for use by other systems may change significantly in the control plane 14, resulting in changes in user plane 12 functionality. Therefore, the prediction validator 500 validates and adjusts the prediction results of the LSTM prediction algorithm 306.

[0065] Therefore, the prediction validator 500 uses the prior knowledge 36 and the user plane time series u(n) and the control plane time series C(n) as input. More specifically, the user plane time series u(n) and the prior knowledge 36 are used as inputs to the statistical prediction model 502. The statistical prediction model 502 can be any of a variety of different statistical prediction model types, including but not limited to: Null Hypothesis Statistical Test (NHST) model, multivariate Kalman filter, multivariate Bayesian analysis, etc. The statistical prediction model 502 generates a statistical prediction 504 or a statistical prediction result p s Similarly, the control plane time series C(n) and user plane time series u(n) data are used as inputs to the LSTM prediction algorithm 306. The LSTM prediction algorithm 306 generates the LSTM statistical prediction 506 or the LSTM prediction result p l . Statistical prediction results p s Used to further reduce the LSTM prediction results pl The error in the and generate the final prediction p f 508, which is defined by the following formula:

[0066] p f =w1·p s +w2·p l Formula 7

[0067] Thereafter, the prediction weights (w1 and w2) are adjusted 510 to minimize the mean squared error (MSE) loss function:

[0068] L(w1,w2)=E(T-(w1·p s +w2·p l )) 2 Formula 8

[0069] The weights w1 and w2 are then updated using the gradient descent algorithm according to the following formula:

[0070]

[0071] The gradient itself is then calculated according to the following formula:

[0072]

[0073]

[0074] Where T is the ground truth and a is the learning rate. The validation process is performed under certain trigger events (n) where the prior knowledge 36 has a significant impact on the change in data rate. It should be understood that the weights w1, w2, w(j-1), w(j) may vary from application to application and may also vary from model to model, but should be understood to assist the predictive validator 500 and PA 30 by selecting which of the single feature, dual feature, or up to N feature models 406, 408, 410 is the appropriate model to address a given trigger event (n).

[0075] Now turn Fig. 7A and Figure 7B To further improve the accuracy of the prediction, the PA 30 utilizes non-uniform or uneven upsampling in the data collection and time series formulation. More specifically, the control plane 14 data is collected from RRC messages at the client or host vehicle 16. The key control plane 14 information (C) is extracted from the RRC messages and monitored according to the following formula:

[0076]

[0077] The user plane time series can be described as:

[0078]

[0079] in is the data throughput or bandwidth of the system. In several aspects, when c(n)=1, Cdetected, the detected C may be a switch between cellular towers 32 or similar cellular towers. Although uniform sampling can be used to some extent, it has several disadvantages in the current situation. Specifically, in uniform sampling, it is difficult to align the sampling point with the precise control plane 14 event, thereby introducing additional measurement uncertainty and increasing the possibility of error. Similarly, the high sampling frequency of uniform sampling leads to a large amount of data, high bandwidth consumption, and increased computational complexity of the prediction model. Non-uniform or uneven sampling provides certain advantages, namely the ability to sample based on the occurrence of control plane 14 events. However, because non-uniform or uneven sampling is only triggered when a specific control plane 14 event occurs, the user plane 12 change may be lost between multiple triggering control plane 14 events. That is, U(n) represents the average data rate between two consecutive control plane 14 events, but fails to capture the changes that occur in between. Therefore, PA 30 applies upsampling at locations that change significantly in the user plane 12.

[0080] In the data collection and time series formulation, the PA 30 double samples the user plane time series u(n) by taking an additional sampling point at time δt after each control plane 14 event. δt is obtained by estimating the time when the most significant fluctuation in throughput occurs after a specific control plane 14 event. Inserting δt-based sampled data into the sampling procedure can improve the prediction accuracy. Now turning more specifically to Figure 7B , showing the upsampling method 600 in more detail. At block 602, the PA 30 obtains the control plane time series c(n) data. At block 604, the PA 30 appends the upsampled control plane time series c_upsampled to the control plane time series c(n). At block 606, the PA 30 compares the value of the control plane time series c(n) at time δt after the trigger event (n) with the value of the control plane time series c(n) at the subsequent trigger event (n+1). When the value of the control plane time series c(n) at time δt after the trigger event (n) is greater than or equal to the value of the control plane time series c(n) at the subsequent trigger event (n+1), the control plane time series c(n) is increased to collect data when the subsequent trigger event n++ occurs. However, when the value of the control plane time sequence c(n) at time δt after the trigger event (n) is less than the value of the control plane time sequence c(n) at the subsequent trigger event (n+1), at box 608, an upsampled control plane time sequence c_upsampled is obtained and then applied to the control plane time sequence c(n).

[0081] Now turn Figure 8 And continue to refer to Figure 1-7B , the models generated by the PA 30 can be shared between the host vehicle 16 and additional vehicles 16' having similar characteristics. For example, additional vehicles 16' traveling along the same or substantially similar path at a particular physical location can deploy the same prediction model to each of the additional vehicles 16' in the similar location during the same time of day, and vice versa. That is, one or more of the multiple vehicles 16 can generate user plane 12 and / or control plane 14 predictions through the PA 30, which are then uploaded to the cloud computing server 28. The cloud computing server 28 can then deploy the user plane 12 and / or control plane 14 predictions to the additional vehicles 16' in the same vicinity. For each vehicle 16 that uploads data to the cloud computing server 28, the additional data is used to further train the model in the cloud. That is, when each vehicle 16 reports to the cloud computing server 28, the upload data rate, GPS data, speed, etc. are uploaded and used as additional training information, which can further refine and improve the accuracy of the predictions generated by the PA 30.

[0082] The disclosed system and method for predicting the user plane 12 using control plane 14 features and Granger causality for feature selection has multiple advantages. These advantages include the ability to accurately capture the complex causal relationships between control plane 14 information and user plane 12 metrics in a highly dynamic 5G environment, where changes in the control plane 14 can directly affect the user plane 12. Granger causality provides a robust and rational method for selecting features for predictive models in a practical, reliable, and low-computational impact manner, while optimizing performance, improving cellular communication reliability and user experience, operating well in dynamic environments, and being easily applicable to new and existing platforms without increasing system complexity.

[0083] The description of the present disclosure is merely exemplary in nature, and variations that do not depart from the gist of the present disclosure are intended to fall within the scope of the present disclosure. Such variations should not be regarded as departing from the spirit and scope of the present disclosure.

Claims

1. A system for predicting a user plane using control plane features and Granger causality, comprising: a master device having one or more sensors that detect telematics information, telecommunications information, master device telemetry information, and master device location information; one or more cloud computing servers; One or more controllers, each of the one or more controllers having a processor, a memory, and one or more input / output (I / O) ports, the I / O ports communicating with the one or more sensors and the one or more cloud computing servers; The memory stores program control logic; The processor executes the program control logic; The program control logic includes a prediction application PA, the PA including: first control logic for obtaining sensor data from the one or more sensors; a second control logic for sending control plane data and user plane data to the infrastructure and the cloud computing server through the I / O port; a third control logic for accessing the prior knowledge data stored in the memory of the cloud computing server; a fourth control logic, configured to perform a Granger causality test on the user plane data and the control plane data; fifth control logic for generating a prediction from the fused user plane data and control plane data using the prediction model; and sixth control logic for applying a prediction validator to predictions from the prediction model; and The PA enables the master device to make reliable predictions even when the best possible model is not fully trained, wherein the reliable predictions allow the master device to adapt to dynamic wireless communication network conditions.

2. The system according to claim 1, wherein: The main device also includes a vehicle that communicates with the cloud computing server via a wireless communication network; and wherein, The first control logic further includes: Telematics information, telecommunications information, vehicle telemetry information, and vehicle location information are obtained from the one or more sensors.

3. The system according to claim 1, wherein: The second control logic further includes: sending a control plane data radio resource control (RRC) message to an infrastructure and one or more cloud computing servers, wherein the infrastructure includes one or more cellular towers; receiving control plane data RRC messages from the one or more cellular towers and the one or more cloud computing servers; sending user plane data to the one or more cellular towers and the one or more cloud computing servers; and User plane data is received from the one or more cellular towers and the one or more cloud computing servers.

4. The system according to claim 1, wherein: The third control logic further includes: accessing prior knowledge data, the prior knowledge data comprising: data obtained from sensors of the host device and additional devices, data obtained from global positioning system GPS satellites, data obtained from infrastructure including one or more cellular towers; and A control plane data RRC message from the master device is compared with the a priori knowledge data to generate a partial input to the prediction validator.

5. The system according to claim 3, wherein: The fourth control logic further includes: applying a Granger causality test to the control plane data RRC message and the user plane data, wherein the Granger causality test outputs a P value that predicts when a change in the control plane data RRC message causes a change in the user plane data; and wherein the Granger causality test is applied by at least three different selections based on the P value, including: Threshold P value selection, fixed number P value selection and conditional selection, where In the threshold P value selection, all control plane data RRC features having a P value less than a predetermined threshold are used; In the fixed number P value selection, the control plane data RRC features are sorted from small to large, and a predetermined number of features with the smallest P values ​​are selected; and In the conditional selection, comparing control plane data RRC features from the threshold P value selection and from the fixed number P value selection; and when it is determined that the P value generated by the fixed number P value selection is less than the P value associated with the threshold P value selection, the fixed number P value selection indicates a Granger causal relationship between the control plane data RRC feature and the user plane data; When determining that the P value generated by the threshold P value selection is less than the P value associated with the fixed number of P value selections, the threshold P value selection indicates a Granger causal relationship between the control plane data RRC feature and the user plane data; and When it is determined that the P values ​​generated by the threshold P value selection and the fixed number P value selection are equal to each other, then the threshold P value selection and the fixed number P value selection equally indicate the Granger causality between the control plane data RRC feature and the user plane data.

6. The system according to claim 5, wherein: The fifth control logic further includes: Using the output of the Granger causality test to reduce the amount of data fed into the prediction model from a first amount of data to a second amount of data that is smaller than the first amount of data and to prevent over-feeding of the prediction model; One or more of a long short-term memory (LSTM) model, a recurrent neural network (RNN) or an autoregressive integrated moving average (ARIMA) model are used to generate a multivariate forecast from the user plane data and the control plane data output by the Granger causality test.

7. The system according to claim 6, further comprising: It is used to train the control logic of the PA online in the following ways: Initializing the PA to a zero-feature model as a baseline; Sending a control plane message with Granger causality test result information and a predicted root mean square error RMSE of the current on-board zero-feature model to the cloud computing server; Evaluate the prediction RMSE training speed and prediction of the on-board zero-feature model; as well as The new model is selectively deployed to the master device.

8. The system according to claim 6, wherein: The sixth control logic further includes: receiving data indicating that a triggering event has occurred; Initializing a prediction validator using prior knowledge and the user plane time series u(n) and the control plane time series C(n) as input; and The weighted fusion statistical prediction is generated according to the following formula: f =w1·p s +w2·p l , where w1 is the statistical prediction p generated by the statistical prediction model applied to the prior knowledge and user plane time series s The first weight, w2, is different from the first weight and is applied to the LSTM prediction p l The second weight, and p f It is a fused prediction applied to fused user plane data and control plane data.

9. The system according to claim 1, wherein: The control logic of the PA also includes: Control logic for applying non-uniform upsampling in data collection and time series formulation for user plane data and control plane data.

10. The system according to claim 9, wherein: The non-uniform upsampling further comprises: control logic that reduces bandwidth consumption and processing complexity from a first level to a second level below the first level by obtaining user plane data and control plane data only when a triggering event occurs and at one or more additional time points after a predefined time δt after the triggering event; and When the value of the control plane time series at time δt after the trigger event is greater than or equal to the value of the control plane time series at a subsequent trigger event, the PA increases the control plane time series to collect data when a subsequent trigger event occurs, and when the value of the control plane time series at time δt after the trigger event is less than the value of the control plane time series at the subsequent trigger event, the PA upsamples the control plane time series.

Images