System and method for generating a path loss propagation model through machine learning

Abstract

Provided are a system and a method for generating a path loss propagation model through machine learning. The present invention provides a system and a method for generating a path loss propagation model through machine learning. This system generates a path loss propagation model for a fifth-generation (5G) network for network planning. The path loss model utilizes fourth-generation (4G) user data to predict the reference signal received power / signal-to-noise and interference ratio (RSRP / SINR).

Description

Technical Field

[0001] Reservation of Rights Part of the disclosure of this patent document includes materials that are the subject of intellectual property rights owned by Jio Platforms Limited (JPL) or its affiliates (hereinafter collectively referred to as the patentee), such as, but not limited to, copyrights, designs, trademarks, integrated circuit (IC) layout designs, and / or trade dress protection. The patentee does not object to the reproduction by a third party of the patent document or patent disclosure described in the patent file or record of the Patent and Trademark Office, but reserves all other rights. All rights with respect to such intellectual property are fully reserved by the patentee.

[0002] Embodiments of the present disclosure generally relate to systems and methods for generating a reference signal received power (RSRP) prediction model in a wireless communication system. More specifically, the present disclosure relates to systems and methods for generating a path loss propagation model through machine learning.

Background Art

[0003] The following description of related art is intended to provide background information associated with the field of the present disclosure. This section may include specific aspects of the technology that may be related to various features of the present disclosure. However, this section is only intended to deepen the reader's understanding of the present disclosure and does not admit prior art.

[0004] For any wireless network, information from a path loss propagation model is required for network planning and, as a result, to provide the optimal service to end users. With the development and introduction of the fifth generation (5G) mobile communication system, a new path loss model with high accuracy is needed.

[0005] Conventionally, path loss prediction models have been constructed based on empirical or deterministic methods. The parameters of the empirical model are extracted from drive test data. Drive testing is a time-consuming and costly process because it is necessary to repeat the drive test multiple times to obtain accurate and reliable data.

[0006] In deterministic models such as ray tracing, radio wave propagation mechanisms and numerical analysis techniques are used to model computational electromagnetics. However, in an actual environment, the computational efficiency is poor and the calculation time is too long, so the implementation of a deterministic model may be difficult.

[0007] Furthermore, the mechanisms of electromagnetic wave propagation in a wireless communication system are diverse and can generally be classified into reflection, diffraction, and scattering. The complex propagation environment makes it difficult to predict the received signal strength.

[0008] Therefore, there is a need in the art to provide a system and method that can mitigate problems associated with the prior art. Object of the Invention

[0009] Some of the objectives of the present disclosure satisfied by at least one embodiment of this specification are listed below.

[0010] An object of the present disclosure is to provide a system and method that provide an intelligent and robust path loss propagation system that predicts the reference signal received power (RSRP) by utilizing actual fourth generation (4G) user data instead of 5G drive test data in the case of a fifth generation (5G) network.

[0011] An object of the present disclosure is to provide a system and method that use a machine learning method to simulate the RSRP and provide an error correction model for accurate prediction.

[0012] The object of the present disclosure is to provide a system and method with improved accuracy by using more accurate actual user data compared to drive test data for prediction.

[0013] The object of the present disclosure is to provide a system and method that utilize an optimized architecture of an artificial neural network (ANN) model to generate performance metrics that are highly accurate compared to conventional methods and provide flexibility.

Summary of the Invention

[0014] This section is provided to introduce, in a simplified form, certain objects and aspects of the present disclosure that will be described in more detail in the detailed description. This summary is not intended to identify key features or to circumscribe the scope of the claimed subject matter.

[0015] In one aspect, the present disclosure relates to a system for estimating a path loss propagation model. The system includes a processor and a memory operably coupled to the processor and storing instructions executed by the processor. The processor receives one or more data parameters associated with a primary network. The one or more data parameters are based on the network configuration of the primary network. The processor predicts a reference signal received power (RSRP) associated with the primary network via a trained learning model based on the one or more data parameters. The trained learning model is based on a trained secondary network model. The processor receives one or more user parameters, and the one or more user parameters are based on the actual RSRP received from a computing device connected to the primary network. The processor generates an error estimate via an error correction model based on the predicted RSRP and the actual RSRP. The processor determines an estimated RSRP associated with the primary network based on the error estimate.

[0016] In one embodiment, one or more data parameters may include at least one of a frequency, one or more physical parameters, and an antenna pattern associated with a primary network.

[0017] In one embodiment, the processor may predict the RSRP via a trained learning model using a naive RSRP prediction technique.

[0018] In one embodiment, one or more user parameters received by the processor may include at least one of label switch router (LSR) data, latitude data, longitude data, one or more radio frequency parameters, and device configuration data.

[0019] In one embodiment, the processor can generate an optimization model via an error correction model. The optimization model may be based on the variation between the predicted RSRP and the actual RSRP.

[0020] In one embodiment, the processor may generate an error estimate via an error correction model using a random forest technique.

[0021] In one embodiment, the trained secondary network model used by the processor may be configured to receive one or more secondary data parameters associated with the secondary network. The one or more secondary data parameters may be based on the network configuration of the secondary network. The trained secondary network model may predict the RSRP associated with the secondary network via the secondary learning model based on the one or more secondary data parameters. The trained secondary network model may receive one or more secondary user parameters. The one or more secondary user parameters may be based on the actual RSRP received from a computing device connected to the secondary network. The trained secondary network model may determine the average RSRP based on the received one or more secondary user parameters and one or more predetermined geographical frameworks associated with the computing device connected to the secondary network. The trained secondary network model may identify one or more computing devices from among one or more predetermined geographical frameworks connected to the secondary network. The trained secondary network model may generate a total average RSRP based on the average RSRP and the RSRP associated with the identified one or more computing devices. The trained secondary network model may generate a secondary error correction model based on the total average RSRP and the predicted RSRP using machine learning techniques.

[0022] In one embodiment, the secondary error correction model may be configured to receive one or more secondary user parameters and generate an activation function for calculating the measured RSRP based on the one or more secondary user parameters and the total average RSRP. The secondary error correction model may calculate the activation function using the average normalization gradient such that the difference between the predicted RSRP and the measured RSRP becomes zero.

[0023] In one embodiment, the machine learning technique can be an artificial neural network (ANN) technique.

[0024] In one embodiment, one or more predetermined geographic frameworks can include at least a geographic area associated with at least another computing device, and a topographic map associated with the at least geographic area.

[0025] In one aspect, the present disclosure relates to a method for estimating a path loss propagation model. The method includes receiving, by a processor associated with the system, one or more data parameters associated with a primary network. The one or more data parameters can be based on the network configuration of the primary network. The method includes predicting, by the processor, a reference signal received power (RSRP) associated with the primary network via a trained learning model based on the one or more data parameters. The trained learning model is based on a trained secondary network model. The method includes receiving, by the processor, one or more user parameters. The one or more user parameters are based on an actual RSRP received from a computing device connected to the primary network. The method includes generating, by the processor, an error estimate based on the predicted RSRP and the actual RSRP via an error correction model. The method includes determining, by the processor, an estimated RSRP associated with the primary network based on the error estimate.

[0026] In one embodiment, the method can include receiving, by the processor, one or more data parameters that can include at least one of a frequency, one or more physical parameters, and an antenna pattern associated with the primary network.

[0027] In one embodiment, the method may include steps of predicting RSRP by a processor through a trained learning model using a naive RSRP prediction technique.

[0028] In one embodiment, the method may include steps of receiving, by a processor, one or more user parameters that may include at least one of label switch router (LSR) data, latitude data, longitude data, one or more radio frequency parameters, and device configuration data.

[0029] In one embodiment, the method may include steps of generating, by a processor, an optimization model through an error correction model. The optimization model may be based on variations between predicted RSRP and actual RSRP.

[0030] In one embodiment, the method may include steps of generating, by a processor, an error estimate through an error correction model using a random forest technique.

[0031] In one aspect, a non - transitory computer - readable medium includes a processor with executable instructions that cause the processor to receive signals from a PHY of NR with OFDM. One or more data parameters may be based on a network configuration of a primary network. The processor predicts an RSRP associated with the primary network through a trained learning model based on the one or more data parameters. The trained learning model is based on a trained secondary network model. The processor receives one or more user parameters. The one or more user parameters are based on an actual RSRP received from a computing device connected to the primary network. The processor generates an error estimate through an error correction model based on the predicted RSRP and the actual RSRP. The processor determines an estimated RSRP associated with the primary network based on the error estimate.

[0032] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the disclosed methods and systems, and like reference numerals refer to like parts throughout the different views. The components in the drawings are not necessarily to scale, emphasis instead being placed upon clearly illustrating the principles of the present disclosure. In some of the drawings, block diagrams are used to show components, which may not represent the internal circuitry of each component. Those of ordinary skill in the art will appreciate that such a disclosure of the drawings includes a disclosure of the electrical, electronic, or circuit components commonly used to implement such components.

Brief Description of the Drawings

[0033]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9A

Figure 9B

Figure 9C

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18A

Figure 18B

Figure 18C

Figure 19

Figure 20

Figure 21A

Figure 21B

Figure 22A

Figure 22B

Figure 22C

Figure 23A

Figure 23B

Figure 23C

Figure 24A

Figure 24B

Figure 24C

Figure 25

Mode for Carrying Out the Invention

[0034] For the purpose of explanation below, various specific details are described so that embodiments of the present disclosure can be fully understood. However, it is clear that embodiments of the present disclosure can be implemented without these specific details. Some of the features described below can be used independently or in combination with other features. By themselves, individual features may not be able to solve all of the problems described above or may only be able to solve some of the problems described above. Some of the problems described above may not be fully solved by any of the features described below.

[0035] The following description provides only exemplary embodiments and is not intended to limit the scope, applicability, or configuration of the present disclosure. Rather, it is intended to provide an effective explanation for those skilled in the art to implement the exemplary embodiments. It should be understood that various changes can be made to the functions and arrangements of the elements without departing from the spirit and scope of the disclosed description.

[0036] To enable a complete understanding of the embodiments, specific details are set forth in the following description. However, those skilled in the art will understand that the embodiments can be practiced without these specific details. For example, circuits, systems, networks, processes, and other components may be presented in block diagram form so as not to obscure the embodiments with unnecessary detail. In other instances, well-known circuits, processes, algorithms, structures, and techniques may be shown without unnecessary detail so as not to obscure the embodiments.

[0037] Also, each embodiment may be described as a process represented as a flowchart, a flow diagram, a data flow diagram, a structure diagram, or a block diagram. Although the operations are described in the flowchart as sequential processes, many of the operations can be executed in parallel or simultaneously. Also, the order of the operations can be changed. When the operations are completed, the process ends, but there may be additional steps not included in the figure. A process can refer to a method, a function, a procedure, a subroutine, a subprogram, etc. When the process refers to a function, its end refers to the function returning to the calling function or the main function.

[0038] As used herein, the expressions "exemplary" and / or "illustrative" mean an example, a case, or an illustration. To avoid doubt, the subject matter disclosed herein is not limited by such examples. Further, any aspect or design described as "exemplary" and / or "illustrative" herein should not necessarily be construed as more preferred (or advantageous) than other aspects or designs, nor is it intended to exclude equivalent exemplary structures and techniques known to those skilled in the art. Further, as long as the expressions "comprising," "having," "including," and other similar expressions are used in any of the detailed description or claims, such expressions are used in an inclusive sense without excluding additional elements or other elements.

[0039] Throughout this specification, references to "one embodiment", "an embodiment", "an example" or "an example" mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present disclosure. Thus, the use of the expressions "in one embodiment" or "in an embodiment" in various places in this specification does not necessarily refer to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0040] The terms used herein are for the purpose of describing particular embodiments only and are not intended to limit the present disclosure. The phrases and expressions used herein shall be construed to include the plural as well, unless otherwise indicated by the context. Further, the expressions "comprises" and / or "comprising" used herein mean the presence of the described features, numbers, steps, operations, elements, or components, but do not preclude the presence or addition of one or more other features, numbers, steps, operations, elements, components, and / or groups thereof. The expression "and / or" used herein includes any and all combinations of one or more of the associated listed items.

[0041] Various embodiments throughout the present disclosure will be described in more detail with reference to FIGS. 1-25.

[0042] FIG. 1 shows an exemplary network architecture (100) for implementing a system (108) proposed according to an embodiment of the present disclosure.

[0043] As shown in FIG. 1, the network architecture (100) may include a system (108). The system (108) may be connected to one or more computing devices (104-1, 104-2... 104-N) via a primary network (106). The one or more computing devices (104-1, 104-2... 104-N) are interchangeably designated as user equipment (UE) (104) and are operated by one or more users (102-1, 102-2... 102-N). Further, the one or more users (102-1, 102-2... 102-N) may be interchangeably referred to as user (102) or users (102).

[0044] In one embodiment, the computing device (104) may include, but is not limited to, mobile, laptop, etc. Further, the computing device (104) may include smartphones, virtual reality (VR) devices, augmented reality (AR) devices, general-purpose computers, desktops, personal digital assistants, tablet computers, and mainframe computers. Additionally, an input device for receiving input from the user (102) such as a touchpad, touch-responsive screen, electronic pen, etc. may be used. Those skilled in the art will understand that the computing device (104) is not limited to the aforementioned devices and that various other devices may be used.

[0045] In one embodiment, the primary network (106) may include at least a portion of one or more networks, by way of example and not limitation, which have one or more nodes that transmit, receive, forward, generate, buffer, store, route, switch, process, or combine one or more messages, packets, signals, waves, voltage or current levels, or combinations thereof. The primary network (106) may include, but is not limited to, a wireless network, a wired network, the Internet, an intranet, a public network, a private network, a packet-switched network, a circuit-switched network, an ad hoc network, an infrastructure network, a public switched telephone network (PSTN), a cable network, a cellular network, a satellite network, an optical fiber network, or combinations thereof.

[0046] In one embodiment, the system (108) may receive one or more data parameters associated with the primary network (106). The one or more data parameters may be based on the network configuration of the primary network (106). The one or more data parameters received by the processor (202) may include, but are not limited to, a frequency, one or more physical parameters, and an antenna pattern associated with the primary network (106).

[0047] In one embodiment, the system (108) may predict a reference signal received power (RSRP) associated with the primary network (106) via a trained learning model based on one or more data parameters. The system (108) may predict the RSRP via the trained learning model using a naive RSRP prediction technique. The trained learning model may be based on a trained secondary network model.

[0048] In one embodiment, the trained secondary network model used by the system (108) may be configured to receive one or more secondary data parameters associated with the secondary network. The one or more secondary data parameters may be based on the network configuration of the secondary network. The trained secondary network model may be configured to predict the reference signal received power (RSRP) associated with the secondary network via a secondary learning model based on the one or more secondary data parameters. The trained secondary network model may be configured to receive one or more secondary user parameters. The one or more secondary user parameters may be based on the actual RSRP received from a computing device (104) connected to the secondary network.

[0049] In one embodiment, the trained secondary network model may be configured to determine an average RSRP based on the received one or more secondary user parameters and one or more predetermined geographical frameworks associated with a computing device (104) connected to the secondary network. The one or more predetermined geographical frameworks may include, but are not limited to, a geographical area associated with the computing device (104) and a topographic map associated with the at least geographical area.

[0050] In one embodiment, the trained secondary network model may be configured to identify one or more computing devices from one or more predetermined geographic frameworks connected to the secondary network. The trained secondary network model may be configured to generate a total average RSRP based on the average RSRP and the RSRP associated with the identified one or more computing devices. The trained secondary network model may be configured to generate a secondary error correction model based on the total average RSRP and the predicted RSRP using machine learning techniques. The machine learning techniques may be artificial neural network (ANN) techniques.

[0051] In one embodiment, the machine learning techniques may incorporate supervised learning in which a machine learning model defines the relationship between input data and a target based on training data. The machine learning model may predict the output variables of the test data based on these relationships. Since the scenario changes dynamically due to user density, changes in clutter and terrain, buildings, and other environmental factors, the machine learning model may construct the relationship between these input variables and output variables in all scenarios.

[0052] In one embodiment, the trained secondary network model may be configured to receive one or more secondary user parameters and generate an activation function for calculating the measured RSRP based on the one or more secondary user parameters and the total average RSRP. Further, the trained secondary network model may be configured to calculate the activation function using the average normalized gradient such that the difference between the predicted RSRP and the measured RSRP becomes zero.

[0053] In one embodiment, the system (108) may receive one or more user parameters. The one or more user parameters may be based on the actual RSRP received from a computing device (104) connected to the primary network (106). The one or more user parameters received by the system (108) may include, but are not limited to, label switch router (LSR) data, latitude data, longitude data, one or more radio frequency parameters, and device configuration data.

[0054] In one embodiment, the system (108) may generate an error estimate via an error correction model based on the predicted RSRP and the actual RSRP. The system (108) may use a random forest technique to generate an error estimate via the error correction model. Further, the system (108) may determine an estimated RSRP associated with the primary network (106) based on the error estimate. The system (108) may generate an optimization model via the error correction model. The optimization model may be based on the variation between the predicted RSRP and the actual RSRP.

[0055] FIG. 1 illustrates the components of the network architecture (100), but in other embodiments of the network architecture (100), the number, type, and arrangement of the components may be different from those in FIG. 1, or may include additional functions not shown in FIG. 1. Further, or alternatively, the functions described herein as being performed by one or more components of the network architecture (100) may be performed by one or more other components of the network architecture (100).

[0056] FIG. 2 shows an exemplary block diagram (200) of the system (108) proposed according to an embodiment of the present disclosure.

[0057] According to FIG. 2, the system (108) may include one or more processors (202) that can be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, logic circuits, and / or any device that processes data based on operation instructions. Among other functions, the one or more processors (202) may be configured to obtain and execute computer-readable instructions stored in the memory (204) of the system (108). The memory (204) may be configured to store one or more computer-readable instructions or routines on a non-transitory computer-readable storage medium, and these instructions or routines may be obtained and executed to create or share data packets via a network service. The memory (204) may include any non-transitory storage device, including, for example, volatile memory such as random access memory (RAM), or non-volatile memory such as erasable programmable read-only memory (EPROM), flash memory, etc.

[0058] In one embodiment, the system (108) may include an interface (206). The interface (206) may include various interfaces, such as, for example, data input / output (I / O) devices and storage device interfaces. The interface (206) may provide a communication path to one or more components of the system (108). Examples of such components may include, but are not limited to, a processing engine (208) and a database (210). The processing engine (208) may include, but is not limited to, a data capture engine (212), a machine learning engine (214), and other engines (216). In one embodiment, the other engines (216) may include, but are not limited to, a data management engine, an input / output engine, a notification engine, etc.

[0059] The processing engine(s) (208) can be implemented as a combination of hardware and programming (e.g., programmable instructions) to implement one or more functions of the processing engine (208). In the examples described herein, such a combination of hardware and programming can be implemented in several different ways. For example, the programming of the processing engine(s) (208) can be processor-executable instructions stored on a non-transitory machine-readable storage medium, and the hardware of the processing engine(s) (208) can include processing resources (e.g., one or more processors) for executing such instructions. In this embodiment, the machine-readable storage medium can store instructions that, when executed by the processing resources, implement the processing engine(s) (208). In such an embodiment, the system (108) can include a machine-readable storage medium that stores instructions and processing resources for executing the instructions, or the machine-readable storage medium can be separate but accessible to the system (108) and the processing resources. In other embodiments, the processing engine(s) (208) can be implemented by an electronic circuit.

[0060] In one embodiment, the processor (202) can receive one or more data parameters via the data ingestion engine (212). The one or more data parameters can be associated with the primary network (106). The processor (202) can store the one or more data parameters in the database (210). The one or more data parameters can be based on the network configuration of the primary network (106). The one or more data parameters received by the processor (202) can include, but are not limited to, frequency, one or more physical parameters, and an antenna pattern associated with the primary network (106).

[0061] In one embodiment, the processor (202) can predict the RSRP associated with the primary network (106) through a trained learning model based on one or more data parameters. The processor (202) can predict the RSRP through a trained learning model using a naive RSRP prediction technique. The trained learning model can be based on a trained secondary network model.

[0062] In one embodiment, the trained secondary network model used by the processor (202) can be configured to receive one or more secondary data parameters associated with the secondary network. The one or more secondary data parameters can be based on the network configuration of the secondary network. The trained secondary network model can be configured to predict the RSRP associated with the secondary network through a secondary learning model based on the one or more secondary data parameters. The trained secondary network model can be configured to receive one or more secondary user parameters. The one or more secondary user parameters can be based on the actual RSRP received from a computing device (104) connected to the secondary network.

[0063] In one embodiment, the trained secondary network model used by the processor (202) can be configured to determine an average RSRP based on the received one or more secondary user parameters and one or more predetermined geographical frameworks associated with a computing device (104) connected to the secondary network. The one or more predetermined geographical frameworks can include, but are not limited to, a geographical area associated with another computing device (104) and a topographic map associated with the at least geographical area.

[0064] In one embodiment, the trained secondary network model used by the processor (202) may be configured to identify one or more computing devices from one or more predetermined geographical frameworks connected to the secondary network. The trained secondary network model used by the processor (202) may be configured to generate a total average RSRP based on the average RSRP and one or more computing devices. The trained secondary network model used by the processor (202) may be configured to generate a secondary error correction model via machine learning techniques based on the total average RSRP and the predicted RSRP. The machine learning technique may be ANN technology generated by the machine learning engine (214).

[0065] In one embodiment, the trained secondary network model used by the processor (202) may be configured to receive one or more secondary user parameters and generate an activation function for calculating the measured RSRP based on the one or more secondary user parameters and the total average RSRP. Further, the trained secondary network model used by the processor (202) may be configured to calculate the activation function using the average normalized gradient such that the difference between the predicted RSRP and the measured RSRP becomes zero.

[0066] In one embodiment, the processor (202) may receive one or more user parameters. The one or more user parameters may be based on the actual RSRP received from the computing device (104) connected to the primary network (106). The one or more user parameters received by the processor (202) may include, but are not limited to, label switch router (LSR) data, latitude data, longitude data, one or more radio frequency parameters, and device configuration data.

[0067] In one embodiment, the processor (202) may generate an error estimate via an error correction model based on the predicted RSRP and the actual RSRP. The processor (202) may generate an error estimate via an error correction model using a random forest technique. Further, the processor (202) may determine an estimated RSRP associated with the primary network (106) based on the error estimate. The processor (202) may generate an optimization model via the error correction model. The optimization model may be based on the variations between the predicted RSRP and the actual RSRP.

[0068] FIG. 2 shows exemplary components of the system (108), but in other embodiments, the system (108) may include a different number, type, and arrangement of components than those shown in FIG. 2, or additional features not shown in FIG. 2. Further, or alternatively, functions described herein as being performed by one or more components of the system (108) may be performed by one or more other components of the system (108).

[0069] FIG. 3 shows an example (300) of a closed-loop workflow of a path loss prediction model according to an embodiment of the present disclosure.

[0070] As shown in FIG. 3, in one embodiment, the closed-loop workflow (300) may include a data collection module (302). The data collection module (300) may include, but is not limited to, data such as user data, clutter and terrain data, building data, one or more cell physical parameters, and antenna patterns.

[0071] In one embodiment, the output from the data collection module (302) may be provided to the data processing module (304). The data processing module (304) may include, but is not limited to, grid creation, free space loss, weighted clutter and building loss, diffraction loss, and directional gain.

[0072] In one embodiment, the output from the data processing module (304) can be provided to the data augmentation module (306). The data augmentation module (306) can include, but is not limited to, RSRP smoothing and local zone coefficients.

[0073] In one embodiment, the output from the data augmentation module (306) can be provided to the feature enhancement module (308). The feature enhancement module (308) can include, but is not limited to, beam bandwidth expansion, sidelobe implementation, indoor / outdoor tagging, wall loss calculation, and line-of-sight (LOS) / non-line-of-sight (NLOS) tagging.

[0074] In one embodiment, the output from the feature enhancement module (308) can be provided to the ANN model module (310). The ANN model module (310) can include, but is not limited to, training of the ANN model, adjustment of hyperparameters, and naive RSRP prediction.

[0075] In one embodiment, the output from the ANN model module (310) can be provided to the fifth-generation (5G) error correction module (312). The 5G error correction module (312) can include, but is not limited to, 5G data processing, one or more 5G parameters, 5G error calculation, and a machine learning (ML) model.

[0076] In one embodiment, the output from the 5G error correction module (312) can be provided to the model prediction module (314). The model prediction module (314) can include, but is not limited to, naive RSRP prediction, 5G error prediction, and boosted naive RSRP with 5G error.

[0077] In one embodiment, the output from the model prediction module (314) can be provided to the result verification module (316). The result verification module (316) can include, but is not limited to, error calculation, correlation, Atoll comparison, and cell match.

[0078] Figure 4 shows an exemplary architecture (400) of a closed-loop workflow according to an embodiment of the present disclosure.

[0079] As shown in Figure 4, the exemplary architecture (400) may include the following features.

[0080] In one embodiment, the geographical area may be divided into a plurality of grids, and each grid may be a square area with size parameters. Further, the geographical area may be mapped using terrain information (including but not limited to terrain, clutter, buildings, etc.), and the loss of clutter, the elevation of the terrain, the height of the buildings, and the covered area of the buildings may be examined.

[0081] In one embodiment, if there are multiple buildings on the same grid among the plurality of grids, the maximum building area may be considered. Further, as part of data collection, the number of collapsed buildings and the height of the buildings may also be recorded.

[0082] In one embodiment, user data for one week, including but not limited to label switch routers (LSRs) / NVPMs, may be collected. One or more samples may be extracted from the user data, which includes geographical properties such as latitude, longitude, environment, etc., but is not limited thereto. Further, network radio frequency (RF) parameters including RSRP, signal-to-interference-plus-noise ratio (SINR), etc. may also be recorded. Device information from the computing device (104) may also be recorded.

[0083] In one embodiment, the user data may be filtered based on a high confidence level to avoid errors based on location inaccuracies. Further, cell physical parameters from the master database may also be obtained as part of data collection.

[0084] In one embodiment, the antenna pattern can be investigated to record the antenna gain in the spatial direction. Additionally, the coverage boundary of each cell can be created based on location, coverage radius, antenna type, and demographic category. Further, when creating the coverage layer of each cell, side lobes are also considered, but their power is halved. Further, samples that do not belong to the coverage layer can be deleted for further processing.

[0085] In one embodiment, the filtered LSR samples can be mapped to a designed grid. Further, the tagging of the source cell and neighboring cells can be determined using the RSRP. Based on environmental parameters, the LSR data grid can be tagged as indoor and outdoor.

[0086] In one embodiment, the height of the user can be considered, and an appropriate coefficient can be added to the height of the user based on the terrain and buildings. Further, the height of the antenna can also be recorded based on the elevation of the terrain.

[0087] In one embodiment, the free-space loss of each grid can be calculated using antenna parameters. Conventional methods can include, but are not limited to, Cost231, ECC-3, Ericsson-9999 model, etc., and can be used based on the frequency and distance of the computing device (104). Additionally, the weighted clutter loss of each grid can be calculated based on the nearby clutter loss. Additionally, the diffraction loss can be calculated based on the number of obstacles between the grid and the cell.

[0088] In one embodiment, when calculating the wall loss effect, the change in the height of the building can be considered. When the height of the building is between or greater than the height of the sound source or receiver, an α-decibel (dB) loss can be considered, and an additional β dB for subsequent walls can be considered. Further, the weighted building loss calculation can be calculated based on the indoor / outdoor parameters.

[0089] In one embodiment, the directional loss can be calculated based on the antenna pattern data and the horizontal / vertical deviation of the grid from the main lobe of the antenna. Accordingly, the RSRP of each grid can be calculated as follows.

[0090] In one embodiment, the system (108) can calculate the weighted clutter loss, building feature extraction, diffraction loss, directional loss, and RSRP.

[0091] In one embodiment, the system (108) can calculate the RSRP as follows: Calculated RSRP = Transmission power + Antenna gain + Port gain - Conventional path loss - Directional loss - Clutter loss - Building wall loss.

[0092] In one embodiment, the system (108) can provide tuning of the path loss model based on training data, scale data, and validation data. Further, the system (108) can use an error correction model and train the error correction model with a hyperparameter tuner. Further, the system (108) can generate a path loss model.

[0093] In one embodiment, the system (108) can generate an RSRP prediction associated with a new area using the path loss model of the new area / geographical location. Further, the system (108) can generate a source / neighbor (NBR) cell prediction and a SINR calculation associated with the RSRP prediction.

[0094] In one embodiment, the system (108) can generate an RSRP prediction using the path loss model associated with the primary network (106) and generate a final prediction associated with the secondary network incorporating additional loss correction. In one embodiment, the primary network (106) can include, but is not limited to, a fourth generation (4G) network, and the secondary network can include, but is not limited to, a 5G network.

[0095] Figure 5 shows an exemplary 5G prediction workflow diagram (500) according to an embodiment of the present disclosure.

[0096] As shown in Figure 5, in one embodiment, parameters of the 5G network configuration can be received by the system (108). The parameters can include, but are not limited to, 5G frequency (502), 5G physical parameters (504), 5G multiple-input multiple-output (MIMO) antenna patterns (506), etc. The system (108) can generate a naive RSRP prediction (512) via a 4G artificial intelligence / machine learning (AI / ML) model (510). Further, the system (108) can receive 5G user data (514) and the predicted naive RSRP data and calculate an error formulation (516) associated with the generated RSRP. Further, the system (108) can receive the error formulation output and generate an error estimation (520) associated with the RSRP via a 5G ML error correction model (518). Thus, the system (108) can generate an RSRP estimation (522) associated with the 5G network.

[0097] Figure 6 shows an example of an RSRP estimation (600) based on neighboring cells within a grid according to an embodiment of the present disclosure.

[0098] As shown in Figure 6, in one embodiment, user history (602) and geographical data (604) can be used to calculate the average RSRP. Outliers (606) can be removed from the user history (602) and geographical data (604) to generate filtered data. Further, a source / NBR cell identification process (608) can be performed on the filtered data. A bucketing process (610) can be implemented and samples can be divided into consecutive incremental buckets of size α1 decibels. The minimum sample coverage area (612) can be observed from the buckets to generate the average RSRP (614).

[0099] Figure 7 shows an example (700) of the assignment of cell identifiers within a grid according to an embodiment of the present disclosure.

[0100] As shown in FIG. 7, in one embodiment, user history data can be mapped to a designed grid, and an RSRP threshold can be set based on cells to remove outliers from each grid. Further, in each grid, a minimum sample S min is expected to be obtained, and cell samples within the grid can be discarded if they do not meet the S min requirements. Further, the assignment of cells (IDs) as source cells for each grid is performed based on the maximum number of samples within that grid.

[0101] In one embodiment, to find neighboring cells, the N1 neighboring cells reported by the neighboring cells can also be considered in the calculation. Further, the neighboring cells can be selected based on the same primary synchronization signal (PSS). If the percentage % of NBR samples exceeds α%, the following procedure can be executed.

[0102] If the source <minimum source sample and NBR> minimum NBR sample and the percentage % of N1 samples > β%, the following steps can be executed. Here, for each source / NBR cell, a bucketing process can be executed to divide the input samples into consecutive increment buckets of size α1 dB. Further, the total number of samples within each bucket can be calculated, and the bucket with the largest number of samples can be selected for processing.

[0103] In one embodiment, if the percentage of samples within the selected bucket < β1%, the bucket size can be sequentially increased by 1 dB until either the % sample criterion or a β2 dB bucket size is observed. Further, the average RSRP of the samples belonging to the selected bucket can be calculated, and an RSRP value can be assigned to the grid based on the samples of the source / NBR cell.

[0104] FIG. 8 shows an exemplary diagram (800) based on a local zone coefficient according to an embodiment of the present disclosure.

[0105] As shown in FIG. 8, when the average RSRP value is assigned to the grid based on the samples of the source / NBR cell, a locally weighted smoothing process can be performed on the RSRP values with reduced high dispersion in the nearby grids. Further, a local zone coefficient can be introduced to suppress unnecessary RSRP values. Therefore, to obtain only the necessary RSRP values, the coverage area of each cell can be divided into a plurality of zones of azimuth and distance, and the RSRP value for each zone can be obtained. Further, the grid including the zone is mapped, and when the RSRP of the grid is α2 dB more or less than the coverage area, the RSRP of the grid can be replaced with the RSRP in units of zones.

[0106] In one embodiment, various conventional models can be used for the calculation of free space loss. The Cost 231 model can be implemented where the path loss of the Cost 231 model is defined as follows. PL = 46.3 + 33.9 log10(f) - 13.82 log10(hb) - ahm+(44.9 - 6.55 log10(hb)) log10(d)+cm The parameter "ahm" in an urban environment is defined as follows. ahm = 3.20(log10(11.75hr))2 - 4.97 In the case of suburban or rural (flat) environments: ahm=(1.1 log10f - 0.7)hr-(1.56 log10f - 0.8

[0107] In one embodiment, the ECC-33 model for path loss is implemented, and the path loss of the ECC-33 model is defined as follows. PL = Afs + Abm - Gb - Gr Afs = 92.4 + 20 log10(d)+20 log10(f) Abm = 20.41 + 9.83 log10(d)+7.894 log10(f)+9.56[log10(f)]^2 Gb = log10(hb / 200){13.958 + 5.8[log10(d)]^2} Gr = [42.57 + 13.7 log10(f)][log10(hr) - 0.585]

[0108] In one embodiment, the Ericsson - 9999 model for path loss is implemented, and the path loss of the Ericsson - 9999 model is defined as follows. PL = a0 + a1 log10(d) + a2 log10(hb) + a3 log10(hb) log10(d) - 3.2(log10(11.75hr)^2) + g(f) g(f) = 44.49 log10(f) - 4.78(log10(f))

[0109] In one embodiment, weighted clutter loss calculation can be used to account for additional weighting from nearby clutter. The loss due to clutter can be estimated over the maximum distance from the receiver.

[0110] In one embodiment, a weighting function can be used to calculate the weight of the clutter loss from each grid and up to the defined maximum distance in the direction of the transmitter from the grid.

[0111] Figures 9A - 9C show exemplary methods (900A, 900B, 900C) for calculating diffraction loss within a grid, according to embodiments of the present disclosure.

[0112] As shown in Figures 9A - 9C, the upper N of the buildings and / or terrain in the line of sight between the transmitter and the receiver b can be considered.

[0113] As shown in Figure 9A, in one embodiment, the Bullington method that can reduce the actual terrain to a single equivalent knife - edge can be used. The position of the equivalent knife - edge is the point where the extension line connects the transmitter and the receiver and each major obstacle meets. The diffraction loss can be calculated using the following formula.

[0114]

Equation

Equation

[0115] As shown in FIG. 9B, in one embodiment, the Deygout method can be used. In the first step, it may include calculating the "ν" parameter for only each edge as if all other edges do not exist. If edge B is the main edge, the diffraction losses J(νD_A) and J(νD_C) of edge A and edge C can be obtained with respect to the line connecting the main edge to Tx and Rx. By adding Tx and Rx to the main edge loss (J(νD_B)), the total approximate diffraction loss (LD) can be obtained. This process can be repeated until all edges are considered for three or more edges.

[0116] As shown in FIG. 9B, in one embodiment, the Causebrook method that uses a correction factor to reduce the overestimation problem associated with the Deygout method can be used. L corrected = L - C1 - C2 Here, L: Diffraction loss by the Deygout method C1 = (6 - L2 + L1)cosα1, C2 = (6 - L2 + L3)cosα2

Number

Number

[0117] As shown in FIG. 9C, in one embodiment, the Giovannelli method can be used when there are two obstacles between the transmitter and the receiver. This method can be described in more detail by the following formula.

[0118]

Number

[0119] In one embodiment, based on the identification of the grid, the weighted indoor coefficient of the building can be incorporated. The grid can be classified as indoor or outdoor. However, to know the depth of the grid within the building, the weighted value of the grid surrounding the basic grid can be calculated. The weighted indoor parameter can be estimated over the maximum distance from the base grid. Further, using a weighting function, the weight of the nearby building structure up to the defined maximum distance can be calculated. The maximum distance can indicate the distance from the base grid where nearby building structures are considered via the weighting function, and the influence of the nearby building structure can be reduced with distance.

[0120] FIG. 10 shows an example of the training (1000) of a path loss model using artificial neural network (ANN) technology according to an embodiment of the present disclosure.

[0121] In one embodiment, ANN technology can be used by a system (108) to generate a path loss model. The ANN technology is composed of an input node layer and an output node layer connected by one or more hidden node layers.

[0122] As shown in FIG. 10, the weights of the path loss model can be updated based on the comparison of the input and the actual output. A loss function can be calculated to check the loss metric, and the further derivative of the loss metric can be used to backpropagate the path loss model, and the weights can be updated until the loss metric ends or converges.

[0123] In one embodiment, in step 1, the system (108) may provide a random initialization of the model. Further, in step 2, the system (108) may receive an input and generate an actual output using a feed-forward process. In step 3, the system (108) may calculate a loss function to generate a loss (error) metric. In step 4, the system (108) may calculate the derivative of the loss (error) metric. In step 5, the system (108) may use the gradients of the previous layer and information from the stack computational graph associated with the input, and use a backpropagation process to generate gradients associated with all layers. In step 6, the system (108) may update the weights based on an update frequency and an optimization function (DELTA). In step 7, the system (108) may train the model until convergence.

[0124] FIG. 11 shows an exemplary forward propagation (1100) of a path loss model according to an embodiment of the present disclosure.

[0125] As shown in FIG. 11, in one embodiment, a forward propagation method in which an input layer node can pass information to a hidden layer node may be used by the system (108). The hidden layer applies a weighting function, and when the value of a specific node or node set in the hidden layer reaches a threshold, the activation of the activation function is started, and the value can be passed to a plurality of layer nodes in the output layer.

[0126] Providing a leveled training instance (x, y)

Number

[0127] Further, in one embodiment, the cost function may be calculated based on the following formula.

[0128]

Number

[0129] FIG. 12 illustrates an example of backward propagation of a path loss model (1200) according to one embodiment of the present disclosure.

[0130] As shown in FIG. 12, in one embodiment, the system 108 calculates the error δ of each hidden node j for each output node to which it connects. j (l) We use backpropagation, which may be responsible for some of the j (l) can be divided according to the strength of the connections between the hidden and output nodes. Blame can then be propagated backward to provide the hidden layer error value.

[0131] Furthermore, in one embodiment, the backpropagation method used by the system 108 may include multiple steps in which weights are updated with the calculated gradients. Backpropagation can be used until the weight values help achieve convergence.

[0132]

number

number

number

Number

[0133] FIG. 13 shows a training example (1300) of an ANN model designed according to an embodiment of the present disclosure.

[0134] As shown in FIG. 13, in one embodiment, the ANN model can receive an input via an input layer that includes, but is not limited to, a label switch router (LSR), building data, and clutter data. The system (108) can process the input and generate an output for transmission. An activation function can be used to generate the output generation. Further, the ANN model can include an α3 hidden layer with a plurality of nodes in each layer, and normalization and early stopping criteria can be added to these layers to avoid overfitting.

[0135] In one embodiment, the ANN model can be evaluated using a holdout evaluation method that tests the model with different data. This can provide an unbiased estimate of the learning performance associated with the trained model. In this method, the dataset can be split into three subsets.

[0136] In one embodiment, a training set can be used, in which data for m cells is selected and a prediction model for each band / demographic category can be constructed. Additionally, a validation set that uses a 20 percent subset of the training set to evaluate the performance of the model constructed during the training phase can also be used. The validation set can provide a test platform for fine-tuning the model's parameters and selecting the model with the highest performance. Further, the model can be trained using the test, and the future performance of the trained model can be evaluated using a different dataset (invisible data different from the training set). When the trained model is adjusted, input test data can be provided to the trained model, and the performance of the trained model can be evaluated using the metrics shown below. ● Mean Absolute Error:

Number

Number

Number

Number

[0137] FIG. 14 shows an exemplary block diagram (1400) for improving the performance of a path loss model, in accordance with an embodiment of the present disclosure.

[0138] As shown in FIG. 14, a baseline model (1402) can be processed to generate an optimized feature set (1404). Further, the optimized feature set (1404) can be further processed using model hyperparameter and architecture search (1406) to generate an improved model (1408).

[0139] In one embodiment, the hyperparameters can include, but are not limited to, learning rate α, number of nodes, number of hidden layers, epoch size, batch size, activation function, and number of optimizers. Further, a dropout rate and an early stopping criterion can be used to avoid overfitting.

[0140] FIG. 15 shows an exemplary diagram (1500) of RSRP prediction based on grid cells and antenna parameters, according to one embodiment of the present disclosure.

[0141] As shown in FIG. 15, in one embodiment, when an ANN model is fully trained, the same model can be used to predict the RSRP of a new area, and as a result, the SINR can be calculated using the predicted RSRP. The geographical area is divided into a plurality of grids, and each grid can be a square area with a size parameter. The geographical area is mapped using terrain information, and clutter loss, terrain elevation, building height, and building coverage area can be determined. When multiple buildings are on the same grid, the largest building area can be considered. Further, the number of buildings can also be considered as an additional feature. Further, in order to use the building features more effectively, a weighted building loss calculation incorporating indoor and outdoor scenarios can also be performed. The height of the user (here, the grid) can be estimated as "h" meters, and an appropriate coefficient can be added to the user's height based on the terrain and buildings.

[0142] FIG. 16 shows an exemplary block diagram for generating RSRP estimation (1600) of a 5G network based on a fourth-generation (4G) ANN model, according to one embodiment of the present disclosure.

[0143] As shown in FIG. 16, in one embodiment, for the prediction of RSRP, cells belonging to the same area and the same band can be considered to be from a cell master database including antenna parameters. Further, the height of the antenna can also be updated based on the elevation of the terrain. In addition, the coverage boundary of each cell can be created based on its location, coverage radius, type of antenna, and demographic category. In the case of a multi-band antenna, the beam width can be considered based on actual sample pattern analysis. When creating the coverage layer of each cell, side lobes can also be considered. Once the coverage layer is created, the free space loss associated with each grid and the combination of cells based on the coverage boundary can be calculated.

[0144] In one embodiment, the weighted clutter loss, diffraction loss, and building wall loss can also be calculated. Further, the directional gain can be calculated based on the antenna pattern data.

[0145] In one embodiment, the RSRP of each grid is calculated and can be provided to a trained model based on its band and demographic category. Further, for each grid, the predicted RSRP values of the source cell and neighboring cells can be reported by user data. If neighboring cell information is not available, up to nb maximum neighboring cells can be used based on the nearest neighbor logic. Further, the SINR can be calculated as follows. If no neighboring cells are found, a -130 dB RSRP value can be considered in the calculation of the SINR. SINR = 10 log10(source RSRP / (total (NBR RSRP) + noise power)) Noise power = -174 + 10 log10(frequency) + 7 dB (UE noise figure)

[0146] As shown in FIG. 16, in one embodiment, a 4GML model (1610) of a predetermined area can be used to calculate 5G RSRP and SINR based on a frequency (1602), a physical parameter (1604), and a 5G antenna pattern (1608). The frequency (1602), the physical parameter (1604), and the 5G antenna pattern (1608) can be provided to the 4GML model (1610) in combination with a network configuration (1606), and a 5G naive RSRP prediction (1612) can be generated. User data (1614) can be received by the system (108), and the system (108) can generate a 5GML error correction model (1618) based on the 5G naive RSRP prediction (1612) using an error formulation (1616) process. Further, the system (108) can generate an RSRP estimate (1622) based on an error estimate (1620) associated with the 5GML error correction model (1618).

[0147] FIG. 17 shows an exemplary representation (1700) of a random forest technique designed for a 5G error correction model according to one embodiment of the present disclosure.

[0148] As shown in FIG. 17, in one embodiment, the system (108) can receive 5G user data and calculate the actual 5G RSRP of each grid. Further, the system (108) can calculate the variation between the actual 5G RSRP and the predicted 5G RSRP based on the trained 4GANN model data of each grid. The system (108) can generate an error correction model using 5G configuration data as an input and dispersion as a required output.

[0149] In one embodiment, the system (108) can use a random forest technique, which can be a meta-estimator that fits a number of decision trees to various subsamples of a dataset, uses averaging to improve prediction accuracy, and controls overfitting.

[0150] In one embodiment, the following pseudo-code can be used to generate an RSRP estimate.

[0151] When b = 1 to B:

[0152] A bootstrap sample Z! of size N can be extracted from the training data.

[0153] Using the bootstrap data, the following procedure can be recursively repeated for each terminal node of the tree to generate a random forest tree Tb until the minimum node size "nmin" is reached.

[0154] "m" variables can be randomly selected from p variables.

[0155] The optimal variable / split point among the m variables can be used for calculation.

[0156] [Number]

[0157] Information gain: Information gain can be a measure of how much information a feature provides about the target. Information gain can contribute to determining the order of attributes within a node of a decision tree.

[0158] [Number]

[0159] The node is split into two daughter nodes, and predictions at a new point x can be generated using the output from the ensemble of trees {Tb}B1.

[0160] [Number]

[0161] Once the model is optimized and adjusted, the model can be used for predicting the 5G RSRP of unknown grids by summing the variance with the naive 5G RSRP prediction to generate the final 5G RSRP estimate.

[0162] Figures 18A - 18C show exemplary graphs (1800A, 1800B, 1800C) representing loss curves according to embodiments of the present disclosure.

[0163] As shown in Figures 18A - 18C, loss curves are shown, which are generated based on the mean absolute error and the epochs.

[0164] Furthermore, Table 1 shows the performance metrics for different demographic categories.

[0165] [Table 1]

[0166] Figure 19 shows an exemplary graph (1900) representing the comparison between the predicted RSRP and the actual RSRP based on the distance of any cell according to an embodiment of the present invention.

[0167] As shown in Figure 19, the comparison between the predicted RSRP and the actual RSRP can be observed based on the distance of any cell.

[0168] Furthermore, the performance metrics of the model for different demographic categories are shown in Table 2.

[0169] [Table 2]

[0170] Figure 20 shows an exemplary graph (2000) representing the RSRP prediction distribution for each demographic category / grid according to an embodiment of the present disclosure.

[0171] As shown in FIG. 20, the RSRP predicted distribution for each demographic category / grid can be observed.

[0172] FIGS. 21A-21B show exemplary representations (2100A, 2100B) of the RSRP error distribution of a grid according to an embodiment of the present disclosure.

[0173] As shown in FIGS. 21A-21B, the RSRP error distribution of the grid can be observed.

[0174] Furthermore, Table 3 shows a comparison of Atoll predictions based on the total number of cells, the calculated total area, the grid size, the total number of user samples, the morphology, the band, and the antenna pattern.

[0175]

Table 3

[0176] FIGS. 22A-22C show exemplary representations (2200A, 2200B, 2200C) of the 5G prediction results of Hyderabad according to an embodiment of the present disclosure.

[0177] As shown in FIGS. 22A-22C, the 5G prediction results of Hyderabad can be observed.

[0178] FIGS. 23A-23C show exemplary representations (2300A, 2300B, 2300C) of the 5G prediction results of Chennai according to an embodiment of the present disclosure.

[0179] As shown in FIGS. 23A-23C, the 5G prediction results of Chennai can be observed.

[0180] FIGS. 24A-24C show exemplary representations (2400A, 2400B, 2400C) of the 5G prediction results of Ahmedabad according to an embodiment of the present disclosure.

[0181] As shown in FIGS. 24A-24C, the 5G prediction results of Ahmedabad can be observed.

[0182] FIG. 25 shows an exemplary computer system (2500) in which embodiments of the present invention can be implemented or are implemented.

[0183] As shown in FIG. 25, the computer system (2500) may include an external storage device (2510), a bus (2520), a main memory (2530), a read-only memory (2540), a mass storage device (2550), communication ports (plural possible) (2560), and a processor (2570). Those skilled in the art will understand that the computer system (2500) may include multiple processors and communication ports. The processor (2570) may include various modules associated with embodiments of the present disclosure. The communication ports (plural possible) (2560) may be any of an RS-232 port used for a modem-based dial-up connection, a 10 / 100 Ethernet port, a gigabit port or a 10 gigabit port using copper wire or optical fiber, a serial port, a parallel port, or other existing or future ports. The communication ports (plural possible) (2560) may be selected according to a network such as a local area network (LAN), a wide area network (WAN), or any network to which the computer system (2500) is connected.

[0184] In one embodiment, the main memory (2530) can be a random access memory (RAM) or any other dynamic storage device commonly known in the art. The read-only memory (2540) can be any static storage device(s), such as a programmable read-only memory (PROM) chip for storing static information such as startup or basic input / output system (BIOS) instructions of the processor (2570), but is not limited thereto. The mass storage device (2550) can be any current or future mass storage device solution that can be used to store information and instructions. Mass storage solutions can include, for example, but are not limited to, parallel advanced technology attachment (PATA) or serial advanced technology attachment (SATA) hard disk drives, or solid state drives (for internal use or external use with a universal serial bus (USB) and / or Firewire interface, etc.).

[0185] In one embodiment, the bus (2520) can communicatively couple the processor(s) (2570) with other memory, storage, and communication blocks. The bus (2520) can be, for example, a peripheral component interconnect (PCI) / PCI extended (PCI-X) bus for connecting expansion cards, drives, and other subsystems, a small computer system interface (SCSI), a (USB), etc., or other buses such as a front side bus (FSB) that connects the processor (2570) to the computer system (2500).

[0186] In another embodiment, an operator and management interface, such as a display, keyboard, cursor control device, is also connected to the bus (2520) and can support direct interaction between the operator and the computer system (2500). Other operator and management interfaces can be provided through network connections connected via communication ports (s) (2560). The above components are only for the purpose of exemplifying various possibilities. The foregoing exemplary computer system (2600) is in no way intended to limit the scope of the present disclosure.

[0187] Although the present disclosure places considerable emphasis on the preferred embodiments, the present invention can take many embodiments without departing from the principles of the disclosure. It is understood that many changes may be made to the preferred embodiments. These and other changes in the preferred embodiments of the present disclosure will be apparent to those skilled in the art from the disclosure herein. It is clearly understood that the foregoing description is merely illustrative of the present disclosure and not limiting.

[0188] The present invention provides a system and method for using a machine learning model to determine a reference signal received power (RSRP) associated with a fifth generation (5G) network that is more accurate than empirical models and more computationally efficient than conventional models.

[0189] The present invention provides a system and method for generating a flexible model architecture for making predictions using a machine learning model based on a wide range of datasets.

[0190] The present invention provides a system and method for using user data together with a machine learning model that is more accurate and in line with real-world scenarios.

[0191] The present disclosure provides a system and method that can use a machine learning path loss model constructed for a fourth generation (4G) network in combination with an error correction model for frequency and antenna pattern in 5G network planning.

[0192] The present invention provides a system and method for fine-tuning a machine learning path loss model to achieve higher accuracy of performance metrics.

Claims

1. A system (108) for estimating a path loss propagation model, the system (108) comprising: a processor (202); a memory (204) operably coupled to the processor (202), the memory (204) storing instructions which, when executed by the processor (202), receive one or more data parameters associated with a primary network (106), the one or more data parameters being based on a network configuration of the primary network (106); predict a reference signal received power (RSRP) associated with the primary network (106) based on the one or more data parameters via a trained learning model, the trained learning model being based on a trained secondary network model; receive the one or more user parameters, the one or more user parameters being based on an actual RSRP received from a computing device (104) connected to the primary network (106); generate an error estimate based on the predicted RSRP and the actual RSRP via an error correction model; and determine an estimated RSRP associated with the primary network (106) based on the error estimate.

2. The system (108) of claim 1, wherein the one or more data parameters include at least one of a frequency, one or more physical parameters, and an antenna pattern associated with the primary network (106).

3. The system (108) of claim 1, wherein the processor (202) predicts the RSRP using a naive RSRP prediction technique.

4. The system (108) of claim 1, wherein the one or more user parameters include at least one of label switch router (LSR) data, latitude data, longitude data, one or more radio frequency parameters, and device configuration data.

5. The processor (202) generates an optimization model through the error correction model, and the optimization model is based on the variation between the predicted RSRP and the actual RSRP, for the system (108) according to claim 1.

6. The processor (202) generates an error estimate using random forest technology, for the system (108) according to claim 1.

7. The trained secondary network model used by the processor (202) receives one or more secondary data parameters associated with the secondary network, and the one or more secondary data parameters are based on the network configuration for the secondary network, predicts the RSRP associated with the secondary network based on the one or more secondary data parameters through a secondary learning model, receives the one or more secondary user parameters, and the one or more secondary user parameters are based on the actual RSRP received from another computing device (104) connected to the secondary network, determines the average RSRP based on the received one or more secondary user parameters and one or more predetermined geographical frameworks associated with another computing device (104) connected to the secondary network, identifies one or more computing devices from among one or more predetermined geographical frameworks connected to the secondary network, generates a total average RSRP based on the average RSRP and the RSRP associated with the identified one or more computing devices, is configured to generate a secondary error correction model based on the total average RSRP and the predicted RSRP using machine learning techniques, for the system (108) according to claim 1.

8. The secondary error correction model receives the one or more secondary user parameters and generates an activation function for calculating the measured RSRP based on the one or more secondary user parameters and the total average RSRP, The system (108) according to claim 7, configured to calculate an activation function using an average normalized gradient such that a difference between the predicted RSRP and the measured RSRP becomes zero. **Claim 9** The system (108) according to claim 7, wherein the machine learning technique is an artificial neural network (ANN) technique. **Claim 10** The system (108) according to claim 7, wherein the one or more predetermined geographical frameworks include a geographical area associated with at least another computing device (104) and a topographical map associated with the at least geographical area. **Claim 11** A method for estimating a path loss propagation model, the method comprising: Receiving, by a processor (202) associated with a system (108), one or more data parameters associated with a primary network (106), the one or more data parameters being based on a network configuration of the primary network (106); Predicting, by the processor (202) via a trained learning model, a reference signal received power (RSRP) associated with the primary network (106) based on the one or more data parameters, the trained learning model being based on a trained secondary network model; Receiving, by the processor (202), the one or more user parameters, the one or more user parameters being based on an actual RSRP received from a computing device (104) connected to the primary network (106); Generating, by the processor (202) via an error correction model, an error estimate based on the predicted RSRP and the actual RSRP; and Determining, by the processor (202) based on the error estimate, an estimated RSRP associated with the primary network (106). **Claim 12** The method according to claim 11, wherein the one or more data parameters include at least one of a frequency, one or more physical parameters, and an antenna pattern associated with the primary network (106).

13. The method according to claim 11, comprising the step of predicting the RSRP using a naive RSRP prediction technique by the processor (202).

14. The method according to claim 11, wherein the one or more user parameters include at least one of label switch router (LSR) data, latitude data, longitude data, one or more radio frequency parameters, and device configuration data.

15. The method according to claim 11, comprising the step of generating an optimization model by the processor (202) via the error correction model, wherein the optimization model is based on a variation between a predicted RSRP and an actual RSRP.

16. The method according to claim 11, comprising the step of generating an error estimate using a random forest technique by the processor (202).

17. A non-transitory computer-readable medium comprising a processor having executable instructions, receiving one or more data parameters associated with a primary network (106), wherein the one or more data parameters are based on a network configuration of the primary network (106), predicting a reference signal received power (RSRP) associated with the primary network (106) based on the one or more data parameters via a trained learning model, wherein the trained learning model is based on a trained secondary network model, receiving the one or more user parameters, wherein the one or more user parameters are based on an actual RSRP received from a computing device (104) connected to the primary network (106), generating an error estimate based on the predicted RSRP and the actual RSRP via the error correction model; and determining an estimated RSRP associated with the primary network (106) based on the error estimate, causing the processor to execute the non-transitory computer-readable medium.

Images