WO2024000036A1

WO2024000036A1 - Signal strength prediction in complex environments

Info

Publication number: WO2024000036A1
Application number: PCT/AU2023/050603
Authority: WO
Inventors: Peng Cheng; Yizhen XU; Wanli OUYANG; Zun YAN; Yonghui Li; Branka Vucetic
Original assignee: The University Of Sydney
Priority date: 2022-06-29
Filing date: 2023-06-29
Publication date: 2024-01-04

Abstract

Disclosed herein are systems and methods for estimating signal strength of wireless communications in different environments, wherein the method may comprise cropping an image of a terrain environment into multiple image units, wherein the image includes a base station point and a measurement point; identifying environmental features, via a convolutional neural network (CNN), wherein the multiple image units are used as inputs to the CNN; applying at least one Long Short-Term Memory (LSTM) neural network, to the identified environmental features, to model a signal propagation pattern in the terrain environment; concatenating a final hidden state output of the LSTM neural network with additional communication features to produce vectors; and inputting the vectors into a deep neural network (DNN) to establish a signal strength prediction model.

Description

SIGNAL STRENGTH PREDICTION IN COMPLEX ENVIRONMENTS

FIELD OF INVENTION

[0001] The present technology pertains to systems and methods for the accurate prediction of wireless signal strength and coverage in complex and outdoor environments. In particular, but not by way of limitation, the present technology provides systems and methods for Signal Strength Prediction in Complex Environments.

BACKGROUND

[0002] The accurate prediction of wireless signal strength and coverage is a fundamental task of an operator during the planning of a wireless communication system. This ensures that a satisfactory quality of service can be provided to the end users.

[0003] Several prediction methods based on propagation path loss estimation and traditional statistical models have been proposed. However, these approaches adopt separate channel models for each individual propagation scenario. These primarily empirical models are inflexible, ignore many key environmental features, and results in significant error gaps between the predicted and real values.

[0004] Accordingly, a need exists for improved prediction of wireless signal strength and coverage in complex and outdoor environments.

[0005] Discussion or mention of any piece of prior art in this specification is not to be taken as an admission that the prior art is widely known or part of the common general knowledge of the skilled addressee of the specification.

SUMMARY

[0006] In some embodiments the present technology is directed to a method to predict signal strength of wireless communication systems in complex environments, the method comprising cropping an image of a terrain environment into multiple image units, wherein the image includes a base station point and a measurement point; identifying environmental features, via a convolutional neural network (CNN), wherein the multiple image units are used as inputs to the CNN; applying at least one Long Short-Term Memory (LSTM) neural network, to the identified environmental features, to model a signal propagation pattern in the terrain environment; concatenating a final hidden state output of the LSTM neural network with additional communication features to produce new vectors; and inputting the new vectors into a deep neural network (DNN) to establish a signal strength prediction model.

[0007] According to a first aspect of the invention, there is provided a method to predict signal strength of wireless communication systems in complex environments, the method comprising: cropping an image of a terrain environment into multiple image units, wherein the image includes a base station point and a measurement point; identifying environmental features, via a convolutional neural network (CNN), wherein the multiple image units are used as inputs to the CNN; applying at least one Long Short-Term Memory (LSTM) neural network, to the identified environmental features, to model a signal propagation pattern in the terrain environment; concatenating a final hidden state output of the LSTM neural network with additional communication features to produce vectors; and inputting the vectors into a deep neural network (DNN) to establish a signal strength prediction model.

[0008] In an embodiment, the method further comprises: generating, by the signal strength prediction model, an output representing a predicted signal strength of a specified location, based on one or more inputs representing the specified location in the terrain environment.

[0009] In an embodiment, the method further comprises: automatically selecting, based on the signal strength prediction model, a location for deployment of a base station of dense 5G mmWave networks.

[0010] In a further embodiment, the additional communication features include one or more of transmission power, transmitter antenna height, antenna pattern, carrier frequency, receiver antenna height, polarization, elevation, and transmission distance.

[0011] In a further embodiment, the convoluted neural network defines four convolution layers. In a yet further embodiment, the convoluted neural network further defines one pooling layer, and three fully connected layers.

[0012] In an embodiment, the layers of the convoluted neural network are each followed by batch normalization and a non-linear activation function.

[0013] In an embodiment, the image is a satellite image, the satellite image includes one or more image bands that identify terrain between at least one base station and at least one measurement point.

[0014] In an embodiment, the identification of environmental features comprises: extracting at least one environmental feature vector from a fully connected layer of the convoluted neural network.

[0015] In an embodiment, the applying of the at least one LSTM neural network comprises: receiving a sequence of input vectors representing the identified environmental features from the convoluted neural network; and mapping the sequence of input vectors to an output sequence, wherein the mapping is undertaken by computing activation of neural network units.

[0016] In a further embodiment, the mapping of the sequence of input vectors to an output sequence comprises: generating, via a forget gate neural network, forget gate vector outputs in the range [0,1], wherein the inputs comprise at least a previous hidden state value and the received sequence of input vectors representing environmental features; determining, an initial pointwise multiplication product of the forget gate vector outputs and a previous cell state value; generating, via an input gate neural network, input gate vector outputs, wherein the inputs comprise at least the previous hidden state value and the received sequence of input vectors; generating, via a new memory neural network, new memory update vector outputs, from inputs comprising at least the previous hidden state value and the received sequence of input vectors; determining, another pointwise multiplication product of the new memory update vector outputs and the input gate vector outputs; deriving, a new cell state value by combining the initial pointwise multiplication product and the another pointwise multiplication product; generating, via an output gate neural network, output gate vector outputs, wherein the inputs for the output gate neural network comprise at least the previous hidden state value and the received sequence of input vectors; and determining, a present hidden state value, from the new cell state value and the output gate vectors.

[0017] In a yet further embodiment, the mapping of the sequence of input vectors to an output sequence further comprises: passing the present hidden state value to another LSTM neural network.

[0018] In another embodiment, one or more of the forget gate neural network, the input gate neural network, and the output gate neural network comprise a sigmoid activation function.

[0019] In another embodiment, one or more of the forget gate neural network, the input gate neural network, and the output gate neural network comprise a previous cell state value as an input.

[0020] According to a second aspect of the invention, there is provided a system for predicting signal strength of wireless communications in complex environments, the system comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, are effective to program the at least one processor to: crop an image of a terrain environment into multiple image units, wherein the image includes a base station point and a measurement point; identify environmental features, via a convolutional neural network (CNN), wherein the multiple image units are used as inputs to the CNN; apply at least one Long Short-Term Memory (LSTM) neural network, to the identified environmental features, to model a signal propagation pattern in the terrain environment; concatenate a final hidden state output of the LSTM neural network with additional communication features to produce new vectors; and input the new vectors into a deep neural network (DNN) to establish a signal strength prediction model.

[0021] In an embodiment, the system further comprises the at least one memory storing instructions that, when executed by the at least one processor, are effective to program the at least one processor to: generate, by the signal strength prediction model, an output representing a predicted signal strength of a specified location, based on one or more inputs representing the specified location in the terrain environment.

[0022] In a further embodiment, the system further comprises the at least one memory storing instructions that, when executed by the at least one processor, are effective to program the at least one processor to:: automatically select, based on the signal strength prediction model, a location for deployment of a base station of dense 5G mmWave networks.

[0023] In a further embodiment, the additional communication features include one or more of transmission power, transmitter antenna height, antenna pattern, carrier frequency, receiver antenna height, polarization, elevation, and transmission distance.

[0024] In a further embodiment, the convoluted neural network defines four convolution layers. In a yet further embodiment, the convoluted neural network further defines one pooling layer, and three fully connected layers.

[0025] In an embodiment, the layers of the convoluted neural network are each followed by batch normalization and a non-linear activation function.

[0026] In a further embodiment, the image is a satellite image, the satellite image includes one or more image bands that identify terrain between at least one base station and at least one measurement point. [0027] In a further embodiment, the instructions to identify environmental features comprise: extract at least one environmental feature vector from a fully connected layer of the convoluted neural network.

[0028] In another embodiment, the instructions to apply the at least one LSTM neural network comprise: receive a sequence of input vectors representing the identified environmental features from the convoluted neural network; and map the sequence of input vectors to an output sequence, wherein the mapping is undertaken by computing activation of neural network units.

[0029] In a further embodiment, the instructions to map the sequence of input vectors to an output sequence comprise: generate, via a forget gate neural network, forget gate vector outputs in the range [0,1], wherein the inputs comprise at least a previous hidden state value and the received sequence of input vectors representing environmental features; determine, an initial pointwise multiplication product of the forget gate vector outputs and a previous cell state value; generate, via an input gate neural network, input gate vector outputs, wherein the inputs comprise at least the previous hidden state value and the received sequence of input vectors; generate, via a new memory neural network, new memory update vector outputs, from inputs comprising at least the previous hidden state value and the received sequence of input vectors; determine, another pointwise multiplication product of the new memory update vector outputs and the input gate vector outputs; derive a new cell state value by combining the initial pointwise multiplication product and the another pointwise multiplication product; generate, via an output gate neural network, output gate vector outputs, wherein the inputs for the output gate neural network comprise at least the previous hidden state value and the received sequence of input vectors; and determine, a present hidden state value, from the new cell state value and the output gate vectors.

[0030] In an embodiment, the instructions to apply at least one LSTM neural network further comprises: passing the present hidden state value to another LSTM neural network.

[0031] In a further embodiment, one or more of the forget gate neural network, the input gate neural network, and the output gate neural network comprise a sigmoid activation function.

[0032] In a further embodiment, one or more of the forget gate neural network, the input gate neural network, and the output gate neural network comprise a previous cell state value as an input.

[0033] According to a third aspect of the invention, there is provided a non-transitory computer-readable storage medium having embodied thereon a program, the program being executable by a processor to perform a method for providing a multi-client network service comprising: cropping an image of a terrain environment into multiple image units, wherein the image includes a base station point and a measurement point; identifying environmental features, via a convolutional neural network (CNN), wherein the multiple image units are used as inputs to the CNN; applying at least one Long Short-Term Memory (LSTM) neural network, to the identified environmental features, to model a signal propagation pattern in the terrain environment; concatenating a final hidden state output of the LSTM neural network with additional communication features to produce new vectors; inputting the new vectors into a deep neural network (DNN) to establish a signal strength prediction model; generating, by the signal strength prediction model, an output representing a predicted signal strength of a specified location, based on one or more inputs representing the specified location in the terrain environment; and automatically selecting, based on the signal strength prediction model, a location for deployment of a base station of dense 5G mmWave networks.

[0034] It is an object of the present invention to overcome or ameliorate at least one of the disadvantages of the prior art, or to provide a useful alternative.

[0035] Other embodiments of the invention will be evident from the following detailed description of various aspects of the invention. Additional features and advantages are described in, and will also be apparent from, the following Detailed Description and the Figures. The features and advantages described herein are not all-inclusive and, in particular, many additional features and advantages will be apparent to one of ordinary skill in the art in view of the figures and description. Also, any particular embodiment does not have to have all of the advantages listed herein and it is expressly contemplated to claim individual advantageous embodiments separately. Moreover, it should be noted that the language used in the specification has been selected principally for readability and instructional purposes, and not to limit the scope of the inventive subject matter.

BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In the description, for purposes of explanation and not limitation, specific details are set forth, such as particular embodiments, procedures, techniques, etc. to provide a thorough understanding of the present technology. However, it will be apparent to one skilled in the art that the present technology may be practiced in other embodiments that depart from these specific details.

[0037] The accompanying drawings, where like reference numerals refer to identical or functionally similar elements throughout the separate views, together with the detailed description below, are incorporated in and form part of the specification, and serve to further illustrate embodiments of concepts that include the claimed disclosure and explain various principles and advantages of those embodiments.

[0038] The methods and systems disclosed herein have been represented where appropriate by conventional symbols in the drawings, showing only those specific details that are pertinent to understanding the embodiments of the present disclosure so as not to obscure the disclosure with details that will be readily apparent to those of ordinary skill in the art having the benefit of the description herein. [0039] FIG. 1 presents a flow diagram of one embodiment of a method to predict and utilize a signal strength prediction model in a complex environment.

[0040] FIG. 2 presents a flow diagram of the steps employed by one embodiment of a Long Short-Term Memory (LSTM) neural network deployed in the signal strength prediction model. [0041] FIG. 3 presents a satellite image of a terrain information used in the signal strength prediction model.

[0042] FIG. 4 presents a schematic diagram that illustrates a preferred embodiment of the overall signal strength prediction training model.

[0043] FIG. 5 presents a schematic diagram to represent the different layers of the convolutional neural network (CNN) utilized in the signal strength prediction model.

[0044] FIG. 6 presents a schematic diagram representing the structure of LSTM neural networks utilized in the signal strength prediction model.

[0045] FIG. 7 presents a diagrammatic representation of an example machine in the form of a computer system capable of running the methods and models described herein.

[0046] FIG 8. illustrates an example of a map incorporating collected data.

[0047] FIG. 9 presents a bar graph of experimental results of the presented methods and models in comparison to a traditional model.

DETAILED DESCRIPTION

Definitions

[0048] The terms “a,” “an,” “the” and similar referents used in the context of describing embodiments (especially in the context of the following claims) are to be construed to cover both the singular and the plural, unless otherwise indicated herein or clearly contradicted by context.

[0049] The use of any and all examples, or exemplary language (e.g., “such as”) provided herein is intended merely to provide better illumination and does not pose a limitation on the scope of any embodiments otherwise claimed. No language in the specification should be construed as indicating any non-claimed element is essential.

[0050] The terms "comprise", "comprises", "comprised" or "comprising", "including" or "having" and the like in the present specification and claims are used in an inclusive sense, that is to specify the presence of the stated features but not preclude the presence of additional or further features.

[0051] The approaches described in this section could be pursued but are not necessarily approaches that have previously been conceived or pursued. Therefore, unless otherwise indicated, it should not be assumed that any of the approaches described in this section qualify as prior art merely by virtue of their inclusion.

[0052] The terms “computer-readable storage medium” and “computer-readable storage media” as used herein refer to any medium or media that participate in providing instructions to a CPU for execution. Such media can take many forms, including, but not limited to, non-volatile media, volatile media, and transmission media. Non-volatile media include, for example, optical or magnetic disks, such as a fixed disk. Volatile media include dynamic memory, such as system RAM. Transmission media include coaxial cables, copper wire and fiber optics, among others, including the wires that comprise one embodiment of a bus. T ransmission media can also take the form of acoustic or light waves, such as those generated during radio frequency (RF) and infrared (IR) data communications. Common forms of computer-readable media include, for example, a flexible disk, a hard disk, magnetic tape, any other magnetic medium, a CD-ROM disk, digital video disk (DVD), any other optical medium, any other physical medium with patterns of marks or holes, a RAM, a PROM, an EPROM, an EEPROM, a FLASHEPROM, any other memory chip or data exchange adapter, a carrier wave, or any other medium from which a computer can read.

Accurate wireless signal strength and wireless coverage prediction

[0053] The accurate prediction of wireless signal strength and wireless coverage is a fundamental task of an operator during the planning of the deployment of a wireless communication system such as a base station structure or cellular tower. Accurately predicting signal strength and quality ensures that a satisfactory level of service can be provided to end users. This is of special importance in complicated terrains and outdoor environments, where environmental features, and terrain topography play a major role in the propagation of signals to users from deployed base towers or other communication structures. The costs of installing communication lines, as well as deploying base stations and cellular communication towers are investments with large overheads intended for long-term use, therefore the prediction of the signal strength generated by such bases, structures, and towers is vital to efficient and effective planning of such long-term projects.

[0054] Current approaches used to predict wireless communication signal strengths based on propagation path loss estimation or traditional statistical models are inflexible, and fail to take into consideration key environmental features in the terrain environment proposed for the deployment of base stations, communication structures, and towers, resulting in significant error gaps between predicted and real-values (The terrain environment for deployment of base stations may hereinafter be referred to as the “deployment environment”.) The technologies presented herein provide systems and methods for the prediction and modelling of signal strength in environments under consideration for the deployment of base stations, communication towers, or other wireless communication structures (hereinafter referred to as “prediction model”, or “the model/models”). The models presented herein take into account the environmental features necessary to producing an accurate prediction of signal strength.

[0055] This application presents a novel and general data-driven propagation modelling framework applicable to various complex outdoor propagation environments of wireless communication systems. The techniques presented herein can achieve a high level of accuracy in the prediction of the signal strength of survey sites based on local measurements. Through this technique, a novel environment model is first developed using high resolution satellite images that capture the propagation environment features based on a convolutional neural network (CNN). In the next stage, a detailed propagation model is developed by combining a long short-term memory (LSTM) network and a deep neural network (DNN), which takes the previously extracted environment model features and the relevant communication network parameters as its input. By training the end-to-end CNN-LSTM-DNN weights, a high-dimensional non-linear relationship between the input and output vectors that represents the signal strength, can be established. This presented framework enables the generation of cellular network propagation models that are much more accurate than the existing theoretical and empirical propagation models for any environment where geographical maps are available, and does so by using only a limited set of data for the initial training of the CNN-LSTM-DNN network. This approach is suitable for developing propagation model software that can be used for automated network planning and, in particular, forthe deployment and optimal base station selection of dense 5G mmWave networks.

[0056] During the training process, the proposed CNN-LSTM-DNN model not only effectively captures the hidden pattern of the time-series satellite image data, but also utilizes the relevant communication network parameters including transmit power, transmitter antenna height, antenna pattern, carrier frequency, receiver antenna height, and transmission distance. By training the end-to-end CNN-LSTM-DNN weights, a high-dimensional non-linear relationship between the input and output vectors that represents the signal strength, can be established. During the inference (prediction) process, given an arbitrary new location along with its satellite image band and relevant communication parameters, this high-dimensional nonlinear relationship is directly exploited to generate an accurate signal strength with low complexity.

[0057] While the present technology is susceptible of embodiment in many different forms, there is shown in the drawings and will herein be described in detail several specific embodiments with the understanding that the present disclosure is to be considered as an exemplification of the principles of the present technology and is not intended to limit the technology to the embodiments illustrated.

[0058] FIG. 1 presents a flow diagram of one embodiment of a method 100 to predict and utilize a signal strength prediction model in a complex environment. The method 100 may utilize any form of image of an environment where wireless communications are proposed to be deployed (hereinafter may be referred to as “deployment environment”), in one embodiment a satellite image of the location/area for the deployment of a wireless communication system is used, and is cropped 105 into multiple different images or image units. The cropped images are then fed into a CNN, whereby the CNN uses the multiple images or image units to identify 1 10 environmental features in the deployment environment. The environmental features identified are then used as vector inputs 1 15 in one or more LSTM neural networks, wherein the LSTM networks produce a model detailing a signal propagation pattern in the deployment environment as a final hidden state output of the LSTM neural networks. This output is then concatenated 120 with additional communication features that may be specific to the deployment environment.

[0059] In various embodiments, including the one represented by method 100, the additional communication features may include transmit power of the base station, transmitter antenna height, antenna pattern, polarization, carrier frequency, receiver antenna height, and transmission distance amongst others. The transmit power of the base station may be denoted by P. Typically, P = 35 W for a 4G LTE base station. The transmitter antenna height may be denoted by HTX which represents the distance from the ground level to the antenna base. The antenna pattern may be represented by k, where k = 1 corresponds to the directional antenna and k = 0 to the omnidirectional antenna. The polarization may be represented by c, where c = 1 corresponds to the horizontal polarization and c = 0 to the vertical polarization. Carrier frequency may be denoted by f, for example, f= 2.6 GHz for a 4G LTE frequency band. A receiver antenna height may be represented as well for example by /-/RX, where d may represent the transmission distance between a base station and a measurement point. In most but not all cases, d is calculated by using the GPS coordinates.

[0060] The new vectors with the concatenated additional communication features may then be input 125 into one or more deep neural networks to establish a signal strength prediction model. This model may then be used to determine the signal quality and/or strength of one or more base stations in different locations in the deployment environment. In preferred embodiments, inputs are derived from one or more image bands overlayed or identified in optional step 101 on the deployment environment image which contains a proposed base station position/point and one or more measurement points where the signal strength is to be predicted that are part of the image from step. By determining these preset locations, or in other embodiments, by inputting the specific location of proposed base stations, cell towers, or other wireless communication structure into the model, the model generates 130 a predicted signal strength of the location in relation to other locations in the deployment environment. In several embodiments, a user of the model may also input the other locations where signal strength prediction is desired, i.e., locations that may represent receiver or client devices and their positions in relation to the deployment environment and the base station. In several embodiments of the invention, the model may also automatically select at least one deployment base stations based on one or more parameters. These parameters may be specified by a user. The model, in various embodiments, may also provide recommendations as to improving the signal strength of proposed base stations, for example by proposing certain heights for the base station, frequencies, carrier signal types, and the other communication features and signal strength variables discussed in this document.

[0061] FIG. 2 presents the steps employed by one embodiment of a Long Short-Term Memory (LSTM) neural network deployed in the signal strength prediction model, for example an LSTM network as deployed in method 100. In the LSTM network method 200, each LSTM network has three gates, the first is a forget gate neural network which receives outputs generated by the CNN that are input into the LSTM which then generates 205 vector outputs (hereinafter “forget gate vector outputs”). The model then determines 210 or calculates an initial pointwise multiplication product of the forget gate vector outputs and a previous cell state value of the LSTM network. A second gate, an input gate neural network, is then used to generate 215 input gate vector outputs. A new memory neural network then generates 220 new memory update vector outputs. Using the generated vectors in steps 215 and 220, a pointwise multiplication product of the input gate vector outputs and the new memory update vector outputs is determined 225 or calculated. The two determined pointwise multiplication products are then combined to derive 230 a new cell state value for the LSTM network. A third gate, the output gate neural network then is used to generate 235 output gate vector outputs. Finally, a present hidden state value is determined 240 from the new cell state value and the output gate vectors. [0062] FIG. 3 presents a satellite image of a terrain environment that is to be used in the signal strength prediction model. Image 300 may include one or more image bands 325, 330 used to capture the propagation environment between base station A 305 and measurement point B 310 (image band 325), or base station A 305 and the measurement point C 315 (image band 330) for example. In one embodiment, the image band 330, for example, may be obtained by first downloading or defining the rectangular area 335 whose outline is defined by dashes and then extracting the diagonal of rectangular area 335 to obtain the image band 330. In various embodiments, elevation data may be obtained along the path defined by image band 330 between base station A 305 and measurement points such as measurement points B 310 or C 315 by using Google Elevation API or other topographical data methods that may be input into the signal strength prediction model.

[0063] FIG. 4 presents a diagram that illustrates a preferred embodiment of the overall signal strength prediction training model. The entire procedure 400 for the signal strength prediction is divided into three parts. In the first part, the original satellite picture or a portion of the original satellite picture, for example as presented in FIG. 3 or a portion such as image band 325, or image band 330, or a combination of bands 325 and 330 are cropped into several basic image units 401. In preferred embodiments the size of each unit may be 96 x 96 pixels. The basic image units 401 are then input 402 into CNN 420. In preferred embodiments, CNN 420 is comprised of four layers 403 including three convolutional layers and a fully connected layer. Via the many layers 403 of the CNN 420, the environmental features in the basic image units 401 are extracted. The extracted environmental features are then input 406 as encoded image feature vectors 405 into one or more LSTM networks 430. Several networks 430 may be used and vectors may be input 406 into one LSTM network 430 and then output by that LSTM and then input 408 into another LSTM network, until the final output 412, which may comprise a decoded features vector, is produced 409 by a final LSTM network 430.

[0064] Other communication features or parameters 410 that may include transmission power, distance, elevation arrays and the other features discussed herein may also be combined 411 to the decoded features vector 412 and/or concatenated 413 with the decoded features vector 412 into a combined new feature vector 414, this new feature vector is then input 415 into one or more deep neural networks (DNN) 440 to generate the final signal strength prediction. The goal of the training is to find the optimal weights in CNN-LSTM-DNN by minimizing the difference between the output of DNN and the labeled data (signal strength). The output 416 of DNN 440 may include a predictive model 450 based on the features in the satellite image and image units 401 and/or it could be a final prediction of the signal strength of one or more measurement points that were initially fed into the procedure 400 via the satellite image and image units 401 .

[0065] FIG. 5 presents a diagram to represent the different layers of the convolutional neural network utilized in the signal strength prediction model. In this embodiment of CNN 500, the environmental features of an image or image unit 501 of a deployment environment are obtained by extracting the features from each image unit using CNN 500. CNN 500 captures all the tiny environmental changes in each image unit 501 , whose impacts on the final received signal strength are further observed through one or more LSTM neural networks. The CNN structure 500 in this preferred embodiment has four convolution layers 505, one pooling layer 506, and three fully connected layers 505. The convolutional layer is used to extract the various features from the input images, and the pooling layer is used to reduce the dimensions of the convolved feature map and decrease the computation costs. The fully connected layers form the last few layers of a CNN architecture. Each layer is followed by batch normalization (BN) 503 and ReLU non-linear activation function 504. The feature vector 510 extracted from the last fully connected layer is 512 x 1 .

[0066] FIG. 6 presents a diagram representing one embodiment of a structure 600 of LSTM neural networks utilized in the signal strength prediction model. LSTM networks are capable of learning long-term dependencies in sequential data. In general, the output of an LSTM at a particular time step is dependent on three things: 1) a cell state which represents the current long-term memory of the network, 2) a hidden state which is known as the short-term memory, and 3) input data at the current time step. The difference between the cell state and hidden state is that the former (cell state) encodes the information that is present in the memory from all previous time steps, on the other hand, the latter (hidden state) is meant to encode the characterization of the previous time step. STM networks utilize three different types of gate, a forget gate 603, input gate 604, and output gate 605, processing data passing on information as it propagates forward. The LSTM networks receive a sequence of the input environmental feature vectors 602 (xi, X2, . . . , XT ) from the CNN, and then it maps the inputs to an output sequence (hi, h2, , h_t) by computing the activation of the network units using the following equations (1 a) - (1 e):

(1 a): ft = σ (Wxfxt +Whfht-t +WcfCt-1 + br), (1 b): it = σ (WxiXt +Whiht-i +WciCt-1 + bi), (1 c): Ct = ftCt-1 + it tanh (WxcXt +Whcht-t + bc), (1d): ot = σ (WxoXt +Whoht-t +Wcoct-t + bo), and (1 e): ht = ot tanh (ct), where t represents the t-th time step, σ is the logistic sigmoid function, tanh is the hyperbolic tangent function, and ft, it, ct, and ot are the activation vectors of the forget gate 603, input gate 604, memory cell, and output gate 605, respectively. Wxf , Whf , Wef are the weight parameters within the forget gate 603. Wx/, Wh,, W_c/, Wxc, Whc is the weight parameters within the input gate 604. Wxo, Who, and Wco is the weight parameters in the output gate 605. bf is the bias parameter in the forget gate 603. b, and bc are the bias parameters in the input gate 604. bo is the bias parameter in the output gate 605.

[0067] In FIG. 6, the first step in the process is handled by the forget gate 603, which determines how the previous cell state is useful by assigning a weight onto it. To achieve this, the previous hidden state 601 and current input 602 is fed into a neural network to generate a vector whose element is in the range [0, 1] (ensured by sigmoid function) (equation (1 a)). These outputs are then pointwise multiplied 611 with the previous cell state (“ftCt-t“ portion in equation (1 c)). Consequently, the closer the output to 0, the less influence from previous cell state on the future time steps.

[0068] The next step involves an input gate 604 and a new memory network 612. The input gate 604 is similar to the forget gate 603, working as a filter with a sigmoid activation function (this is equation (1 b)). This step essentially determines whether the new input data is worth remembering. A new memory network 612 (“tan h (WxcXt +Whcht-i + b_c)” portion in equation (1 c)) is a tanh activated neural network, learning how to combine the previous hidden state and new input data to generate a ‘new memory update vector’. The output of the input gate 604 and new memory network 612 is pointwise multiplied 613 (/.e., the “it tanh (WxcXt +Whcht-t + b_c)” portion of equation (1 c)) and then added 614 to the previously calculated pointwise multiplied produet to generate a new cell state (equation (1 c)).

[0069] Finally, the output gate 605 plays a significant role in the final step (equations (1d) and (1 e)), which determines the present hidden state (short memory) 607 that will be passed to the next LSTM unit 609. These steps repeat recursively from t = 1 to t = T, while only the last result h T 610 is used to predict the signal strength. LSTM networks process data passing on information as it propagated forward, choosing which information is relevant to remember or forget during sequence processing.

[0070] FIG. 7 is a diagrammatic representation of an example machine in the form of a computer system 1 , within which a set of instructions for causing the machine to perform any one or more of the methodologies discussed herein may be executed. In various example embodiments, the machine operates as a standalone device or may be connected (e.g., networked) to other machines. In a networked deployment, the machine may operate in the capacity of a server or a client machine in a server-client network environment, or as a peer machine in a peer-to-peer (or distributed) network environment. The machine may be a personal computer (PC), a tablet PC, a set-top box (STB), a personal digital assistant (PDA), a cellular telephone, a portable music player (e.g., a portable hard drive audio device such as an Moving Picture Experts Group Audio Layer 3 (MP3) player), a web appliance, a network router, switch or bridge, or any machine capable of executing a set of instructions (sequential or otherwise) that specify actions to be taken by that machine. Further, while only a single machine is illustrated, the term “machine” shall also be taken to include any collection of machines that individually or jointly execute a set (or multiple sets) of instructions to perform any one or more of the methodologies discussed herein.

[0071] The example computer system 1 includes a processor or multiple processor(s) 5 (e.g., a central processing unit (CPU), a graphics processing unit (GPU), or both), and a main memory 10 and static memory 15, which communicate with each other via a bus 20. The computer system 1 may further include a video display 35 (e.g., a liquid crystal display (LCD)). The computer system 1 may also include an alphanumeric input device(s) 30 (e.g., a keyboard), a cursor control device (e.g., a mouse), a voice recognition or biometric verification unit (not shown), a drive unit 37 (also referred to as disk drive unit), a signal generation device 40 (e.g., a speaker), and a network interface device 45. The computer system 1 may further include a data encryption module (not shown) to encrypt data.

[0072] The components provided in the computer system 1 are those typically found in computer systems that may be suitable for use with embodiments of the present disclosure and are intended to represent a broad category of such computer components that are known in the art. Thus, the computer system 1 can be a server, minicomputer, mainframe computer, or any other computer system. The computer may also include different bus configurations, networked platforms, multi-processor platforms, and the like. Various operating systems may be used including UNIX, LINUX, WINDOWS, QNX ANDROID, IOS, CHROME, TIZEN, and other suitable operating systems. [0073] The disk drive unit 37 includes a computer or machine-readable medium 50 on which is stored one or more sets of instructions and data structures (e.g., instructions 55) embodying or utilizing any one or more of the methodologies or functions described herein. The instructions 55 may also reside, completely or at least partially, within the main memory 10 and/or within the processor(s) 5 during execution thereof by the computer system 1 . The main memory 10 and the processor(s) 5 may also constitute machine-readable media.

[0074] The instructions 55 may further be transmitted or received over a network 70 via the network interface device 45 utilizing any one of several well-known transfer protocols (e.g., Hyper Text Transfer Protocol (HTTP)). While the machine-readable medium 50 is shown in an example embodiment to be a single medium, the term "computer-readable medium" should be taken to include a single medium or multiple medium (e.g., a centralized or distributed database and/or associated caches and servers) that store the one or more sets of instructions. The term "computer-readable medium" shall also be taken to include any medium that is capable of storing, encoding, or carrying a set of instructions for execution by the machine and that causes the machine to perform any one or more of the methodologies of the present application, or that is capable of storing, encoding, or carrying data structures utilized by or associated with such a set of instructions. The term "computer-readable medium" shall accordingly be taken to include, but not be limited to, solid-state memories, optical and magnetic media, and carrier wave signals. Such media may also include, without limitation, hard disks, floppy disks, flash memory cards, digital video disks, random access memory (RAM), read only memory (ROM), and the like. The example embodiments described herein may be implemented in an operating environment comprising software installed on a computer, in hardware, or in a combination of software and hardware.

[0075] One skilled in the art will recognize that Internet service may be configured to provide Internet access to one or more computing devices that are coupled to the Internet service, and that the computing devices may include one or more processors, buses, memory devices, display devices, input/output devices, and the like. Furthermore, those skilled in the art may appreciate that the Internet service may be coupled to one or more databases, repositories, servers, and the like, which may be utilized to implement any of the embodiments of the disclosure as described herein.

[0076] The computer program instructions may also be loaded onto a computer, a server, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.

[0077] Suitable networks may include or interface with any one or more of, for instance, a local intranet, a PAN (Personal Area Network), a LAN (Local Area Network), a WAN (Wide Area Network), a MAN (Metropolitan Area Network), a virtual private network (VPN), a storage area network (SAN), a frame relay connection, an Advanced Intelligent Network (AIN) connection, a synchronous optical network (SONET) connection, a digital T1 , T3, E1 or E3 line, Digital Data Service (DDS) connection, DSL (Digital Subscriber Line) connection, an Ethernet connection, an ISDN (Integrated Services Digital Network) line, a dial-up port such as a V.90, V.34 or V.34bis analog modem connection, a cable modem, an ATM (Asynchronous Transfer Mode) connection, or an FDDI (Fiber Distributed Data Interface) or CDDI (Copper Distributed Data Interface) connection. Furthermore, communications may also include links to any of a variety of wireless networks, including WAP (Wireless Application Protocol), GPRS (General Packet Radio Service), GSM (Global System for Mobile Communication), CDMA (Code Division Multiple Access) or TDMA (Time Division Multiple Access), cellular phone networks, GPS (Global Positioning System), CDPD (cellular digital packet data), RIM (Research in Motion, Limited) duplex paging network, Bluetooth radio, or an IEEE 802.1 1-based radio frequency network. The network 215 can further include or interface with any one or more of an RS-232 serial connection, an IEEE-1394 (Firewire) connection, a Fiber Channel connection, an IrDA (infrared) port, a SCSI (Small Computer Systems Interface) connection, a USB (Universal Serial Bus) connection or other wired or wireless, digital or analog interface or connection, mesh or Digi® networking. [0078] In general, a cloud-based computing environment is a resource that typically combines the computational power of a large grouping of processors (such as within web servers) and/or that combines the storage capacity of a large grouping of computer memories or storage devices. Systems that provide cloud-based resources may be utilized exclusively by their owners or such systems may be accessible to outside users who deploy applications within the computing infrastructure to obtain the benefit of large computational or storage resources.

[0079] The cloud is formed, for example, by a network of web servers that comprise a plurality of computing devices, such as the computer device 1 , with each server (or at least a plurality thereof) providing processor and/or storage resources. These servers manage workloads provided by multiple users (e.g., cloud resource customers or other users). Typically, each user places workload demands upon the cloud that vary in real-time, sometimes dramatically. The nature and extent of these variations typically depends on the type of business associated with the user.

[0080] It is noteworthy that any hardware platform suitable for performing the processing described herein is suitable for use with the technology. Various forms of computer-readable media may be involved in carrying one or more sequences of one or more instructions to a CPU for execution. A bus carries the data to system RAM, from which a CPU retrieves and executes the instructions. The instructions received by system RAM can optionally be stored on a fixed disk either before or after execution by a CPU.

[0081] Computer program code for carrying out operations for aspects of the present technology may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++, or the like and conventional procedural programming languages, such as the "C" programming language, Go, Python, or other programming languages, including assembly languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider).

Examples

[0082] FIG 8. presents an example map 800 incorporating collected data. The data was collected through an antenna mounted on a vehicle and is provided by an Australian telecommunications company. This dataset contains nine surveyed 4G LTE sites, including Naraellan, Camden Golf Course, Narrabri Te, Redfern Te, Springwood Wb, Fyshwick Te, Mount Panorama, Dubbo, and North Parramatta Te. The total number of data samples is 9523, which is divided into training and testing data sets. The CNN-LSTM-DNN model was trained on a batch size of 10 using adaptive moment estimation (Adam) optimizer with [ β1 = 0.9, β2 = 0.98 and ∈ = 10-8. Wherein β1 and β2 represent decay rates and e represents a constant. The training is to minimize the mean square error between the prediction and the ground truth by updating the parameters (weight and bias) in the model. Unlike traditional gradient-base optimization algorithm, Adam computes adaptive learning rates for each parameter by storing an exponentially decaying average of past gradients and squared gradients, controlled by β1 and [32, respectively. They are the estimates of the first moment and the second moment of the gradients.

[0083] FIG. 9 presents a graph 900 of experimental results illustrating the improvement in predicting Signal Strength Accuracy by the presented methods and models over a traditional Hata model. Different training/testing sample ratios are used in applying the presented methods and models to predict signal strength, each of these different ratios are represented by their own bar on the graph. A root mean square error (RMSE) is used to measure the performance of these different methods and is presented on the y- axis. Bar graph 910 presents a training model where 90% of the data is used for training and 10% for testing, the RMSE is the lowest at 5.1 . Bar 920 presents a training method where 20% of the data samples are used for testing, and 80% are used for training, with an RMSE of 5.61. Bar 930 has an RMSE of 5.95 and is a result of a training sample using 70% of the data, with 30% of the data used for testing. All three methods are superior to the traditional Hata method, represented by bar 940, for estimating signal strength which has an RMSE of 20.71.

[0084] The foregoing detailed description includes references to the accompanying drawings, which form a part of the detailed description. The drawings show illustrations in accordance with exemplary embodiments. These example embodiments, which are also referred to herein as “examples,” are described in enough detail to enable those skilled in the art to practice the present subject matter.

[0085] The embodiments can be combined, other embodiments can be utilized, or structural, logical, and electrical changes can be made without departing from the scope of what is claimed. The following detailed description is, therefore, not to be taken in a limiting sense, and the scope is defined by the appended claims and their equivalents. In this document, the terms “a” or “an” are used, as is common in patent documents, to include one or more than one. In this document, the term “or” is used to refer to a nonexclusive “or,” such that “A or B” includes “A but not B,” “B but not A,” and “A and B,” unless otherwise indicated. Furthermore, all publications, patents, and patent documents referred to in this document are incorporated by reference herein in their entirety, as though individually incorporated by reference. In the event of inconsistent usages between this document and those documents so incorporated by reference, the usage in the incorporated reference(s) should be considered supplementary to that of this document; for irreconcilable inconsistencies, the usage in this document controls.

[0086] The corresponding structures, materials, acts, and equivalents of all means or step plus function elements in the claims below are intended to include any structure, material, or act for performing the function in combination with other claimed elements as specifically claimed. The description of the present technology has been presented for purposes of illustration and description, but is not intended to be exhaustive or limited to the invention in the form disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the invention. Exemplary embodiments were chosen and described to best explain the principles of the present technology and its practical application, and to enable others of ordinary skill in the art to understand the invention for various embodiments with various modifications as are suited to the particular use contemplated.

[0087] While specific embodiments of, and examples for, the system are described above for illustrative purposes, various equivalent modifications are possible within the scope of the system, as those skilled in the relevant art will recognize. For example, while processes or steps are presented in a given order, alternative embodiments may perform routines having steps in a different order, and some processes or steps may be deleted, moved, added, subdivided, combined, and/or modified to provide alternative or subcombinations. Each of these processes or steps may be implemented in a variety of different ways. Also, while processes or steps are at times shown as being performed in series, these processes or steps may instead be performed in parallel or may be performed at different times.

[0088] The various embodiments described above, are presented as examples only, and not as a limitation. The descriptions are not intended to limit the scope of the present technology to the forms set forth herein. To the contrary, the present descriptions are intended to cover such alternatives, modifications, and equivalents as may be included within the spirit and scope of the present technology as appreciated by one of ordinary skill in the art. Thus, the breadth and scope of a preferred embodiment should not be limited by any of the above-described exemplary embodiments.

Claims

CLAIMS What is claimed is:

1 . A method to predict signal strength of wireless communication systems in complex environments, the method comprising: cropping an image of a terrain environment into multiple image units, wherein the image includes a base station point and a measurement point; identifying environmental features, via a convolutional neural network (CNN), wherein the multiple image units are used as inputs to the CNN; applying at least one Long Short-Term Memory (LSTM) neural network, to the identified environmental features, to model a signal propagation pattern in the terrain environment; concatenating a final hidden state output of the LSTM neural network with additional communication features to produce vectors; and inputting the vectors into a deep neural network (DNN) to establish a signal strength prediction model.

2. The method of claim 1 , further comprising: generating, by the signal strength prediction model, an output representing a predicted signal strength of a specified location, based on one or more inputs representing the specified location in the terrain environment.

3. The method of claim 1 , further comprising: automatically selecting, based on the signal strength prediction model, a location for deployment of a base station of dense 5G mmWave networks.

4. The method of claim 1 , wherein the additional communication features include one or more of transmission power, transmitter antenna height, antenna pattern, carrier frequency, receiver antenna height, polarization, elevation, and transmission distance.

5. The method of claim 1 , wherein the convoluted neural network defines four convolution layers.

6. The method of claim 4, wherein the convoluted neural network further defines one pooling layer, and three fully connected layers.

7. The method of claim 1 , wherein the layers of the convoluted neural network are each followed by batch normalization and a non-linear activation function.

8. The method of claim 1 , wherein the image is a satellite image, the satellite image includes one or more image bands that identify terrain between at least one base station and at least one measurement point.

9. The method of claim 1 , wherein the identification of environmental features comprises: extracting at least one environmental feature vector from a fully connected layer of the convoluted neural network.

10. The method of claim 1 , wherein the applying of the at least one LSTM neural network comprises: receiving a sequence of input vectors representing the identified environmental features from the convoluted neural network; and mapping the sequence of input vectors to an output sequence, wherein the mapping is undertaken by computing activation of neural network units.

11. The method of claim 10, wherein the mapping of the sequence of input vectors to an output sequence comprises: generating, via a forget gate neural network, forget gate vector outputs in the range [0,1], wherein the inputs comprise at least a previous hidden state value and the received sequence of input vectors representing environmental features; determining, an initial pointwise multiplication product of the forget gate vector outputs and a previous cell state value; generating, via an input gate neural network, input gate vector outputs, wherein the inputs comprise at least the previous hidden state value and the received sequence of input vectors; generating, via a new memory neural network, new memory update vector outputs, from inputs comprising at least the previous hidden state value and the received sequence of input vectors; determining, another pointwise multiplication product of the new memory update vector outputs and the input gate vector outputs; deriving, a new cell state value by combining the initial pointwise multiplication product and the another pointwise multiplication product; generating, via an output gate neural network, output gate vector outputs, wherein the inputs for the output gate neural network comprise at least the previous hidden state value and the received sequence of input vectors; and determining, a present hidden state value, from the new cell state value and the output gate vectors.

12. The method of claim 11 , wherein the mapping of the sequence of input vectors to an output sequence further comprises: passing the present hidden state value to another LSTM neural network.

13. The method of claim 11 wherein one or more of the forget gate neural network, the input gate neural network, and the output gate neural network comprise a sigmoid activation function.

14. The method of claim 11 wherein one or more of the forget gate neural network, the input gate neural network, and the output gate neural network comprise a previous cell state value as an input.

15. A system for predicting signal strength of wireless communications in complex environments, the system comprising: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, are effective to program the at least one processor to: crop an image of a terrain environment into multiple image units, wherein the image includes a base station point and a measurement point; identify environmental features, via a convolutional neural network (CNN), wherein the multiple image units are used as inputs to the CNN; apply at least one Long Short-Term Memory (LSTM) neural network, to the identified environmental features, to model a signal propagation pattern in the terrain environment; concatenate a final hidden state output of the LSTM neural network with additional communication features to produce new vectors; and input the new vectors into a deep neural network (DNN) to establish a signal strength prediction model.

16. The system of claim 15, further comprising the at least one memory storing instructions that, when executed by the at least one processor, are effective to program the at least one processor to: generate, by the signal strength prediction model, an output representing a predicted signal strength of a specified location, based on one or more inputs representing the specified location in the terrain environment.

17. The system of claim 15, further comprising the at least one memory storing instructions that, when executed by the at least one processor, are effective to program the at least one processor to:: automatically select, based on the signal strength prediction model, a location for deployment of a base station of dense 5G mmWave networks.

18. The system of claim 15, wherein the additional communication features include one or more of transmission power, transmitter antenna height, antenna pattern, carrier frequency, receiver antenna height, polarization, elevation, and transmission distance.

19. The system of claim 15, wherein the convoluted neural network defines four convolution layers.

20. The system of claim 19, wherein the convoluted neural network further defines one pooling layer, and three fully connected layers.

21. The system of claim 15, wherein the layers of the convoluted neural network are each followed by batch normalization and a non-linear activation function.

22. The system of claim 15, wherein the image is a satellite image, the satellite image includes one or more image bands that identify terrain between at least one base station and at least one measurement point.

23. The system of claim 15, wherein the instructions to identify environmental features comprise: extract at least one environmental feature vector from a fully connected layer of the convoluted neural network.

24. The system of claim 15, wherein the wherein the instructions to apply the at least one LSTM neural network comprise: receive a sequence of input vectors representing the identified environmental features from the convoluted neural network; and map the sequence of input vectors to an output sequence, wherein the mapping is undertaken by computing activation of neural network units.

25. The system of claim 15, wherein the instructions to map the sequence of input vectors to an output sequence comprise: generate, via a forget gate neural network, forget gate vector outputs in the range [0,1], wherein the inputs comprise at least a previous hidden state value and the received sequence of input vectors representing environmental features; determine, an initial pointwise multiplication product of the forget gate vector outputs and a previous cell state value; generate, via an input gate neural network, input gate vector outputs, wherein the inputs comprise at least the previous hidden state value and the received sequence of input vectors; generate, via a new memory neural network, new memory update vector outputs, from inputs comprising at least the previous hidden state value and the received sequence of input vectors; determine, another pointwise multiplication product of the new memory update vector outputs and the input gate vector outputs; derive a new cell state value by combining the initial pointwise multiplication product and the another pointwise multiplication product; generate, via an output gate neural network, output gate vector outputs, wherein the inputs for the output gate neural network comprise at least the previous hidden state value and the received sequence of input vectors; and determine, a present hidden state value, from the new cell state value and the output gate vectors.

26. The system of claim 15, wherein the instructions to apply at least one LSTM neural network further comprises: passing the present hidden state value to another LSTM neural network.

27. The system of claim 15 wherein one or more of the forget gate neural network, the input gate neural network, and the output gate neural network comprise a sigmoid activation function.

28. The system of claim 15 wherein one or more of the forget gate neural network, the input gate neural network, and the output gate neural network comprise a previous cell state value as an input.

29. A non-transitory computer-readable storage medium having embodied thereon a program, the program being executable by a processor to perform a method for providing a multi-client network service comprising: cropping an image of a terrain environment into multiple image units, wherein the image includes a base station point and a measurement point; identifying environmental features, via a convolutional neural network (CNN), wherein the multiple image units are used as inputs to the CNN; applying at least one Long Short-Term Memory (LSTM) neural network, to the identified environmental features, to model a signal propagation pattern in the terrain environment; concatenating a final hidden state output of the LSTM neural network with additional communication features to produce new vectors; inputting the new vectors into a deep neural network (DNN) to establish a signal strength prediction model; generating, by the signal strength prediction model, an output representing a predicted signal strength of a specified location, based on one or more inputs representing the specified location in the terrain environment; and automatically selecting, based on the signal strength prediction model, a location for deployment of a base station of dense 5G mmWave networks.