Machine learning based antenna array verification, prototyping, and optimization

Abstract

An example method of estimating an angular resolution of an antenna array includes receiving a plurality of values of amplitude and phase of a radio frequency (RF) signal for each of a plurality of antenna elements included by the antenna array, performing, by a machine learning model, a feature extraction operation to transform the plurality of values of amplitude and phase to a plurality of data points in a reduced dimension space, clustering, by the machine learning model, the plurality of data points into a plurality of clusters, and computing an angular resolution value of the antenna array based on the clustered data points.

Description

[0001] Related Applications

[0002] This application is an international application of U.S. Non-Provisional Application No. 16 / 584,383, filed September 26, 2019, which claims the benefit of U.S. Provisional Application No. 62 / 878,232, filed July 24, 2019, the entire contents of both of which are incorporated herein by reference. BACKGROUND

[0003] An antenna array can include a plurality of antenna elements having certain geometrical shapes and sizes. In practice, antenna design parameters can include the number, type, and geometrical properties (e.g., shape, size, area) of the antenna elements, as well as the size of the spaces between the antenna elements. BRIEF DESCRIPTION OF DRAWINGS

[0004] The disclosure is illustrated in the figures of the accompanying drawings, which are presented by way of example and not limitation, in which:

[0005] Figure 1 A phase response of an ideal antenna in the spatial domain is shown.

[0006] Figure 2 A phase response of an example area-optimized antenna array is shown.

[0007] Figure 3 A correlation of radio frequency (RF) signal phases of an angle of arrival measured by a first element and a second element of an example antenna array is depicted.

[0008] Figure 4 A two-dimensional projection of a set of multi-dimensional data points representing input RF signal measurements is shown.

[0009] Figure 5 A structure of an example neural network is shown schematically.

[0010] Figure 6 An example automated workflow for antenna design is shown schematically.

[0011] Figure 7 A phase response of an example prototype antenna array is shown.

[0012] Figure 8 is a block diagram of an example system for implementing the methods described herein.

[0013] Figure 9 is a flowchart of an example method of estimating an angular resolution value for an antenna array.

[0014] Figure 10 is a flowchart of an example method of automating an antenna design workflow based on estimating an angular resolution value for an antenna array. DETAILED DESCRIPTION

[0015] The embodiments described herein pertain to systems and methods for using machine learning models for antenna array verification, prototyping, and optimization.

[0016] Antenna design parameters can include the number, type, and geometric properties of antenna elements (e.g., shape, size, area), as well as the dimensions of the space between antenna elements. Thus, in one example, the design objective of an antenna might involve optimizing certain operational characteristics of the antenna, such as the resolution of the angle of arrival estimation (also referred to herein as angular resolution), while satisfying physical design constraints, which can be specified by the range or maximum value of certain antenna design parameters (e.g., the number, type, and geometric properties of antenna elements, and / or the space between antenna elements). In another example, the design objective of an antenna might involve optimizing certain physical design parameters (e.g., the number, type, and geometric properties of antenna elements, and / or the space between antenna elements) while satisfying operational constraints (e.g., providing at least the desired angular resolution). Additional constraints can be imposed on the design process itself, for example, reducing the number of physical prototypes and experiments involved in the design process.

[0017] Unlike various common implementations that analyze antenna gain distribution in the spatial and / or frequency domains, the systems and methods disclosed herein analyze the phase response of multi-element antennas in order to estimate their angular resolution. Figure 1 The diagram illustrates the phase response of an ideal antenna in the spatial domain (the X-axis corresponds to the number of antenna elements, and the Y-axis corresponds to the phase of the received RF signal). Each curve 110A-110Z corresponds to a certain value of the angle of arrival (AoA) of the received RF signal (in...). Figure 1 (represented as Θ).

[0018] As from Figure 1 As can be seen, the angular resolution of an antenna exhibiting the described linear phase response will depend solely on the spatial domain characteristics (i.e., the number of antenna elements). However, antenna design requirements may limit the physical size of the antenna, for example, for mobile or wearable devices. Therefore, the phase response of an example area-optimized antenna may become significantly nonlinear, such as... Figure 2 The curves 210A-210Z are shown.

[0019] Estimating the angular resolution of antenna arrays exhibiting largely nonlinear phase responses can be extremely challenging, at least in terms of computational complexity. Therefore, the system and method described in this paper rely on machine learning techniques to build and train efficient models for estimating the angular resolution of antenna arrays.

[0020] The raw data used to estimate the angular resolution may include multiple RF signal amplitude and phase values ​​measured by multiple antenna elements at one or more RF channels. Therefore, each input data point is represented by a vector in a 2*N dimension hyperspace, where N is the number of antenna elements, and the dimension represents the amplitude and phase values ​​measured by each antenna element. In some implementations, the RF signal amplitude and phase values ​​can be generated by an antenna simulation model based on specified antenna design parameters.

[0021] The system and method disclosed herein employ a machine learning model that performs feature extraction to reduce the dimensionality of the input data, and then performs clustering on the extracted features in the reduced-dimensional space to estimate the angular resolution of the prototype antenna array. Once the angular resolution of the prototype antenna array is determined, the system can formulate design optimization feedback to improve the suboptimal parts of the antenna array response, as described in more detail below.

[0022] Various aspects of the methods and systems described herein are illustrated by way of example rather than limitation. The methods described herein may be implemented by hardware (e.g., general-purpose and / or special-purpose processing devices, and / or other devices and associated circuitry), software (e.g., instructions executable by the processing devices), or a combination thereof.

[0023] As noted above, the systems and methods of this disclosure analyze the phase response of multi-element antennas in order to estimate their angular resolution. Figure 3 Curves 310A-320N and 320A-320N depict the phase of the RF signal measured by the first and second elements of the example antenna array, respectively (the X-axis corresponds to the AoA of the received RF signal, and the Y-axis corresponds to the phase of the received RF signal). Each of curves 310A-310N and 320A-320N corresponds to a specific RF channel. (The remaining text appears to be a fragment from a different source and is not directly related to the previous sentence.) Figure 3 As can be seen, at some values ​​of AoA, the phase response may not be sufficiently distinguishable to provide the necessary angular resolution.

[0024] Since the phase response of an antenna is essentially nonlinear, the computational complexity of direct phase response analysis increases exponentially with the number of antenna array elements. Therefore, machine learning models can be used to perform feature extraction to reduce the dimensionality of the input data, thereby significantly reducing the computational complexity of the task at hand.

[0025] Then, the machine learning model can use the extracted features (i.e., the representation of the input data in the reduced-dimensional space) to cluster the data points representing the phase response into multiple clusters, such that each cluster corresponds to a certain value of AoA. For example... Figure 4The schematic illustration depicts a two-dimensional projection of a set of multidimensional data points representing the input data. The data points are grouped into multiple overlapping clusters, such that each cluster corresponds to a certain value of the AoA of the received RF signal (in...). Figure 4 (represented as Θ1-Θ5).

[0026] In some implementations, the models used to perform feature extraction and clustering can be implemented by neural networks (e.g., convolutional neural networks (CNNs) or recurrent neural networks (RNNs)). A neural network is a computational model that implements multiple connected nodes called "artificial neurons," such that each artificial neuron processes one or more input signals (including bias signals) and transmits its output signal to one or more neighboring artificial neurons. The output of an artificial neuron can be computed by applying its activation function to a linear combination of its inputs. Neural networks can be trained by processing examples ("training datasets") to perform feature extraction, regression, and / or classification tasks (typically without programming using any task-specific rules), as described in more detail below.

[0027] like Figure 5 As schematically illustrated, the neural network employed in the systems and methods of this disclosure can be represented by a multilayer perceptron (MLP) 500, whose artificial neurons are grouped into several layers, including an input layer 510, one or more hidden layers 520A-520L, and an output layer 530. The input layer 510 includes one or more neurons 540A-540N connected to one or more neurons 550A-550K of the first hidden layer 520A. The first hidden layer neurons 550A-550K are further connected to one or more neurons 560A-560M of the second hidden layer 520L. The second hidden layer neurons 560A-560M are further connected to one or more neurons 570A-570Z of the output layer 530. At least some nodes in the artificial neural network 500 can utilize non-linear activation functions, while the remaining nodes (e.g., nodes in the output layer) can utilize linear activation functions. Although... Figure 5 A single hidden layer 550 is schematically shown, but the number of hidden layers can vary in various implementations of the systems and methods disclosed herein. In some embodiments, the number of hidden layers is a hyperparameter of the model, i.e., a parameter whose value is specified before the training process. Other hyperparameters of the model may include the number of nodes in each layer, the type of activation function, etc.

[0028] Figure 5Each edge in the diagram represents a connection used to transmit signals from one artificial neuron to another in a manner similar to the operation of synapses in the human brain. During the network training phase, edge weights are defined based on a training dataset that includes multiple labeled inputs (i.e., inputs with known classifications). These edge weights either increase or decrease the signal transmitted through the corresponding connection.

[0029] The neural network training process can begin by initializing all edge weights and neuron biases to random or predetermined values. The systems and methods disclosed herein utilize an unsupervised training procedure to train neural networks that perform feature extraction and clustering tasks. Unlike supervised training, which requires labeling the training dataset with known output values, unsupervised training uses an unlabeled training dataset.

[0030] In some implementations, clustering techniques (e.g., k-means clustering) are used to assign samples from the training set to one of k clusters in a manner that minimizes the average distance between the samples and the cluster centroids. After completing the clustering iterations, a predetermined quality metric (e.g., cluster density) is computed and compared to an expected value, and the error is propagated back through the first few layers of the neural network, where weights and biases are adjusted to minimize a predetermined loss function (e.g., the difference between the observed and expected cluster densities). This process can be repeated until the quality metric meets predetermined conditions (e.g., below or above a predetermined threshold). In various other implementations, other clustering techniques may be employed.

[0031] In some implementations, feature extraction, clustering, and regression operations can be performed by the same neural network, where a first subset of the layers performs feature extraction, a second subset performs clustering, and the remaining layers perform regression. In other implementations, multiple separately trained neural networks can be used to perform each of the feature extraction, clustering, and regression tasks.

[0032] Refer again Figure 4 Machine learning models can use extracted features (i.e., the representation of the input data in a reduced-dimensional space) to cluster data points representing the phase response into multiple clusters, such that each cluster corresponds to a certain value of AoA. Therefore, the angular resolution of the antenna array can be visualized by the degree of cluster overlap: relatively small overlap or no overlap indicates that the AoA values ​​of points belonging to adjacent clusters are distinguishable, and the angular resolution is at least the difference between the AoA values ​​of the cluster centroids, while significant overlap indicates that the angular resolution cannot be accurately estimated based on the available input data. Therefore, the angular resolution of the antenna array can be inferred from the observed RF signal parameter values ​​by a neural network performing a regression task.

[0033] Supervised training of a neural network involves sequentially processing labeled data items from the training dataset (i.e., multiple vectors of RF signal power levels, such that each vector, including amplitude and phase values ​​of the RF signal measured by multiple antenna elements at a given frequency channel, is labeled with its corresponding AoA value, and a set of vectors is labeled with their corresponding angular resolution values). The observed neural network output is compared to the desired output specified by the labels associated with the vectors being processed, and the error is propagated back through the first few layers of the neural network, where the weights and biases are adjusted accordingly. This process can be repeated until the output error falls below a predetermined threshold.

[0034] Therefore, the machine learning model can receive input data including multiple RF signal amplitude and phase values ​​(measured by multiple antenna elements at one or more RF channels or generated by an antenna simulation model based on specified antenna design parameters), perform feature extraction to reduce the dimensionality of the input data, and cluster the extracted feature values ​​to estimate the angular resolution of the antenna array. While the examples described herein refer to neural networks, the systems or methods disclosed herein can employ other machine learning techniques.

[0035] The machine learning model described in this paper can be used in automated antenna design workflows by estimating the angular resolution of a prototype antenna array, identifying suboptimal parts of the antenna array response, and providing feedback to antenna designers.

[0036] Figure 6 An example automated workflow for antenna design is described. Workflow 600 can be initiated by antenna design module 610 feeding an initial set of antenna design parameters 610 to antenna array simulation model 620. In some implementations, antenna design module 610 can be implemented by a computer-aided design (CAD) system that can be fully automated and / or supervised by a human designer. Antenna design parameters may include the number, type, and geometric properties (e.g., shape, size, area) of antenna elements, as well as the dimensions of the space between antenna elements.

[0037] The antenna array simulation model 620 generates raw data for estimating angular resolution, which may include multiple RF signal amplitude and phase values ​​measured by multiple antenna elements at one or more RF channels. Therefore, each input data point is represented by a vector in a 2*N dimension hyperspace, where N is the number of antenna elements, and the dimension represents the amplitude and phase values ​​measured by each antenna element. Note that the antenna array model 620 is optional, and in some implementations, physical measurements of the received RF signals by a physical antenna prototype may be performed instead.

[0038] The amplitude and phase values ​​of the RF signal are then fed into a machine learning model 630, which performs a feature extraction operation 632 to reduce the dimensionality of the input data. The machine learning model 630 can then use the extracted features (i.e., the representation of the input data in the reduced-dimensional space) to perform a clustering operation 634 to group the data points into multiple clusters, such that each cluster corresponds to a certain value of AoA. The machine learning model 630 can then perform a regression operation 636 to estimate the angular resolution 640 and / or the AoA value 650, as described in more detail above.

[0039] The processing apparatus implementing workflow 600 can further identify one or more suboptimal portions and / or one or more satisfactory portions of the antenna array response at block 660. Suboptimal portions of the antenna array response may include portions corresponding to feature clusters that substantially overlap based on a selected overlap metric (which therefore adversely affect angular resolution). Satisfactory portions of the antenna array response may include portions corresponding to non-intersecting or weakly overlapping feature clusters.

[0040] Figure 7 The phase response of the example prototype antenna is schematically illustrated. Each of curves 710A, 710B, and 710N corresponds to a certain AoA value of the received RF signal. (As shown from...) Figure 7 As can be seen, within regions 720A and 720B, the phase response may not be sufficiently distinguishable to provide the necessary angular resolution. Therefore, refer again... Figure 6 The processing device implementing workflow 600 can further identify antenna array elements corresponding to the identified suboptimal and / or satisfactory portions of the antenna array response. The processing device can further identify design parameters (e.g., dimensions) of the antenna array elements that may have led to a suboptimal antenna response.

[0041] Output data, including angular resolution 640, AoA value 650, and the identified suboptimal and / or satisfactory portion 660 of the antenna array response, can be fed back to antenna design module 610, which can modify the values ​​of one or more antenna design parameters and initiate a new iteration of workflow 600 by feeding the modified antenna design parameters to antenna model 620.

[0042] Figure 8This is a block diagram of a system implementing the methods described herein. System 800 may include processing device 806, which may optionally include front-end circuitry 808 coupled to antenna array 804. Front-end circuitry 808 may include transceiver 812 and analog-to-digital converter (ADC) 814. Transceiver 812 coupled to antenna array 804 may transmit or receive RF signals via antenna array 804. Processing device 806 may implement angular resolution estimation tool 120 as described herein. Angular resolution estimation tool 120 may implement one or more machine learning models (e.g., neural networks) to perform feature extraction operation 816 to reduce the dimensionality of input data, perform clustering operation 818 to classify data points into several clusters corresponding to different AoA values, and perform regression operation 820 to estimate angular resolution and / or AoA values ​​812 and 823, as described in more detail above.

[0043] The processing device may include one or more application processors, one or more host processors, one or more microcontrollers, and / or other processing components. In some embodiments, system 800 may be implemented as a system-on-a-chip (SoC), including a transceiver and a processor for processing digital values ​​representing RF signals received or transmitted by the transceiver. The transceiver and processor may reside on a common carrier substrate or may be implemented in separate integrated circuits. Alternatively, system 800 may be implemented as a mobile or wearable device (e.g., a smartphone or smartwatch). Alternatively, system 800 may be implemented as a desktop computer, a portable computer, or a server.

[0044] In some implementations, the processing device may receive multiple values ​​of amplitude and phase of the radio frequency (RF) signal for each of the multiple antenna elements included in the antenna array. The processing device 806 may then employ a machine learning model to perform feature extraction operations to transform the multiple values ​​of amplitude and phase into multiple data points in a reduced-dimensional space. The processing device 806 may then employ a machine learning model to classify the multiple data points into multiple clusters. The processing device 806 may then calculate the angular resolution value of the antenna array and / or the AoA value for each cluster based on the data points from the multiple clusters, as described in more detail above.

[0045] Figure 9This is a flowchart of an example method for estimating the angular resolution values ​​of an antenna array. Each of method 900 and / or its individual functions, routines, subroutines, or operations can be executed by processing logic including hardware (circuit, dedicated logic, etc.), software (e.g., running on a general-purpose computing system or a dedicated machine), firmware (embedded software), or any combination thereof. Two or more functions, routines, subroutines, or operations of method 900 can be executed in parallel or in an order that may differ from the order described below. In some implementations, method 900 can be executed by a single processing thread. Alternatively, method 900 can be executed by two or more processing threads, each thread executing the operation of one or more individual functions, routines, subroutines, or methods. In the illustrative example, the processing threads implementing method 900 can be synchronized (e.g., using semaphores, critical sections, and / or other thread synchronization mechanisms). Alternatively, the processing threads implementing method 900 can execute asynchronously relative to each other. In one embodiment, the operation of method 900 can be performed by… Figure 8 The example system 800 is executed by the processing device 806.

[0046] At block 910, the processing device implementing the method can receive multiple values ​​of amplitude and phase of a radio frequency (RF) signal for each of the multiple antenna elements included in the antenna array, as described in more detail above.

[0047] At box 920, the processing device can employ a machine learning model to perform feature extraction operations, transforming multiple values ​​of amplitude and phase into multiple data points in a reduced-dimensional space. In some implementations, the model used to perform feature extraction and clustering can be implemented by a neural network, such as a convolutional neural network (CNN), a recurrent neural network (RNN), or a multilayer perceptron (MLP), as described in more detail above.

[0048] At box 930, the processing device can classify multiple data points into multiple clusters, as described in more detail above.

[0049] At box 940, the processing device can calculate the angular resolution value of the antenna array based on data points from multiple clusters. The angular resolution of the antenna array can be characterized by cluster overlap: relatively small overlap or no overlap indicates that the AoA values ​​of points belonging to adjacent clusters are distinguishable, and the angular resolution is at least the difference between the AoA values ​​of the cluster centroids, while significant overlap indicates that the angular resolution cannot be accurately estimated based on the available input data, as described in more detail above.

[0050] At box 950, the processing device may optionally perform a regression operation to infer the AoA value for each cluster based on data points from multiple clusters. In some implementations, the same model may be used to perform the feature extraction, clustering, and regression operations of boxes 920, 930, and 950. Alternatively, multiple separate models may be used to perform each of these tasks, as described in more detail above.

[0051] At box 960, the processing device may output (e.g., by displaying the calculated values ​​and / or transmitting the calculated values ​​to one or more receivers via one or more communication networks) the calculated angular resolution values ​​and / or AoA values, and the method may terminate.

[0052] Figure 10 This is a flowchart of an example method for automating an antenna design workflow based on estimating the angular resolution values ​​of an antenna array. Each of method 1000 and / or its individual functions, routines, subroutines, or operations can be executed by processing logic including hardware (circuit, dedicated logic, etc.), software (e.g., running on a general-purpose computing system or a dedicated machine), firmware (embedded software), or any combination thereof. Two or more functions, routines, subroutines, or operations of method 1000 can be executed in parallel or in an order that may differ from the order described below. In some implementations, method 1000 can be executed by a single processing thread. Alternatively, method 1000 can be executed by two or more processing threads, each thread executing the operation of one or more individual functions, routines, subroutines, or methods. In the illustrative example, the processing threads implementing method 1000 can be synchronized (e.g., using semaphores, critical sections, and / or other thread synchronization mechanisms). Alternatively, the processing threads implementing method 1000 can execute asynchronously relative to each other. In one embodiment, the operation of method 1000 can be performed by… Figure 8 The example system 800 is executed by the processing device 806.

[0053] At block 1010, the processing device implementing the method can receive multiple values ​​of design parameters for an antenna array comprising multiple antenna elements. Antenna design parameters may include the number, type, and geometric properties (e.g., shape, size, area) of the antenna elements, as well as the dimensions of the space between the antenna elements, as described in more detail above.

[0054] At box 1020, the processing device can apply an antenna simulation model to calculate multiple values ​​of the amplitude and phase of the RF signal received by each antenna element of the antenna array, each antenna element being characterized by specified design parameters, as described in more detail above.

[0055] At box 1030, the processing device can employ a machine learning model to perform feature extraction operations, transforming multiple values ​​of amplitude and phase into multiple data points in a reduced-dimensional space. In some implementations, the model used to perform feature extraction and clustering can be implemented by a neural network, such as a convolutional neural network (CNN), a recurrent neural network (RNN), or a multilayer perceptron (MLP), as described in more detail above.

[0056] At box 1040, the processing device can classify multiple data points into multiple clusters, as described in more detail above.

[0057] At box 1050, the processing device can employ a machine learning model to perform a regression operation to infer the angular resolution value of the antenna array based on data points from multiple clusters. The angular resolution of the antenna array can be characterized by cluster overlap: relatively small overlap or no overlap indicates that the AoA values ​​of points belonging to adjacent clusters are distinguishable, and the angular resolution is at least the difference between the AoA values ​​of the cluster centroids, while significant overlap indicates that the angular resolution cannot be accurately estimated based on the available input data. In some implementations, the same model can be used to perform the feature extraction, clustering, and regression operations of boxes 1030, 1040, and 1050. Alternatively, multiple separate models can be used to perform each of these tasks, as described in more detail above.

[0058] In response to determining at block 1060 that the angular resolution value is less than or equal to a predetermined threshold, the processing device may output the calculated angular resolution value of the antenna array at block 1070 (e.g., by displaying the calculated angular resolution value and / or transmitting the calculated angular resolution value to one or more receivers via one or more communication networks), and the method may terminate.

[0059] Otherwise, in response to determining at block 1060 that the angular resolution value exceeds a predetermined threshold, the processing device may identify at block 1080 the antenna array elements and / or antenna array design parameters that have caused the angular resolution value to exceed the predetermined threshold. In some implementations, the processing device may identify one or more suboptimal portions of the antenna array response as portions in which the phase response is not sufficiently distinguishable to provide the necessary angular resolution. The processing device may further identify the antenna array elements corresponding to the identified suboptimal portions of the antenna array response. The processing device may further identify the design parameters (e.g., dimensions) of the antenna array elements that may have caused the suboptimal antenna response, as described in more detail above.

[0060] At box 1090, the processing device may notify the antenna design module (e.g., by displaying a notification and / or transmitting a notification to one or more receivers via one or more communication networks) of the calculated angular resolution value and antenna elements and / or antenna design parameters that may have adversely affected the angular resolution. When remedial measures are taken (e.g., modifying the identified antenna design parameters), the antenna designer or the antenna design CAD module may resubmit the modified antenna design parameters to input box 1010 of method 1000.

[0061] The embodiments described herein can be implemented by an electronic system including a processing device that performs the same or similar functions as described with respect to the above figures, or vice versa. In another embodiment, the processing device may be a microprocessor or a microcontroller. The angular resolution estimation tool 120 may be implemented as firmware executing on the microcontroller or microprocessor. The microcontroller may report the estimates described herein to an application processor. The electronic system may include a host processor for a computer system that utilizes the microcontroller to provide the estimates described herein. The host processor may include one or more processing devices, memory, and other hardware or software components that perform operations for the electronic system.

[0062] The processing device may include analog and / or digital general purpose input / output (“GPIO”) ports. GPIO ports may be programmable. GPIO ports may be coupled to programmable interconnect and logic (“PIL”), which acts as an interconnect between the GPIO ports and the digital block array of the processing device. The processing device may include analog blocks that can be programmed and reprogrammed in some cases to implement various analog functions. The processing device may also include a digital block array. In one embodiment, the digital block array may be configured to implement various digital logic circuits (e.g., DACs, digital filters, or digital control systems) using configurable user modules (“UM”). The digital block array may be coupled to a system bus. The processing device may also include memory devices, such as random access memory (“RAM”) and program flash memory. RAM may be static RAM (“SRAM”), and program flash memory may be a non-volatile storage device that can be used to store firmware (e.g., control algorithms executable by the processor to implement the operations described herein). The processing device may also include a memory controller unit (“MCU”) coupled to the memory and the processor. The processor may be a processing element (e.g., a processor core) configured to execute instructions or perform operations. The processor may include other processing elements, as will be understood by those skilled in the art who benefit from this disclosure. It should also be noted that the memory may be internal to or external to the processing device. Where the memory is internal, it may be coupled to a processing element, such as a processing core. Where the memory is external to the processing device, the processing device is coupled to another device in which the memory resides, as will be understood by those skilled in the art who benefit from this disclosure.

[0063] In one embodiment, the processing device further includes processing logic. Some or all of the operations of the processing logic may be implemented in firmware, hardware, or software, or a combination thereof. As described herein, the processing logic may receive signals from an antenna array. The processing device may also include an array of analog blocks (e.g., a field-programmable analog array). The analog block array may also be coupled to a system bus. In one embodiment, the analog block array may also be configured to implement various analog circuits (e.g., an ADC or analog filter) using configurable UMs. The analog block array may also be coupled to GPIO ports. The processing device may include an internal oscillator / clock and a communication block (“COM”). In another embodiment, the processing device includes a spread-spectrum clock. The oscillator / clock block provides clock signals to one or more components of the processing device. The communication block may be used to communicate with external components, such as a host processor (also known as an application processor), via an application interface (“I / F”) line.

[0064] The processing device may reside on a common carrier substrate, such as an integrated circuit (“IC”) die substrate, a multi-chip module substrate, etc. Alternatively, components of the processing device may be one or more individual integrated circuits and / or discrete components. In one exemplary embodiment, the processing device is a programmable system-on-a-chip developed by Cypress Semiconductor Corporation, San Jose, California. Processing device. Alternatively, the processing device may be one or more other processing devices known to those skilled in the art, such as a microprocessor or central processing unit, controller, dedicated processor, digital signal processor (“DSP”), application-specific integrated circuit (“ASIC”), field-programmable gate array (“FPGA”), etc.

[0065] It should also be noted that the embodiments described herein are not limited to configurations having a processing device coupled to an application processor, but may include systems that measure RF signals and transmit raw data to a host computer (where the raw data is analyzed by an application). In fact, the processing performed by the processing device can also be performed within the application processor.

[0066] The angular resolution estimation tool 120 can be integrated into an IC of a processing device, or alternatively, in a separate IC. Alternatively, a description of the angular resolution estimation tool 120 can be generated and compiled for incorporation into other integrated circuits. For example, behavioral-level code, or portions thereof, describing the parameters used for the angular resolution estimation tool 120 can be generated using a hardware description language such as VHDL or Verilog and stored on a machine-accessible medium (e.g., CD-ROM, hard disk, floppy disk, etc.). Furthermore, the behavioral-level code can be compiled into register-transfer-level (“RTL”) code, netlists, or even circuit layouts and stored on a machine-accessible medium. Behavioral-level code, RTL code, netlists, and circuit layouts can represent various levels of abstraction describing the angular resolution estimation tool 120. It should be noted that components of an electronic system may include all, some, or none of the components described above.

[0067] In one embodiment, the electronic system can be used in a tablet computer. Alternatively, the electronic device can be used in other applications, such as a laptop computer, mobile phone, personal data assistant (“PDA”), keyboard, television, remote control, monitor, handheld multimedia device, handheld media (audio and / or video) player, handheld gaming device, signature input device for point-of-sale transactions, e-book reader, medical device, warehouse tracking device (e.g., scanner used by a shipping company), automotive device (e.g., car keys and in-vehicle electronics), global positioning system (“GPS”), or control panel. Some embodiments can be implemented as a computer program product that may include instructions stored on a machine-readable medium. These instructions can be used to program a general-purpose or special-purpose processor to perform the described operations. A machine-readable medium includes any mechanism for storing or transmitting information in a machine-readable form (e.g., software, processing application). Machine-readable media may include, but are not limited to, magnetic storage media (e.g., floppy disks); optical storage media (e.g., CD-ROMs); magneto-optical storage media; read-only memory (ROM); random access memory (RAM); erasable programmable memory (e.g., EPROMs and EEPROMs); flash memory; or another type of medium suitable for storing electronic instructions.

[0068] Furthermore, some embodiments can be implemented in a distributed computing environment where machine-readable media are stored on and / or executed by more than one computer system. Additionally, information transmitted between computer systems can be pulled or pushed across communication media connecting the computer systems.

[0069] Although the operations of the methods described herein are shown and described in a specific order, the order of operations of each method can be changed so that certain operations can be performed in reverse order or that certain operations can be performed at least partially in parallel with other operations. In another embodiment, instructions or sub-operations of different operations may be intermittent and / or alternating. The terms “first,” “second,” “third,” “fourth,” etc., as used herein are intended as labels to distinguish different elements and their numerical designation does not necessarily imply order. As used herein, the term “coupled” may mean a direct connection or an indirect connection via one or more intermediate components. Any signal provided by the various buses described herein may be time-multiplexed with other signals and provided via one or more common on-die buses. Furthermore, interconnections and interfaces between circuit components or blocks may be shown as buses or single signal lines. Each bus in a bus may alternatively be one or more single signal lines, and each of the single signal lines may alternatively be a bus.

[0070] The foregoing description sets forth numerous specific details, such as examples of specific systems, components, methods, etc., to provide an understanding of several embodiments of the invention. However, it will be apparent to those skilled in the art that at least some embodiments can be practiced without these specific details. In other instances, well-known components or methods are not described in detail or presented in a simple block diagram format to avoid unnecessarily obscuring the embodiments. Therefore, the specific details set forth are merely exemplary. Specific implementations may differ from these exemplary details and are still considered within the scope of these embodiments.

[0071] Embodiments of the claimed subject matter include, but are not limited to, the various operations described herein. These operations may be performed by hardware components, software, firmware, or a combination thereof.

[0072] The foregoing description sets forth numerous specific details, such as examples of specific systems, components, methods, etc., to provide an understanding of several embodiments of the claimed subject matter. However, it will be apparent to those skilled in the art that at least some embodiments of this disclosure can be practiced without these specific details. In other instances, well-known components or methods are not described in detail or presented in a simple block diagram format. Therefore, the specific details set forth are merely exemplary. Specific implementations may differ from these exemplary details and are still considered to be within the scope of the claimed subject matter.

Claims

1. A method comprising: The processing device receives multiple values ​​of amplitude and phase of the radio frequency (RF) signal for each of the multiple antenna elements included in the antenna array; The processing device applies a machine learning model to perform feature extraction operations, transforming multiple values ​​of the amplitude and phase into multiple data points representing the phase response in a reduced-dimensional space; The processing device applies the machine learning model to cluster the multiple data points into multiple clusters, wherein the angular resolution of the antenna array is characterized by cluster overlap, and each cluster corresponds to the value of the angle of arrival (AOA) of the radio frequency (RF) signal. The processing device calculates the angular resolution value of the antenna array based on the plurality of clusters, wherein the angular resolution value reflects the difference between a first angle of arrival (AoA) value and a second AoA value, the first AoA value being associated with the centroid of a first cluster of a first cluster in the plurality of clusters, and the second AoA value being associated with the centroid of a second cluster of a second cluster in the plurality of clusters; as well as The angular resolution value is output through the processing device.

2. The method according to claim 1, further comprising: Receive multiple values ​​of the design parameters of the antenna array; Multiple values ​​for the amplitude and phase of the RF signal for each of the multiple antenna elements are calculated by applying a simulation model to multiple values ​​of the design parameters.

3. The method according to claim 1, wherein, Calculating the angular resolution value further includes: The machine learning model is applied to perform a regression operation to infer the angular resolution value based on the plurality of data points.

4. The method according to claim 1, further comprising: The machine learning model is applied to perform a regression operation to compute the corresponding value of the angle of arrival (AoA) for each of the plurality of clusters.

5. The method according to claim 1, further comprising: Identify subsets of overlapping clusters among the multiple clusters; as well as Based on the subsets of the overlapping clusters, identify the physical parameters of the antenna array that have caused the subsets of the overlapping clusters to overlap.

6. The method according to claim 1, wherein, The machine learning model is represented by a neural network.

7. The method according to claim 6, further comprising: An unsupervised training process is performed to train the neural network to perform the feature extraction operation.

8. The method according to claim 6, further comprising: An unsupervised training process is performed to train the neural network to perform clustering operations.

9. The method according to claim 6, further comprising: A supervised training process is performed to train the neural network to perform a regression operation to infer the angular resolution value based on the plurality of data points.

10. The method according to claim 1, further comprising: In response to determining that the angular resolution value is less than or equal to a predetermined threshold, the calculated angular resolution value is output as the angular resolution value of the antenna; as well as In response to determining that the angular resolution value exceeds a predetermined threshold, the antenna array element corresponding to the identified suboptimal and / or satisfactory portion of the antenna array response is identified, and the design parameters of the antenna array that cause the angular resolution value to exceed the predetermined threshold are identified. The feedback includes the angular resolution, AoA value, and output data of the suboptimal and / or satisfactory portions of the identified antenna array response, in order to modify the design parameters of the identified antenna array.

11. A method comprising: The processing device receives multiple values ​​of the design parameters of an antenna array, which includes multiple antenna elements. The processing device applies a simulation model to multiple values ​​of the design parameters to calculate multiple values ​​of the amplitude and phase of the RF signal for each of the multiple antenna elements. The processing device applies a machine learning model to calculate the angular resolution value of the antenna array, including: performing a feature extraction operation using the machine learning model to transform multiple values ​​of amplitude and phase into multiple data points representing the phase response in a reduced-dimensional space; clustering the multiple data points into multiple clusters using the machine learning model, wherein the angular resolution of the antenna array is characterized by cluster overlap, and each cluster corresponds to the value of the angle of arrival (AOA) of the RF signal; and calculating the angular resolution value of the antenna array based on the multiple clusters using the processing device; and In response to determining that the angular resolution value exceeds a predetermined threshold, the design parameters of the antenna array that have caused the angular resolution value to exceed the predetermined threshold are identified.

12. The method of claim 11, further comprising: In response to determining that the angular resolution value is less than or equal to a predetermined threshold, the angular resolution value of the antenna array is output.

13. The method according to claim 11, wherein, Calculating the angular resolution value further includes: The machine learning model is used to perform regression operations to infer the angular resolution value based on the multiple clusters.

14. The method of claim 13, further comprising: An unsupervised training process is performed to train the machine learning model to perform the feature extraction operation.

15. The method of claim 13, further comprising: An unsupervised training process is performed to train the machine learning model to perform clustering operations.

16. The method of claim 13, further comprising: A supervised training process is performed to train the machine learning model to perform regression operations to infer the angular resolution values ​​based on the multiple clusters.

17. The method of claim 13, further comprising: For each of the multiple clusters, calculate the corresponding value of the angle of arrival (AoA).

18. The method according to claim 11, wherein, The machine learning model is represented by a neural network.

19. An apparatus comprising: A transceiver configured to be coupled to an antenna comprising a plurality of antenna elements, the transceiver being used to receive a plurality of values ​​for the amplitude and phase of a radio frequency (RF) signal for each of the plurality of antenna elements; as well as A processor, coupled to the transceiver, is used for: Feature extraction is performed using a machine learning model to transform multiple values ​​of the amplitude and phase into multiple data points representing the phase response in a reduced-dimensional space. The processing device applies the machine learning model to classify the multiple data points into multiple clusters, wherein the angular resolution of the antenna array is characterized by cluster overlap, and each cluster corresponds to the value of the angle of arrival (AOA) of the radio frequency (RF) signal. as well as The angular resolution value of the antenna is calculated based on the plurality of clusters, wherein the angular resolution value reflects the difference between a first angle of arrival (AoA) value and a second AoA value, the first AoA value being associated with the first centroid of the first cluster in the plurality of clusters, and the second AoA value being associated with the second centroid of the second cluster in the plurality of clusters.

20. The device according to claim 19, wherein, The processor is also used for: For each of the multiple clusters, calculate the corresponding value of the angle of arrival (AoA).

21. The device according to claim 19, wherein, The machine learning model is represented by a neural network.

Images